‘Subordination’ versus ‘Coordination’ in Sentence and Text

Studies in Language Companion Series (SLCS)

This series has been established as a companion series to the periodical
Studies in Language.

Editors
Werner Abraham Michael Noonan
University of Vienna University of Wisconsin, Milwaukee

Editorial Board
Joan Bybee Christian Lehmann
University of New Mexico University of Erfurt
Ulrike Claudi Robert E. Longacre
University of Cologne University of Texas, Arlington
Bernard Comrie Brian MacWhinney
Max Planck Institute, Leipzig Carnegie-Mellon University
University of California, Santa Barbara
Marianne Mithun
William Croft University of California, Santa Barbara
University of New Mexico
Edith Moravcsik
Östen Dahl University of Wisconsin, Milwaukee
University of Stockholm
Masayoshi Shibatani
Gerrit J. Dimmendaal Rice University and Kobe University
University of Cologne
Russell S. Tomlin
Ekkehard König University of Oregon
Free University of Berlin

Volume 98
‘Subordination’ versus ‘Coordination’ in Sentence and Text
A cross-linguistic perspective
Edited by Cathrine Fabricius-Hansen and Wiebke Ramm
‘Subordination’ versus ‘Coordination’
in Sentence and Text
A cross-linguistic perspective

Edited by

Cathrine Fabricius-Hansen
Wiebke Ramm
University of Oslo

John Benjamins Publishing Company

Amsterdam / Philadelphia
 TM
The paper used in this publication meets the minimum requirements of
8

American National Standard for Information Sciences – Permanence of

Paper for Printed Library Materials, ansi z39.48-1984.

Library of Congress Cataloging-in-Publication Data

Subordination versus coordination in sentence and text : a cross-linguistic perspective /

edited by Cathrine Fabricius-Hansen, Wiebke Ramm.
p. cm. (Studies in Language Companion Series, issn 0165-7763 ; v. 98)
Includes bibliographical references and index.
1. Grammar, Comparative and general--Subordinate constructions. 2. Grammar,
Comparative and general--Coordinate constructions. 3. Grammar,
Comparative and general--Clauses. 4. Grammar, Comparative and general--
Sentences. I. Fabricius-Hansen, Cathrine. II. Ramm, Wiebke.
P294.S83    2008
415--dc22 2007052231
isbn 978 90 272 3109 3 (Hb; alk. paper)

© 2008 – John Benjamins B.V.

No part of this book may be reproduced in any form, by print, photoprint, microfilm, or any
other means, without written permission from the publisher.
John Benjamins Publishing Co. · P.O. Box 36224 · 1020 me Amsterdam · The Netherlands
John Benjamins North America · P.O. Box 27519 · Philadelphia pa 19118-0519 · usa
 Table of contents

1. Editor’s introduction: Subordination and coordination from different

perspectives 1
   Cathrine Fabricius-Hansen & Wiebke Ramm

Part I. General and theoretical issues

2. RST revisited: Disentangling nuclearity 33

   Manfred Stede

3. Subordination and coordination in syntax, semantics,

and discourse: Evidence from the study of connectives 59
   Hardarik Blühdorn

Part II. Cross-linguistic approaches

4. A corpus-based perspective on clause linking patterns in

English, French and Dutch 89
   Christelle Cosme

5. Sentence splitting – and strategies to preserve discourse

structure in German-Norwegian translations 115
   Kåre Solfjeld

6. Upgrading of non-restrictive relative clauses in translation:

A change in discourse structure? 135
   Wiebke Ramm

7. Subordination in narratives and macro-structural planning:

A comparative point of view 161
  Mary Carroll, Antje Rossdeutscher, Monique Lambert &
Christiane von Stutterheim
 ‘Subordination’ verses. ‘Coordination’ in sentence and text

Part III. Monolingual studies

8. German dependent clauses from a constraint-based perspective 187
   Anke Holler

9. To the right of the clause: Right dislocation vs. afterthought 217
   Maria Averintseva-Klisch

10. Exploring the role of clause subordination in discourse structure:

The Case of French avant que 241
   Laurence Delort

11. Pseudo-imperatives and other cases of conditional conjunction

and conjunctive disjunction 255
   Michael Franke

12. From discourse to “odd coordinations”:

On Asymmetric Coordination and Subject Gaps in German 281
   Ingo Reich

Part IV. Diachronic perspectives

13. Old Indic clauses between subordination and coordination 307
   Rosemarie Lühr

14. Rhetorical relations and verb placement in the early

Germanic languages: A cross-linguistic study 329
   Svetlana Petrova & Michael Solf

Index of subjects 353

Index of names 357

 Editors’ introduction
Subordination and coordination from different
­perspectives

Cathrine Fabricius-Hansen & Wiebke Ramm

University of Oslo

1.  Preliminaries

The present collection of papers addresses semantic, pragmatic or discourse-oriented

aspects of coordination and subordination in a broad sense. Five of the papers approach
the topics from a cross-linguistic perspective. The collection is the outcome of a work-
shop (AG8) on ‘Subordination’ versus ‘Koordination’ in Satz und Text aus sprachverglei­
chender Perspektive / ‘Subordination’ versus ‘Coordination’ in Sentence and Text – from a
Cross-linguistic Perspective, which was organised by the editors during the 28th Annual
Meeting of the Deutsche Gesellschaft für Sprachwissenschaft in February 2006.1
Coordination and subordination are well-established grammatical terms, but like
many other linguistic terms with a long history, they are somewhat fuzzy, both being
used in a variety of – mutually related – senses depending on the theoretical context.
This holds for other, closely related notions such as parataxis and hypotaxis as well.
Thus Lehmann (1988) observes in a paper on the typology of clause linkage:
The term subordination is applied, in different schools of linguistics, to different
kinds of phenomena. In the broadest use, which may be found in certain trends
of ­European structuralism, the size and nature of the subordinate element is of no
concern. Here subordination practically means the same as dependency.

.  The research underlying this introduction has been carried out in connection with the project
SPRIK (Språk i kontrast / Languages in Contrast) at the University of Oslo, Faculty of Humanities,
funded by the Norwegian Research Council (NFR) under project number 158447/530 (2003–2007).
Editing was financed by the SPRIK project and by grants from the Department of Literature, Area
Studies and European Languages and the Faculty of Humanities. We thank Maria F. Krave and
Signe Oksefjell Ebeling for their meticulous work with the manuscripts. And we thank the editors
of the Studies in Language Companion Series for giving us and the other workshop participants
this opportunity to publish our research presented at the DGfS 06 conference.
 Cathrine Fabricius-Hansen & Wiebke Ramm

In the most narrow use, characteristic of classical philology, only finite clauses
can be said to be subordinate. Here subordination practically means the same as
hypotaxis; and consequently the two latter terms are mostly used interchangeably.
 (Lehmann 1988: 221)
In what is probably their most widespread application, ‘subordination’ and ‘coordina-
tion’ – along with their adjectival cognates ‘subordinate’, ‘coordinate’, etc. – are syntactic
notions denoting relations between parts of a complex syntactic unit. That is, they con-
cern the structure of sentences or clauses and their parts.
Subordination is an asymmetric relation – both in linguistics and in everyday life –
and by that token intimately correlated with the notion of hierarchical structure: If A
is subordinate to B, then B cannot be subordinate to A; but B may, in its turn, be sub-
ordinate to a third entity C, and so on. This means that B has a ‘higher’ position in the
domain structured by subordination, i.e., it is nearer to the ‘topmost’ element of that
structure, which is conventionally identified with an element that is not subordinate
to any other element within the domain. In social hierarchies, this kind of asym-
metry is typically correlated or associated with (social) ‘importance’, ‘prominence’
and the like. Similarly, the subordinate clause in clausal subordination is often said
to contain less prominent or salient information, to have less communicative weight
(‘kommunikatives Gewicht’ in German) than the ‘superordinate’ clause (typically the
so-called main clause) (see e.g., Aarts 2006; Hartmann 1984; Hetland & Molnár 2001;
Peyer 1997; Reis 1993). However, while importance and prominence may be quite
transparent concepts with respect to social hierarchies, it is far from clear how the
functional notions of prominence, salience or communicative weight can be mapped
onto specific structural-syntactic categories as defined by the grammars of different
languages.
As far as the domain of natural discourse and texts is concerned, it is a common
observation in various theoretical approaches that entities of this domain too can be
organised hierarchically (‘subordinating’, ‘hypotactically’) or non-hierarchically (‘coor-
dinating’, ‘paratactically’). Relevant examples are Rhetorical Structure Theory (Mann &
Thompson 1988) and Segmented Discourse Representation Theory (Asher 1993; Asher
& Lascarides 2003). However, relatively few attempts have been made so far to answer
the following questions (but see references in Matthiessen & Thompson 1988: 275):

i. What kind of relationship holds between the discourse-related and the sentence-re­
lated (syntactic) notions of subordination and coordination or related distinctions? Is
the parallelism as close as the terminology may suggest? What impact does the choice
between syntactic subordination and coordination have on discourse structure?
ii. By what means are hierarchical (‘subordinating’) and non-hierarchical (‘coordi-
nating’) discourse relations signalled across languages? To what degree, and how, do
languages differ in this respect? Do languages exploit syntactic subordination (and
coordination) differently as a means of organizing information in discourse – ‘informa-
tion packaging’ – and if so, why and in what ways?
 Editors’ introduction 

The phrases in sentence and text and a cross-linguistic perspective occurring in the
title of the present volume relate to these two sets of questions. An additional important
issue concerns methodology:
iii. Since the questions raised above are partly of an empirical nature, extensive stud-
ies of natural texts – ‘wildlife’ texts, so to speak2 – are needed, including studies of
parallel corpora (of original texts and their translations) and comparable texts in dif-
ferent languages: Explorative analyses indicate that text- and corpus-oriented cross-
linguistic studies can make an important contribution to the issues specified in (i) and
(ii) and thus improve our general understanding of how information packaging on
sentence and text level are related (see Doherty 1996, 1999, 2005 and 2006; Granger et
al. 2003; Hasselgård et al. 2002; Johansson 2007; Johansson & Oksefjell 1998).
These issues were the motivation for our workshop and the articles collected in
this volume. We invited contributions based on parallel corpora and/or language com-
parison, including, but not limited to, topics such as the following:

– Syntactically adjoined structures from the perspective of information structuring on

discourse level
– ‘Subordinating’ versus ‘coordinating’ discourse relations / clause combining, and
their realisation
– Connectives and punctuation as a means of structuring discourse and signalling
discourse relations

Apart from punctuation, these topics are all addressed more or less directly in the
papers below.
The collection is divided into four parts: The two contributions in Part I are con-
cerned with central theoretical questions. Part II contains four corpus-based studies
of clause combining and discourse structure, comparing at least two of the languages
German, English, French, Dutch, and Norwegian. The five papers in Part III address
specific – predominantly semantic – topics relating to German, English or French.
Finally, the two papers in Part IV approach the topic of subordination, coordination
and rhetorical relations from a diachronic perspective, involving Old Indic and early
Germanic languages, respectively.
Sections 2–4 of this introduction address some of the general issues raised above
more thoroughly, referring to individual papers where relevant. The final section (5)
gives a brief presentation of each contribution, outlining how it fits into the overall
topic of study. First, however, we need to comment on our distinction between ‘sen-
tence’ and ‘text’.

.  We have coined the term ‘wild(life) texts’ inspired by the Norwegian newspaper Aftenposten,
which after a very mild winter (2005) introduced the compound vill-snø ‘wild-snow’ referring
to natural snow.
 Cathrine Fabricius-Hansen & Wiebke Ramm

2.  ‘Sentence’ versus ‘text’

What ‘is’ a sentence? As is well known, this question cannot be answered in a theory-
independent and at the same time precise manner. From a syntactic point of view the
sentence is the domain of syntactic theory: a unit that is governed by syntactic princi-
ples (constituent structure, dependency, government, binding, movement, etc.). From
a discourse perspective a sentence might be a unit representing a single illocutionary
act on the part of the speaker/reader and intended to be understood as such by the
hearer/reader. Thus, it is a unit the reader/hearer is expected to process ‘in a single step’,
assigning to it a communicative purpose and a content to be integrated into her/his
representation of the current discourse, i.e., what corresponds to an utterance in spo-
ken language. As mentioned in Section 1, we shall primarily be concerned with written
texts. And for the present purpose we shall assume without further discussion that
sentence boundaries in the discourse-oriented sense are marked by full stop in written
discourse – with question and exclamation marks as non-assertive alternatives. That
is, we take the orthographic sentence as defined e.g., by Huddleston & Pullum (2002:
1728) to be the basic processing unit in written texts, and full stop to be the written
default counterpart of declarative terminal fall in intonation, signalling to the reader/
hearer that the end of the current communicative unit has been reached and that the
common ground can be updated with the content expressed by that unit in the given
context (Büring 2007). Comma may signal the end of a syntactic sentence (clause) but
cannot by itself mark a sentence boundary in the discourse-oriented sense. For the
rest of this introduction we shall use the term ‘sentence’ in that sense unless otherwise
indicated, assuming that discourse-sentences by default manifest themselves as ortho-
graphic sentences in written language.
The terminological decisions made above have some important implications:

i. A text consists of a finite number of sentences in succession, with one-sentence

texts as a marked category.
ii. A sequence of syntactically independent clauses separated by commas is not a sen-
tence sequence but constitutes a single (complex3) sentence if properly demarcated.4
iii. If clause combining (Haiman & Thompson 1988), or  clause linkage (Lehmann
1988), is to be understood as connecting clauses rather than sentences (which in our

.  Sentences having the form of two or more coordinated main clauses are sometimes termed
compound rather than complex.

.  As Huddleston & Pullum (2002: 1728) put it: “[…] there will often be no syntactically
marked distinction between a sentence with the form of a combination of two successive main
clauses and a sequence of sentences each of which has the form of a main clause. In writing, one
function of punctuation is precisely to indicate whether successive clauses belong together or
are to be treated as separate.”
 Editors’ introduction 

opinion is reasonable) it must be confined to the sentence level. That is, simple jux-
taposition of syntactically independent clauses separated by comma, without overt
coordination, represents a special case of (paratactic) clause linkage. Corresponding
full-stop sentences, on the other hand, are, strictly speaking, not related by any kind of
clause combing; they simply succeed each other.5
According to Huddleston & Pullum (2002: 1735f) colon and semicolon, like comma,
“normally mark boundaries within a sentence, and hence can be regarded as second-
ary boundary marks” located between comma and full stop in a hierarchy of “relative
strength”. To our knowledge, however, the discourse functions of semicolon and colon
have not yet been thoroughly investigated from the perspective of discourse structure
or discourse processing, let alone in a cross-linguistic setting (a detailed account of the
use and functions of the colon in German, however, can be found in Karhiaho 2003).
So, since the topic of punctuation is not addressed in the present collection of papers,
we shall refrain from further speculation on this matter.

3.  Coordination and subordination as syntactic notions

3.1  Coordination
The literature on the syntax, semantics and pragmatics of coordination – clausal or not –
is extensive (see for instance Bátori et al. 1975; Carston 2002; Carston & Blakemore 2005;
Haspelmath 2004; Pasch et al. 2003; Redder forthcoming, Schwabe & Zhang 2000 for
recent overviews and discussions), so we shall confine ourselves to issues of immediate
interest in the present context.
As a syntactic relation, coordination (or conjunction, as it is often called6) is tradi-
tionally said to hold if the units in question are syntactically ‘equivalent’, ‘have the same
status’, ‘play the same role’ in the given syntactic context (see e.g., Lang 1984; Lehmann
1988; Zifonun et al. 1997: 2362ff; Huddleston & Pullum 2002: 1281ff; Haspelmath 2004
with further references; Carston & Blakemore 2005: 353 9; Crysman 2006). Haspelmath
(2004), for instance, proposes the definition in (a) below, which he later refines as in
(b), leaving out the reference to relative salience (compare Sections 1 above and 4.2
below), on the one hand, and stressing the semantic nature of ‘sameness’ of type, on the
other (cf. also Reich forthcoming). This in its turn distinguishes Haspelmath’s second
proposal from Lehmann’s definition in (c).

.  The literature on clause combining (e.g., Lehmann 1988; Matthiessen & Thompson 1988) is
not very explicit in this respect.

.  But see e.g., Cormack & Smith (2005: 396), who distinguish notionally between (logical)
conjunction and (syntactic) coordination.
 Cathrine Fabricius-Hansen & Wiebke Ramm

a. A construction [A B] is considered coordinate if the two parts A and B have the

same status (in some sense that needs to be specified further), whereas it is not
coordinate if it is asymmetrical and one of the parts is clearly more salient or im-
portant, while the other is in some sense subordinate. (Haspelmath 2004: 3f.)
b. The term coordination refers to syntactic constructions in which two or more
units of the same type are combined into a larger unit and still have the same
semantic relation with other surrounding elements. (Haspelmath 2004: 34)
c. Coordination is a relation of sociation [i.e., non-dependency] combining two syn-
tagms of the same type and forming a syntagm which is again of the same type.
(Lehmann 1988: 182)

In his pioneering work on the semantics of coordination, Lang (1984) explicitly dis-
tinguishes between genuine “coordinate conjoining” and “subordinate conjoining”,
stressing that the conjuncts have to fulfil both the grammatical condition of “homo-
geneousness” and semantic conditions concerning the relation between conjunct
meanings in order for the construction to qualify as (acceptable) “coordinate con-
joining”.
The coordinated entities (the conjuncts, coordinates, coordinands, conjoins) may
be clauses in a broad sense, i.e., finite or non-finite verb phrases or full clauses, or they
may be lower-level categories (phrasal coordination). In this book, we shall predomi-
nantly be concerned with the former variety, i.e., clausal coordination (including verb
phrase coordination).
Coordination, as traditionally understood, is signalled by a coordinating connec-
tive (‘conjunction’, e.g., and, or, but), which in the languages studied here introduces
non-initial constituents of coordinating constructions;7 that is, coordination represents
a syndetic variety of paratactic clause combining (Lehmann 1988). The archetypical co-
ordinating conjunction is and together with its counterparts in other languages – the
general terms ‘coordination’ and ‘conjunction’ are, in fact, often used in the restricted
default sense of and-coordination/-conjunction. And this is how they will be used in
this introduction, as in most of the individual contributions, if nothing is said to the
contrary (see Section 3.3).
Coordination is a symmetric relation in the formal sense: if A is coordinated to B,
then B is also coordinated to A. However, in the literature on coordination the term
‘symmetric’ often seems to be used in a less formal sense, referring to the criterion that
the coordinated entities must be of the same type and must have the same relation with
surrounding elements. In yet other contexts symmetry is related to the permutativity
property of logical conjunction (&) and other operators (intersection and summation)
that are taken to represent the meaning of and, depending on the semantic nature of

.  See Haspelmath (2004) for a typological overview. In multiple coordination, comma is a
semantically neutral alternative to and and or between all but the last two conjuncts.
 Editors’ introduction 

the conjuncts (Partee & Rooth 1983). However, the symmetry condition, in one sense
or the other, is not always met by constructions that are yet classified as coordinative
according to diagnostic coordination tests, as summarized e.g., by Lang (1984), te Velde
(2000), Haspelmath (2004: 34f), and Reich (forthcoming).8 In other words, the concept
of (syntactic) coordination – like subordination (see 3.2) and many other traditional
grammatical notions – has prototype structure: In addition to ‘canonical’ or ‘standard’
coordination, which is symmetric by definition, the literature has registered an impres-
sive variety of asymmetric construction types.9
Opinions vary somewhat as to where exactly one should draw the borderline
between coordination (parataxis) and subordination (hypotaxis) in the continuum
represented by non-canonical varieties of these categories, i.e., and-constructions
exhibiting crucial subordinate properties (“subordinate conjoining” (Lang 1984) or
“pseudo-coordination” (e.g., Johannessen 1998)), on the one hand, and coordina-
tion-like clause combining formally marked as subordinating (see Section 3.2), on
the other hand. At the other end of the continum we find what Zifonun et al. (1997:
2362) call Quasikoordination: “adjacent conjuncts” that do not form a single ortho-
graphic sentence or a single “intonation unit”, i.e., cases where a clause contain-
ing a coordinating connective follows a major boundary mark in written or spoken
language (e.g., full stop and falling boundary tone, respectively). So-called (and-)
parenthetical coordination (Blakemore 2005; see also Zifonun et al. 1997: 2363f)
represents yet another non-canonical variety of coordination that we cannot take
up here.
Asymmetry in coordination has been an object of much theoretical discussion,
and, as far as clausal coordination is concerned, certain types of asymmetry, in fact,
occur so frequently that what is called canonical coordination may seem an abstract
ideal rather than a ‘real’ prototype. Thus it has been repeatedly observed that clausal
coordination in practice often deviates from standard logical conjunction by being
sensitive to permutation of the conjuncts, i.e., that the interpretation may change with
the order of the conjuncts, even if overt anaphoric dependences between the second
and the first conjunct are adjusted; cf. (1)–(2), taken from Levinson (2000: 121) and
Blakemore & Carston (2005: 570), respectively.
(1) John turned the key and the engine started.
(≠ The engine started and John turned the key.)
(2) She [[jumped on the horse] and [rode into the sunset]].
(≠ She rode into the sunset and jumped on the horse.)

.  Coordinate Structure Constraint (CSC), Across the Board Movement (ATB), No Backward
Anaphora, the possibility of multiple conjuncts.

.  “Unbalanced” coordination according to Johannessen (1998).

 Cathrine Fabricius-Hansen & Wiebke Ramm

Here the conjuncts are ‘equal’ at the syntactic level, they are full clauses of the same
‘type’ in (1) and verb phrases in (2); that is, the asymmetry concerns the level of
interpretation alone. This phenomenon of conjunction buttressing (Levinson 2000)
can be explained as pragmatic enrichment based on the principle of informativeness
­(I-principle) which allows the hearer/reader to choose the strongest interpretation co-
herent with what is explicitly said: “[…] conjunction is ripe for I-enrichment: when
events are conjoined, they tend to be read as temporally successive and, if at all plau-
sible, as causally connected.” (Levinson 2000: 122)
Asymmetries of this semantic-pragmatic kind have played a prominent role in
Relevance Theory as an argument for the central distinction between what is said and
what is implicated (see Blakemore & Carston 2005; Carston 2002: 222ff). In his con-
tribution to the present volume, Solfjeld (Chapter 5) exploits certain aspects of the
relevance-theoretic approach to clausal coordination to account for a characteristic
translation strategy in Norwegian translations from German.
Other asymmetries in clausal coordination overtly concern morpho-syntactic as-
pects of the initial and non-initial conjuncts: asymmetries with respect to syntactic cat-
egory (full clause vs. e.g., verb phrase), type of clause (e.g., imperative vs. declarative,
verb-end vs. verb- second or verb-first in German), choice of verb form, extraction
and binding phenomena, gapping, etc. (see e.g., Johannessen 1998, te Velde 2000, Hud-
dleston & Pullum 2002: 1323ff., Reich to appear). Some examples are seen in (3) – (5):
(3) a. Give me your address, and I’ll see what I can do.
b. Give me your money, or I’ll shoot.
(4) Gestern [hat er wieder verschlafen und ist zwei
Yesterday has he again overslept and is two
Stunden zu spät gekommen].
hours too late come.
‘Yesterday he overslept again and arrived two hours late.’
(5) Here is the whisky which I [went to the store and bought].

In (3) an imperative and a declarative clause are coordinated, and in the German exam-
ple (4) the second conjunct has a subject gap which must be interpreted as co-referential
with – or bound by – the overt subject of the first conjunct. The fronted temporal ad-
verbial gestern ‘yesterday’ can, from a semantic perspective, be analysed as ‘common’ to
both conjuncts, i.e., as standing outside the coordinate construction.10 (5) is a ‘classical’
example (from J. R. Ross 1967) of asymmetric extraction leaving an object gap in the

.  It should be noted that when we front the subject instead of the adverbial, we get what
­appears to be a syntactically canonical VP-coordination:

(4)’ Er [hat gestern verschlafen und ist zwei Stunden zu spät gekommen].
he has yesterday overslept and is two hours too late arrived
‘He overslept yesterday and arrived two hours late.’
 Editors’ introduction 

second conjunct alone and thus violating the so-called Coordinate Structure Constraint
(see e.g., the discussion in Johannessen 1998: 214ff; te Velde 2000; Kehler 2002; Crysman
2006 and Reich forthcoming, with relevant references). Asymmetries in clausal coordi-
nation are the subject of Franke’s and Reich’s contributions (Chapters 11 and 12).
To conclude our discussion of asymmetries in clausal coordination, the following
points should be made:

i. On the surface, clauses (or sentences) may be represented as rather simple syn-
tactic structures, depending on syntactic theory. However, from a semantic per-
spective, they must be conceived as multi-level or multi-layered structures: a full
sentence contains a series of nested operators or modifiers (‘functional projec-
tions’) – polarity, aspect, tense, modality, illocutionary force – above the predicate
expressed by the lexical (main) verb and head of the clause. In the case of clausal
coordination, then, it has to be decided at which semantic level the second con-
junct attaches to the first – and whether it does in fact constitute a full indepen-
dent clause from a semantic point of view or rather a semantically reduced clausal
variant, i.e., a more or less ‘desententialized’ entity in the sense of Lehmann (1988:
200). For this reason the difference between full clausal and verb phrase coordina-
tion should probably be taken more seriously than has hitherto been the case.
ii. It is by now generally acknowledged that semantic interpretation is more flexible
and that pragmatics plays a more prominent and pervasive role in interpretation
than was envisaged some decades ago (Sperber & Wilson 1986, Levinson 2000,
Carston 2002, Kehler 2002, Zeevat forthcoming). Against that background it is
conceivable that a semantically adequate conjunctive interpretation may be ex-
tracted from what seems somewhat ‘odd’ on the surface. At some level the con-
juncts must be interpreted as semantic objects of the same type if the meaning of
the coordinator is as presumed; otherwise the interpretation would break down.11
iii. Since the first conjunct necessarily precedes and is not structurally subordinated
to or embedded in the second conjunct, anaphoric elements occurring in the sec-
ond conjunct – including zero anaphors – may find their antecedents in the first
conjunct. More generally stated: the first conjunct is an accessible context when

But even so, i.e., when placed within the first conjunct, the temporal adverbial gestern ‘yesterday’
is interpreted as setting a temporal frame for both conjuncts. Thus asymmetries involving subject
gaps – type (4) ­– might after all be special instances of VP-coordination, despite the counter
arguments put forward by Reich (fortcoming).

.  “To sum up, extraction out of parts of conjuncts is possible when a link can be established
between the conjuncts which can be described in terms of consequence, despiteness, or time
(there may be other relationships, too)” (Johannessen 1998: 233). Cf. Kehler (2002, 2004), who
argues for a coherence-driven, i.e., pragmatic, approach to asymmetries in coordination. See
also e.g., Vuillaume’s (2000) account of asymmetric coordination in German.
 Cathrine Fabricius-Hansen & Wiebke Ramm

the second is processed, but not the other way round (“No Backward Anapho-
ricity”; cf. Lang 1984; te Velde 2000; Haspelmath 2004 and Reich forthcoming,
among others). Consequently, permutativity between conjuncts is to be expected
only in the – presumably quite rare cases – where the second conjunct is not
linked, explicitly or implicitly, to the first in any other way than by being con-
joined with it.
iv. As argued e.g., by Lang (1977, 1984), Carston (2002), Blakemore & Carston (2005)
and Jasinskaja (2006), coordination itself can be taken to signal that the conjoined
clauses are more tightly linked than a corresponding sentence sequence. The co-
ordinating connective has a “procedural” or “operational” meaning triggering
the construction of a what Lang (1984: 71) terms a Common Integrators, i.e., “a
conceptual entity which encompasses the conjunct-meanings”, on the basis of the
conjunct meanings themselves, the linguistic and situational contexts and extra-
linguistic knowledge systems etc. That is, the coordinator following the first con-
junct makes clear that the sentence is not yet finished but that more ‘of the same
kind’ as the structure processed thus far will have to be integrated in the interpre-
tation (cf. Redder forthcoming with further references).

(i) – (iv) go a long way towards explaining why syntactic-semantic mismatches

may occur in natural language coordination and why they are interpreted as they are.
But (i) – (iv) cannot, of course, account for language-specific restrictions on syntactic-
semantic asymmetry.
For the reasons mentioned above, canonical symmetric coordination is probably
the exception rather than the norm in language use. (6) and (7) are cases in point.
(6) Tom did his homework and Anna went to the cinema.
(7) I had a paper to finish and my parents needed help.

Here the conjuncts can hardly be understood as connected in any other way than
as partial – and jointly exhaustive (Jasinskaja 2006) – answers to a given ‘Question
under Discussion’12 like What did the children/ Tom and Anna do last night? for (6)
and Why didn’t you join us? for (7). That is, unlike the asymmetric examples discussed
above, the coordinate constructions in (6) and (7) cannot be interpreted as self-con-
tained coherent texts. The conjuncts are not linked at the level of eventualities (causally
or otherwise); the relation between them is of a purely rhetorical nature.
In symmetric coordination, the order of the conjuncts has no bearing on truth
conditions. However, this is not to say that the order is arbitrary or irrelevant in general.
There may be other reasons for the speaker/writer to prefer one order over the other,
e.g., relative importance or salience from a psychological/emotional – or rhetorical –
point of view: Only the second, or last, conjunct or the whole coordinate sentence

.  Or Common Integrator in the sense of Lang (1984).

 Editors’ introduction 

may function as an attachment point for a subsequent sentence; the first conjunct is
blocked (see Section 4.2).
Like sentence interpretation in general, the interpretation of coordinate construc-
tions interacts with information structure at sentence level (focus-background and
topic-comment partition, cf. Krifka (2007)). In spoken language prosody and intona-
tion play a prominent role in this respect (see Lang (2004) for coordination). To our
knowledge, however, coordination, in written language at least, has not been investi-
gated specifically from that perspective, except for symmetric coordination exhibit-
ing the characteristic pattern of parallel and contrast seen in (6) (see e.g., Lang 2004,
Abraham 2006, Büring 2007). The interplay between information structure / prosody
and subordination/coordination in a historical language (Old Indic) is central in the
contribution by Lühr (Chapter 13).

3.2  Subordination
As pointed out in Section 1, the term subordination – like coordination – may be used
in a general sense, as defined e.g., by Lehmann (1988):
(a) A grammatical relation R connecting syntagms X and Y is a relation of depen­
dency iff X occupies a grammatical slot of Y or vice versa. In a dependency relation,
Y depends on X iff X determines the grammatical category of the complex and
thus its external relations. […] Embedding is the dependency of a subordinate
syntagm. (Lehmann 1988: 182)

But often – as will be the case here – ‘subordination’ is applied in the restricted sense
of clausal subordination, as the asymmetric counterpart of clausal coordination (cf. te
Velde 2000): A clause β is subordinate to another clause β if the former is syntactically
dependent on the latter (in a sense to be further specified) but not vice versa. In tra-
ditional structural terms dependency can be conceived as structural (and functional)
embedding: the subordinate clause β is embedded into the superordinate or matrix
clause β with a specific syntactic function (complement, adverbial, attribute, etc.) at
some level of that clause. Other definitions of clausal subordination rest more directly
on the functional properties of the subordinate clause (e.g., Longacre 1985; Matthies-
sen & Thompson 1988; Cristofaro 2003). Cristofaro (2003) presents a comprehensive
typological study of subordination in 80 different languages based on what she terms
the Asymmetry Approach. She defines subordination in terms of the cognitive-lin-
guistic notion of ‘profiling’ (Langacker 1991), explicitly equating (non-)profiling with
(non-)assertion:13

.  According to Langacker (1991: 436) “[…] a subordinate clause can also be understood more
broadly as one that is not a main clause. Is there, then, something that all such clauses have in
common that sets them apart from main clauses? I suggest that there is, and that the distinction
rests on profiling. Specifically, the main clause is the head of at a particular level of organization,
 Cathrine Fabricius-Hansen & Wiebke Ramm

(b) We are now in a position to propose a functionally based definition of

subordination, resting on cross-linguistically applicable and consistent criteria.
By subordination will be meant a situation whereby a cognitive asymmetry is
­established between linked SoAs [States of Affairs], such that the profile of one
of the two (henceforth, the main SoA) overrides that of the other (henceforth,
the dependent SoA). This is equivalent to saying that the dependent SoA is
(pragmatically) non-asserted, while the main one is (pragmatically) asserted.
(Cristofaro 2003: 33)

In practice Cristofaro’s investigation seems to support approaches that consider subor-

dination to be a multidimensional phenomenon (Lehmann 1988, Verstraete 2007).
The definition in (c) below represents a third variant, based primarily on the no-
tion of prominence (cf. Section 1):
(c) In a general sense, if an element α is subordinate to an element β, it is less
prominent than β and usually α is dependent on β. […] It is a defining characteristic
of subordination (also called hypotaxis) that the subordinate element is syntactically
at a lower level in the overall structure than the element or string it is subordinate to.
 (Aarts 2006: 249)

As is well known, subordinate clauses in the canonical sense of complement, (restric-

tive) adverbial and (restrictive) relative clauses tend to exhibit specific formal charac-
teristics: constraints on verb form or word order (e.g., verb-final in German), the pres-
ence of specific subordination markers – complementizers, subordinating connectives
(subordinating conjunctions, subordinators, subjunctions), relativizers – etc.14
(8) She doubted that she would gain access.
(9) If you can’t beat them, join them!
(10) Keiner, der es gesehen hat, wird es je vergessen.
nobody rel.pron it seen has will it ever forget
‘Nobody who has seen it will ever forget it’

i.e., the clause that lends it profiling to the composite structure of a multiclausal expression.
A subordinate clause is then describable as one whose profile is overridden by that of a main
clause.”
.  It should be noted that from a structural point of view the entity consisting of subjunction
+ clause should not, strictly speaking, be considered a clause but a phrase of a different category,
headed by the subjunction, i.e., a ‘subjunction phrase’ or the like (cf. Pasch et al. 2003). Simi-
larly for coordination: the second conjunct including the conjunction itself by all relevant phrase
structure criteria constitutes a ‘conjunction phrase’ (ConjP) rather than a clause (see in particular
Johannessen 1998). Viewed as devices of so-called clause combining, conjunctions and subjunc-
tions are also not considered parts of the clauses they introduce. In this introduction, however, we
adhere to the traditional, albeit sloppy use of the term ‘subordinate/coordinate clause’.
 Editors’ introduction 

Hence the term ‘subordinate’ is sometimes also used with reference to clauses having
the relevant formal properties without being syntactically embedded in the strict sense
suggested above, i.e., without apparently having the function that is typically associ-
ated with those properties. So-called ‘supplementary’ (Huddleston & Pullum 2002:
1356) or ‘sentential’ relative clauses and resultative clauses introduced by so that are
cases in point; cf. (11) and (12) from Huddleston & Pullum (2002: 1356, 969). 15
(11) We called in to see Sue’s parents, which made us rather late.
(12) Most primary teachers are women so that suitable ‘role’ models, to use the trendy
phrase, are more abundant for girls than for boys.

Clauses having subordinate form may even be used as independent sentences with a
marked (exclamatory) illocutionary function, and conversely: clauses may be struc-
turally subordinated (embedded) without being formally marked as subordinate (see
e.g., Fabricius-Hansen 1992, Reis 1999). Such mismatches between form and function
have given rise to much discussion – and terminological confusion – in particular,
it seems, with respect to German (see e.g., Fabricius-Hansen 1992 and forthcoming;
Holler 2005; Lefèvre 2000; Pasch et al. 2003; Peyer 1997; Reis 1997; Zifonun et al. 1997:
2233ff). They are, however, clear indications of the multidimensionality of the notion
of clausal subordination. This is the issue addressed by Holler (Chapter 8) within the
framework of Head Driven Phrase Structure Grammar (HPSG). In what follows, we
shall use the term ‘subordinate clause’ to refer to clauses that are formally character-
ized as such, irrespective of whether they are properly embedded or not.
From a semantic point of view complement clauses, being arguments of predi-
cates, represent a very different case from adverbial and relative clauses; since none
of the papers in this volume is concerned with this category we shall not dwell on
it here (see e.g., Noonan 1985 for details). Adverbial and relative subordinate clause
types are often subsumed under the notion of (syntactic) adjuncts, together with e.g.,
non-clausal adverbials and attributive adjectival phrases (Austin et al. 2004; Lang et al.
2003). Semantically, such adjuncts are – grosso modo – said to modify clauses and noun
phrases (including determiner phrases), respectively. As for adverbial clauses,
[t]his modification can occur at various levels (such as verb phrase, tense phrase,
mood phrase) and in various dimensions (such as times and worlds). These varia-
tions give rise to a categorization of adverbial clauses (temporal, modal, …) and
a subcategorization according to a range of relations within these dimensions,
depending on the subjunction. (Sæbø forthcoming)

Modification often amounts to restricting the denotation of the modified constituent.

Thus the noun phrase student that passes the exam denotes a proper subset of the set

.  See Zifonun et al. (1997) for a detailed account of such mismatches between function and
form.
 Cathrine Fabricius-Hansen & Wiebke Ramm

denoted by the head noun (phase) student, i.e., the property of being both a student
and passing the exam – but with priority assigned to the head noun property (Bier-
wisch 2003). Similarly, the temporal clause in (13) restricts the temporal frame de-
noted by the future tense of the main clause.
(13) I’ll leave when you arrive.

This means that adjunction, including clausal adjunction, may involve the same type
of semantic operation as coordination, viz. logical conjunction and related operations
(see Section 3.1), but ‘asymmetrically’ restricted to the domain of the modified entity,
i.e., semantically ‘downgraded’.
In addition to restrictive relative clauses we find non-restrictive (parenthetical,
supplementary, appositive) relative clauses as exemplified in (11) above and (14):
(14) Dr. Brown, who lives next door, comes from Australia. (Quirk et al. 1985: 290)

However, while there seems to be general consensus as to the syntactic-semantic anal-

ysis of the canonical restrictive relative clause this, to the best of our knowledge, does
not hold for the non-restrictive varieties. It is not quite clear what their syntactic sta-
tus is, e.g., whether they should be considered adjuncts to the noun (or determiner)
phrase or to the matrix clause itself, nor is it clear how their semantic contribution
should be accounted for compositionally.
For adverbial clauses (in a broad ragbag sense) we can similarly distinguish be-
tween, on the one hand, canonical varieties that are properly integrated into their matrix
clause as modifiers at various levels (temporal, conditional, causal, illocutionary, …; cf.
the quotation above), on a par with ordinary non-clausal adverbials, and, on the other
hand, ‘non-restrictive’ varieties having a more supplementary relation to the subordi-
nate clause, like (12) above and (15):
(15) White motorists make up 78 per cent of Maryland highway traffic, while black
drivers account for about 17 per cent.

Integration as a dimension structuring the subordination-coordination (parataxis)

continuum is discussed in detail in the paper by Holler (Chapter 7); see also Lehmann
(1988). However, subordinate clauses that are not properly integrated into their matrix
clause, e.g., as a temporal or causal modifier, are ‘less subordinate’ (hypotactic) and
‘more coordinate’ (paratactic) than canonical restrictive varieties. This makes this type
of clause linkage particularly interesting in the present context: In order to under-
stand the rationale behind such ‘intermediate’ patterns of clause combining we have to
transgress the sentence level and investigate their use, i.e., how they function in natural
discourse (see Section 4.2) and across languages. This challenge is taken up by Ramm
(Chapter 6).
Being non-integrated or somehow detached from the rest of the sentence is a prop-
erty that certain subordinate clause types share with e.g., (clausal and non-clausal) par-
enthetical constructions (see Dehé & Kavalova (2007) for a recent collection of articles
on parentheticals), appositions and various ‘dislocated’ elements at the (left or right)
 Editors’ introduction 

sentence periphery. The discourse-structural functions of right dislocation are the topic
of Averintseva-Klisch’s contribution (Chapter 9).
In restricting our use of the term subordination to clausal subordination we devi-
ate terminologically from Lehmann (1988), who uses hypotaxis and parataxis for the
clausal varieties of coordination and subordination in the more general sense. In Mat-
thiessen & Thompson (1988), on the other hand, ‘hypotaxis’ and ‘hypotactic clause
(combining)’ refer to a subclass of what is traditionally subsumed under subordina-
tion, viz. subordinate clauses that are not embedded (or integrated). Matthiessen &
Thompson even suggest that
there is no advantage to postulating a grammatical category of ‘subordinate’
clause; rather the grammar of English at least, and perhaps of other grammars as
well, suggests that a distinction between what we have been calling ‘hypotaxis’
and embedding is crucial. (Matthiessen & Thompson 1988: 317)

It is, however, not quite clear how embedding is defined in their (systemic-functional)
theoretical framework. In practice, the notion seems to subsume complement clauses
and clauses functioning as restricting modifiers in noun phrases. This leaves us with
hypotaxis covering the heterogeneous group of adverbial clauses in the broadest tra-
ditional sense, whether semantically restrictive or not, as well as non-restrictive or
‘appositive’ noun-modifying clauses. It is far from evident that hypotaxis, thus under-
stood, is a natural category. So we prefer equating hypotaxis etc. with subordination
etc. in the general sense; parataxis, then, could be a cover term for explicitly marked
coordination, as it is understood here (Section 3.1), and mere juxtaposition of mu-
tually independent clauses. But on the whole, we refrain from using these terms. It
should be noted, however, that terminology varies somewhat in the present volume
as a whole. Thus coordination in the paper by Cosme (Chapter 4) is used in the broad
sense of Lehmann (1988), covering syndetic paratactic clause combining (with and
etc.) as well as asyndetic juxtaposition of independent clauses (separated by comma)
or sentences (separated by full stop) whereas e.g., Solfjeld (Chapter 5) understands
‘coordination’ in the narrower sense outlined in Section 3.1.

4.  ‘Coordination’ and ‘subordination’ at text level

Following our survey of current conceptions of syntactic coordination and subordina-

tion, this section briefly introduces relevant approaches to the description and repre-
sentation of discourse structure (4.1) and the discourse functions of syntactic subordi-
nation and coordination (4.2).

4.1  Relevant discourse-theoretical distinctions

There exists at present a whole series of more or less different conceptions of discourse
structure and the relations between discourse units. We shall confine ourselves to
models that are applied or discussed in the papers below.
 Cathrine Fabricius-Hansen & Wiebke Ramm

Segmented Discourse Representation Theory (SDRT), as outlined in Asher (1993)

and Asher & Lascarides (2003), is an extension of ‘classical’ Discourse Representation
Theory (DRT, Kamp & Reyle 1993). While DRT incrementally collapses the content
of sentence sequences into one single discourse representation structure (DRS), SDRT
aims at modelling discourse coherence and the incremental (dynamic) construction
of ‘segmented’ discourse representations (SDRSs) that ‘keep track of ’ the units mak-
ing up the text in question. This is achieved by representing such units – the discourse
constituents – as discourse referents of their own, elementary discourse constituents
being identified roughly with (DRSs assigned to) clauses. Discourse relations (Asher
1993, 2004), or rhetorical relations (Asher & Lascarides 2003), are defined for pairs of
discourse units immediately succeeding each other.
Like many approaches to discourse structure, SDRT acknowledges the fact that
discourse units may be organised hierarchically or non-hierarchically by distinguish-
ing between two types of discourse relations, ‘subordinating’ and ‘coordinating’.16 The
function of a discourse unit connected to its preceding discourse unit by a subordinat­
ing discourse relation (the prototypical example being Elaboration) is to change the
‘granularity’ of description in the text in the sense of providing more detail to some
element present in the preceding discourse unit (Asher & Lascarides 2003: 8, 146;
Asher 2004: 172), whereas discourse units connected to the preceding context by a
coordinating discourse relation continue the description without changing granular-
ity. Central to the definitions of subordinating vs. coordinating discourse relations is
the (formal) notion of a discourse topic (Asher 2004) and of discourse dominance (in
subordinating discourse relations the preceding discourse unit dominates the current
one).
As noted, SDRS construction is understood as an incremental (dynamic) online
process. Consequently, it has to be decided for each ‘new’ or ‘incoming’ discourse seg-
ment (≈ clause) how it attaches to the SDRS established thus far, i.e., at which point
and by which discourse relation(s). In this process the so-called Right Frontier Con­
straint (RFC) is of central importance, in particular with respect to the options for the
use of anaphoric expressions (Asher 1993: 270–271; Asher & Lascarides 2003: 10–12):
Anaphors can only be resolved if their antecedents can be found on the ‘right frontier’
of the incrementally growing discourse structure which covers the proposition intro-
duced by the prior sentence and any propositions that dominate it (i.e., are in a sub-
ordinating discourse relation to it). The concepts of ‘coordinating’ and ‘subordinating’
discourse relations are applied in the papers by Ramm (Chapter 6), Averintseva-Klisch
(Chapter 9), Delort (Chapter 10) and Petrova & Solf (Chapter 14), and discussed by
Blühdorn (Chapter 3).

.  We shall use the notation ‘coordinating’, ‘subordinating’ when referring to discourse rela-
tions rather than syntactic relations.
 Editors’ introduction 

Rhetorical Structure Theory (RST), as originally presented by Mann & Thompson

(1988), aims at modelling textual coherence by assigning discourse relations17 recur-
sively to pairs of neighbouring text spans of increasing length, starting with elementary
discourse units of (typically) clause length. The text (passage) is assumed to be coherent
if it is possible to assign a discourse relation to all discourse units contained in the text.
In RST, the idea that discourse units may be connected hierarchically as well as
non-hierarchically is accounted for by the distinction between nucleus-satellite and
multinuclear discourse relations. The former are weighted in the sense that the nucleus
is conceived as more central to the writer’s purpose than the satellite (i.e., the relation
between the discourse units is asymmetric). In multinuclear relations the discourse
units are not weighted against each other, i.e., they are of equal importance to the
writer’s purpose (they represent a symmetric relation). The notion of nuclearity in RST
is the central topic of Stede’s contribution to the present volume (Chapter 2).
Seemingly, the fundamental distinction between nucleus-satellite and multinu-
clear discourse relations is similar to the ‘subordinating’/‘coordinating’ dichotomy in
SDRT, and some of the discourse relations do indeed have the same labels in both
theories (e.g., Elaboration, Background, Contrast). The underlying criteria for the as-
signment of a discourse relation, however, are quite different: In SDRT, which aims at
an incremental model-theoretic interpretation of discourse, the definitions of SDRT
relations are closely related to the construction and maintenance of discourse topics
and to the temporal relations between eventualities. In RST, on the other hand, the
definitions of discourse relations are based on the writer’s intentions as to which com-
municative effects s/he wants to reach by uttering a piece of discourse.
Yet another concept of discourse organisation can be found in the Quaestio model,
as outlined e.g., in Klein & v. Stutterheim (1991, 1992) and Stutterheim (1997), which
was originally developed to account for text production rather than discourse analysis.
According to this approach, a text answers a specific text question, quaestio, which
may be divided into a sequence of sub-questions.18 The global structure or develop-
ment of the text will be constrained in different ways depending on the nature of the
quaestio, e.g., whether it is a narrative or a descriptive task. Sentences or clauses that
contribute to answering the text quaestio (by answering a sub-quaestio) are said to be-
long to the main structure of the text. In addition, a text may contain various kinds of
side structures consisting of clauses, sentences or larger text spans that do not directly
contribute to the quaestio, but fulfil other functions.

.  The set of discourse relations is not absolutely fixed, but many RST applications operate
with a set of approx. 30 relations (see e.g., the relation definitions on the ‘official’ RST website
http: //www.sfu.ca/rst/01intro/definitions.html.
.  In essence, the notion of quaestio corresponds to the notion of a Question under Discussion
(QuD), which was developed in question-answer analysis and which is applied in Jasinskaja’s
(2006) approach to discourse relations and the discourse function of and-coordination.
 Cathrine Fabricius-Hansen & Wiebke Ramm

Since it is not quite clear how the side structures are linked to the main structure
(if at all), a text may not be formally represented as a connected hierarchical structure
(e.g., tree structure) in this model, as is the case in SDRT and RST. But, obviously, side
structures are ‘secondary’ to the main structure.
The quaestio approach has primarily been applied to narrative texts, i.e., text pas-
sages that are constrained by a principle of temporal movement; the paper by Carroll
et al. (Chapter 7) provides an example. As observed by Ramm (Chapter 6) the model
may not be applied quite as easily to non-narrative texts.

4.2  Syntactic coordination and subordination from a textual perspective

Discourse theories differ also with respect to whether, and to what degree, they assume
and model a (more or less) direct correlation between coordination and subordination
in syntax and on the level of discourse representation.
As for syntactic coordination, SDRT is relatively explicit in assuming that syntac-
tic and-coordination is a reliable indicator of coordination on discourse level, i.e., a
‘coordinating’ discourse relation like Narration or Result. Asher & Vieu 2005 (follow-
ing Txurruka 2000), for example, use and-coordination as a test to distinguish be-
tween ‘coordinating’ and ‘subordinating’ discourse relations, although they do point
to data suggesting that the inference from and-coordination to discourse coordina-
tion may be defeasible in some cases (Asher & Vieu 2005: 598–599; see also Asher
2004: 183–185).
RST is less explicit than SDRT in associating syntactic coordination with a specific
type of discourse relation, since the theory generally refrains from directly mapping
discourse relations to concrete syntactic realisations in a particular language. However,
examples given on the RST website (http://www.sfu.ca/rst/), illustrating the multinu-
clear relation Conjunction, suggest that also RST assumes that the syntactic symmetry
of and-coordination mirrors a multinuclear (symmetric, non-weighted) relation on
discourse level.
Whether such a correlation between syntactic and discourse coordination is a fea-
ture that holds universally across languages, however, has been questioned in recent
contrastive studies (e.g., Ramm & Fabricius-Hansen 2005; Ramm forthcoming). The
studies show that coordination is used somewhat differently in Norwegian versus Ger-
man and English. In particular, syntactic coordination seems to be compatible with
some (‘subordinating’) discourse relations in Norwegian that are blocked in German
or English.
As concerns possible correlations between syntactic subordination and discourse
subordination or asymmetry, RST is slightly more explicit, e.g., in Matthiessen &
Thompson (1988: 317), who hold “that hypotactic clause combining is best under-
stood as a grammaticalization of the Nucleus-Satellite relations which characterize
the rhetorical organization of certain types of written discourse”. Of course, due to
the recursive nature of constructing discourse representations in RST, the concept of
 Editors’ introduction 

asymmetric nucleus-satellite relations is not restricted to syntactically subordinate

structures: a nucleus-satellite relation may also hold between independent sentences
or even between paragraphs in a text. But again there seems to be some agreement
that syntactic subordination typically implies subordination/asymmetry on discourse
level as well. The issue of possible non-correlations between subordination on the syn-
tactic level and on the level of discourse relations is addressed in the papers by Delort
(Chapter 10) and Ramm (Chapter 6). Also Stede (Chapter 2) shows that automatically
equating syntactic subordinating with discourse asymmetry (in the RST sense) is a
simplification that does not do justice to the complexity of natural discourse (cf. Stede
2004).
The Relevance-Theoretic work on and-coordination by Blakemore and Carston
(Blakemore 1987, 2002; Blakemore & Carston 2005), pointing to the possibility of
asymmetric interpretations, has already been mentioned above (Section 3.1; see also
Solfjeld, Chapter 5 in this volume). Blakemore and Carston also show that coordi-
nation is possible in certain cases while blocked in others and, in particular, that
using coordination instead of a sequence of non-coordinated (‘full stop’) sentences
sends two types of signal to the reader: (i) The two conjuncts should be processed
as a unit, both conjuncts functioning together as premises in the derivation of a
joint cognitive effect; (ii) specific inferences are licensed regarding the semantic-
pragmatic relations holding between them, the first conjunct always functioning as
a background to the processing of the second. Thus, in narratives a temporal-causal
relation is often inferred (see Section 3.1). A non-narrative use of coordination can
be seen in argumentative examples, where the conjuncts make a joint contribution
as steps in an argumentation (Blakemore & Carston 2005). Relevance Theory does
not distinguish between ‘coordinating’ and ‘subordinating’ discourse relations – and
prefers to avoid the notion of discourse relations altogether (Blakemore 2002: Sec-
tion 5.3) – but most of the narrative, as well as the argumentative examples given in
Blakemore & Carston (2005), would probably be classified as ‘coordinating’ relations
in the SDRT framework.
Klein and v. Stutterheim’s quaestio approach also dispenses with an explicit theo-
retical concept of discourse relations. But at least in narration, clause coordination
with and etc. is generally taken to continue the current quaestio, i.e., mostly the main
story line, corresponding to the ‘coordinating’ discourse relations Narration and Con-
tinuation in SDRT. Syntactic subordination, on the other hand, is conceived as a means
of ‘downgrading’ information (Stutterheim 1997; see also Carroll et al. in this volume),
without necessarily assigning the downgraded information to some side structure.19
Associating syntactic subordination with a downgrading or backgrounding func-
tion at text level is common practice in more traditional linguistic description (see e.g.,

.  Side structures may be realized in different ways: as subordinate clauses, as parenthetical
insertions or other kinds of ‘orphans’, and as independent sentences or text passages.
 Cathrine Fabricius-Hansen & Wiebke Ramm

Aarts 2007; Brandt 1996; Cristofaro 2000; Hartmann 2004; Hetland & Molnár 2001;
Peyer 1997). Unfortunately, however, these notions are understood in varying ways,
having being explained alternately – and mostly not very explicitly – in terms of e.g.,
information structure (focus – background), illocutionary function (assertion vs. non-
assertion), temporal anchoring (event versus co-temporal state), (lack of) relevance in
the given discourse context, attachment options for subsequent discourse units, or in
terms of other general terms like salience, prominence or communicative weight.
There are exceptions such as Tomlin (1985) and Schecker (2000), who define the
notions of foreground(ed) and background(ed), or downgraded, information in ex-
perimental text reproduction independently of linguistic form. And they both con-
clude that (certain types of) subordinate clauses do tend to encode information that is
backgrounded or downgraded in the defined sense.
On the whole, however, empirical investigations into discourse-functional aspects
of subordination and coordination in natural discourse are rare. There is a conspicu-
ous lack of comprehensive studies combining a precise theoretical framework and a
sound methodology with extensive empirical data from ‘wildlife’ texts including, but
not restricted to, narratives.
Matthiessen and Thompson conclude their article on “The structure of discourse
and ‘subordination’ ” as follows:

[I]f hypotaxis in English is a grammaticalization of rhetorical relations, then

it follows that the grammar of clause combining may differ radically from one
language to another. Indeed, preliminary discourse-based investigations of
such phenomena in unrelated languages strongly suggest that this is the case.
If the basic approach to clause combining taken in this paper is correct, then
the interesting cross-linguistic issue is how and to what extent the grammar of
clause-combining in a given language reflects the rhetorical organization
of discourse in that language […]. (Matthiessen & Thomspon 1988: 317)

Since then some empirically oriented contrastive research has been done, indicating
that even quite closely related languages (or language communities) may differ con-
siderably in their use of syntactic coordination and subordination (in a broad sense)
as a means of information packaging at text level; see the papers by Cosme, Solfjeld and
Carroll et al. (Chapters 4–7), with further references. Part of this picture is the differ-
ence between a more ‘incremental’ and a more ‘hierarchical’ style of writing, which is
mentioned by Cosme (Chapter 4), and taken up more in detail by Ramm (Chapter 6);
see also Fabricius-Hansen (2007).
Of course, more studies are needed, based on more data, more languages and a
more refined assessment of language-specific structural constraints and language-spe-
cific means of expressing syntactic subordination and coordination, on the one hand,
and (discourse) relations between independent sentences, on the other. But it is our
hope that the papers presented here will help reduce the research gaps outlined above,
either directly or by encouraging further research.
 Editors’ introduction 

5.  Outline of contents

The overall structure of the following collection of papers is outlined in Section 1.

Below we briefly present the individual contributions (in the order in which they ap-
pear below): their research topics, theoretical framework and methods of investiga-
tion, their relation to the general topics of the present volume, as outlined above, and
their relation to each other.
Manfred Stede (‘RST revisited: Disentangling nuclearity’, Chapter 2) discusses the
central notion of nuclearity in RST (see Section 4.1 above) and other discourse-theoretical
approaches, relating it to the broader concept of salience. Reviewing three problematic as-
pects of RST nuclearity – its relationship to syntactic subordination, its direct association
with coherence relations, and its purported pervasiveness – he argues that the complexity
of textual coherence and discourse organization is not adequately modelled by hierarchi-
cal (tree) structures alone, but demands a multi-level representation. The last part of his
paper outlines an annotation framework for multi-level discourse representation. Like
Blühdorn (Chapter 3), Stede directly addresses theoretical issues of primary importance
in the present context. The RST notion of nuclearity also plays a role in the paper by
Ramm (Chapter 6).
Hardarik Blühdorn (‘Subordination and coordination in syntax, semantics, and
discourse: Evidence from the study of connectives’, Chapter 3) addresses the basic ques-
tion of whether the syntactic and the discourse notions of subordination and coordi-
nation are as closely related as the parallel terminology may suggest (cf. Sections 1 and
4.1 above). Using the syntax and semantics of connectives (in the broad sense of e.g.,
Pasch et al. 2003) as his point of departure, he argues that the nature of the syntactic
and the discourse domains are too different to warrant such parallelism and that syn-
tactic hierarchy should not be considered a general model for the conceptualization of
discourse hierarchy, nor vice versa. Blühdorn’s contribution relates to those by Solfjeld
(Chapter 5) Ramm (Chapter 6) and Delort (Chapter 10) by pointing to mismatches
between syntactic subordination or coordination and ‘subordination’/‘coordination’ at
the discourse level. Further, it ties in well with Stede’s paper (Chapter 2) in that both
address a primarily theoretical issue.
Christelle Cosme (‘A corpus-based perspective on clause linking patterns in English,
French and Dutch’, Chapter 4) compares clause-linking patterns and discourse infor-
mation packaging in English, French and Dutch. Investigating comparable (authentic)
as well as translational parallel corpora she concludes that the three languages employ
clausal subordination and paratactic clause combining differently: On the scale between
a more incremental and a more hierarchical information organization Dutch tends to-
wards the former and French towards the latter, with English situated somewhere in
the middle. From a theoretical perspective, Cosme’s study supports the conception of
(clausal) coordination and subordination as gradient categories (cf. Section 3 above);
these aspects of her paper link it to Holler’s study (Chapter 8). Her contribution is relat-
ed to the three subsequent papers – Solfjeld, Ramm and Carroll et al. – by its contrastive
 Cathrine Fabricius-Hansen & Wiebke Ramm

perspective and its focus on the use of subordination and paratactic clause combining.
It shares two of its object languages (English and French) with Carroll et al. (Chapter 7)
but is more closely connected to Solfjeld (Chapter 5) and Ramm (Chapter 6) as far as its
research questions, data and methods are concerned.
Kåre Solfjeld’s study (‘Sentence splitting – and strategies to preserve discourse
structure in German-Norwegian translations’, Chapter 5) is related to the papers by
Cosme (Chapter 4) and, in particular, Ramm (Chapter 6): it is contrastive (German-
Norwegian), based on parallel (translational) corpora, and concerned with the relation
between syntactic subordination/coordination and discourse interpretation. The par-
ticular question Solfjeld addresses is to what extent and by what means the discourse
structure of the (German) source text (ST) is preserved in the (Norwegian) target text
(TT) when a phrasal adjunct is rendered as an independent sentence or clause. Fol-
lowing a primarily relevance-theoretic line of argumentation, he concludes that simply
preserving the linear order of the relevant information units often suffices to ensure an
adequate interpretation of the TT, but that clause coordination with og ‘and’ may be
used as compensation for missing grammatical options to express ‘backgrounding’ by
syntactic subordination (adjunction); and connectives may be added in the second TT
clause/sentence if the ST discourse relation is not inferable otherwise.
Wiebke Ramm (‘Upgrading of non-restrictive relative clauses in translation – a
change in discourse structure?’, Chapter 6) analyses the discourse functions of non-
restrictive relative clauses (see Section 3.1 above) and the discourse-functional effect
of ‘upgrading’ (German) non-restrictive relative clauses to independent sentences in
(Norwegian) translations. She investigates whether the distinction between appositive/
discontinuative and continuative relative clauses is relevant for how a non-restrictive
relative clause is translated into Norwegian (in cases where translation by a corre-
sponding relative clause is not an option). Ramm concludes that upgrading is more
problematic with respect to discontinuative/appositive relative clauses, since this gives
them too much weight in the discourse structure compared to the original text. The
study also reveals certain shortcomings of approaches based on SDRT and the quaes-
tio model. Furthermore, as in Cosme’s paper, the topic of different discourse organisa-
tion strategies across languages is taken up: The upgrading of relative clauses makes
the translations less hierarchically organised than the original texts – an observation
which calls for reflections on the cross-linguistic assessment of coherence within and
across sentence boundaries.
Mary Caroll, Antje Rossdeutscher, Monique Lambert & Christiane von Stutter­
heim (‘Subordination in narratives and macrostructural planning: A comparative
point of view’, Chapter 7) present a comparative study of ‘macrostructural planning’
in English, French and German retellings of a silent animation. Following the quaes-
tio model (cf. Section 4.1 above) they take macrostructural planning to manifest
itself in information selection (Which events to narrate?) and strategies of down-
grading selected information (e.g., clausal subordination and passive) in accordance
with the primary narrative task of advancing the story line. Part of this picture is
 Editors’ introduction 

the choice of (protagonist versus narrator) perspective, which involves the temporal
frame of reference and potential mapping of narrated events onto subordinate rather
than main clauses (i.e., downgrading). The first part of the study presents empirical
generalizations relating observed significant language-specific differences between
the three sets of narratives to well-defined morphosyntactic parameters like sub-
ject vs. topic prominence, aspectual marking and word order flexibility. The sec-
ond part constitutes a first step towards a formal (decision-hierarchical) model of
the language-specific principles guiding macrostructural narrative planning. As a
cross-linguistic investigation of information packaging at discourse level this paper
is linked, in particular, to Ramm (Chapter 6) and Cosme (Chapter 4).
The paper by Anke Holler (‘German dependent clauses from a constraint-based
perspective’, Chapter 8) focuses on the general issue of non-canonical subordination
(see Section 3.1 above), proposing a fine-grained systematization and explication
within the constraint-based framework of Head-driven Phrase Structure Grammar
(HPSG). She argues that dependent clauses must be distinguished regarding their de-
gree of integration into a putative matrix clause and exemplifies this by means of five
instances of non-canonical clause types in German: weil-verb second clauses, continu-
ative wh-relative clauses, verb second relative clauses, free dass-clauses, and dependent
verb second clauses. The distinctions between these clause types are modelled in the
HPSG framework as constraints and sorts partially ordered in multiple-inheritance
hierarchies. By not depending on the position of the finite verb, the analysis goes be-
yond earlier constraint-based approaches implementing a strict dichotomy between
main and subordinate clauses and relying on the V2 property alone. It is linked to
other papers concerned with detached constructions (cf. Averintseva-Klisch, Chapter
9) and mismatches between formal and functional aspects of subordination, i.e., De­
lort (Chapter 10) and Solfjeld (Chapter 5).
Maria Averintseva-Klisch (‘To the right of the clause: Right dislocation vs. af-
terthought’, Chapter 9) investigates the discourse functions of two apparently simi-
lar constructions at the right sentential edge in German, right dislocation (RD) and
afterthought (AT). She shows that RD is syntactically adjoined to its host sentence
and functions as a discourse-structuring device marking the topic for the following
discourse segment, whereas AT is syntactically independent of its host sentence and
serves as a local repair strategy in discourse. Furthermore, she shows that these two
functions can be identified in French and Russian data as well. Her paper ties in with
other papers in this volume investigating possible mismatches between syntactic and
discourse structure, e.g., Ramm (Chapter 6) and Delort (Chapter 10), by pointing out
that AT exemplifies a subordinating discourse relation (according to the tests devel-
oped by Asher & Vieu 2005), but is detached syntactically as an orphan to its host
sentence.
Laurence Delort (‘Exploring the role of clause subordination in discourse structure –
the case of French avant que’, Chapter 10) investigates discourse relational aspects of
a specific subordinating connective, viz. French avant que (‘before’). She ­argues that
 Cathrine Fabricius-Hansen & Wiebke Ramm

avant que, depending on the (sentence-internal and -external) context, may convey at
least two different discourse relations between the connected clauses. Adopting rel-
evant SDRT definitions (see Section 4.1), she identifies these two relations with Nar­
ration, which is a ‘coordinating’ relation, and Background, which she classifies as a
‘subordinating’ relation in accordance with Vieu & Prévot (2004).20 The difference
manifests itself in different substitution and translation options. Delort’s contribution
relates to other SDRT-inspired contributions, viz. Averintseva-Klisch (Chapter 9) and
Petrova & Solf (Chapter 14), and to Blühdorn (Chapter 3), Ramm (Chapter 6) and Hol­
ler (Chapter 8), which are also concerned with the mismatch between syntactic and
discourse ‘subordination’.
Michael Franke (‘Pseudo-imperatives and other cases of conditional conjunc-
tion and conjunctive disjunction’, Chapter 11) investigates interpretational asym-
metries (‘pragmatic puzzles’) exhibited by so-called pseudo-imperatives (PIs), i.e.,
and-conjunction and or-disjunction as exemplified in example (3) in Section 3.1.
Analyzing PIs on the background of related non-standard conjunction and disjunc-
tion, he proposes a pragmatic explanation of PI interpretation that is based on the
different discourse segmentation behaviour of the two connectors: Disjunctive PIs are
discourse separating, associating the two connected clauses with separate communi-
cative events, while conjunctive PIs may be either discourse integrating or discourse
separating, with different consequences for the interpretation of the construction as
a whole. In distinguishing between discourse integrating and discourse separating
(instances of) conjunctive or disjunctive coordination, Franke implicitly addresses
not only the fundamental issue of discourse segmentation (see Section 2 above) but
also the question of where, i.e., at which meaning level, coordinate clause combining
may take place (see Section 3.1 above). The topic of his contribution – asymmetry in
clausal coordination – links it to the paper by Reich (Chapter 12).
Ingo Reich (‘From discourse to “odd coordinations” – on asymmetric coordination
and subject gaps in German’, Chapter 12) addresses syntactic-semantic asymmetries
relating to word order (‘clause type’) and subject gaps in German – so-called Asym-
metric Coordination (AC) and SLF-Coordination (SLFC)21 – as illustrated in example
(4) in Section 3.1. Combining generative feature-based syntax and formal semantics
with discourse-relational considerations, he argues that the conjuncts in both cases are
semantically linked by a relation of event subordination, termed Occasion. He defines
this relation as a semantic explication of the discourse (‘Contiguity’) relation Occa-
sion as understood by Kehler (2002), introducing it syntactically by way of a specific

.  Asher (1993) and Asher & Lascarides (2003) view Background as a ‘coordinating’ rela-
tion; but as argued by Vieu & Prévot (2004) and Behrens & Fabricius-Hansen (forthcoming), it
fails to meet decisive (RFC-based) criteria of discourse ‘coordination’. In Asher (2004: 179–180),
Background is characterized as ‘subordinating’.
.  SLF stands for ‘Subject Lacking in F(ronted) structure’ (Reich forthcoming: 2).
 Editors’ introduction 

functional projection. Thus, the syntactic-semantic characteristics of the AC and SLFC

constructions are explained as a ‘grammaticalization’ of properties that are typically
attributed to coordinate structures in discourse. Reich’s contribution is related to other
contributions concerned with subordination-coordination mismatches, i.e., Blühdorn
(Chapter 3) and Holler (Chapter 8), and to the immediately preceding paper (Franke)
on pragmatic asymmetries in coordination.
Rosemarie Lühr (‘Old Indic clauses between subordination and coordination’,
Chapter 13) addresses the interplay between prosody and subordination/coordina-
tion in Old Indic. Her paper aims to explain the distribution of sentences with and
without stress on the initial verb and, in particular, the status of sentences that have
the prosodic properties of subordinate clauses but are not introduced by a subordina-
tion marker. Lühr proposes a solution in terms of information structure, viewing the
latter construction type as the Old Indic counterpart of German coordinate structures
showing the rise-fall contour that goes along with contrastive coordination. That is,
contrary to what has often been suggested in the literature, she takes stress on the
first verb in sentences lacking a subordinator to be a main clause phenomenon signaling
that a continuation will follow. Lühr’s contribution is the only one in the collection that
explicitly takes up prosodic and information-structural aspects of subordination-
coordination mismatches. However, it is linked to the paper by Petrova & Solf (Chapter 14)
in applying theories of information structure and discourse relations to historic
languages.
Svetlana Petrova & Michael Solf (‘Rhetorical relations and verb placement in the
early Germanic languages. A cross-linguistic study’, Chapter 14) investigate discourse-
related properties of verb placement (V1 versus V2) in Old High German (OHG) and
other early Germanic (declarative) main clauses. Adopting the SDRT distinction be-
tween ‘subordinating’ and ‘coordinating’ discourse relations, they argue that V1 is a
common correlate of discourse ‘coordination’ in all early Germanic languages while
‘subordinating’ discourse relations are signalled by different syntactic means. These
differences may be responsible for the development of different word order patterns
in the modern languages. Petrova & Solf ’s contribution is closely related to the paper
by Lühr (Chapter 13) in being concerned with information or discourse structural
aspects of main clause word order in historic Indo-European languages.

References

Aarts, Bas. 2006. Subordination. In Encyclopedia of Language & Linguistics, 2nd edn, Keith
Brown (ed.), 248–254. Amsterdam: Elsevier.
Abraham, Werner. 2006. Topic, focus and default vs. contrastive accent: Typological differences
with respect to discourse prominence. In On Information Structure, Meaning and Form.
Generalizations across Languages [Linguistik Aktuell / Linguistics Today 100], Kerstin
Schwabe & Susanne Winkler (eds), 183–203. Amsterdam: John Benjamins.
 Cathrine Fabricius-Hansen & Wiebke Ramm

Asher, Nicholas. 1993. Reference to Abstract Objects. Dordrecht: Kluwer.

Asher, Nicholas. 2004. Discourse topic. Theoretical Linguistics 30: 163–202.
Asher, Nicholas & Lascarides, Alex. 2003. Logics of Conversation [Studies in Natural Language
Processing]. Cambridge: CUP.
Asher, Nicholas & Vieu, Laure. 2005. Subordinating and coordinating discourse relations. Lin­
gua 115: 591–610.
Austin, Jennifer R., Engelberg, Stefan & Rauh, Gisa. 2004. Adverbials. The Interplay between
Meaning, Context, and Syntactic Structure [Linguistik Aktuell/Linguistics Today 70].
Amsterdam/Philadelphia: John Benjamins.
Bátori, István, Pusch, Luise F., Levin, Jurij L., Abraham, Werner, Bublitz, Wolfram & Roncador,
Manfred von. 1975. Syntaktische und semantische Studien zur Koordination [Studien zur
deutschen Grammatik 2]. Tübingen: Niemeyer.
Behrens, Bergljot & Fabricius-Hansen, Cathrine. Forthcoming. The relation Accompanying Cir-
cumstance across languages. Conflict between linguistic expression and discourse subor-
dination? In Proceeedings from the Third International Conference on Contrastive Semantics
and Pragmatics, Shanghai, September 15–19, 2005.
Bierwisch, Manfred. 2003. Heads, complements, adjuncts: Projection and saturation. In Mod­
ifying Adjuncts [Interface Explorations 4], Ewald Lang, Claudia Maienborn & Cathrine
Fabricius-Hansen (eds), 113–162. Berlin: Mouton de Gruyter.
Blakemore, Diane. 1987. Semantic Constraints on Relevance. Oxford: Blackwell.
Blakemore, Diane. 2002. Relevance and Linguistic Meaning. Cambridge: CUP.
Blakemore, Diane. 2005. And-parentheticals. Journal of Pragmatics 35: 1165–1181.
Blakemore, Diane & Carston, Robyn. 2005. Introduction to coordination: Syntax, semantics and
pragmatics. Lingua 115: 353–358.
Brandt, Margareta. 1996. Subordination und Parenthese als Mittel der Informationsstrukturie-
rung in Texten. In Ebenen der Textstruktur. Sprachliche und kommunikative Prinzipien,
Wolfgang Motsch (ed.), 211–240. Tübingen: Niemeyer.
Büring, Daniel. 2007. Semantics, intonation and information structure. In The Oxford Handbook
of Linguistic Interfaces, Gillian Raamchand & Charles Reiss (eds), Ch. 15. Oxford: OUP.
Carston, Robyn. 2002. Thoughts and Utterances. London: Blackwell.
Carston, Robyn & Blakemore, Diane (eds). 2005. Coordination: Syntax, Semantics and Pragmat­
ics [Lingua 115].
Cormack, Annabel & Smith, Neil. 2005. What is coordination? Lingua 115: 395–418.
Cristofaro, Sonia. 2003. Subordination. Oxford: OUP.
Crysman, Berthold. 2006. Coordination. In Encyclopedia of Language & Linguistics, 2nd edn,
Keith Brown (ed.), 183–196. Amsterdam: Elsevier.
Dehé, Nicole & Kavalova, Yordanka (eds). 2007. Parentheticals [Linguistik Aktuell/Linguistics
Today 106]. Amsterdam: John Benjamins.
Doherty, Monika (ed.). 1996. Information Structure: A Key Concept for Translation Theory [Lin-
guistics 34].
Doherty, Monika (ed.). 1999. Sprachspezifische Aspekte der Informationsverteilung [Studia Gram-
matica 47]. Berlin: Akademie-Verlag.
Doherty, Monika (ed.). 2005. Language-specific Discourse Linking [Linguistics 43].
Doherty, Monika. 2006. Structural Propensities: Translating Nominal Word Groups from English
into German [Benjamins Translation Library 65]. Amsterdam: John Benjamins.
Fabricius-Hansen, Cathrine. 1992. Subordination. In Deutsche Syntax. Ansichten und Aussich­
ten, Ludger Hoffmann (ed.), 458–483. Berlin: de Gruyter.
 Editors’ introduction 

Fabricius-Hansen, Cathrine. 2007. Dreimal (nicht) dasselbe: Sprachliche Perspektivierung im

Deutschen, Norwegischen und Englischen. LiLi: Zeitschrift für Literaturwissenschaft und
Linguistik 145: 61–86.
Fabricius-Hansen, Cathrine. Forthcoming. Subjunktor. In Handbuch der deutschen Wortarten,
Ludger Hoffmann (ed.), article C22. Berlin: de Gruyter.
Granger, Sylviane, Lerot, Jacques & Petch-Tyson, Stephanie (eds). 2003. Corpus-based Approach­
es to Contrastive Linguistics and Translation Studies [Approaches to Translation Studies 20].
Amsterdam: Rodopi.
Haiman, John & Thompson, Sandra D. (eds). 1988. Clause Combining in Grammar and Dis­
course. Amsterdam: John Benjamins.
Hartmann, Dietrich. 1984. Reliefgebung: Informationsvordergrund und Informations-
hintergrund in Texten als Problem von Textlinguistik und Stilistik. Zur Verwendung
sprachlicher Mittel zum Ausdruck von Haupt- und Nebeninformation. Wirkendes Wort
4: 305–323.
Haspelmath, Martin. 2004. Coordinating constructions: An overview. In Coordinating Construc­
tions [Typological Studies in Language 58], Martin Haspelmath (ed.), 3–40. Amsterdam:
John Benjamins.
Hasselgård, Hilde, Johansson, Stig, Behrens, Bergljot & Fabricius-Hansen, Cathrine (eds). 2002.
Information Structure in a Cross-linguistic Perspective [Language and Computers – Studies
in Practical Linguistics 39]. Amsterdam: Rodopi.
Hetland, Jorunn & Molnár, Valéria. 2001. Informationsstruktur und Reliefgebung. In Language
Typology and Language Universals [HSK 20(2)], Martin Haspelmath & Ekkehard König
(eds), 617–633. Berlin: Mouton de Gruyter.
Holler, Anke. 2005. Weiterführende Relativsätze. Empirische und theoretische Aspekte [Studia
Grammatica 60]. Berlin: Akademie Verlag.
Huddleston, Rodney & Pullum, Geoffrey K. 2002. The Cambridge Grammar of the English Lan­
guage. Cambridge: CUP.
Jasinskaja, Ekaterina. 2006. Pragmatics and Prosody of Implicit Discourse Relations: The Case
of Restatement. PhD dissertation, University of Tübingen.
Johannessen, Janne Bondi. 1998. Coordination. Oxford: OUP.
Johansson, Stig & Oksefjell, Signe (eds). 1998. Corpora and cross-linguistic research: Theory,
method and case studies [Language and Computers – Studies in Practical Linguistics 24].
Amsterdam: Rodopi.
Johansson, Stig. 2007. Seeing through Multilingual Corpora. On the Use of Corpora in Contrastive
Studies [Studies in Corpus Linguistics 26]. Amsterdam: John Benjamins.
Kamp, Hans & Reyle, Uwe. 1993. From Discourse to Logic. Introduction to Modeltheoretic Se­
mantics of Natural Language, Formal Logic and Discourse Representation Theory [Studies in
Linguistics and Philosophy 42]. Dordrecht: Kluwer.
Karhiaho, Izabela. 2003. Der Doppelpunkt im Deutschen: Kontextbedingungen und Funktionen.
PhD dissertation, Göteborg University.
Kehler, Andrew. 2002. Coherence, Reference and the Theory of Grammar. Stanford: CSLI.
Kehler, Andrew. 2004. Discourse topics, sentence topics and coherence. Theoretical Linguistics
30: 227–240.
Klein, Wolfgang & Stutterheim, Christiane von. 1991. Text structure and referential movement.
Sprache und Pragmatik 22: 2–32.
Klein, Wolfgang & Stutterheim, Christiane von. 1992. Textstruktur und referentielle Bewegung.
LiLi 86: 67–92.
 Cathrine Fabricius-Hansen & Wiebke Ramm

Krifka, Manfred. 2007. Basic notions of information structure. In Interdisciplinary Studies of

Information Structure 6, Caroline Fery & Manfred Krifka (eds). Potsdam University 2007.
(Also at URL: http://amor.rz.hu-berlin.de/~h2816i3x/).
Lang, Ewald. 1977. Semantik der koordinativen Verknüpfung [Studia Grammatica XIV]. Berlin:
Akademie-Verlag.
Lang, Ewald. 1984. The Semantics of Coordination [Studies in Language Companion Series 9].
Authorized English translation from Lang (1997). Amsterdam: John Benjamins.
Lang, Ewald. 2004. Schnittstellen der Konnektoren-Beschreibung. In Brücken schlagen. Grundla­
gen der Konnektorensemantik, Hardarik Blühdorn, Eva Breindl & Ulrich Hermann Waßner
(eds), 45–92. Berlin: Mouton de Gruyter.
Lang, Ewald, Maienborn, Claudia & Fabricius-Hansen, Cathrine (eds). 2003. Modifying Ad­
juncts. Berlin: Mouton de Gruyter
Langacker, Ronald W. 1991. Foundations of Cognitive Linguistics Vol. II. Stanford CA: Stanford
University Press.
Lefèvre, Michel (ed.). 2000. Subordination in Syntax, Semantik und Textlinguistik. Tübingen:
Stauffenburg.
Lehmann, Christian. 1988. Towards a typology of clause linkage. In Clause Combining in Gram­
mar and Discourse, John Haiman & Sandra D. Thompson (eds), 181–226. Amsterdam:
John Benjamins.
Levinson, Stephen C. 2000. Presumptive Meanings. The Theory of Generalized Conversational
Implicature. Cambridge MA: The MIT Press.
Longacre, Robert E. 1985. Sentences as combinations of clauses. In Language Typology and Syn­
tactic Description II, Timothy Shopen (ed.), 235–286. Cambridge: CUP.
Mann, William C. & Thompson, Sandra A. 1987. Rhetorical structure theory: A framework for
the analysis of text. IPRA Paper in Pragmatics 1: 79–102.
Mann, William C. & Thompson, Sandra A. 1988. Rhetorical Structure Theory: Toward a func-
tional theory of text organization. Text 8(3): 243–281.
Matthiessen, Christian & Sandra A. Thompson. 1988. The structure of discourse and ‘subor-
dination’. In Clause Combining in Grammar and Discourse, John Haiman & Sandra A.
Thompson (eds), 275–329. Amsterdam: John Benjamins.
Noonan, Michael. 1985. Complementation. In Language Typology and Syntactic Description II,
Timothy Shopen (ed.), 42–140. Cambridge: CUP.
Pasch, Renate, Brauße, Ursula, Breindl, Eva & Waßner, Ulrich Hermann. 2003. Handbuch der
deutschen Konnektoren: Linguistische Grundlagen der Beschreibung und syntaktische Merk­
male der deutschen Satzverknüpfer (Konjunktionen, Satzadverbien und Partikeln). Berlin:
de Gruyter.
Partee, Barbara H. & Rooth, Mats. 1983. Generalized conjunction and type ambiguity. In Mean­
ing, Use and Interpretation of Language, Rainer Bäuerle, Christoph Schwarze & Arnim von
Stechow (eds), 361–386. Berlin: de Gruyter.
Peyer, Ann. 1997. Satzverknüpfung – syntaktische und textpragmatische Aspekte [RGL 178].
Tübingen: Niemeyer.
Quirk, Randolph, Greenbaum, Sidney, Leech, Geoffrey & Svartvik, Jan. 1985. A comprehensive
Grammar of the English Language. London: Longman.
Ramm, Wiebke. Forthcoming. Discourse-structural salience from a cross-linguistic perspective:
Coordination and its contribution to discourse (structure). In Salience. Multidisciplinary
Perspectives on its Function in Discourse [Trends in Linguistics. Studies and Monographs
(TiLSM)], Michael Grabski et al. (eds). Berlin: Mouton de Gruyter.
 Editors’ introduction 

Ramm, Wiebke & Fabricius-Hansen, Cathrine. 2005. Coordination and discourse-structural

salience from a cross-linguistic perspective. In Salience in Discourse: Multidisciplinary Ap­
proaches to Discourse 2005, Manfred Stede, Christian Chiarcos, Michael Grabski & Luuk
Lagerwerf (eds), 119–128. Münster: Stichting/Nodus.
Redder, Angelika. Forthcoming. Konjunktor. In Handbuch der deutschen Wortarten, Ludger
Hoffmann (ed.), article C12. Berlin: de Gruyter.
Reich, Ingo. Forthcoming. “Asymmetrische Koordination” im Deutschen. Habilitationsschrift.
University of Tübingen 2007. To appear in [Studien zur deutschen Grammatik]. Stauffen-
burg: Tübingen.
Reis, Marga. 1993. Satzfügung und kommunikative Gewichtung: Zur Grammatik und Prag-
matik von Neben- vs. Unterordnung am Beispiel ‘implikativer’ und-Konstruktionen im
Deutschen. In Wortstellung und Informationsstruktur [Linguistische Arbeiten 306], Marga
Reis (ed.), 203–249. Tübingen: Niemeyer.
Reis, Marga. 1997. Zum syntaktischen Status unselbständiger Verbzweit-Sätze. In Syntax
im Fokus. Festschrift für Heinz Vater, Christa Dürscheid, Karl Heinz Ramers & Monika
Schwarz (eds). Tübingen: Niemeyer.
Reis, Marga. 1999. On sentence types in German: An enquiry into the relationship between
grammar and pragmatics. Interdisciplinary Journal for Germanic Linguistics and Semiotic
Analysis 4: 195–236.
Schwabe, Kerstin & Zhang, Ning. 2000. Ellipsis in Conjunction [Linguistische Arbeiten 418].
Tübingen: Niemeyer.
Schecker, Michael. 2000. Zur kommunikativ-funktionalen Leistung von Nebensätzen. In Subor­
dination in Syntax, Semantik und Textlinguistik, Michel Lefèvre (ed.), 115–122. Tübingen:
Stauffenburg.
Shopen, Timothy (ed.). 1985. Language Typology and Syntactic Description II. Cambridge: CUP.
Sperber, Dan & Wilson, Deirdre. 1986. Relevance: Communication and Cognition. Oxford:
Blackwell. (2nd edn, with postface, 1995).
Stede, Manfred. 2004. Does discourse processing need discourse topics? Theoretical Linguistics
30: 241–254.
Stutterheim, Christiane von. 1997. Einige Prinzipien der Textproduktion: Empirische Untersu-
chungen zur Produktion mündlicher Texte [Germanistische Linguistik 184]. Tübingen: Nie-
meyer.
Sæbø, Kjell Johan. Forthcoming. Adverbial clauses. In Semantics: An International Handbook of
Natural Language Meaning [Handbücher zur Sprach- und Kommunikationswissenschaft/
Handbooks of Linguistics and Communication Science], Claudia Maienborn, Klaus von
Heusinger & Paul Portner (eds), Ch. 62. Berlin: Mouton de Gruyter
Tomlin, Russell S. 1985. Foreground-background information and the syntax of subordination.
Text 5: 85–122.
Txurruka, Isabel G. 2000. The semantics of  ‘and’ in discourse. Technical Report ILCLI-00–
LIC-9, University of the Basque Country.
Velde, John te. 2000. On the structure of coordination. In Ellipsis in Conjunction[Linguistische
Arbeiten 418], Kerstin Schwabe & Ning Zhang (eds), 51–78. Tübingen: Niemeyer.
Verstraete, Jean-Christophe. 2007. Rethinking the Coordinate-Subordinate Dichotomy. Interper­
sonal Grammar and the Analysis of Adverbial Clauses in English [Topics in English Linguis-
tics 55]. Berlin: Mouton de Gruyter.
Vieu, Laure & Prévot, Laurent. 2004. Background in SDRT. In Proceedings of the conference
Traitement Automatique du Langage Naturel, 485–494. Fez, Morocco.
 Cathrine Fabricius-Hansen & Wiebke Ramm

Vuillaume, Marcel. 2000. Asymmetrische Koordination in subordinierten Strukturen. In Sub­

ordination in Syntax, Semantik und Textlinguistik, Michel Lefèvre (ed.), 45–54. Tübingen:
Stauffenburg Verlag.
Zeevat, Henk. Forthcoming. Discourse structure in optimality theoretic pragmatics. In Struc­
turing Information in Discourse: The Explicit/Implicit Dimension [Special issue of Linguis-
tics], Cathrine Fabricius-Hansen & Bergljot Behrens (eds).
Zifonun, Gisela, Hoffmann, Ludger, Strecker, Bruno et al. 1997. Grammatik der deutschen Sprache
Bd. 1–3 [Schriften des Instituts für Deutsche Sprache 7]. Berlin: Walter de Gruyter.
part i

General and theoretical issues

 RST revisited
Disentangling nuclearity1

Manfred Stede
University of Potsdam

The paper discusses the notion of nuclearity as put forward by Rhetorical

Structure Theory as a general principle of text organization. On the basis of an
inquiry into different kinds of salience phenomena in texts, several problems
with the purported role of nuclearity are identified. It is argued that RST trees
conflate too much information from different realms of description in a single
structure. As an alternative to a more detailed investigation of coherence
phenomena, an approach toward multi-level analysis and annotation of text
is outlined, which also keeps the various sources of salience apart.

Keywords: RST, nuclearity, coherence, multi-level analysis

1.  Introduction

Among the various theories of discourse structure, Rhetorical Structure Theory (RST,
Mann & Thompson 1988) can probably claim to be the most empirical one: It was
developed at the time on the basis of thorough analyses of a variety of texts, and it
has since been used for quite different purposes and for different types of text. To
characterize it very briefly,2 the idea of RST is to postulate some 20 coherence rela-
tions, defined in terms of the effect they are meant to cause in the reader’s beliefs and
attitudes, and to claim that a coherent text can be analysed in terms of these relations.
This would lead to a tree structure that recursively connects all the “minimal units”
and the resulting larger text spans. Following popular practice in later publications, we
henceforth call the minimal units EDUs, for “elementary discourse units”. Mann and
Thompson acknowledge that coherence relations are “pre-realizational” in the sense

.  For their very helpful comments on earlier versions of this paper, I am grateful to Bonnie
Webber, John Bateman and two anonymous reviewers.

.  A good general introduction to RST, also discussing other work that has been inspired by
the theory, is given by Taboada & Mann (2006). See also the website www.sfu.ca/rst.
 Manfred Stede

that they connect mental representations of text spans – but for the purposes of text
analysis one regards them as connecting actual portions of surface text. As for the
syntactic shape of the EDUs, the authors deliberately do not go into much detail but
characterize them as being typically clauses. Mann and Matthiessen (1991: 234) elabo-
rate on this decision:
The units of an RST analysis are chosen to fit the purposes of the analysis,
and are not theoretically prescribed. Our usual practice in analyzing a text is to
regard clauses as the realizations of units, but to merge restrictive relative and
complement clauses with their parents and to treat elliptical clauses as if they
were non-elliptical.

In the following, we use “text segments” and “text spans” interchangeably as referring
either to EDUs or to larger compositions of them.
A relation may only join adjacent text spans, and any span can only have one “par-
ent node” in the tree. In the relation set postulated by Mann and Thompson (1988),
21 of the 24 coherence relations join a nucleus and a satellite segment, where the latter
is less “important” (to be clarified in the next section). The remaining three relations
are multinuclear, i.e., they join segments of equal importance. An important claim
is another partitioning of relations into two groups: Instances of presentational rela-
tions are supposed to change the reader’s mind in some way, such as motivate her to
take a particular action, enable her to perform an action, or encourage her to believe
a certain proposition. Subject-matter relations, on the other hand, describe kinds of
connections between events happening in the world; here, the effect is merely that the
reader “recognizes” the relationship between the two units. As an example, consider
the following definition of the (presentational) relation Evidence:
– Constraints on nucleus (N): Reader might not believe N to a degree satisfactory to
writer
– Constraints on satellite (S): Reader believes S or will find it credible
– Constraints on the N+S combination: Reader’s comprehending S increases Reader’s
belief of N
– The effect: Reader’s belief of N is increased
– Locus of the effect: N

The empirical foundation was the main reason for RST’s considerable popularity
among researchers concerned with authentic discourse. And yet, the very fact that the
same rhetorical relations can be applied to so many different pairs of text spans im-
plies the danger that definitions become so vague that at many points in a text analysis,
several relations can be assumed to hold, which equally well conform to the defini-
tions. Ambiguity is not a priori problematic for text analysis – it is not surprising that
portions of text and their relationships can have different readings – but it should be
possible for the assumed representations to take notice of such genuine ambiguities
 RST revisited 

rather than to hide them under the covers of a single tree representation. In earlier
work (Grote et al. 1997; Stede 2004b), colleagues and myself argued that both the
Concession and the Contrast relation in RST are victims of the tendency toward too
unspecific definitions; they conflate a variety of different ways in which two states of
affairs can be contrasted with one another. In this paper, I try to make essentially the
same point for RST’s notion of nuclearity. Mann and Thompson, in a nutshell, char-
acterized the nucleus of a relation as being “more central to the writer’s purposes”, and
my point will be that being central to one’s purposes can be a very different thing from
case to case, and that it is worth distinguishing certain notions that occasionally give
rise to conflicting predictions on nuclearity. As a consequence, it will be questioned
whether it is helpful to assume a single hierarchical structure as an account of text
coherence.
The paper proceeds as follows. In order to do justice to the original proposals on
nuclearity by Mann, Matthiessen and Thompson in the late 1980s, their position will
first be summarized in Section 2. Also, some interesting extensions proposed by other
authors will be mentioned there. I then turn to a more general discussion of “salience”
in text and investigate different ways in which certain segments can be more impor-
tant/prominent/salient than others (Section 3). This will be my starting point for a
critique of RST-nuclearity in Section 4, which will be followed in Section 5 by a pro-
posal for a multi-level approach to discourse structure, which confines different types
of information to distinct layers.

2.  Nuclearity in RST

2.1  The original idea: Mann, Matthiessen, and Thompson

Towards the end of the most-cited original RST paper, Mann & Thompson (1988)
devote a section to clarifying their notion of nuclearity, which they in the previous
sections had used more on an intuitive basis when defining the various coherence
relations. The trigger for proposing nuclearity as a general principle of text organiza-
tion was the observation that all but three (Sequence, Contrast, and Joint) of their
coherence relations were quite obviously asymmetric. A clear example is Evidence:
When A is evidence for B, then B cannot be evidence for A. In carving out this
asymmetry for all the relations, Mann and Thompson noticed three commonalities,
which for them warranted the general division into nucleus and satellite: (i) Often,
one segment depends on the other for the discourse to be comprehensible – for
Evidence, the nuclear claim is necessary for the satellite to be understood correctly;
without the nucleus, it would be a non-sequitur. Consider the following example,
where the relation holds between nucleus (a) and satellite (b), and (a) is more central
for the text function than (b):
 Manfred Stede

(1) a. Tomorrow the weather will be nice.

b. I listened to the latest forecast on the radio.
c. So we can go to the picnic in the afternoon.

(ii) Often, the satellite is prone to substitution – for Evidence, one might replace the
satellite with a different piece of evidence, without changing the overall purpose of
the text. (iii) Often, the nucleus is more essential to the writer’s purposes than the
satellite.
The much-cited criterion for making the distinction between nuclei and satellites is
Mann and Thompson’s deletion test. The first argument concerns the overall text func-
tion: We can determine the “most nuclear” minimal unit of a text by tracing down from
the root node to the nucleus at each level all the way to the minimal units. For illustra-
tion, see Figure 1,3 which gives the RST analysis of a short text that extends example
(1). Nuclei are attached with straight lines to their mother nodes, whereas satellites are
linked to their nuclei with curved lines. Moving down from the root node (1–6) to a leaf
node (i.e., an EDU) without ever following a satellite branch leads to unit (3), the “most
nuclear” one. The deletion test predicts that when removing the most nuclear unit, the
overall message of the text typically becomes quite difficult, if not impossible, to infer.
This diagnostic is then extended to all minimal units that serve as nuclei in the tree (in
Figure 1, these are 1, 3, 4, and 5): When they are removed from the text, the result is an
incoherent sequence of utterances. On the other hand, with removal of those minimal
units that have been assigned satellite status (2 and 6 in Figure 1), the remainders (even
though lacking some cohesion) still convey an idea of what the text is about and often
even manage to acceptably communicate the main idea.

1–6
Volitional-result

1–2 3–6
Evidence Elaboration

Tomorrow the I listened to the So we can go to 4–6

Evaluation
weather will be latest forecast on the picnic in the
nice. the radio. afternoon.
4–5 That leaves plenty
Sequence of time for beer
and sausages!
We can take the and come back by
bus at 1pm 6pm.

Figure 1.  RST analysis of a sample text.

As for the relationship between nuclearity and linguistic structure, Mann and Thomp-
son mention the idea that the nucleus/satellite distinction directly corresponds to
the syntactic phenomenon of hypotactic clause combining: “Grammars in many

.  The figure was created with the RSTTool software (http://www.wagsoft.com/RSTTool).
 RST revisited 

languages draw a distinction between hypotactic and main clauses because of the
nucleus-satellite distinction in discourse” (p. 269). We do not elaborate here on the
even stronger proposition conveyed in that quotation – that nuclearity might be re-
sponsible for grammars having evolved hypotactic clause combining; instead we will
later discuss merely the (weaker) correlation that is proposed to hold between gram-
mar and discourse structure. The claim is expanded in more detail by Matthiessen
and Thompson (1988), who argue for a direct mapping between subordinate clauses
and satellites of coherence relations.
Mann and Thompson acknowledge two cases where the nucleus/satellite distinc-
tion is rather meaningless: in “enveloping structures” such as conventionalized begin-
nings and endings of letters, and in parallel structures as discussed by Fries (1981). His
examples involve repeated comparisons and contrasts between the same entities. The
shape of such structures comparing A and B can be sketched in a simplified way as fol-
lows: While A is X1, B is rather Y1. A has a X1, and B has a X2. X3 is highly relevant for
A, whereas X4 is more important for B.
Regarding the functional interpretation of nuclearity, Mann and Thompson make
another distinction between two groups of relations, based on their notion of locus of
effect of the relation, which is part of each definition and states the segment that the
described effect should arise from:4

– When the locus of effect is only the nucleus (as in Evidence), nuclearity rep-
resents the qualitative difference in role between the essential and the ines-
sential, thought and afterthought. The satellite supports the nucleus but does
not contribute to it, and the writer intends the reader to notice the distinction
introduced by nuclearity.
– When the locus of effect is both nucleus and satellite (as in Condition or Elabo-
ration), the structural difference between nucleus and satellite represents some
distinction in the organization of the subject matter. The distinction is presented
as important to the reader, and the significance of the satellite tends to be found in
the nucleus – both nucleus and satellite contribute to the result.

The role of locus of effect is discussed in more detail in a less-cited paper by Mann &
Matthiessen (1991). Interestingly, here the authors propose a more balanced view of
nucleus and satellite in subject-matter relations, for which “it seems quite reasonable
that the locus of effect should be the nucleus plus the satellite. The purpose of such
relations is to represent the kind of connection prevailing between the nucleus and the
satellite, not just to achieve the function of the nucleus” (Mann & Matthiessen 1988:

.  This is a part of the definitions that in fact did not always receive enough attention by other
authors referring to RST relations and applying them to sample texts. One reason might be that
the “locus of effect” field is not present in the definitions that were chosen for the RST website
(www.sfu.ca/rst).
 Manfred Stede

244). Mann and Matthiessen proceed to state that the division of relations according
to locus of effect and that of presentational versus subject-matter orientation is in fact
identical. This minimally diverges from the definitions given in Mann and Thomp-
son (1988), where Concession is in conflict with this view – maybe not surprisingly,
because Concession can indeed be seen as a somewhat problematic presentational
relation (Grote et al. 1997).

2.2  Nuclei on the run: extensions by other authors

The role of nuclearity for discourse structure has been strengthened further by Marcu
(2000) who proposed the “strong nuclearity hypothesis”: When a relation is postulat-
ed to hold between two spans of text, then it should also hold between the nuclei of
these two spans. This move ensures the “upward compatibility” of the idea of nucle-
arity, from minimal units to arbitrary text spans. At the same time, it generates new
constraints for nuclearity decisions at the lower levels: The annotator is encouraged to
assign nuclearity status to EDUs in such a way that the combinations of larger units
work out correctly. Thus, if in doubt about assigning a relation with a particular nu-
cleus/satellite distribution on a lower level, considering the wider context will often
assist in making the decision. Mann and Thompson had argued their case on the basis
of small examples with elementary units only; the upward extension was implicitly al-
luded to but not explicitly proposed. Marcu took this step and supported it, inter alia,
with the argument that the application of automatic text summarization would benefit
from it: The nuclear portions of a text are assumed to be the more important ones that
should become part of an (extractive) summary of the text. In Marcu (1999), he sug-
gests to turn the notion of nuclearity into a scalar one when considering complete texts;
the measure for assigning a degree of nuclearity to an EDU considers to what extent it
is embedded in larger nuclei and satellites, which results in a partial ordering of all the
units according to their “importance”. For example, the aforementioned “most nuclear”
EDU(s) will have only nuclear segments along the path from the EDU to the root node
of the tree; others also appear in a number of satellites, determining their degree. For
our sample tree in Figure 1, this measure would assign the same degree to EDUs 1, 4
and 5, as they all involve one satellite connection when moving up to the root node.
Another notable promotion of the role of nuclearity was suggested by Cristea et
al. (1998). In their “veins theory”, they argue that antecedents to anaphoric expres-
sions are more likely to be found in nuclear material of the preceding context rather
than in satellite material. This appears to echo the distinction between subordinating
and coordinating coherence relations in Segmented Discourse Representation The-
ory (Asher & Lascarides 2003), which to some extent was motivated by constraining
anaphoric accessibility – an idea that probably originated with Polanyi’s (1988) “right-
frontier constraint”. While it had originated in this fashion as a linguistically-motivated
criterion, SDRT in more recent work seems to regard it as a close relative of nuclearity in
RST (cf. Danlos 2008).
 RST revisited 

An interesting move was made in the annotation guidelines for the RST Treebank
(Carlson et al. 2003). The number of coherence relations used in that corpus is 78, in-
cluding 25 multinuclear ones. This amounts to 32% of all relations, compared to 12.5%
in the original Mann/Thompson relation set (where only 3 out of 24 relations are
multinuclear). 17 of those 25 multinuclear relations have mononuclear counterparts,
i.e., relations with identical definitions except for the assignment of one or two nuclei,
respectively. In the RST Treebank, annotators made equal use of both versions: They
chose the multinuclear version of a relation 1,905 times (51.4%) and the mononuclear
one 1,801 times (48.6%).5 Moreover, according to the guidelines, 8 of the mononu-
clear relations occur in both variants of nucleus/satellite distribution (as Mann and
Thompson had proposed merely for the Cause and Result relations). This indicates
that Carlson et al. were acutely aware of the problem that annotators often experience
when asked to decide which of two segments should be labelled nuclear. Carlson et al.’s
overall solution is to work with a much larger set of coherence relations; to my mind,
this hides the fact that coherence arises from distinct sources, which also should be
represented separately, and that nuclearity as a purportedly “general” phenomenon is
largely orthogonal to the description of the specific coherence relations.

2.3  Other concepts of coherence relations

When considering other proposals that explain text coherence in terms of relations,
it turns out that not many of them include a notion that corresponds to nuclearity
as discussed in RST. None of the works of Hobbs (1979), Sanders et al. (1992), or Kehler
(2002) systematically assign different prominence to the segments of relations. On the
other hand, Bateman & Rondhuis (1997), who more explicitly draw on RST, propose
to regard nuclearity not as a static aspect of the relation definitions, but as a distinct
feature that can combine with a set of other features to dynamically create a coherence
relation (similar to the decompositional approach of Sanders et al. (1992)). That is,
they transfer the idea underlying the Cause/Result relation pair of Mann and Thomp-
son to the entire set of relations, so that any relation can assign nucleus status to either
span.6 This amounts to a more radical solution to the above-mentioned problem noted
by Carlson et al. (2003), whose response was to partition the overall relation set ac-
cording to the distribution of nuclearity.
Another approach that borrowed nuclearity from RST is “relational discourse anal-
ysis” (Moser & Moore 1995), which aims to merge RST with the approach of Grosz &

.  These numbers of annotations in the treebank were provided by Bonnie Webber (personal
communication).

.  Similarly, a recent proposal of a relation set by Golebiowski (2006) states that the majority of
her relations “are basically neutral, and thus able to function both paratactically and hypotacti-
cally” (p. 261).
 Manfred Stede

Sidner (1986). These authors use the terms core and contributor for nucleus and satellite,
respectively. A recent approach is that of Wolf & Gibson (2005) who draw a distinction
between directed and undirected relations, and while they mention RST as related work,
they do not seem to regard their directionality as a general principle of text organiza-
tion. Rather, they treat it as one feature of the semantics of the individual relations.
The same holds for the more surface-oriented conjunctive relations proposed by Martin
(1992), which will be mentioned again in Section 5.
In summary, the idea of nuclearity as a principle of discourse organization based
on “centrality for the writer’s purposes” is largely confined to RST and approaches that
directly draw on it; a range of other theories of discourse coherence do not postulate
an equivalent notion.

3.  Salience in text

Having stated the RST position on nuclearity, we now turn to a somewhat broader
viewpoint and examine different kinds of relative salience of segments in text – where
salience is now meant as a neutral cover term for a variety of phenomena to be investi-
gated. It is beyond doubt that a reader of a text perceives different portions as more or
less salient; the crucial question is whether this is due to a single, underlying notion –
nuclearity – that should be postulated as an elementary principle of text organization.
In this section, I will challenge this idea with observations that resulted from our anno-
tation work with the Potsdam Commentary Corpus (Stede 2004), a collection of German
newspaper commentaries, 175 of which have been annotated with RST trees by students
trained on the basis of (Mann & Thompson 1988). From time to time, we asked the an-
notators on what grounds they had made their decision of nucleus assignment between
adjacent text spans. After adding some of our own observations we get the following list
(a–j) of reasons that annotators can put forward when labelling a text segment as a nucleus.
To illustrate the phenomena, in several cases we refer to a sample text, the beginning of the
introductory section of a research paper, shown in Figure 2. Sentences are numbered, but
note that most sentences contain more than one “minimal unit” of an RST-style analysis.

a. One segment turns out to be more supportive of the text’s overall purpose.
Constructed example:
(2) a. The Labour Party supported the strike.
b. The Liberals strongly voted against it.
c. Thus we have seen another example of the weakness of the Labour.

Considering (a) and (b) in isolation, none of the two appears to deserve nuclearity
status more than the other. But (c) indicates that the text is meant to encourage readers
to support the position of the Liberals, thus rendering (b) nuclear.
b. One segment supports the intention of the other segment. This is the “classical”
situation of RST relations such as Evidence or – in this example – Enablement:
 RST revisited 

(1) It is widely believed that the best human tutors are more effective than the best
computer tutors, (…). (2) A major difference between human and computer tutors is
that human tutors use face-to-face spoken natural language dialogue, whereas com-
puter tutors typically use menu-based interaction or typed natural language dialogue.
(3) This raises the question of whether making the interaction more natural, such as
by changing the modality of the computer tutoring to spoken natural language dia-
logue, would decrease the advantage of human tutoring over computer tutoring.
(4) In fact, as will be detailed below, several potential benefits of spoken tuto-
rial dialogue with respect to increasing learning have already been hypothesized in
the literature. (5) One hypothesis is that spoken dialogue may be better at eliciting
student behaviors that are believed to accelerate learning, such as student knowl-
edge construction. (6) A second hypothesis is that speech allows tutors to infer a
more accurate student model, which similarly is believed to accelerate learning. (7)
A third hypothesis is that speech primes a more social interpretation of the tutorial
environment, which again is hypothesized to accelerate learning.
(8) It is thus important to test whether a move to spoken dialogues is likely to
yield increased benefits with respect to learning and other performance measures.
(9) Furthermore, if the addition of speech can indeed increase learning gains, it is
also important to understand why spoken dialogue accelerates learning.

Figure 2.  Excerpt from D. Litman et al.: “Spoken versus typed dialogue tutoring.” Int’l Journal
of Artificial Intelligence in Education 16(2), 2006, p. 146. Sentence numbers inserted by MS.

(3) Open the printer cover carefully. The lever is located at the lower right of the back-
side.
In the sample text in Figure 2, sentence 8 states the main question the paper is set to an-
swer, and 4–7 are meant to support the claim that the question is actually a relevant one.
c. Recurrence: Material in one of the two spans is taken up later in the text, and
hence appears to be more important for the development of the text:
(4) a. Jim was opening a wine bottle when
b. his mother rushed in.
c. She held a letter in her hands.
Choosing a relation joining (a) and (b) can amount to the decision which proposi-
tion is meant to be in the foreground and which in the background; since the mother
­appearing in (b) is also the subject of (c), this local configuration suggests that (b) is
more nuclear than (a).
d. Repetition: When longer material is repeated, as is the case in both the second
clauses of sentences 6 and 7 of the sample text, it does not provide new information
but merely reminds the hearer, which renders it less nuclear than a potentially more
informative adjacent unit (cf. the deletion test mentioned in Section 2).
 Manfred Stede

e. Digression: When a text segment constitutes a temporary digression from the

main topic of the text, it is likely to be only a satellite, which is also supported by the
deletion test.
f. Meta-discursive elements: A segment can appear to be decidedly non-nuclear
when it is a text-internal directive. An example is the clause as will be detailed below in
sentence 4 of the sample text.
g. Connective: Sometimes, the connective chosen by the author can mark the nucle-
us/satellite distinction, as with the German zwar (sat) … aber (nuc) (‘true … but’), or
nicht nur (sat) … sondern (nuc).” Example:

(5) Der Bürgermeister hat zwar keinen Erfolg beim

the mayor has true no success with.the
Bürokratieabbau, aber er ist ein eifriger
bureaucracyreduction but he is an eager
Wirtschaftsförderer
economysupporter
‘The mayor has no success in cutting red tape, but he is a vivid supporter of the
local economy.’
Similarly, punctuation marks can indicate parenthetical, and hence satellite,
information.
h. Other lexical marking of salience: Authors can use lexical items to explicitly as-
sign “weight” to utterances, as the phrase it is important to does in sentences 8 and 9 of
the sample text.
i. Syntactic structure: A main clause often seems more nuclear than an associated
subordinate clause, as mentioned in Section 2.
(6) While the speaker struggled through his manuscript, the audience gradually
turned to sleep.
Several examples can also be found in the sample text, where subordination correlates
with other criteria, e.g., with (f) in sentence 4.
j. RST definition: The relation is stated by the RST definition to assign nuclearity in a
specific way. For instance, Sentence 9 of the sample text expresses a Condition relation,
which according to Mann and Thompson (1988) consists of a condition-satellite and
a consequence-nucleus. Similarly, Carlson et al. (2003) had to fix the nucleus assign-
ment for the relation Attribution used in the RST Treebank. It relates a statement to its
source, as in The spokesman reported that the company lost two important contracts last
year. Or consider sentence 1 of the sample text, where the source is not an individual
but “public” opinion. Rather than having annotators decide whether the source’s act
of reporting or the content of the report is more important, Carlson et al. define the
content portion of such sentences generally to be the nucleus.
Notice that these factors, except for (j), are not meant to be hard-and-fast rules for
making the nuclearity decision; rather, they are evidential factors. In a suitable context,
 RST revisited 

any one of them could be cited by an annotator as the decisive factor for labelling one
of the adjacent text segments as nuclear. (a)–(j) are obviously not mutually exclusive,
but – and that is the point here – they are largely independent of each other: Any one
of them alone can in some specific context be taken as the decisive evidence.
The observation that the factors are not mutually exclusive suggests that they op-
erate on different levels of description. (a) and (b) are intention-based criteria: When
the intention of one segment “dominates” that of the adjacent one, or when the seg-
ment intention supports the overall text intention, we have evidence for nuclearity.
(c)–(f) have, in different ways, to do with the thematic development of the text. In (c),
co-reference and topic-continuity trigger the decision on the level of minimal units,
and in (e) topic (dis-)continuity, among minimal or larger units, is responsible. Re-
peating information in (d) and supplying text-organizing meta-information in (f) can
be seen as ways of interrupting the “main” flow of information. Criteria (g)–(i) are the
most surface-oriented ones, and they are likely to co-occur with some of the others.
Underneath the surface, (h) again is related to intentions: The author, for whatever
reason, chooses to emphasize some portion of the text. In contrast to (a) and (b), how-
ever, this is not relational but a feature of a single text unit – the reader shall recognize
that the author regards the contents of this unit as central for her purposes. This might
co-occur with (a) – the unit can also support the intention of a larger segment – but it
need not necessarily do so. Finally, (j) is of an entirely different kind: It is an inherent
necessity postulated by the underlying theory to assign nuclearity in one particular
way once a specific relation has been found applicable.
The fact that such a multitude of factors can influence nuclearity assignment is not
per se problematic. The question is, however, whether those distinct factors indeed are
reflexes of a common underlying notion, a “general principle of text organization”. In
the next section I will argue that we have reason to question this.

4.  Problems with nuclearity in RST

We now proceed to synthesize the exposition of RST-nuclearity and our inventory of

types of salience into a critique and then to an outline of a different approach towards in-
vestigating discourse phenomena. To that end, this section critically reviews three central
aspects of the RST view of nuclearity: its relationship to syntactic subordination, its direct
association with coherence relations, and the purported pervasiveness of nuclearity.

4.1  Syntactic subordination and nuclearity

Undoubtedly, the idea to link syntactic subordination to discourse nuclearity has in-
tuitive appeal, and moreover it is not difficult to accumulate evidence in support of
such a link: Quite often, a syntactically subordinate clause is in fact “less central to the
writer’s purposes” than the matrix clause is. But on the other hand, one can also gather
 Manfred Stede

evidence for the opposite position – the writer’s purposes running against syntactic
subordination, which altogether suggests that matters are more complicated, and that
the view accentuated especially in Matthiessen & Thompson (1988) generalizes too
boldly from a mere tendency to a principle (see Blühdorn, this volume). For one thing,
we have pointed out in the previous section that in the absence of any over-arching
intentional structure, it is the continuation of discourse – in terms of referential and/
or thematic continuity – which often governs nuclearity assignment. If a discourse
referent retains prominence in the subsequent text, the clause containing its original
introduction has a good reason to claim nuclearity status. This clause, however, might
quite well be a subordinate one. This point is also made by Golebiowski (2006: 262),
who uses the sentence While I was feeling really well on Saturday, I was ill on Sunday
to illustrate that in a suitable context, either both clauses can be of equal functional
prominence (7a) or one can be more important (7b):

(7) a. W
 hile I was feeling really well on Saturday, I was ill on Sunday. On Sat-
urday, I had a lot of energy and I (…) managed to have a beautiful meal
with my friends and even spent an enjoyable evening with my family. On
Sunday, however, I woke up with a splitting headache. In spite of taking
some pain relief tablets, my head did not clear up till the evening. I had to
cancel all arrangements made for the day and ended up spending most of
Sunday in bed.
b. While I was feeling really well on Saturday, I was ill on Sunday. I woke up
with a splitting headache. In spite of taking some pain relief tablets, my
head did not clear up till the evening. I had to cancel all arrangements
made for the day and ended up spending most of Sunday in bed.

It seems worthwhile to relate this point to a different issue raised in Mann and Thomp-
son (1988), that of a “standard ordering” of the spans of relations. The authors state
that many relations have a clear tendency as to the linear ordering of their nucleus und
satellite; for example, both Antithesis and Condition would normally have their satel-
lites occur before their nuclei. Notice that this statement in conjunction with frequent
linguistic realization of such relations (“although A, B” and “if A, then B”, respective-
ly) indeed lends support to the purported link between satellite (A) and subordinate
clause. But, of course, this “standard” ordering is by no means mandatory – “B, al-
though A” and “B, if A” are perfectly natural. In these cases, however, the subordinate A
segments (i.e., the satellites qua relation definition) due to their position at the end of
the sentence can play different roles for the continuation of the discourse; for the case
of temporal clauses, ordering has been investigated by de Swart (1999) and Schilder
and Tenbrink (2001).
In summary, for text portions where intentions provide the structural scaffolding,
subordination seems to usually mirror the relative weights of such intentions:
(8) While Smith’s approach to the free-will problem was not entirely wrong, the
general direction of his philosophy has been completely refuted by now.
 RST revisited 

But where intentions do not play the critical role,7 linear order takes over as an im-
portant force in assigning different degrees of prominence to text segments. Syntactic
subordination also plays a role here, but the interactions become complicated, and
RST does not offer the means to account for such effects.
In addition to the more difficult relationship between subordinate clauses and
nuclearity, it is neither the case that, conversely, co-ordinating clauses would naturally
lead to multinuclearity. It is well-known that in clauses joined by the contrastive con-
junction but the second clause typically carries more “weight” than the first one. An
illustrative example was given by Elhadad & McKeown (1990):
(9) a. He failed the exam, but he is smart. Let’s hire him.
b. ?He is smart, but he failed the exam. Let’s hire him.

4.2  Nuclearity tied to relations

For the seven presentational, or intention-based, coherence relations in RST, nuclear-
ity plays the role of “domination” between discourse segment purposes in the theory
of Grosz & Sidner (1986). RST generalized this notion to the 16 subject-matter rela-
tions as well and defined for each relation which elements would play the roles of
nucleus and satellite, respectively. The one exception is the set of causal relations,
where RST distinguishes Cause from Result, with the only difference being the map-
ping from cause/effect to nucleus/satellite. Mann and Thompson thus acknowledge
that describing a causal relation between two events does not inherently make either
cause or effect more salient than the other. But is such an inherent ascription war-
ranted for the other subject-matter relations? Given the list of factors contributing to
nuclearity decisions in Section 3, it is not clear that, for any instance of a relation such
as Interpretation or Evaluation, the interpreted or evaluated segment would consis-
tently be more important than the interpreting/evaluating one. Consider again our
example text in Figure 1. If the text were continued with a sentence like For me, eating
and drinking have always been the major attraction at picnics, the weight distribution
for the preceding three segments would change, because now it turns out that the
beer-and-sausage sentence is more relevant for the text function than the Sequence
segment giving the timing of the trip. Similarly, for the Interpretation relation, it is
not difficult to conceive contexts that render the interpretation more prominent than
what is interpreted.
With other relations, predefining the nuclearity assignment is equally problem-
atic. A case in point (noted also by Bateman and Rondhuis (1997)) is the Purpose

.  This does not mean that intentions can be entirely absent; it is to be understood in terms of
the distinction drawn between subject-matter and presentational relations by Mann and Thomp-
son (and also by other authors), as sketched at the very beginning of the paper.
 Manfred Stede

relation, where according to the RST definition the underlying goal of the activity is
the satellite and the activity itself is the nucleus. But, given a sentence like (10), why
would buying the sports car and impressing the girl-friend not be equally able to play
the role of nucleus?
(10) Jim bought a red sports car to impress his new girl-friend.

Adopting again the argument of continuation possibilities, both elements can easily be
picked up by a subsequent sentence – even more so when the linear order of the Purpose
sentence is in line with the continuation. Or, independent of a particular continuation,
either the sports car or the necessity of impressing the girl-friend can be the main topic of
the larger discourse unit, so that criterion (a) from Section 3 would favour it as nucleus. It
is evident that the elements of a Purpose relation are clearly distinguishable on semantic
grounds, but this should not entail a strict mapping to a nuclearity assignment.

(11) If I win the lottery next week, the first thing I’ll buy is a red sports car.

The same point can be made for the Condition relation. In (11), the two clauses clearly
express propositions of different epistemic status, and thus the relation is “asymmet-
ric” – but it is not clear in what sense the two clauses should be distinguished on the
grounds of a general notion of nuclearity. Example (9) above, at least, demonstrates a
difference in salience that is clearly not the same as the one found in such Condition
examples. And besides, the “continuation argument” applies to Condition just as well:
The following clause can either elaborate the chances of winning the lottery, or the
prospective purchase of a sports car (or neither one).
Mann and Thompson apparently saw these problems and pointed out that the
definitions of relations such as Condition and Purpose would specify as “locus of
effect” both the nucleus and the satellite, and therefore the deletion test would not
produce satisfactory results for them (which is correct). Resorting to a “locus of effect”,
however, brings a new parameter into play and in effect undermines the general nature
of the nuclearity idea. It seems more appropriate to question the strict association be-
tween (subject-matter) relations and nuclearity assignment altogether.
Finally, while Condition and Purpose are clearly marked at the linguistic surface
and thus do not create ambiguity for annotators, the situation is different with the
aforementioned Interpretation and Evaluation. We pointed out that their fixed nucleus
assignment can create conflicts in the annotation process. Notice however, that these
relations are much less “visible” at the surface and hence rely more on context-driven
interpretation. Unfortunately, when the definition of Interpretation or Evaluation is
met for a text span, chances are that other relation definitions are also met, e.g., the
notorious Background or the rather vague Circumstance. These two offer a way to turn
the nucleus assignment around, and thus annotators can select a “weaker” relation
rather than the specific Evaluation/Interpretation, merely in order to achieve a suitable
nucleus assignment. This might serve to get the overall discourse structure right (in
 RST revisited 

terms of degrees of nuclearity of segments; see above), but it is clearly not optimal for
describing local coherence.

4.3  Enforced nuclearity

One source of complaints from RST annotators is the requirement that nucleus assign-
ment is a “must” for every pair of segments, unless one wants to resort to the multi-
nuclear Joint relation, which is void of any specific content. On the level of minimal
units, such cases are encountered, for example, in descriptive texts that characterize
various aspects of a topic in an enumerative way. Among those aspects, two may be
causally related, but there need not be a consequence of the causality for the writer’s
intentions, nor for thematic continuity. The relationship can be marked by a causative
adverbial, leaving the surface structure “neutral”:8
(12) Th
 e courtyard was very quiet that day. Two teenagers read their books in one
­corner, and a new car was parked beside them. The trees still had all their leaves,
and thus the flowers were blooming in a cool shade. A little bench stood in the
middle of the lawn. (…)

Teenagers, car, trees, flowers and bench all contribute to the description. The marked
causal relation between the third and the fourth clause leaves no doubt as to the
“semantic directionality” from cause to effect, but from a textual perspective, it appears
to be an artificial move to grant either of the two clauses a nuclear status, for on what
grounds would it be more central to the writer’s purposes? Accordingly, an annotator
would struggle here to decide between RST’s Cause and Result relations. Recall that
the relation set used by Carlson et al. (2003) employs many more multinuclear rela-
tions than the original RST set, which indicates that those researchers encountered the
same problem (for our specific example, a multinuclear Cause-Consequence relation
is provided by Carlson et al.).
As indicated earlier, Mann & Thompson (1988) acknowledged that envelope
structures (conventionalized beginnings and endings) and certain parallel struc-
tures of comparison and contrast are to be exempt from the need to assign nuclei. My
sugges­tion here is that this list needs to be extended, which questions the explanatory
power of the nuclearity notion. Besides some problematic cases on the level of minimal
units as just illustrated, the problem is also prominent higher up in the tree structure,
between larger segments, where structural clues indicating a salience distribution
are often absent. The severity of this problem differs along the dimension of text type
(in the sense of, e.g., Werlich 1975), or what Smith (2003) calls discourse mode. For

.  In German, where a causative adverbial such as deshalb can float quite freely throughout the
clause, the impression of perceived “neutrality”, or the lack of a structure-induced weighting,
might be somewhat clearer than in English.
 Manfred Stede

argumentative text, the overall structure is often characterized quite well by the RST
presentational relations and their nuclearity assignments, which reflect the relative
weight and hierarchical structure of the points argued by the author. For narrative,
presentational relations are much less relevant, and all the problems with subject-
matter relations, as discussed in the previous section, become especially prominent.
A central aspect in understanding narrative is reconstructing the temporal structure
of reported events, which cannot be characterized sufficiently with RST relations. And
relative “centrality to writer’s purposes” is often not an issue, as illustrated in example
(12) above. A similar judgement applies to expository text, whose function is to explain
some (concrete or abstract) entity to the reader, by outlining its decomposition into
parts, its function, and the like. Here, the reader is typically invited to form a mental
image not of a temporal event sequence but of a spatial configuration of objects (even
if they are abstract). As with narrative, the linearity of the exposition is very impor-
tant for ease of processing, while internal structure according to “importance” is only
occasionally relevant; nuclearity thus does not explain nearly as much as it does in
argumentative text, where the linear order is much less critical.9

5.  Multi-level discourse representation and annotation

Researchers investigating the different text types (cf. the citations above) point out
that, on the one hand, the types can be distinguished by (relative frequencies of) indi-
vidual surface-linguistic features, and, on the other, coherence is created in different
ways. For example, both Lötscher (1987) and Smith (2003) characterize type-specific
strategies of thematic development. And we noted above that the overall function of a
text varies according to the type: convey the temporal structure of events; convince the
reader by presenting an argument; enable the reader to form a mental image of some,
possibly complicated, object or state of affairs, etc., and along with the text function
varies the shape of the functional description of the text, which is supposed to capture
the phenomena that make the reader experience the text as coherent. To some extent,
this can be explained in terms of RST relations: While argumentative text can be char-
acterized by presentational relations, expository text features much fewer instances of
those and instead is rich in Elaboration, as the text moves from (sub-) topic to (sub-)
topic. In narrative, we might expect an abundance of Sequence relations, but of course
the temporal structure can be much more complicated and require different means of

.  We conducted many re-ordering experiments with the texts in the Potsdam Commentary
Corpus, indicating that the text function is relatively independent of the linear order of the
­segments, as long as referential chains are not broken. Roughly speaking: In presenting your
argument, you have quite a bit of freedom in sequencing and arranging your points.
 RST revisited 

representation (see, e.g., Mani & Pustejovsky 2004). Thus, different text types favour
different means of describing text structure.
However, unlike genre, text type is not a category that texts typically belong to in
their entirety. While for most texts we can clearly state that it is a recipe, a news report,
an instruction manual, etc. (genre), it is relatively rare for a text to be an instance of
just one type. Narratives typically contain descriptive, expository or portions of other
types; a largely argumentative text may contain a short narration of events that are
then being commented on, and so forth. Therefore, a single type-specific description
will normally not be sufficient to cover a text. Instead, we need arrangements of de-
scriptive devices that are able to characterize the various dimensions of text coher-
ence simul­taneously. One undisputed facet of this notion is the long-recognized fact
that both relation-based and reference-based means play a role in creating coherence.
Then, taking the step from co-reference to the more abstract level of thematic develop-
ment, Knott et al. (2001) voiced their discomfort with RST’s mixing the more “stan-
dard” coherence relations with Elaboration (in particular, the object-attribute variant),
which does not convey a relationship between propositions, but instead indicates a
shifting focus of attention. Accordingly, Knott et al. dismissed that coherence relation
and introduced a different representational device, which leads to “entity chains”: Indi-
vidual portions of the text are captured with standard RST trees, and these portions are
linked by a focus shift indicator.
My proposal is to take a further step and to seek a text representation that distin-
guishes various dimensions of coherence from one another. Whereas Knott et al. use
a single representation that alternates between RST relations and topic shifts, it seems
more appropriate to consider the dimensions in parallel. After all, some referential
phenomena are bound to occur also within an RST-portion of an entity chain in the
style of Knott et al. (2001) – thematic development and RST-like relations are by no
means mutually exclusive. Similarly, temporal relations can easily overlap with non-
temporal ones:
(13) Tim’s performance at the piano was impeccable. When he had walked onto the
stage, people seemed sceptical, but as soon as he finished the sonata, the whole
audience burst into applause.

Understanding (13) involves both reconstructing the right order of events (on the basis
of verb tense and with the help of world knowledge) and grasping the contrast between
the “before” and “after” situations.
Motivated by related observations, Wolf and Gibson (2005) responded to the multi-
faceted nature of local coherence by dismissing RST’s assumption that text structure be
represented by a tree; instead, they allow coherence relations to quite freely connect non-
adjacent segments, which can also lead to cross-dependencies. Importantly, though, they
hold on to coherence relations as the single representational device, merely substitut-
ing a graph for the constrained tree structure. It seems to me that incrementally adding
more relations, as phenomena are being recognized, is not a promising direction. Rather,
 Manfred Stede

one should explicitly distinguish the various dimensions of discourse structure, so that,
for example, the phenomena listed in Section 3 can be attributed to their respective
realms. Incidentally, in one RST paper, Mann & Matthiessen (1991) point in this direc-
tion when they relate RST relations to the metafunctions of systemic-functional linguis-
tics10 (SFL, Halliday 2004), but they neglect the possibility of allowing different realms of
relationships to operate simultaneously, which is the underlying idea of characterizing
sentence structure in SFL. Bateman & Rondhuis (1997) took the step to decompose rhe-
torical relations systematically along the three metafunctions (interpersonal, ideational,
textual). A more cautious step in this direction had also been taken by Moore & Pollack
(1992) with their suggestion that text needs to be analyzed with subject-matter and pre-
sentational relations in parallel, which can possibly lead to two divergent tree structures.
However, the thesis that both types of relations always apply simultaneously appears to
be too strong, unless one includes in the relation set some very unspecific relations that
can be used in cases where just one of the two groups is clearly the appropriate one.
In order to make progress with defining an explanatory set of representation levels
that accounts for the relevant dimensions of coherence, I propose a data-oriented ap-
proach that emphasizes systematic annotation of authentic text, so that ideas can be
tested immediately with “real” data and be revised when necessary.

5.1  Annotation framework

Research on discourse (in particular: text) structure can be broadly divided into two
kinds of approaches:
– Work that aims at extending existing theories of sentence syntax and/or semantics
to the discourse level. Emphasis is on rigorous formalization and faithfulness to
the theories. Usually, such approaches work with constructed examples and are
gradually extended to cover more phenomena and, eventually, gain a certain cov-
erage of “real” data (corpora of actual language use).
– Work that aims at handling “real” data with as much coverage as possible, thus
emphasizing an empirical foundation. These accounts suffer from the fact that the
theories are much less formalized and definitions are often vague.
Representatives of the first approach are SDRT (Asher & Lascarides 2003), D-LTAG
(Webber et al. 2003), and LDM (Polanyi 1988). For the second approach, RST has prob-
ably been the most influential; recently, an alternative corpus-oriented approach has
been proposed by Wolf & Gibson (2005). In general, recent years have seen a surge of
interest in annotating text corpora with discourse-related information. One example is
the Penn Discourse TreeBank (Prasad et al. 2004) that builds on a popular Wall Street
Journal corpus and provides annotations of connectives and their arguments. For RST,

.  The authors correlate presentational relations with the interpersonal, and subject-matter
relations with the ideational metafunction; as for the textual metafunction, they vaguely indicate
that the order of text spans in a relation functions textually.
 RST revisited 

the RST Treebank (Carlson et al. 2003) has been made available, which also covers
Wall Street Journal text. These are important steps, because discourse research urgently
needs an impetus similar to that which sentence-related annotations (so-called sen-
tence treebanks) have given to syntax research. Such investment of annotation efforts
should be carefully reflected, though. Text is a highly multi-faceted means of transmit-
ting information (in the broadest sense), and our discussion of various salience phe-
nomena was meant to illustrate that. Annotations according to RST are a good starting
point for exploring coherence phenomena with “real” data, but they clearly have their
limitations, as the discussion in the preceding sections demonstrated. Building RST
trees is a very complex task that asks annotators to make a range of quite different deci-
sions (segmentation, nuclearity assignment, relation choice) whose individual results
are then amalgamated into a single representation. In our experience, annotators are
quite often unhappy after finishing a text analysis because of the many ambiguities
encountered along the way. In many cases, one could just as well have decided oth-
erwise, and the alternative result would have been equally defendable in terms of the
RST definitions. The trouble is that those choice points are all hidden in the tree, not
allowing the inspection of the various individual micro-decisions and for considering
possible consequences of alternative answers to them.
In order to make text annotations maximally useful for research purposes, it
therefore seems more helpful to separate the various phenomena from each other
and annotate them individually, in the shape of a systematic multi-level annotation
(henceforth MLA, see Stede 2007). These levels can then be “mined” for correlations
and inform the search for more comprehensive theories of discourse representation
and processing. Until recently, this was extremely cumbersome to do (if possible at
all), but now with the advances in XML-based software technology, creating and que-
rying MLA corpora is a realistic and fruitful research avenue. One possibility is the
architecture developed at Potsdam University, with a generic data exchange format
for diverse, task-specific annotation tools and a database that reads and aligns the in-
dividual annotations, and allows for manual cross-level analyses, as well as statistical
evaluations.11 Importantly, our MLA framework is flexible enough to allow for new
levels to be added incrementally to an existing corpus. In the following, I sketch one
possible configuration of annotation levels, which is at present being implemented for
the Potsdam Commentary Corpus (PCC, Stede 2004), and then I relate this to RST.

5.2  Levels of annotation

Sentence syntax. A representation of syntactic structure serves as building blocks for
the subsequent levels. In PCC, we use the TIGER schema (Brants et al. 2002), which is
designed to be as theory-neutral as possible. It includes a constituent structure as well

.  http://www.ling.uni-potsdam.de/~stede/MLA.html. Technical details on the database can

be found in Dipper et al. (2004).
 Manfred Stede

as information on grammatical functions. The dedicated annotation tool that allows

for efficient, semi-automatic construction of syntax trees is Annotate.12
Referential structure. Besides discourse relations, co-reference is generally taken
to be the most important source of coherence in text. For the time being, our annota-
tions are restricted to links between anaphors (pronouns and definite descriptions)
and nominal antecedents; both “bridging” relations and event anaphora are excluded.
The units related here – referring expressions – must be licensed by the level of senten-
tial syntax. Technically, the syntax trees are automatically mapped to the input format
for the annotation tool (MMAX2,13 specifically designed for co-reference) so that the
range of possible “markables” is already presented to the annotator.
Thematic structure. As a step of abstraction over the referential chains, we parti-
tion the text into zones that annotators identify as dealing with distinct topics. These
zones can be embedded, and they can bear subtopic relations to one another, which
the annotators assign. This builds on the idea of the “focus stack” used in the atten-
tional structure representation of Grosz and Sidner (1986). In addition, a specific label
“Formulaic” is reserved for meta-utterances whose topic is the text itself (e.g., the first
clause of sentence 4 of the text in Figure 1) or that contain other, genre-specific, con-
ventionalized formulae – in other words, for utterances that do not have a standard,
“content” topic. In future work, the level of thematic structure will be linked to our
work on annotating information structure in sentences (Dipper et al. 2007).
Conjunctive relations. Along the lines of Martin (1992), this level identifies the
“minimal units” of higher-level discourse structure, as well as the connectives linking
those units in the text. From the relations proposed by Martin, we chose 10 abstract
ones, and hence a set much smaller than that of RST, for labelling the type of con-
nection. Following Martin, this does not include any salience/nuclearity assignment,
but relations are in general directed to reflect the (semantically) different roles of the
units.14 The range of possible minimal units is not arbitrary but constrained by con-
stituents built on the sentential syntax level.15 For every connective (as defined by the
criteria in Pasch et al. (2003: 331), and adding certain prepositions), annotators can
mark its semantic scope (similar to the PDTB, Prasad et al. 2004), but they can also
leave one or both segments unspecified, if they find it hard to decide on the scope of an

.  http://www.coli.uni‑saarland.de/projects/sfb378/negra‑corpus/annotate.html

.  http://www.eml-research.de/english/research/nlp/download/mmax.php

���.  Bateman (2001) proposed a discourse representation combining RST and Conjunctive
­Relations, where the former is in charge of capturing abstract intentions and content, and the
latter serves to model local cohesion and thematic development; hence, Bateman’s goal is some-
what similar to ours.
.  See Dinesh et al. (2005) for a discussion of possible mismatches between the units of
­syntactic analysis and discourse structure.
 RST revisited 

adverbial connective such as so. Technically, we have implemented a dedicated annota-

tion tool that allows for marking connectives and scopes semi-automatically (Stede &
Heintze 2004).
Intentional structure. This most abstract level of representation largely corresponds
to the “presentational” relations of RST, with exceptions such as our treatment of Evalu-
ation (see below). Thus it is usually only a partial structure covering not the entire text
but only those portions where segments are indeed related via “dominance” to achieve
a particular intention. Marking this can involve introducing new segments (larger than
minimal units), which have not been found at the conjunctive-relation level because
there is no connective present. Technically, we use RSTTool for this step.16

5.3  Example
For the short text given in Figure 1, the annotation would proceed as follows. After the
syntactic analysis of the individual sentences, co-referential NPs are being identified,
which in this case leads only to chain I (sentence 2), we (3), we (4), elided we (5). The
event anaphor that (6) and the bridging connections (weather – forecast, afternoon –
1pm, picnic – beer and sausages) are at present not being represented. On the thematic
level, the text is divided into segments (1–2), dealing with weather, and (3–6), dealing
with picnic. The latter has a subtopic segment (4–5), timing. The assignment of conjunc-
tive relations is a flat, surface-oriented annotation, which here identifies the relations
Addition (1, 2), Consequence (1–2, 3), Temporal-successive (4, 5) and Addition (4–5,
6). Finally, the intentional structure corresponds in part to the presentational relations of
the RST tree: Evidence (2, 1) and Motivation (1–2, 3). As regards RST-Evaluation, notice
that in the example (and in many other cases), there is no clear “dominance” relation
between the evaluated and the evaluating segment. In this case, we would annotate 6 as
dominating (4–5), in contrast to the RST nuclearity assignment.

5.4  RST trees versus multi-level representation

For a short and “friendly” text such as the one in Figure 1, MLA might not seem par-
ticularly advantageous when compared to the RST tree, which serves as a compact
representation of the text structure and also indicates its function (by showing the
“most-nuclear unit”, here 3). But recall, for one thing, that MLA is richer in informa-
tion (syntax, co-reference), which can be used to analyse correlations between levels.
Also, the partial intentional structure of MLA conveys the most “important” segments:
the roots of the individual intention-based trees, here 3 and 6. The fact that these two
are disconnected seems to do justice to the text, as neither of the two speech acts sup-
ports the other.

.  http://www.wagsoft.com/RSTTool
 Manfred Stede

More importantly, the picture changes when we turn to the realistic case of longer
texts. In general, when comparing an RST analysis with a representation following
MLA, at first sight two kinds of information appear to be missing in the latter: the
subject-matter relations, and the existence of a complete tree structure. As for the
subject-matter relations, they are in part replaced by the conjunctive relations. These
are explicitly intended to be less specific than the RST relations, which have been criti-
cized for their essentially open-ended nature, for example by Grosz and Sidner (1986).
After all, the range of possible semantic relationships between events in the world or
within people’s minds is not likely to be small. From the discourse perspective, seek-
ing a “complete” inventory of relationships seems neither fruitful nor necessary: In
terms of discourse function, reporting a semantic relation between two eventualities
is not so different from reporting a single eventuality, so that a fine-grained inventory
of semantic relations would not correspond to an equally fine-grained inventory of
consequences for text coherence. In short, a dedicated inventory of subject-matter (or
semantic) relations seems to belong to the realm of domain knowledge but not to that
of text-oriented coherence relations.
The loss of a complete tree, which after all was supposed to be the main criterion
for coherence postulated by RST, might look like a rather dramatic move. However,
the completeness of an RST tree (or that of other discourse representation theories) is
often achieved with the help of coherence relations of a somewhat dubious status. RST
offers the multinuclear Joint relation for cases where a more specific connection can-
not be found; the main reason why Joint is relatively rarely used seems to be the wide
applicability of the Elaboration relation, which, as annotators report, fits in very many
situations. Elaboration, however, has the disadvantage of masking exactly what is being
elaborated on. Typically it is not a relationship between the propositions or eventu-
alities, but between individual entities taking part in those eventualities. As noted by
Knott et al. (2001), this amounts to a regular shift of focus of attention, which should
be represented as such – in our framework on the levels of referential structure (exactly
which entities are co-referent?) and thematic structure (does the overall topic change
or stay the same?). The latter has the advantage of explicitly modelling the flow of top-
ics through the discourse, which cannot be read off an RST tree. And recall that we
emphasized the need for representing focus shift and coherence relations in parallel (in
contrast to the suggestion by Knott et al.), because focus shifts occur throughout the
text, and not just in places where no intentional or subject-mater relation is present.
In the version of MLA sketched above, the hierarchy information encoded in
an RST tree is distributed to three levels: the partial intentional structure (where the
role of the hierarchy is largely in analogy to that of an RST tree), the flat segment an-
notation on the conjunctive-relations level, and the partitioning into content zones,
which allow for embedding. The claim here is that this tripartite representation is in
fact an advantage, as RST annotators regularly report that the precise assignment of
the single hierarchy is very difficult and time-consuming – and the result is typically
ambiguous anyway.
 RST revisited 

When we proceed to merge the three levels (which have been annotated individu-
ally) into a single data structure, the result is, in principle, a graph that in various ways
violates the constraints of a tree. There can be multiple roots, nodes can have multiple
parents, and there may be crossing edges (which is disallowed in RST). To what extent
these violations are either annotation artefacts, or indeed reflections of phenomena in
the linguistic structure, is subject to empirical investigation. From the corpus-oriented
perspective, we should decide on the most appropriate formal shape of text structure
not based on a priori considerations of elegance or processing efficiency, but based on
a comprehensive picture of the variety of structuring phenomena in discourse, given
analyses of authentic data.
We argued that the main goal of the multi-level approach is to separate distinct kinds
of information from one another, so that they can be individually annotated (resulting
in easier coding procedures) and reviewed, and ultimately set into correspondence with
one another. Obviously, the amount of work necessary to produce such annotations is
larger than with an RST tree, but the gain in useful information should outweigh that
additional effort. In other words, a smaller number of texts annotated with a multi-level
approach are likely to be more useful for discourse research than a larger number of
RST trees, which are prone to the problems indicated throughout this paper.

6.  Conclusion: Salience on different levels of description

The possible relationships between distinct levels of discourse representation had been
widely discussed in the computational linguistics community in the early 1990s; it had
been moved forward especially by research in text generation, where the successive map-
ping from deep representations of knowledge to the surface text must be made explicit.
This discussion ended without a clear result, as the attention shifted – in accordance
with the general trend in computational linguistics – to decidedly surface-based meth-
ods of discourse processing, then focusing on understanding rather than generation
(e.g., Marcu 2000). Today, it is opportune to re-open the discussion of levels and give
it a decidedly data-oriented twist: Developments in annotation and retrieval software
over the past years have made it possible to systematically collect text annotations on
different levels and relate them to one another, either manually, or by employing statis-
tical techniques of pattern recognition. Annotated corpora (in particular the so-called
treebanks) have tremendously helped sentence-oriented research over the past decade;
on the text level, this development has only started. It will, among other things, enable
us to investigate research hypotheses on level interactions that have been put forward,
for example, for rhetorical structure and constituent ordering, rhetorical structure and
co-reference, or discourse structure and sentential information structure.
Finally, we need to consider what happens to the idea of nuclearity in the pro-
posed MLA approach. To some extent, it is being preserved in its original flavour,
namely on the level of intentional structure. Here, it should also be possible to identify
 Manfred Stede

the central units of a text (with respect to its overall purpose), as it has been men-
tioned as an advantage of RST trees. In text portions where intention-based nuclearity
is absent, salience can arise from other levels. The referential chains indicate what are
the prominent entities under discussion in the text, and in which units they are being
mentioned; the thematic structure identifies meta-discursive statements or formulaic
portions, as well as temporary topic switches (digressions); and the syntactic structure
identifies subordinate clauses, which might generate a mood of non-salience in case
none of the other levels points to a different effect. Nuclearity thus becomes an epiphe-
nomenon resulting from different types of sources of salience and non-salience – but
it might just as well be absent in certain regions of the text, where segments are equally
salient to the author’s purposes.

References

Asher, Nicholas & Lascarides, Alex. 2003. Logics of Conversation. Cambridge: CUP.
Bateman, John. 2001. Between the leaves of rhetorical structure: static and dynamic aspects of
discourse organization. Verbum 23(1): 31–58.
Bateman, John & Rondhuis, Jan-Klaas. 1997. ‘Coherence relations’: Towards a general specifica-
tion. Discourse Processes 24: 3–49.
Brants, Sabine, Dipper, Stefanie, Hansen, Silvia, Lezius, Wolfgang & Smith, George. 2002. The
TIGER Treebank. In Proceedings of the Workshop on Treebanks and Linguistic Theories,
Sozopol, 24–41.
Carlson, Lynn, Marcu, Daniel & Okurowski, Mary Ellen. 2003. Building a discourse-tagged cor-
pus in the framework of rhetorical structure theory. In Current Directions in Discourse and
Dialogue, Jan van Kuppevelt & Ronnie Smith (eds), 85–112. Dordrecht: Kluwer.
Cristea, Dan, Ide, Nancy & Romary, Laurent. 1998. Veins Theory: A model of global discourse
cohesion and coherence. In Proceedings of COLING/ACL’98, Montréal, 281–285.
Danlos, Laurence. 2008. Strong generative capacity of RST, SDRT and dependency DAGs for
Discourse. To appear in Constraints in Discourse, Anton Benz & Peter Kühnlein (eds). Am-
sterdam: John Benjamins.
Dinesh, Nikhil, Lee, Alan, Miltsakaki, Eleni, Prasad, Rashmi, Joshi, Aravind & Webber, Bonnie.
2005. Attribution and the (Non-) Alignment of Syntactic and Discourse Arguments of Con-
nectives. In Proceedings of the ACL Workshop “Frontiers in Corpus Annotation”, 29–36. Ann
Arbor MI.
Dipper, Stefanie, Götze, Michael, Stede, Manfred & Wegst, Tillmann. 2004. ANNIS: A Linguistic
Database for Exploring Information Structure. In Working Papers of the SFB632, Interdisci-
plinary Studies on Information Structure (ISIS) 1, Shinichiro Ishihara, Michaela Schmitz &
Anne Schwarz (eds), 245–279. Potsdam: Universitätsverlag.
Dipper, Stefanie, Götze, Michael & Skopeteas, Stavros (eds). 2007. Information structure in cross-
linguistic corpora. Interdisciplinary Studies on Information Structure (ISIS) 9. Potsdam: Uni-
versitätsverlag.
Elhadad, Michael & McKeown, Kathy. 1990. Generating Connectives. In Proceedings of the 13th
International Conference on Computational Linguistics (COLING), 97–101.Helsinki.
 RST revisited 

Fries, Peter. 1981. On the status of theme in English: Arguments from discourse. Forum Linguis-
ticum VI(1): 1–38.
Golebiowski, Zofia. 2006. The distribution of discoursal salience in research papers: relational
hypotaxis and parataxis. Discourse Studies 8(2): 259–278.
Grosz, Barbara & Sidner, Candace. 1986. Attention, Intention and the Structure of Discourse.
Computational Linguistics 12(3):175–204.
Grote, Brigitte, Lenke, Nils & Stede, Manfred. 1997. Ma(r)king concessions in English and Ger-
man. Discourse Processes 24(1): 87–118.
Halliday, Michael. 2004. An introduction to functional grammar, 3rd edn. London: Arnold.
Hobbs, Jerry. 1979. Coherence and co-reference. Cognitive Science 1:67–82.
Kehler, Andrew. 2002. Coherence, Reference and the Theory of Grammar. Stanford CA: CSLI.
Knott, Alistair, Oberlander, Jon, O’Donnell, Michael & Mellish, Chris. 2001. Beyond elabora-
tion: The interaction of relations and focus in coherent text. In Text Representation: Linguis-
tic and Psycholinguistic Aspects, Ted Sanders, Jost Schilperoord & Wilbert Spooren (eds),
181–196. Amsterdam: John Benjamins.
Lötscher, Andreas. 1987. Text und Thema. Studien zur thematischen Konstituenz von Texten.
Tübingen: Niemeyer.
Mani, Inderjeet & Pustejovsky, James. 2004. Temporal discourse models for narrative structure.
In Proceedings of the ACL Workshop on Discourse Annotation, 57–64. Barcelona.
Mann, William & Matthiessen, Christian. 1991. Functions of language in two frameworks. Word
42(3): 231–249.
Mann, William & Thompson, Sandra. 1988. Rhetorical structure theory: Towards a functional
theory of text organization. Text 8 (3): 243–281.
Marcu, Daniel. 1999. Discourse trees are good indicators of importance in text. In Advances in
Automatic Text Summarization, Inderjeet Mani & Mark Maybury (eds), 123–136. Cam-
bridge MA: The MIT Press.
Marcu, Daniel. 2000. The Theory and Practice of Discourse Parsing and Summarization. Cam-
bridge MA: The MIT Press.
Martin, James. 1992. English Text – System and Structure. Amsterdam: John Benjamins.
Matthiessen, Christian & Thompson, Sandra. 1988. The structure of discourse and “subordina-
tion”. In Clause Combining in Discourse and Grammar, John Haiman & Sandra Thompson
(eds), 275–330. Amsterdam: John Benjamins.
Moore, Johanna & Pollack, Martha. 1992. A problem for RST: The need for multi-level discourse
analysis. Computational Linguistics 18(4): 537–544.
Moser, Megan & Moore, Johanna. 1995. Using discourse analysis and automatic text genera-
tion to study discourse cue usage. In AAAI Spring Symposium on Empirical Methods in
Discourse Interpretation and Generation, 92–98. Stanford University.
Pasch, Renate, Brauße, Ursula, Breindl, Eva & Waßner, Hermann Ulrich. 2003. Handbuch der
deutschen Konnektoren. Berlin: Walter de Gruyter.
Polanyi, Livia. 1988. A formal model of the structure of discourse. Journal of Pragmatics 12:
601–638.
Prasad, Rashmi, Miltsakaki, Eleni, Joshi, Aravind & Webber, Bonnie. 2004. Annotation and data
mining of the Penn Discourse TreeBank. In Proceedings of the ACL Workshop on Discourse
Annotation, 88–95. Barcelona.
Sanders, Ted, Spooren, Wilbert & Noordman, Leo. 1992. Towards a taxonomy of coherence rela-
tions. Discourse Processes 15:1–35.
 Manfred Stede

Schilder, Frank & Tenbrink, Thora. 2001. Before and after: sentence-internal and -external dis-
course relations. In Proceedings of the Workshop From Sentence Processing to Discourse In-
terpretation: Crossing the Borders, 1–10. Utrecht.
Smith, Carlota. 2003. Modes of Discourse. The Local Structure of Texts. Cambridge: CUP.
Stede, Manfred. 2004. The Potsdam Commentary Corpus. In Proceedings of the ACL Workshop
on Discourse Annotation, 96–102. Barcelona.
Stede, Manfred. 2004b. Kontrast im Diskurs. In Brücken schlagen – Grundlagen der Konnekto-
rensemantik, Hardarik Blühdorn, Eva Breindl & Hermann Ulrich Waßner (eds), 255–286.
Berlin: deGruyter.
Stede, Manfred. 2007. Korpusgestütze Textanalyse. Grunalzüge der Ebenen-orientierte Textlin-
guistik. Tübingen: Narr.
Stede, Manfred & Heintze, Silvan. 2004. Machine-assisted rhetorical structure annotation. In
Proceedings of the International Conference on Computational Linguistics (Coling-04), 425–
431.Geneva.
de Swart, Henriette. 1999. Position and meaning. Time adverbials in context. In Focus. Linguis-
tic, Cognitive, and Computational Perspectives, Peter Bosch & Rob van der Sandt (eds),
336–361. Cambridge: CUP.
Taboada, Maite & Mann, William. 2006. Rhetorical structure theory: Looking back and moving
ahead. Discourse Studies 8(3): 423–459.
Webber, Bonnie, Joshi, Aravind, Stone, Matthew & Knott, Alistair. 2003. Anaphora and dis-
course structure. Computational Linguistics 29(4): 545–587.
Werlich, Egon. 1975. Typologie der Texte. Heidelberg: Quelle und Meyer.
Wolf, Florian & Gibson, Edward. 2005. Representing discourse coherence: A corpus-based anal-
ysis. Computational Linguistics 31(2): 249–287.
 Subordination and coordination in syntax,
semantics and discourse
Evidence from the study of connectives1

Hardarik Blühdorn
Institut für Deutsche Sprache, Mannheim

This article discusses the question whether the distinction between subordination
and coordination is parallel in syntax and discourse. Its main thesis is that
subordination and coordination, as they are commonly understood in the
linguistic literature, are genuinely syntactic concepts. The distinction between
hierarchical and non-hierarchical connection in discourse structure, as far as
it is defined clearly in the literature, is of a quite different nature. The syntax
and semantics of connectives (as the most prominent morphosyntactic means
by which subordination and coordination are encoded) offers little evidence to
support the assumption of a structural parallelism between syntax and discourse.
As a methodological consequence, sentence and discourse structure should not
be mixed up in linguistic analysis.

Keywords: Coherence, relation, conjunction, adposition, adverbial connective.

1.  Introduction

The distinction between linguistic subordination and coordination, i.e., between hi-
erarchical and non-hierarchical connection in language, has been investigated during
the last decades by many authors within various research paradigms. Both types of
connection have been examined on sentence level as well as on the level of text and dis-
course (for syntax-oriented approaches see e.g., Foley & Van Valin 1984: 238ff; Shopen
1985; Wesche 1995; Kortmann 1996; Haumann 1997; Van Valin & LaPolla 1997: 441ff;
Johannessen 1998; Cristofaro 2003; Haspelmath (ed.) 2004; for discourse-oriented
approaches see e.g., Thompson & Longacre 1985; Mann & Thompson 1988; Polanyi
1988; Günthner 1996; Lefèvre 2000; Asher & Vieu 2005).

.  I am grateful to Bernd Wiese, Anke Holler, Manfred Stede, Ingolfur Blühdorn, Marina Fo-
schi Albert, and Cathrine Fabricius-Hansen, as well as to three anonymous reviewers for their
helpful comments on earlier versions of this paper.
 Hardarik Blühdorn

One of the questions addressed in the literature concerns the relation between
syntactic connection and discourse connection. Should it be assumed that hierarchi-
cal connections of clauses reflect hierarchical connections of discourse units and that
non-hierarchical connections of clauses reflect non-hierarchical connections of dis-
course units (see Hopper & Thompson 1984: 736f; Quirk et al. 1985: 919f; Matthies-
sen & Thompson 1988; O’Dowd 1992; Schecker 2000; Wegener 2000; Cristofaro 2003:
45ff)? Or should it be assumed that the two domains of connection are structurally
independent of each other?
The present article contributes to the theoretical discussion of this question from
the point of view of the study of connectives, the examples being taken from Modern
German. In many languages, connectives (“linkers”; see Quirk et al. 1985: 921) are
among the most important means used to establish subordinative and coordinative
relations in syntax. Most theories of discourse structure pay particular attention to
their role in establishing coherence relations (see e.g., Halliday & Hasan 1976: 226ff;
Polanyi 1988: 605; Knott & Dale 1994: 45ff; Knott et al. 2001). In terms of traditional
grammar, they belong to the following word classes: subordinating and coordinating
conjunctions, adpositions, and adverbs (see Pasch et al. 2003: 38ff; Blühdorn 2007a).
There are several other means that serve to encode syntactic subordination in the
languages of the world: complementizers, relative pronouns and relative particles, spe-
cialized converbs, non-finite verb forms like infinitives, gerunds and participles, as
well as inflectional case forms like locatives, instrumentals or ablatives; coordinative
relations may also be encoded by asyndetic juxtaposition, plurals or collectives (see
O’Dowd 1992; Muller 1996; Kortmann 1996: 5, 73; Zifonun 2001; Cristofaro 2003:
51ff; Heath 2004; Breindl & Waßner 2006; Breindl 2007a, b). These other means will
not be looked at in this article, the focus being on prototypical coordination and on
adjunct relations (adverbial adjuncts and adverbial subordinate clauses). In particular,
I will not be interested in complement relations (subject or object clauses), nor in at-
tributive relations (relative clauses), which are not connective relations in the sense
in which the term is used here (see Pasch et al. 2003: 1ff, 38f; on complements and
adjuncts see Bierwisch 2003).
The research project team Handbuch der deutschen Konnektoren (HdK) at the In-
stitut für Deutsche Sprache in Mannheim has been investigating the syntax and se-
mantics of German connectives for several years (see Pasch et al. 2003; Blühdorn et
al. 2004; Pasch 2004). One of the recurring issues discussed in that work has been
the syntactic and semantic nature of subordination and coordination (see Pasch et al.
2003: 230ff, 267ff).
Based on experience from the HdK project, the thesis to be presented in this
paper is:

The study of connectives offers little evidence in favour of a general structural

parallelism between hierarchical and non-hierarchical connections in syntax and
 Subordination and coordination: Evidence from connectives 

discourse. Rather, it should be assumed that hierarchical as well as non-

hierarchical discourse relations may, in principle, be encoded by both coordination
and subordination in syntax. Whatever the relevant factors that control the choice
between syntactic subordination and coordination, they should not be identified with
the distinction between hierarchical and non-hierarchical connection in discourse.

In support of this thesis, I will present three main arguments:

i. The syntactic distinction between coordination and subordination is neutralized
at levels higher than the sentence. On the levels of text and discourse it does not
play any relevant role (section 2).
ii. In semantics, similarly as in syntax, we can distinguish between hierarchical and
non-hierarchical connections. Both types of semantic connection can be encoded
by both types of syntactic connection (section 3).
iii. Syntax and semantics provide similar models for non-hierarchical connection,
but contrasting models for hierarchical connection. There is no reason to believe
that the structure of hierarchical relations in discourse should be generally more
similar to hierarchical relations in syntax than to hierarchical relations in seman-
tics (section 4).
In syntax, hierarchical connection of clauses is traditionally called subordination, and
non-hierarchical connection of clauses is called coordination. In line with this tradi-
tion, I will use the terms subordination and coordination for the two main types of
connection in syntax. In semantics, I will distinguish between symmetrical (non-hier-
archical) and asymmetric (hierarchical) connection of conceptual entities (spatial ob-
jects, events, propositions, or acts). The terms hierarchical and non-hierarchical them-
selves will be reserved for the connection of rhetorical units (utterances and speech
acts) in discourse.
The main question of this article can therefore be reformulated as follows:
Is there a parallelism between coordinative vs. subordinative connection of claus-
es in syntax, symmetrical vs. asymmetric connection of concepts in semantics
and non-hierarchical vs. hierarchical connection of rhetorical units in discourse?
Or are these three levels of connection independent of each other?
In section 2, this question is approached from a syntactic point of view, in section 3
from a semantic point of view, and in section 4 from the point of view of discourse.
Section 5 summarizes the conclusions to be drawn from the arguments presented.

2.  Syntactic connections

Recent studies on syntactic coordination and subordination in the languages of the

world suggest that there is no sharp dichotomic distinction between these two types of
 Hardarik Blühdorn

connection (see Cristofaro 2003: 15ff). Rather, they should be viewed as prototypical
poles on some sort of gradient (see Quirk et al. 1985: 927f; O’Dowd 1992: 68f; Kortmann
1996: 56ff; Johannessen 1998: 237ff). In this section the focus is on the prototypical poles
rather than the various intermediate types which occur in German (see Pasch 2000), as
well as in many other languages (see Haspelmath 2004: 33ff).

2.1  Coordinative connections

Coordinative connections are realized by coordinating conjunctions. The prototype of
a coordinating conjunction is und (‘and’):
(1) Es gab da überhaupt keine Action-Abteilung,
und ich war von Tag zu Tag mieser gelaunt.
Und dann erwog ich auch bald meine Abseilung
in den ewigen Underground.
Und sehn wir uns nicht in dieser Welt,
dann sehn wir uns in Bielefeld!
(Udo Lindenberg, Rätselhaftes Bielefeld)
they didn’t even have an action department / and my mood got worse by the
day / and soon I began to think about abseiling / to the eternal underground /
and if we won’t meet in this world / then we’ll catch up in Bielefeld

The expressions linked by a coordinating conjunction (coordinator) are typically of

the same formal and/or functional category (for exceptions see Johannessen 1998;
Osborne 2003: 114ff). In example (1), all conjuncts (or coordinands; see Haspelmath
2004) are V(erb)2-sentences.
Opinions differ about what may be the most adequate representation of the syntac-
tic structure of coordination (see e.g., Dik 1972; Wiese 1980; Wesche 1995; Johannessen
1998; Camacho 2003; Osborne 2003, 2006; Eisenberg 2004: 205ff, 377ff). Most but not
all approaches assume structures in which both coordinands have equal status in rela-
tion to the coordinator or some other category. One of the exceptions is Johannessen
(1998: 108ff), who proposes a structure in which one coordinand is the complement of
the coordinator, the other being its specifier. Her main interest, however, is in so-called
unbalanced, i.e., non-prototypical, coordination.
In prototypical coordination, the morphosyntactic format of the coordinands is
defined independently of the coordinator. Coordinators neither select coordinands of
a specific category nor do they require or attribute specific morphosyntactic features.
German has some connectives that behave similarly to coordinators, but neverthe-
less select relata of a certain morphosyntactic format: denn (‘for’), e.g., can only con-
nect V2-clauses (see Duden 2005: 628), whereas sowie (‘as well as’) can only connect
V-final-clauses and constituents less complex than a clause (words or phrases) (see
Breindl 2007b). Connectives with a similar non-prototypical behaviour can be found
in several languages: for and as well as in English, car and ainsi que in French, gdyż and
oraz in Polish, etc.
 Subordination and coordination: Evidence from connectives 

Prototypical coordinators and similar elements are strongly constrained as to

their linear position in relation to the coordinands. In German, they must be posi-
tioned in the middle between the coordinands, with a slightly stronger affinity to the
right one:
(2) a. Ihr kauft ein und wir warten hier an der Ecke.
you can go shopping and we’ll wait here at the corner
b. *Und wir warten hier an der Ecke ihr kauft ein.
and we’ll wait here at the corner you can go shopping

If two coordinate clauses are separated by a comma or period, then it is invariably put
to the left of the coordinator and not to its right:

c. Ihr kauft ein. Und wir warten hier an der Ecke.

you can go shopping. and we’ll wait here at the corner
d. * Ihr kauft ein und. Wir warten hier an der Ecke.
you can go shopping and. we’ll wait here at the corner

In other languages, coordinators may take other linear positions (e.g., Latin ‑que,
which is a suffix added to the second coordinand). In general their positions are much
more constrained than the positions of any other class of connectives (see Haspelmath
2004: 6ff; also Osborne 2006). We can therefore say that coordinators connect their
coordinands basically by linear sequence.

2.2  Subordinative connections

One of several means to establish subordinative connections between clauses are sub-
ordinating conjunctions (adverbial subordinators; see Kortmann 1996), like während
(‘while’):
(3) Die Pinguine waren braun-gelb, während die Giraffen schwarz-weiß waren.
the penguins were yellow-brown, while the giraffes were black and white

Subordinating conjunctions influence the morphosyntactic format of one of their re-

lata (the subordinate clause). The authors of the Handbuch der deutschen Konnekto-
ren (Pasch et al. 2003: 8ff, 106ff) call this relatum the internal argument. The relation
­between the subordinating conjunction and its internal argument is described as a type
of government: in German, subordinating conjunctions select V-final order of their
internal argument; in many languages they require certain tense and/or mood forms of
the subordinate verb. On the other hand, subordinating conjunctions do not have any
influence on the morphosyntactic format of their external argument (the main clause)
(see Pasch et al. 2003: 361, 416f). Subordinative connections are therefore structurally
asymmetric.
While the relata of coordinators are typically of the same morphosyntac-
tic ­category, the relata of subordinating conjunctions typically belong to different
 Hardarik Blühdorn

­ orphosyntactic categories. They can be distinguished in functional terms as main

m
and subordinate clauses, or formally as clauses with certain morphosyntactic proper-
ties, e.g., V2 and V-final. But categorical differences between the relata are by no means
obligatory in subordinative connections. Both relata can be of the same category, if,
for independent reasons, the external argument is a subordinate clause as well:
(4) Maria erzählte, dass [die Pinguine braun-gelb waren, während [die Giraffen
schwarz-weiß waren]V-final]V-final.
Mary told us that [the penguins were yellow-brown, while [the giraffes were
black and white]]
In such cases the morphosyntactic form of the external argument is never determined
by the connective.
Prototypical coordinators can connect expressions of any morphosyntactic cat-
egory. Subordinating conjunctions, in contrast, can only connect clauses. This limita-
tion is largely compensated for by adpositions, which are formally and functionally
very similar to subordinating conjunctions, but which take noun phrases instead of
clauses as at least one of their relata (see Kortmann 1996: 25, 58ff, 66ff):
(5) die Vorkommnisse während der Abschlussfeier
the occurrences during the leaving party
It is a well-known fact that in German, as in many other languages, clauses can be
transformed into noun phrases (see Hopper & Thompson 1984: 737f, 744ff; O’Dowd
1992; Eisenberg 2004: 252ff). The expressions resulting from nominalization must
then be connected by adpositions instead of subordinating conjunctions. Adpositions
require their internal argument to adopt a specific case form, but they have no influ-
ence on the morphosyntactic form of their external argument.
Subordinating conjunctions and adpositions have a fixed serial position in rela-
tion to their internal, but not in relation to their external argument. In German, they
typically take a position at the left margin of their internal argument. They structurally
embed their internal argument into the external argument. In the linear structure of
the external argument they can be moved rather freely together with their internal
argument. In particular, they can be postposed (as in (6/7a)), preposed (as in (6/7b))
and – at least for subordinating conjunctions – even be interposed to the external
argument (as in (6c)).

Subordinating conjunction:
(6) a. Wir warten hier an der Ecke, solange ihr einkauft.
we’ll wait here at the corner, while you go shopping
b. Solange ihr einkauft, warten wir hier an der Ecke.
while you go shopping, we’ll wait here at the corner
c. Wir warten, solange ihr einkauft, hier an der Ecke.
we’ll wait, while you go shopping, here at the corner
 Subordination and coordination: Evidence from connectives 

Adposition:
(7) a. die Aufräumarbeiten nach der Abschlussfeier (waren ermüdend)
the cleaning work after the leaving party (was exhausting)
b. nach der Abschlussfeier die Aufräumarbeiten (waren ermüdend)
after the leaving party the cleaning work (was exhausting)

Thus, subordinating conjunctions and adpositions (subordinators) do not link their

relata by linear sequence, but by government and embedding.

2.3  Adverbial connections

Coordinative and subordinative connections of the kinds discussed so far are estab-
lished by syntactic means such as linear ordering, government and embedding. They
clearly fall within the scope of syntax.
Semantically equivalent connections can also be encoded by means of adverbial
connectives. In the following pairs of examples, the (a)-variants encode the connec-
tion by means of a subordinator or coordinator, whereas the (b)-variants encode a
semantically equivalent connection by means of an adverbial connective:
(8) a. Die Pinguine waren braun-gelb, während die Giraffen schwarz-weiß waren.
(subordinating conjunction)
the penguins were yellow-brown, while the giraffes were black and white
b. Die Pinguine waren braun-gelb. Die Giraffen dagegen waren schwarz-weiß.
(adverbial connective)
the penguins were yellow-brown. the giraffes, in contrast, were black and
white
(9) a. Die Aufräumarbeiten nach der Abschlussfeier waren sehr anstrengend.
(adposition)
the cleaning work after the leaving party was very exhausting
b. Die Abschlussfeier war ein großer Erfolg. die Aufräumarbeiten danach waren
sehr anstrengend. (adverbial connective)
the leaving party was a great success. The cleaning work afterwards was
very exhausting
(10) a. Es gab da überhaupt keine Action-Abteilung, und ich war von Tag zu Tag
mieser gelaunt. (coordinating conjunction)
they didn’t even have an action department, and my mood got worse by
the day
b. Es gab da überhaupt keine Action-Abteilung. Zudem war ich von Tag zu Tag
mieser gelaunt. (adverbial connective)
they didn’t even have an action department. in addition, my mood got
worse by the day
Adverbial connectives are syntactic constituent of one of their semantic relata. They
are attached to that relatum as adverbial adjuncts. They may influence the tense and/
 Hardarik Blühdorn

or mood of the verb of that relatum, but they do not influence the morphosyntactic
format of their other relatum. To their other relatum, they do not bear any syntactic
relation (see Pasch et al. 2003: 485).
In relation to both connected expressions, adverbial connectives do not have a
fixed linear position. Like most adverbials, they can be moved relatively freely within
the relatum of which they are a constituent, and though they typically occur in the
right (subsequent) relatum they can also occur in the left (antecedent) relatum:
(11) Wir warten hier so lange. Ich meine, bis ihr mit dem Einkaufen fertig seid.
we’ll wait here for the time being. I mean until you have finished shopping

Adverbial connectives link their relata neither by government and embedding nor by lin-
ear sequence. Instead, they connect them semantically or, more precisely, by reference.
The semantic representation of an adverbial connective contains a slot for a referent that
cannot be identified on grounds of the information provided by the sentence in which the
adverbial is a constituent. In order to identify that referent, the interpreter must look for
the necessary information in the preceding or following context. Depending on where
the required information is placed, we can distinguish between anaphoric (backward
oriented) and cataphoric (forward oriented) adverbial connections. Thus, the adverbial
connectives dagegen (‘in contrast’), danach (‘afterwards’), and zudem (‘in addition’), in
(8b), (9b) and (10b) respectively, connect their relata anaphorically, whereas so lange (‘for
the time being’) in (11) connects its relata cataphorically.
In many adverbial connectives of German, the referential element is morphologi-
cally visible. Such connectives are results of word formation processes in which an
adpositional and a pronominal component have been contracted into one word form.
In the following examples, the pronominal component is boldfaced, the other one be-
ing the adpositional component: da-gegen (‘in contrast’, lit. ‘there-against’), da-nach
(‘afterwards’, lit. ‘there-after’), hier-bei (‘on this occasion’, lit. ‘here-at’), hier-für (‘for
this’, lit. ‘here-fore’), zu-dem (‘in addition’, lit. ‘to-that’), außer-dem (‘moreover’, lit.
‘outside-that’), in-dessen (‘however’, lit. ‘in-that’), während-dessen (‘in the meantime’,
lit. ‘during-that’) etc. Thus, in (8b) the pronominal component da- of the adverbial
connective dagegen contained in the second sentence refers anaphorically to the prop-
osition encoded by the first sentence. Similarly in (10b), the pronominal component
-dem of the adverb zudem contained in the second sentence refers anaphorically to the
proposition encoded by the first sentence. In (9b), the pronominal component da- of
the adverb danach refers to an event (the leaving party) described in the preceding
sentence. In (11), the pronominal component so refers to an event (finishing shop-
ping) described in the subsequent sentence (on the referential function of da ‘there’
and so ‘so’ see Blühdorn 2003).
Some adverbial connectives of German, such as bestenfalls (‘at best’), wenigstens
(‘at least’) or anschließend (‘afterwards’) do not contain morphologically explicit pro-
nominal components. We cannot go into the details here, but probably all adverbial
connectives can be traced back historically to expressions that involve some sort of
 Subordination and coordination: Evidence from connectives 

referential element, and all of them are used in exactly the same referential way (see
Webber et al. 2003: 548ff). We are therefore justified in assuming that the semantic
representation of all adverbial connectives contains a referential slot, even if their mor-
phological form has not preserved a corresponding pronominal element.
It is interesting to observe that the pronominal element, where it is visible, quite
often maintains case morphology within the contracted form, as in zu-dem (lit.:
to-that-DAT), außer-dem (lit.: outside-that-DAT), in-dessen (lit.: in-that-GEN) or
während-dessen (lit.: during-that-GEN). The complex morphology of these connec-
tives reveals that their referential linking does not simply substitute syntactic subor-
dination. Rather, the linking force of adverbial connectives builds on a subordinative
relation which has become incorporated in their structure. The pronominal compo-
nent was originally case-governed by the adpositional component: it is, in fact, its
internal argument. Consequently, the clause to which the adverbial connective is at-
tached as an adjunct must be its external argument. In the contracted form of the
connective, the subordinative force of the adposition has become encapsulated, so that
it is no longer able to contribute actively to syntactic structure. But at the same time,
the semantic scope of the connective is extended beyond the limits of the sentence by
the referential force of the pronominal component. Structurally, the internal argument
is incorporated within the connective, but its referent must be found in the context.
Thus, in a sense, adverbial connectives are closer in syntax to their external argument
(the clause of which they are a constituent) than to their internal argument (the clause
to which they establish a link by reference).
Table 1 provides an overview of the linking properties of the connectives that have
been discussed so far:

Table 1.  Overview of the linking properties of connectives

Subordinators Coordinators Adverbial connectives
Linking by government + ±
and embedding
Linking by linear + ±
sequence
Linking by reference +

Subordinating conjunctions and adpositions (subordinators) link their relata hierarchi-

cally, by government and embedding, whereas coordinators link them non-hierarchically,
by linear sequence. Adverbial connectives neutralize this difference (see Quirk et al. 1985:
927f). With subordinators, they share government and embedding, but they encapsulate
these relations in their morphology, so that they cannot take effect within sentence struc-
ture. With coordinators, they share a positional affinity to the right (subsequent) relatum
(see ibid.: 921f). But in coordinators this affinity amounts to a strict syntactic rule, while
in adverbial connectives it is only a pragmatic preference. The particular linking force of
 Hardarik Blühdorn

adverbial connectives is based neither on government and embedding nor on serial posi-
tion, but on reference, i.e., on a principle that does not play a crucial role in syntactic coor-
dination or subordination.
Reference is a discourse relation, not a syntactic relation. When using referential ex-
pressions such as adverbial connectives as cohesive devices, speakers refer to discourse
entities, not to other syntactic expressions. This means that adverbial connectives establish
connections on discourse level, while subordinators and coordinators establish connec-
tions on sentence level. Connections established by adverbial connectives are outside the
scope of syntactic rules. The syntactic distinction between coordination and subordina-
tion makes sense only within sentence boundaries, i.e., as a means to construct complex
and compound sentences (see Quirk et al. 1985: 719). It cannot be transferred to connec-
tions established by adverbial connectives. As a consequence, the distinction between hi-
erarchical and non-hierarchical connections in discourse cannot reasonably be supposed
to mirror the distinction between coordination and subordination in syntax. A very con-
siderable part of discourse connections is established on a level higher than the sentence,
i.e., beyond the scope of syntactic rules. At least for those connections, whether they count
as hierarchical or non-hierarchical can only be decided on other than syntactic grounds.
An additional observation may be made here about two groups of subordinators
with pronominal components. Some German subordinating conjunctions contain
such components, e.g., nachdem (‘after’), indem (‘as’) or seitdem (‘since’). Subordina-
tors of this kind derive historically from adverbs. Their pronominal components are
residues with a very weak (if any) referential function left in present day use (on nach-
dem see Blühdorn 2004). A second group are relative adverbs like weshalb (‘where-
fore’), wobei (‘whereby’), worauf (‘whereupon’) etc., which contain the pronominal
components wes- and wo-. Adverbs of this kind can be used to introduce adverbial
relative clauses. They establish a type of connection that combines referential linking
with linking by government, though not by embedding (see Pasch et al. 2003: 241ff,
422ff). The existence of this type of linking does not weaken my argument. I do not
claim, in fact, that reference has no role to play on sentence level, but that its role is not
crucial for syntactic connections. On the other hand, I do claim that syntactic linking
mechanisms such as government, embedding and syntactic serialization do not work
beyond the boundaries of the sentence, i.e., are neutralized on the level of discourse.

3.  Semantic connections

Strictly speaking, the term connective does not refer to a syntactic, but to a semantic
category. Connectives are elements of several syntactic classes that share the func-
tion of encoding semantically characterized relations between conceptual entities such
as events and propositions (see Pasch et al. 2003: 1ff, 38f; Blühdorn 2003; Lohnstein
2004). In semantics, as well as in syntax, we can distinguish between hierarchical and
non-hierarchical connections.
 Subordination and coordination: Evidence from connectives 

According to Lang (1984: 69ff), the semantic relata of coordinative connections

must be tied up by a common integrator. This term refers to a superordinate concep-
tual category, under which both relata can be subsumed, and under which they are in
contrast with each other. In example (12), we can construe something like “colours of
zoo animals” as the common integrator:
(12) Die Pinguine waren braun-gelb und die Giraffen waren schwarz-weiß.
the penguins were yellow-brown, and the giraffes were black and white

For the verses by Udo Lindenberg, a possible common integrator seems to be “motives
that suggest abseiling to the eternal underground”:

(13) Es gab da überhaupt keine Action-Abteilung, und ich war von Tag zu Tag mieser
gelaunt.
they didn’t even have an action department, and my mood got worse by the
day

The examples show that the common integrator need by no means pertain to generally
accessible world knowledge. On the contrary, it may be construed ad hoc, according to
the demands of each context in discourse.
It might seem as if the requirement of a common integrator could be a good crite-
rion to distinguish between hierarchical and non-hierarchical connections in seman-
tics. But on a closer look we find that this requirement is not restricted to semantically
symmetrical nor to syntactically coordinative connections. Rather, it is a characteristic
of a much more extensive class of connections, which may be symmetrical or asymmet-
ric and which may be encoded by coordination, subordination or any other syntactic
mechanism. Also, all kinds of adverbial connections belong to this class, i.e., require a
common integrator.
According to a well-known model proposed by Lyons (1977: 442ff; 791ff), con-
ceptual entities can be divided into four general categories: spatial (first order) entities,
temporal (second order) entities, logical/epistemic (third order) entities and deontic
(fourth order) entities (see also Kortmann 1996: 28ff). First order entities are spatial
objects, second order entities are states of affairs (including states and events), third
order entities are propositions, and fourth order entities are intentional entities, which,
for lack of a better term, I will call acts (see Blühdorn 2003: 16ff). The four general
categories define four conceptual domains: space, time, logic/episteme (the domain of
knowledge) and ethics/deontics (the domain of intentions and acting).
It is a general semantic requirement on adverbial connections (subordinative or
not), as well as on coordinative connections, that their relata must belong to the same
conceptual domain and, consequently, be of the same general category. Thus, in order
to connect two relata spatially, both must be spatial objects; in order to connect them
temporally, both must be states of affairs; in order to connect them logically, both must
be propositions; and in order to connect them deontically, both must be acts (for more
details see Blühdorn 2007a, b).
 Hardarik Blühdorn

It can be concluded that the requirement of a common integrator is not an ap-

propriate criterion for distinguishing between hierarchical and non-hierarchical
connections in semantics. A better criterion seems to be relational symmetry. Non-
hierarchical semantic connections are symmetrical. Their relata have equal semantic
functions and equal semantic weight (see Breindl 2007b: 144). One of the syntactic
consequences of semantic symmetry is the possibility of inverting the sequence of the
relata without a significant change of meaning. The relata in the following example are
symmetrically connected. They can be inverted without semantic consequences:

(14) a. Die Pinguine waren braun-gelb und die Giraffen waren schwarz-weiß.
the penguins were yellow-brown, and the giraffes were black and white
b. Die Giraffen waren schwarz-weiß und die Pinguine waren braun-gelb.
the giraffes were black and white, and the penguins were yellow-brown

Hierarchical semantic connections, in contrast, are asymmetric. Changing the syn-

tactic sequence of their relata will significantly change the meaning. Where an inver-
sion of the relata gives rise to such a change of meaning, it can be concluded that the
connection is not understood as symmetrical, even if it is syntactically encoded by a
coordinator:

(15) a. Maria ging in die Bibliothek und sie bekam Hunger.

Mary went to the library, and she began to feel hungry
b. Maria bekam Hunger und sie ging in die Bibliothek.
Mary began to feel hungry, and she went to the library

In the most plausible reading of these examples, the connected events are ordered in
a temporal sequence, which is inverted from (15a) to (15b). It is not uncommon for
syntactically coordinative connections to be interpreted in semantically asymmetric
ways – an effect that can be explained by very general cognitive and pragmatic prin-
ciples (see Grice 1981: 185f; Posner 1980: 182ff; Lang 1984: 80ff; Blakemore & Carston
2005; Breindl 2007a). The examples show that syntactic coordination and semantic
symmetry must be carefully distinguished. By no means can they be identified with
each other.
The relata of asymmetric connections cannot be inverted without significant se-
mantic consequences. They have different relational (thematic) roles. One of them is
being connected (like a ship that drops its anchor), the other is what it is being connect-
ed to (like the sea ground in which the anchor is fixed). In Ronald Langacker’s (1987:
231ff) terminology, the former is called trajector (T) and the latter landmark (L).
Three types of asymmetric connections can be distinguished (see Blühdorn 2003:
19f; Blühdorn 2005: 315f):

– situating connections
– conditional connections
– causal connections
 Subordination and coordination: Evidence from connectives 

Situating connections are stative. They assign to the trajector a place in a conceptual
domain, which is described by a relation to the landmark:
(16) Bevor du nach Hause gehst (L), lösch bitte das Licht (T).
before you go home (L), please switch off the light (T)

Switching off the light is the trajector that is situated on the time scale in relation to
the event of going home (landmark). Switching off the light and going home them-
selves are evidently non-stative, but their sequential relation on the time scale, en-
coded by the conjunction before, is stative. Each of the events has its fixed position
in time, and the position of the trajector is defined on grounds of the position of the
landmark.
Conditional connections are dynamic: the landmark event not only situates the
trajector event, but it also influences the value to be taken by the trajector event. It is
not yet clear if the trajector event will in fact become real (or come true). This depends
on the value to be taken by the landmark event:
(17) Und sehn wir uns nicht in dieser Welt (L), dann sehn wir uns in Bielefeld (T).
and if we won’t meet in this world (L), then we’ll catch up in Bielefeld (T)

The example tells us that the meeting in Bielefeld will take place on the condition that
the meeting in this world does not.
Causal connections are dynamic as well, but in a causal connection the value of
the trajector event is already fixed. The trajector event is real, and the landmark event
has influenced the fixing of its value:
(18) Ich war von Tag zu Tag mieser gelaunt (T), weil es gab da überhaupt keine
Action-Abteilung (L).
my mood got worse by the day (T), for they didn’t even have an action
department (L)

The inexistence of an action department is the cause which led to the change of the
speaker’s mood. The result is presented as factual, i.e., as a state of affairs with a fixed
reality value.
The distinction between semantically symmetrical and asymmetric connections
can be made within the boundaries of the sentence (on intra-sentence level) and also
between sentences in discourse (on inter-sentence level). Both kinds of semantic con-
nection can be encoded by coordinators as well as by subordinators and also by adver-
bial connectives. Some further examples may illustrate this point:
(19) a. Das ist kein Selbstbedienungsladen, und sonntags ist hier zu.
this is no self-service store, and on Sundays we are closed
b. Sonntags ist hier zu, und das ist kein Selbstbedienungsladen.
on Sundays we are closed, and this is no self-service store
c. Das ist kein Selbstbedienungsladen. Außerdem ist hier sonntags zu.
this is no self-service store. besides, we are closed on Sundays
 Hardarik Blühdorn

d. Sonntags ist hier zu. Außerdem ist das kein Selbstbedienungsladen.

on Sundays we are closed. besides, this is no self-service store

The connections in (19a-d) are symmetrical. The sequence of their relata can be in-
verted without a significant change of meaning. In (19a/b), the connection is encoded
by a coordinator (intra-sentence level); in (19c/d), a semantically equivalent connec-
tion is encoded by an adverbial connective (inter-sentence level). My argument for
considering coordination of two main clauses an intra-sentence connection is the
syntactic constraint on the position of the coordinator, discussed in section 2. In ad-
dition, we can consider punctuation and intonation. Coordinated main clauses may
be separated by a comma instead of a period and may even not be separated by any
punctuation mark. In spoken utterances, they can be included into the same intona-
tional phrase.

(20) a. Wenn du keine Lust hast (L), gehe ich allein (T).
if you don’t feel like it (L), I’ll go on my own (T)
b. Wenn ich allein gehe (L), hast du keine Lust (T).
if I go on my own (L), you won’t feel like it (T)
c. Du hast keine Lust? (L) Dann gehe ich allein. (T)
you don’t feel like it? (L) then I’ll go on my own (T)
d. Ich gehe allein? (L) Dann hast du keine Lust. (T)
I’ll go on my own? (L) then you won’t feel like it (T)

The examples in (20a-d) illustrate asymmetric connections. Inverting the sequence of

their relata leads to a significant change of meaning. The connections in (20a/b) are
encoded by a subordinator (intra-sentence level). In (20c/d), semantically equivalent
connections are encoded by an adverbial connective (inter-sentence level).
On intra-sentence level symmetrical and asymmetric connections can be encoded
by coordinators as well as subordinators:

(21) a. Die Pinguine waren braun-gelb und die Giraffen waren schwarz-weiß.
the penguins were yellow-brown, and the giraffes were black and white
b. Die Giraffen waren schwarz-weiß und die Pinguine waren braun-gelb.
the giraffes were black and white, and the penguins were yellow-brown
c. Die Pinguine waren braun-gelb, während die Giraffen schwarz-weiß waren.
the penguins were yellow-brown, while the giraffes were black and white
d. Die Giraffen waren schwarz-weiß, während die Pinguine braun-gelb waren.
the giraffes were black and white, while the penguins were yellow-brown
(22) a. Ich war von Tag zu Tag mieser gelaunt (T), weil es gab da überhaupt keine
Action-Abteilung (L).
my mood got worse by the day (T), for they didn’t even have an action
department (L)
 Subordination and coordination: Evidence from connectives 

b. Es gab da überhaupt keine Action-Abteilung (T), weil ich war von Tag zu Tag
mieser gelaunt (L).
they didn’t even have an action department (T), for my mood got worse by
the day (L)
c. Ich war von Tag zu Tag mieser gelaunt (T), weil es da überhaupt keine
Action-Abteilung gab (L).
my mood got worse by the day (T), because they didn’t even have an action
department (L)
d. Es gab da überhaupt keine Action-Abteilung (T), weil ich von Tag zu Tag
mieser gelaunt war (L).
they didn’t even have an action department (T), because my mood got
worse by the day (L)

The examples in (21a-d) illustrate symmetrical connections. In (21a/b), the connec-

tions are encoded by a coordinator. In (21c/d), the same relata are symmetrically con-
nected by a subordinator. The examples in (22a–d) illustrate asymmetric connections.
Inversion of the relata leads to a change of meaning. The connections in (22a/b) are
encoded by a (non-prototypical) coordinator. Both relata are syntactically realized as
main clauses. In (22c/d), the same relata are asymmetrically connected by a subordi-
nator. One of the relata is realized as a main clause, the other as a subordinate clause.
The data discussed in this section are from Modern German, but data from other
languages could just as well have been used. From the observations made, it can be
concluded that the distinctions between syntactic coordination and subordination and
between semantic symmetry and asymmetry are independent of each other. Connec-
tions of syntactic units and the connections of the encoded conceptual entities can be
either parallel or non-parallel in structure.

4.  Discourse connections

Since the 1980s, several models and theories of discourse structure have been pro-
posed, which have tried to give an explicit account of coherence relations in written
and spoken text. Most of them distinguish in some way or another between hierarchi-
cal and non-hierarchical discourse relations. But they differ considerably in how they
motivate this distinction.

4.1  Syntax and semantics as models for understanding discourse structure

One obvious hypothesis is that discourse structure might be parallel to syntactic and/
or semantic structure (see Hopper & Thompson 1984: 736f; O’Dowd 1992; Matthiessen
& Thompson 1988; Mann & Thompson 1988: 269; Taboada & Mann 2006: 427). The
and-variant of this hypothesis has often been assumed for coordinative relations. Coor-
dinated syntactic units are typically of the same formal and functional category, and their
 Hardarik Blühdorn

linear order can be inverted without semantic consequences. Symmetrically connected

conceptual units are of the same semantic category and have a common semantic func-
tion. Non-hierarchically connected discourse units should belong to the same rhetorical
category and have a common discourse function.
Many examples discussed in the literature seem to support the view that semantic sym-
metry and syntactic coordination are natural linguistic means to encode non-hierarchical
discourse relations. Yet we have seen in section 3 above that coordinative connections in
syntax are typically underspecified for semantic interpretation. Depending on the context
of the utterance, they may (or even must) receive an asymmetric reading:
(23) Maria ging zu McDonald’s, und sie bekam Hunger.
Mary went to McDonald’s, and she began to feel hungry
a.→ Außerdem bekam sie Hunger.
besides, she began to feel hungry
b.→ Dann bekam sie Hunger.
then she began to feel hungry
c.→ Deshalb bekam sie Hunger.
therefore she began to feel hungry

(23a) to (23c) are possible interpretations of the second part of (23). (23a) is a sym-
metrical reading. The two connected propositions have equal status: two predicates
which are true of Mary. (23b) and (23c) are asymmetric readings of (23). (23b) is a
situating interpretation: Mary’s going to McDonald’s is the landmark, her beginning to
feel hungry is the trajector. (23c) is a causal interpretation: going to McDonald’s is the
causal landmark, feeling hungry the causal trajector. Most theories of discourse struc-
ture have paid little attention to the multiple interpretability of coordinative syntactic
connections.
Some authors have assumed structural parallelism between syntax, semantics and
discourse also for subordinative adverbial connections (e.g., Matthiessen & Thompson
1988). For most of these connections, however, for general reasons, only the or-variant
of the hypothesis is possible, because the syntactic and semantic structures of typical
subordinative adverbial connections are inverse to each other. A crucial property of
their syntactic structure is embedding. Hierarchically higher relata, e.g., main clauses,
embed hierarchically lower relata, e.g., subordinate clauses:
(24) main clause

adverbial subordinate clause

Semantically asymmetric connections, too, can be characterized, in a sense, as embed-

ding relations. The landmark is the embedding part: it forms the conceptual background
framework into which the trajector is inserted. Using the terminology of Gestalt psy-
chology, Langacker (1987: 231ff) characterizes the landmark as ground. The trajector,
 Subordination and coordination: Evidence from connectives 

on the other hand, is the embedded part: it takes a position in relation to the conceptual
background. Langacker (ibid.) characterizes it as figure. In terms of embedding, then,
the landmark’s position in the semantic hierarchy is higher than the position of the
trajector, for the trajector is embedded by the landmark:
(25) landmark

trajector

Looking at the mapping relations between hierarchical connections in syntax and se-
mantics, we realize that the semantically embedding landmark is invariably encoded
by the syntactically embedded expression (e.g., the subordinate clause) and the seman-
tically embedded trajector by the syntactically embedding expression (e.g., the main
clause):
(26) Solange ihr einkauft (L), warten wir hier an der Ecke (T).
while you go shopping (L), we’ll wait here at the corner (T)

Lohnstein (2004: 143) states, from the point of view of model-theoretic semantics:
“Das Wahrheitsintervall des Hauptsatzes [...] wird relativ zum Wahrheitsintervall des
Nebensatzes [...] bestimmt, so dass der Nebensatz die Auswertungsdomäne für den
[...] Hauptsatz determiniert.” (The truth interval of the main clause is fixed in relation
to the truth interval of the subordinate clause, so that the subordinate clause deter-
mines the domain of interpretation for the main clause.)
Bierwisch (2003) sees the crucial difference between complements and adjuncts in
the direction of the attribution of thematic roles. Both complements and adjuncts are
syntactically subordinate to their heads, but whilst complements receive their thematic
roles from their heads, adjuncts attribute thematic roles to their heads. If we interpret
attribution of thematic roles as a manifestation of semantic superordination, we can
state that complements are semantically subordinate and adjuncts semantically super-
ordinate to their heads. Applied to adverbial subordinate clauses, this means that they
are semantically superordinate to their main clauses.
The hierarchies of subordinative syntactic connections and asymmetric semantic
connections are thus inverse to each other: the syntactically embedding part is the
semantically embedded part and vice versa:
(27) main clause landmark

subordinate clause trajector

This application of the landmark-trajector distinction on adverbial clauses differs from
Langacker’s own proposal (see Langacker 1991: 436). When explaining the subordinate
 Hardarik Blühdorn

status of adverbial clauses, Langacker abandons his analysis in terms of landmark and
trajector and stipulates, instead, a general iconicity between syntactic and semantic sub-
ordination in terms of profiling (see ibid.: 436f; also Cristofaro 2003: 29ff). This solu-
tion appears somewhat ad hoc to me, and Langacker himself does not explain it. As a
consequence, questions may also be raised about Cristofaro’s (ibid.) claim that semantic
asymmetry provides a more reliable starting point for the analysis of subordination than
traditional morphosyntactic asymmetries do.

4.2  Some formal accounts of discourse structure

4.2.1  Rhetorical Structure Theory
The hypothesis of a structural parallelism between syntax, semantics and discourse has
been of some importance in Rhetorical Structure Theory (RST; see Mann & Thompson
1988; Taboada & Mann 2006; Stede in this volume). RST distinguishes between nuclear
information and satellite information within a discourse. Nuclear information is main
information, satellite information is secondary information. The difference between
them becomes clearest when they are deleted. The deletion of nuclear information will
make the discourse less coherent, the remaining parts becoming more difficult to com-
prehend. The deletion of satellite information will make the discourse less explicit, but
the remaining information will still be coherent. Thus, the omission of satellite informa-
tion may play an important role in summarizing (see Mann & Thompson 1988: 267f).
RST distinguishes between two types of relations: nucleus-satellite relations and
nucleus-nucleus relations. The former are hierarchical, the latter are non-hierarchical
(Mann & Thompson 1988: 246ff, 266). Nucleus-satellite relations are more frequently
dealt with in RST-related studies. The examples analysed by Mann & Thompson (ibid.:
252, 261ff) show that they can be encoded by both coordinative and subordinative
connections in syntax. Both nuclear (N) and satellite (S) information can be encoded
by both main (M) and adverbial subordinate clauses (A):
(28) I’ll post more details later (S–M), but this is a good time to reserve the place on
your calendar (N–M). (concessive relation; coordinative connection)
(29) As your floppy drive writes or reads (S–A), a Syncom diskette is working four
ways (N–M). (circumstance relation; subordinative connection 1)
(30) A carbon additive drains away static electricity (S–M), before it can attract dust
or lint (N–A). (antithesis relation; subordinative connection 2)
Nucleus-nucleus relations are not in the focus of interest in RST. The examples
­given by Mann and Thompson (ibid.: 278f) are not conclusive, but they suggest that
­nucleus-nucleus relations can also be encoded by both coordinative and subordina-
tive connections:
(31) Peel oranges (N–M) and slice crosswise (N–M). Arrange in a bowl (N–M)
and sprinkle with rum and coconut (N–M). (sequence relation; coordinative
connection)
 Subordination and coordination: Evidence from connectives 

(32) Chill (N–M) until ready to serve (N–A). (sequence relation; subordinative
connection)
Satellite-satellite relations (see Asher & Vieu 2005: 592, 594f), which are a second
type of symmetrical relations, are not recognized in RST. But the text analysis given in
Mann & Thompson (1988: 261ff) shows that they can at least be encoded by syntactic
coordination:
(33) Strong binders hold the signal-carrying oxides tightly within the coating (S). And
the non-woven jacket liner (...) provides thousands of tiny pockets to keep what it
collects (S).
One of the main practical problems in RST seems to be how to reliably recognize main
information and how to distinguish it from secondary information (see Stede in this
volume). As a consequence, the distinction between hierarchical and non-hierarchical
discourse connections remains problematic as well.
Mann & Thompson (1988: 249f) insist that the definitions of RST relations “do not
rely on morphological or syntactic signals. Recognition of the relation always rests on
functional and semantic judgements alone. [...] We have found no reliable, unambigu-
ous signals for any of the relations.” Although this affirmation seems to be somewhat
idealized (compare the section on explicit signalling of discourse relations in Taboada
& Mann 2006: 438ff), RST thus provides at least some evidence for considering dis-
course structure as independent of syntactic structure. On the other hand, Mann &
Thompson (1988: 269) explicitly suggest a functional link between asymmetric dis-
course relations and hypotaxis in syntax.
An unresolved theoretical problem in RST has to do with the concept of a dis-
course relation. Syntactic subordination and coordination are defined as relations
between syntactically categorized units (such as phrases, clauses or sentences) with
certain syntactic functions (such as head, complement or adjunct). Symmetrical and
asymmetric semantic connections are defined as relations between semantically cat-
egorized entities (such as spatial objects, events or propositions) bearing semantic or
thematic roles (such as agent, theme, cause, goal etc.).
Similarly, hierarchical and non-hierarchical discourse relations should be defined
as relations between rhetorically categorized units (such as utterances or speech acts)
with certain rhetorical functions (such as topic and focus, or intended communicative
effects). But that is not how discourse relations are defined in RST. The basic units of
analysis in RST are clauses, sentences or phrases, i.e., syntactic units. The relations
between these units are divided into “subject matter” and “presentational” ones (Mann
& Thompson 1988: 256f; also Taboada & Mann 2006: 435f). The former are clearly
semantic relations such as cause, condition, result etc.; the latter are pragmatic relations
between speech acts such as evidence, motivation, justify etc. Only the latter deserve to
be called discourse relations in the strict sense of the term.
RST is a hybrid model that incorporates syntactic, semantic and rhetorical concepts
and categorizations (see Stede in this volume). RST tree diagrams contain information
 Hardarik Blühdorn

of different domains. This makes them suggestive and at the same time non-conclusive
in relation to a possible parallelism between sentence and discourse structure. They are
suggestive to the extent that their basic units are syntactic instead of rhetorical catego-
ries, and they are non-conclusive to the extent that these units are associated with partly
semantic and partly rhetorical functions.

4.2.2  The Linguistic Discourse Model

A more homogeneous account of discourse structure is offered by Polanyi’s (1988)
Linguistic Discourse Model (LDM). In LDM, discourse is segmented into discourse
constituent units of different levels of complexity. The units at the elementary level are
clauses and so-called discourse operators (assigners, connectives and discourse mark-
ers) (Polanyi 1988: 605f). Constituents at the levels of higher complexity are genuine
discourse units such as interactions, speech events, stories, plans, question-answer
sequences, lists etc. (ibid.: 603). Representations of the elementary level of discourse
structure in LDM are relatively close to syntax; representations of higher levels are
much closer to conversational analysis.
A non-hierarchical (“coordinating”) discourse relation is defined in LDM as a
relation between two or more discourse constituents that are all linked by the same
relation to a common higher order constituent. The linear sequence in which these
constituents are uttered is viewed as motivated by cognitive or communicative prin-
ciples and therefore as non-random. Thus, the possibility of inverting the sequence of
the relata is not a criterion for non-hierarchical discourse connection in LDM. The
decisive criterion is the common discourse function in relation to a superordinate con-
stituent (ibid.: 606f).
A hierarchical (“subordinating”) discourse relation is defined as a relation in which
the information conveyed by a discourse constituent S1 further specifies information
conveyed by a preceding constituent S0. More specifically, the same inferences can be
drawn from S0 and S1, but some of the information which can be inferred from S1
is more detailed than the information which can be inferred from S0 (Polanyi 1988:
609). Discourse constituents which interrupt preceding constituents or which consti-
tute thematic or interactional digressions are also treated as subordinate, even if no
information related to the preceding constituent can be inferred from them (Polanyi
1988: 611, 619). Thus, the concept of discourse subordination in LDM is much broader
than the concept of a nucleus-satellite relation in RST. It is striking that in LDM, for
purely technical reasons, a subordinate discourse unit can never precede its superordi-
nate unit (Polanyi 1988: 613ff). This constraint seems somewhat unrealistic, when we
think of the structural possibilities of natural discourse.
Hierarchical discourse relations are also characterized as “embedding” relations
in LDM (Polanyi 1988: 613). But the underlying notion of discourse embedding is not
related to syntactic embedding as it was discussed in section 2 above, nor to seman-
tic embedding as it was discussed in section 4.1. Although LDM considers clauses as
 Subordination and coordination: Evidence from connectives 

­elementary units of discourse and distinguishes between coordinating and subordi-

nating connectives as discourse operators, the relation between syntactic structure and
discourse structure is not explicitly discussed. In LDM, an incoming discourse unit is
processed as subordinate or coordinated in relation to a preceding discourse unit in
accordance with contextual information and general world knowledge (Polanyi 1988:
616ff).
On the whole, LDM is much more concerned with genuine discourse units and
functions than with syntax and semantics. Syntactic coordination and subordination
are never explicitly mentioned as criteria for considering a given discourse relation as
hierarchical or non-hierarchical. On the other hand, the role of contextual and general
world knowledge in the selection of a hierarchical or non-hierarchical discourse rela-
tion remains somewhat fuzzy. It seems not impossible that the recognition of syntactic
coordination and subordination may contribute indirectly or implicitly to the selec-
tion of discourse relations.

4.2.3  Segmented Discourse Representation Theory

The approach of Asher & Vieu (2005) is more abstract than both RST and LDM. It
does not aim primarily at describing discourse relations in real data, but rather at giving
general characterizations of categories of discourse relations such as Narration and
Elaboration. Asher & Vieu do not believe in the possibility of defining the distinction
between hierarchical and non-hierarchical discourse relations in semantic terms (Asher &
Vieu 2005: 598). Instead, they look for syntactic definitions of the two types of rela-
tions within the framework of Segmented Discourse Representation Theory (SDRT).
They propose formal criteria for testing whether relations of different categories are
“subordinating” or “coordinating” (Asher & Vieu: 599ff).
The first criterion exploits the so-called right-frontier-constraint on anaphoric
linking (see Polanyi 1988: 602, 613ff). If in a sequence of discourse units α, β and γ a
discourse relation R connects α and β, and γ can be anaphorically attached to α, then R
is subordinating. If γ can only be attached to β, then R is coordinating.
The second criterion is based on an SDRT principle called Continuing Discourse
Patterns (CDP). If discourse units β and γ are connected by a coordinating discourse
relation R2 which requires that its relata “bear the same discourse relation to a domi-
nant constituent” (Asher & Vieu 2005: 595) and if discourse unit α is connected to β
by a discourse relation R1 which is different from R2, then it can be inferred that R1
is a subordinating relation and that α and γ are connected by a relation R3 of the same
subordinating type as R1. If R2 is not a coordinating relation that requires R1 and R3
to be relations of the same type, then R1 is coordinating.
The third criterion is very similar to the first one and can therefore be left out
here. The fourth criterion is based on the SDRT assumption that two discourse units
α and β cannot be connected simultaneously by two discourse relations one of which
is subordinating and the other coordinating. Therefore, if a relation R between α and
 Hardarik Blühdorn

β is proved to be coordinating, α and β cannot be connected simultaneously by a sub-

ordinating relation.
The second and fourth tests seem to be somewhat circular, since what can be “in-
ferred” from them is little more than a reformulation of their input conditions. The
first test is more interesting, but the authors warn that it might not be conclusive in
all cases.
When applying the tests to some concrete examples, Asher & Vieu recognize that
there are more difficulties. An analysis of the discourse relation Result (ibid.: 604ff)
suggests that it can be “used” either in a coordinating or in a subordinating manner.
The same possibility might exist for other discourse relations as well. Asher & Vieu
do not explain what exactly it means for a discourse relation to be “used”. But they
draw the conclusion that “the concepts of subordinating and coordinating discourse
relations” might be founded neither on semantics nor on syntax. Instead they suggest
that these properties might be a matter of information packaging (ibid.: 600, 609).

4.3  Future perspectives

An important conclusion to be drawn from the discussion of different models of dis-
course structure is that discourse should be viewed as a system sui generis. Discourse
structure is the arrangement of discourse units (see Polanyi 1988: 603ff), which can be
ordered on different levels of constituency as well as on different levels of focality. On
higher levels of constituency, the units are structurally more complex, on lower levels
they are structurally more simple. On higher levels of focality, units are more central
to the communicative goals of the speaker, on lower levels of focality, units have sup-
porting or secondary functions (see Klein & von Stutterheim 1992). Non-hierarchical
discourse relations connect units of the same level, either of constituency or of focal-
ity (intra-level connection); hierarchical discourse relations connect units of different
levels of constituency or of focality (inter-level connection).
Discourse should not be viewed from the beginning as something derived from
syntax and/or semantics. Hybrid conceptualizations make it more difficult to recog-
nize genuine discourse properties and make discourse studies more susceptible to
precipitate conclusions. The interaction between hierarchical and non-hierarchical
connections in syntax, semantics and discourse can only be described in appropriate
detail on the basis of independent and explicit theories of each of the three domains
of language structure.
Recent research in the areas of intonational phonology and conversational analysis
(see e.g., Chafe 1988; Selting 1995; Peters 2005; Büring 2006; Moroni 2006) has shown
ever more clearly that (in non-tone languages) a considerable part of discourse structure
is encoded by intonation, even in written discourse (“silent prosody”: Féry 2006). This
suggests that the widespread convention of taking the clause as the elementary unit of
discourse structure must be seriously questioned. The clause is a syntactic unit. The corre-
sponding discourse unit is the utterance. Clauses have subjects and predicates; utterances
 Subordination and coordination: Evidence from connectives 

have an information structure which can be described in terms of focus, background and
topic (see Jacobs 1988, 2001; Klein & von Stutterheim 1992; Büring 2006; Moroni 2006).
Subject, predicate, adjunct etc. are categories that cannot be directly related to the com-
municative and interactional goals of language users. Focus, background and topic are
immediately related to this kind of goals and are therefore more appropriate concepts for
analysing discourse structure.
As long as we have no independent and comprehensive theory of discourse struc-
ture, we should refrain from formulating hypotheses about possible parallelisms be-
tween discourse and syntax or other subsystems of language. The examples of text anal-
ysis given within different research paradigms (e.g., Mann & Thompson 1988; Klein &
von Stutterheim 1992; Asher & Vieu 2005) all indicate that both main and secondary
discourse units can, in principle, be encoded by both main and secondary syntactic
expressions and that both hierarchical and non-hierarchical discourse connections can
be encoded by both coordinative and subordinative syntactic connections.
The relation between syntax, semantics and discourse should be viewed as a mat-
ter of rhetorical options to be taken by speakers and writers in individual acts of en-
coding. Languages may fix some of these options in their structure and leave only
some of them open to the speakers’ preference, but it seems very improbable that any
language should generally identify discourse hierarchy with syntactic subordination
and/or with semantic asymmetry. Most discourse connections, at any rate, could not
be captured by such a deterministic system: all relations that are encoded referentially,
across sentence boundaries, by adverbs and proforms. The connecting function of
those elements, as we have seen in section 2, is generally independent of the syntactic
distinction between coordination and subordination.

5.  Conclusion

The concepts of subordination and coordination in language are ultimately termino-

logical metaphors. The notional content of such metaphors depends very much on the
nature of the domain to which they are applied. Syntax and discourse are domains of
quite different nature. While discourse structure is about presenting information in
order to achieve communicative goals, syntactic structure is about arranging formal
expressions in order to facilitate parsing. An appropriate account of discourse struc-
ture should preferably be given in a terminology sufficiently different from the one
used in syntax, in order to avoid misconceptions.
The arguments presented in this paper suggest that syntactic hierarchy should not
be considered a general model for the conceptualization of discourse hierarchy, nor
vice versa. The data from the study of connectives do not support a view that consid-
ers syntactic subordination as a generally specialized means for encoding rhetorical
hierarchy. It seems slightly more plausible to expect that syntactic coordination should
be a preferable means for encoding non-hierarchical discourse connections. But as we
 Hardarik Blühdorn

have seen, syntactic coordination can (or must) in many cases receive an asymmetric
semantic interpretation, and the literature contains numerous examples in which syn-
tactic coordination encodes hierarchical discourse relations.
The final conclusion is therefore that we should be careful in assuming too many
parallels between syntax and discourse structure. Syntax seems to be designed in such
a way that the structural variants it offers can be employed very flexibly to encode the
structural variants of discourse. This flexibility in the relationship between syntax and
discourse seems to be one of the features of natural languages that render them suit-
able for infinite use in communication.

References

Asher, Nicholas & Vieu, Laure. 2005. Subordinating and coordinating discourse relations. Lingua
115: 591–610.
Bierwisch, Manfred. 2003. Heads, complements, adjuncts: Projection and saturation. In Modify-
ing adjuncts, Ewald Lang, Claudia Maienborn & Cathrine Fabricius-Hansen (eds), 113–159.
Berlin: Mouton de Gruyter.
Blakemore, Diane & Carston, Robyn. 2005. The pragmatics of sentential coordination with and.
Lingua 115: 569–589.
Blühdorn, Hardarik. 2003. Zur Semantik der Konjunktion als. Paradigmatische und syntagma-
tische Aspekte. Linguistik online 13: 11–53, URL: http://www.linguistik-online.de/13_01/
bluehdorn.html.
Blühdorn, Hardarik. 2004. Die Konjunktionen nachdem und bevor. In Brücken schlagen:
Grundlagen der Konnektorensemantik, Hardarik Blühdorn, Eva Breindl & Ulrich Hermann
Waßner (eds), 185–211. Berlin: Walter de Gruyter.
Blühdorn, Hardarik. 2005. Zur Semantik kausaler Satzverbindungen: Integration, Fokussierung,
Definitheit und modale Umgebung. Studi Linguistici e Filologici Online. Rivista Telematica
del Dipartimento di Linguistica dell‘Università di Pisa (SLiFO) 3.2: 311–338, URL: http://
www.humnet.unipi.it/slifo/2005vol2/Bluh­dorn3.2.pdf.
Blühdorn, Hardarik. 2007a. Epistemische Lesarten von Satzkonnektoren – Wie sie zustande
kommen und wie man sie erkennt. In Semantik vs. Pragmatik oder Semantik und Pragma-
tik?, Inge Pohl (ed.). Frankfurt: Peter Lang (in print).
Blühdorn, Hardarik. 2007b. Kausale Satzverknüpfungen im Deutschen. Pandaemonium Germani-
cum 10: 253–282. São Paulo: Humanitas, FFLCH/USP.
Blühdorn, Hardarik, Breindl, Eva & Waßner, Ulrich Hermann (eds). 2004. Brücken schlagen:
Grundlagen der Konnektorensemantik. Berlin: Walter de Gruyter.
Breindl, Eva. 2007a. Die Bedeutung additiver Konnektoren (und, sowohl als auch, sowie). In
Semantik vs. Pragmatik oder Semantik und Pragmatik?, Inge Pohl (ed.). Frankfurt: Peter
Lang (in print).
Breindl, Eva. 2007b. Additive Konjunktoren und Adverbien im Deutschen. In Feldergrammatik
in der Diskussion. Funktionaler Grammatikansatz in Sprachbeschreibung und Sprachvermitt-
lung, Joachim Buscha & Renate Freudenberg-Findeisen (eds), 141–164. Frankfurt: Peter
Lang.
Breindl, Eva & Waßner, Ulrich Hermann. 2006. Syndese vs. Asyndese. Konnektoren und andere
Wegweiser für die Interpretation semantischer Relationen in Texten. In Text – Verstehen:
 Subordination and coordination: Evidence from connectives 

Grammatik und darüber hinaus, Hardarik Blühdorn, Eva Breindl & Ulrich Hermann Waßner
(eds), 46–70. Berlin: Walter de Gruyter.
Büring, Daniel. 2006. Intonation und Informationsstruktur. In Text – Verstehen: Grammatik
und darüber hinaus, Hardarik Blühdorn, Eva Breindl & Ulrich Hermann Waßner (eds),
144–163. Berlin: Walter de Gruyter.
Camacho, José. 2003. The Structure of Coordination: Conjunction and Agreement Phenomena in
Spanish and Other Languages. Dordrecht: Kluwer.
Chafe, Wallace. 1988. Linking intonation units in spoken English. In Clause Combining in
Grammar and Discourse, John Haiman & Sandra A. Thompson (eds), 1–27. Amsterdam:
John Benjamins.
Cristofaro, Sonia. 2003. Subordination. Oxford: OUP.
Dik, Simon C. 1972. Coordination: Its Implications for the Theory of General Linguistics, 2nd edn.
Amsterdam: North-Holland.
Duden. 2005. Die Grammatik [Duden Vol. 4], Dudenredaktion (ed.). 7th edn. Mannheim:
Dudenverlag.
Eisenberg, Peter. 2004. Grundriss der deutschen Grammatik, Vol. 2: Der Satz. 2nd edn. Stuttgart:
Metzler.
Féry, Caroline. 2006. Laute und leise Prosodie. In Text – Verstehen: Grammatik und darüber
hinaus, Hardarik Blühdorn, Eva Breindl & Ulrich Hermann Waßner (eds), 164–183. Ber-
lin: Walter de Gruyter.
Foley, William A. & Van Valin Jr., Robert D. 1984. Functional Syntax and Universal Grammar.
Cambridge: CUP.
Grice, Paul. 1981. Presupposition and conversational implicature. In Radical Pragmatics, Peter
Cole (ed.), 183–198. New York NY: Academic Press.
Günthner, Susanne. 1996. From subordination to coordination? Verb-second position in Ger-
man causal and concessive constructions. [Arbeitspapier Nr. 75]. Konstanz: Fachgruppe
Sprachwissenschaft der Universität Konstanz.
Halliday, Michael A.K. & Hasan, Ruqaiya. 1976. Cohesion in English. London: Longman.
Haspelmath, Martin. 2004. Coordinating constructions: An overwiew. In Coordinating Con-
structions, Martin Haspelmath (ed.), 3–39. Amsterdam: John Benjamins.
Haspelmath, Martin. (ed.). 2004. Coordinating Constructions. Amsterdam: John Benjamins.
Haumann, Dagmar. 1997. The Syntax of Subordination. Tübingen: Max Niemeyer.
Heath, Jeffrey. 2004. Coordination: An adaptionist view. In Coordinating Constructions, Martin
Haspelmath (ed.), 67–88. Amsterdam: John Benjamins.
Hopper, Paul J. & Thompson, Sandra A. 1984. The discourse basis of lexical categories in univer-
sal grammar. Language 60: 703–752.
Jacobs, Joachim. 1988. Fokus-Hintergrund-Gliederung und Grammatik. In Intonationsfor-
schungen, Altmann, Hans (ed.), 89–134. Tübingen: Max Niemeyer.
Jacobs, Joachim. 2001. The dimensions of topic-comment. Linguistics 39: 641–681.
Johannessen, Janne Bondi. 1998. Coordination. Oxford: OUP.
Klein, Wolfgang & von Stutterheim, Christiane. 1992. Textstruktur und referentielle Bewegung.
Zeitschrift für Literaturwissenschaft und Linguistik 86: 67–92.
Knott, Alistair & Dale, Robert. 1994. Using linguistic phenomena to motivate a set of coherence
relations. Discourse Processes 18: 35–62.
Knott, Alistair, Sanders, Ted & Oberlander, Jon. (eds). 2001. Special issue: Levels of Representa-
tion in Discourse Relations. Cognitive Linguistics 12: 197–332.
Kortmann, Bernd. 1996. Adverbial Subordination: A Typology and History of Adverbial Subordi-
nators Based on European Languages. Berlin: Mouton de Gruyter.
 Hardarik Blühdorn

Lang, Ewald. 1984. The Semantics of Coordination [Studies in Language Companion Series 9].
Amsterdam: John Benjamins.
Langacker, Ronald W. 1987. Foundations of Cognitive Grammar. Vol. 1: Theoretical Prerequisites.
Stanford CA: Stanford University Press.
Langacker, Ronald W. 1991. Foundations of Cognitive Grammar. Vol. 2: Descriptive Application.
Stanford CA: Stanford University Press.
Lefèvre, Michel (ed.). 2000. Subordination in Syntax, Semantik und Textlinguistik [Eurogerman-
istik 15]. Tübingen: Stauffenburg.
Lohnstein, Horst. 2004. Variable und invariante Strukturmerkmale von Satzkonnektoren. In
Brücken schlagen: Grundlagen der Konnektorensemantik, Hardarik Blühdorn, Eva Breindl
& Ulrich Hermann Waßner (eds), 137–160. Berlin: Walter de Gruyter.
Lyons, John. 1977. Semantics. 2 vols. Cambridge: CUP.
Mann, William C. & Thompson, Sandra A. 1988. Rhetorical Structure Theory: Toward a func-
tional theory of text organization. Text 8(3): 243–281.
Matthiessen, Christian & Thompson, Sandra A. 1988. The structure of discourse and ‚subor-
dination‘. In Clause Combining in Grammar and Discourse, John Haiman & Sandra A.
Thompson (eds), 275–329. Amsterdam: John Benjamins.
Moroni, Manuela. 2006. Modalpartikeln an der Schnittstelle zwischen Syntax und Prosodie.
Doctoral thesis, Università degli studi di Verona, Facoltà di lingue e letterature straniere.
Muller, Claude. (ed.). 1996. Dépendance et intégration syntaxique: Subordination, coordination,
connexion [Linguistische Arbeiten 351]. Tübingen: Max Niemeyer.
O’Dowd, Elizabeth. 1992. The syntactic metaphor of subordination: A typological study. Lingua
86: 46–80.
Osborne, Timothy. 2003. The Third Dimension: A Dependency Grammar Theory of Coordina-
tion for English and German. PhD dissertation, Pennsylvania State University. Ann Arbor
MI: UMI.
Osborne, Timothy. 2006. Shared material and grammar: Toward a dependency grammar theory
of non-gapping coordination for English and German. Zeitschrift für Sprachwissenschaft
25: 39–93.
Pasch, Renate. 2000. Vorschlag für eine Neuordnung der “subordinierenden” Konjunktionen
des Deutschen. In Subordination in Syntax, Semantik und Textlinguistik, Michel Lefèvre
(ed.), 23–32. Tübingen: Stauffenburg.
Pasch, Renate, Brauße, Ursula, Breindl, Eva & Waßner, Ulrich Hermann. 2003. Handbuch
der deutschen Konnektoren: Linguistische Grundlagen der Beschreibung und syntaktische
Merkmale der deutschen Satzverknüpfer (Konjunktionen, Satzadverbien und Partikeln).
Berlin: Walter de Gruyter.
Pasch, Renate. 2004. Das “Handbuch der deutschen Konnektoren”. In Brücken schlagen: Grundla-
gen der Konnektorensemantik, Hardarik Blühdorn, Eva Breindl & Ulrich Hermann Waßner
(eds), 11–44. Berlin: Walter de Gruyter.
Peters, Jörg. 2005. Intonation. In Duden, Vol. 4: Die Grammatik, Dudenredaktion (ed.), 95–128.
Mannheim: Dudenverlag.
Polanyi, Livia. 1988. A formal model of the structure of discourse. Journal of Pragmatics 12:
601–638.
Posner, Roland. 1980. Semantics and pragmatics of sentence connectives in natural language.
In Speech Act Theory and Pragmatics, John R. Searle, Ferenc Kiefer & Manfred Bierwisch
(eds), 169–203. Dordrecht: D. Reidel.
 Subordination and coordination: Evidence from connectives 

Quirk, Randolph, Greenbaum, Sidney, Leech, Geoffrey & Svartvik, Jan. 1985. A Comprehensive
Grammar of the English Language. London: Longman.
Schecker, Michael. 2000. Zur kommunikativ-funktionalen Leistung von Nebensätzen. In Subor-
dination in Syntax, Semantik und Textlinguistik, Michel Lefèvre (ed.), 115–122. Tübingen:
Stauffenburg.
Selting, Margret. 1995. Prosodie im Gespräch. Tübingen: Max Niemeyer.
Shopen, Timothy (ed.). 1985. Language Typology and Syntactic Description, Vol. II: Complex
Constructions. Cambridge: CUP.
Taboada, Maite & Mann, William C. 2006. Rhetorical Structure Theory: looking back and mov-
ing ahead. Discourse Studies 8(3): 423–459.
Thompson, Sandra A. & Longacre, Robert E. 1985. Adverbial clauses. In Language Typology
and Syntactic Description, Vol. II: Complex Constructions, Timothy Shopen (ed.), 171–234.
Cambridge: CUP.
Van Valin Jr., Robert D. & LaPolla, Randy J. 1997. Syntax: Structure, Meaning and Function.
Cambridge: CUP.
Webber, Bonnie, Stone, Matthew, Joshi, Aravind & Knott, Alistair. 2003. Anaphora and dis-
course structure. Computational Linguistics 29: 545–587.
Wegener, Heide. 2000. Asymmetrische Koordination in subordinierten Strukturen. In Subor-
dination in Syntax, Semantik und Textlinguistik, Michel Lefèvre (ed.), 33–44. Tübingen:
Stauffenburg.
Wesche, Birgit. 1995. Symmetric Coordination: An Alternative Theory of Phrase Structure. Tübin-
gen: Max Niemeyer.
Wiese, Bernd. 1980. Grundprobleme der Koordination. Lingua 51: 17–44.
Zifonun, Gisela. 2001. Grammatik des Deutschen im europäischen Vergleich: Der Relativsatz.
Mannheim: Institut für Deutsche Sprache.
part ii

Cross-linguistic approaches
 A corpus-based perspective on clause
linking patterns in English, French and Dutch*

Christelle Cosme
Université catholique de Louvain

The two traditional clause linking devices, viz. coordination and

subordination, are particularly interesting to study from a cross-linguistic
perspective. It has often been claimed that principles of information
packaging, of which clause linking forms an integral part, are – at least
partly – language-specific. Thus, while some languages favour a hierarchical
discourse information structure making intensive use of subordinating
devices, others more readily employ an incremental discourse information
structure, favouring coordinating devices. Close inspection of the contrastive
literature reveals that English, French and Dutch exhibit preferred clause
linking patterns. The claims found in the contrastive literature are tested
empirically on the basis of multilingual corpora. Clause linking, it is
ultimately argued, is best treated as a gradient phenomenon.

Keywords: clause linking, coordination, subordination, stylistic preferences,

multilingual corpora

1.  Introduction

Clause linkage strategies have traditionally been divided into two basic types, i.e.,
coordination and subordination. While subordination organizes discourse hierarchi-
cally, coordinating relations are believed to structure discourse in a non-hierarchical,
more incremental fashion (see Asher & Vieu’s (2005) treatment within Segmented
Discourse Representation Theory and Mann & Thompson’s (1988) Rhetorical Struc-
ture Theory). The two clause linking devices are particularly interesting to study from
a cross-linguistic perspective. Principles of discourse organization or information

*  This study has been made possible thanks to the support of the Belgian Fund for Scientific
Research (Research Fellowship), which I gratefully acknowledge. My thanks also go to Sylviane
Granger for her insightful comments on the first draft of the article. Last but not least, I wish to
thank the three anonymous reviewers for providing useful feedback.
 Christelle Cosme

packaging, of which clause linking forms an integral part, have indeed been claimed
to be – at least partly – language-specific (e.g., Hasselgård et al. 2002; Moder & Mar-
tinovic-Zic 2004) (Section 2).
The present study takes a cross-linguistic view on sentence-level coordination and
subordination in three languages, namely English, French and Dutch. The aim is to
test a number of intuition-based claims made in the contrastive literature (Section 3).
Such claims will be tested empirically on the basis of multilingual corpus data (Section 4).
Another implicit aim of the research presented in this article is to gain new insights
into the relationship between coordination and subordination. It will be argued, in line
with among others Lehmann (1988) and Cristofaro (2003), that clause linking is best
treated as a gradient phenomenon (Section 5). In other words, we hold the view that
coordination and subordination should be regarded as the two prototypical poles of a
clause linking continuum allowing for a number of in-between constructions.

2.  Information packaging across languages

As witnessed by a number of recently published titles such as Discourse across languages

and cultures (Moder & Martinovic-Zic 2004) or Information structure in a cross-linguistic
perspective (Hasselgård et al. 2002), there is clear evidence that discourse organization
differs across languages. In other words, principles of information packaging are – at
least partly – language-specific. It appears that the language-specific ways of structuring
information are manifest at various layers of discourse organization.
To begin with, languages may differ with regard to focus assignment, thematic struc-
ture and word order patterns. Comparing English and German, Doherty (1999: 135)
argues that German, unlike English, permits the identification of syntactic functions
on a strictly morphological basis thanks to its case distinctions. As a result of its non-
configurational character, German allows for greater variability in word order than
English. By comparison, the latter, being a typically configurational language, requires
specific structural configurations to distinguish subjects from objects. English subjects
are therefore mainly fixed to the position before the verb (i.e., the initial position). As
a result, while English sentences opening with an adverbial are marked, beginning a
sentence with an adverbial is the unmarked pattern in German. Thus, a word-for-word
translation of the German sentence given in example (1), though it is strictly speaking
grammatically acceptable, would not be very likely in English. A translation placing
the adverbial in sentence-final position appears to be stylistically more appropriate, as
it adheres more tightly to the preferences of the target language (i.e., English).
(1) Zehn Jahre nach dem Röhrenversuch konnten Jaffe und Ken die
Beobachtungen bestätigen.
Jaffe and Ken confirmed the observations ten years after the tube experiment.
[Doherty 1999: 125]
 Clause linking in English, French and Dutch 

Language-specific preferences have also been shown to be manifest at the level of

informational density. It turns out that what is expressed by a phrase in one language
may be expressed by a clause in another language, and vice versa. German, for in-
stance, is generally assumed to have a more ‘nominal’ (as opposed to ‘sentential’) style
than English and Norwegian (e.g., Solfjeld 1996). In other words, German is char-
acteristically thought to make more intensive use of phrasal constituents than Eng-
lish and Norwegian, the latter favouring clausal constituents. Accordingly, translating
from English into German frequently triggers so-called ‘phrasal reduction’ (Doherty
1998: 239), which can be described as a phenomenon whereby a clause in the source
language is turned into a phrase in the target language (see also Doherty 1999 for a
similar study). Example (2) may serve as illustration.
(2) Auch werden bei einem verlangsamten Zellwachstum nur geringe Mengen an
natürlichem Interferon gefunden.
Also when cell growth is slowing down, only small amounts of naturally
produced interferon are found.
 [Doherty 1999: 116]

Last but not least – and this is of special interest in the present article, languages are
also said to differ with regard to their preferred clause linking patterns. Fabricius-Hansen
(1998), for instance, observes that some languages (like German) favour a hierarchical
(i.e., vertical) discourse information structure making intensive use of hypotactic (i.e.,
subordinating) devices, while other languages (like English and, to an even larger ex-
tent, Norwegian) prefer an incremental (i.e., horizontal or linear) discourse information
structure, typically “organiz[ing] the discourse information structure in smaller chunks
that can each be processed easily, with the result that a representation of the whole dis-
course is built up very gradually, step by step” (Fabricius-Hansen 1998: 203). Coordinat-
ing or paratactic constructions are typical ways of structuring discourse in an incremen-
tal fashion. Translating from a hierarchical language (like German) into a relatively more
incremental language (like English or Norwegian) thus, as a rule, involves some degree
of so-called ‘information splitting’, i.e., a shift from a high degree of syntactic complex-
ity to a less complex, more paratactical style (Fabricius-Hansen 1996: 521). Example (3)
shows how a typical complex sentence in German is turned into two independent sen-
tences divided by a colon in English, and how a German subordinating construction
(i.e., in this case, a relative clause) is rendered by means of a more horizontal, incremen-
tal construction (i.e., ‘and’-coordination) in English.
(3) Wenn die unter jener anderen Landungsbrücke beheimateten Fische mein Kommen
bemerkten, kam erschreckend
When the fish under the other landing-stage noticed my coming, a startling
thing happened:
aus dem Dunkel unter dem Steg hervor ein mehrere Meter breites und fast ebenso
hohes und viele Male längeres, auf dem besonnten Grund einen tiefschwarzen
Schatten werfendes Ungeheuer auf mich zugeschossen,
 Christelle Cosme

from the darkness of the stage emerged a monster several yards high and wide,
and many times this length, throwing a deep black shadow on the sunlit sea
bottom as it shot towards me,
das sich erst beim nahen Herankommen in eine Unzahl freundlicher Purpurmäuler
auflöste.
and only as it drew very near did it become resolved into a crowd of friendly
grunts and snappers.
[Fabricius-Hansen 1999: 135]

The above-mentioned language-specific preferences in terms of information packag-

ing have an obvious impact on translation. A good translation has to follow the stylis-
tic norms and preferences of the target language, which sometimes implies a certain
amount of information restructuring. As argued by Chuquet & Paillard (1987: 135),
following the language-specific strategies of information distribution is of paramount
importance in translation, as it contributes substantially – to an even larger extent than
lexical and grammatical choices – to the ‘natural-soundingness’ of a translation.

3.  Close-up on English, French and Dutch

3.1  Review of the contrastive literature

A review of the contrastive literature dealing with English, French and Dutch brings out
major contrasts in the way the three languages typically package information. It should
be noted that the contrasts surveyed are located at the level of stylistic preferences. As
rightly pointed out by Doherty (2005: 1), the cross-linguistic differences observed are
thus “rarely grammaticalized, but determine only preferences in the distribution of in-
formation.” It will therefore be argued that, though devices like coordination and sub-
ordination are universal, they are employed with varying frequencies across languages.
Thus, to convey a given content x, some languages may show a preference for device A
(e.g., coordination), while others may favour device B (e.g., subordination).

3.1.1  English vs. French

English and French have been claimed to diverge markedly in the way they prefer to
package information in discourse. The contrastive literature abounds in claims, such
as the following by Chuquet & Paillard (1987: 151), that the explicit linking of clauses
tends to be realized through coordination in English and through subordination in
French:
Lorsque les relations entre les procès sont explicitées dans les deux langues, il est
fréquent que ce soit sous forme de coordination en anglais, de subordination en
français. On retrouve ici une des manifestations de la plus grande actualisation
des procès en anglais […], marquée par l’emploi de formes verbales principales
pour désigner des procès repères.
 Clause linking in English, French and Dutch 

Throughout the contrastive literature the term ‘coordination’ is taken in its very broad
sense, as referring not only to explicit coordination by means of a conjunction (e.g.,
‘and’), but also to implicit (so-called ‘asyndetic’) coordination by means of a punctua-
tion mark (i.e., a comma, a semi-colon, a colon, or even a full stop). Since it presents
the major advantage of allowing for the inclusion of sequences of two independent
sentences in the analysis, this broad definition of coordination will also be adopted in
the present article. The term ‘coordination’ as used in the present study thus broadly
overlaps with the concept of ‘parataxis’ (see e.g., Martin 1992; Halliday & Matthiessen
2004). Two independent sentences strung together in sequence by means of a punc-
tuation mark (i.e., a comma, a colon, a semi-colon, and even a full stop) will therefore
be considered as coordination on an equal footing with explicit linking of clauses by
means of a coordinating conjunction.1
As illustrated in example (4), the different preferences of English and French are
particularly manifest when the relationship holding between the clauses is of a tempo-
ral nature. Thus, the English version of example (4), which contains four coordinated
predicates in chronological sequence (see bold), is rendered in French by means of a
blend of coordinating and subordinating devices.
(4) He stepped aside for me to enter the house, took my overcoat and hat,
guided me to a room on the second floor – Gantvoort’s library – and
left me.
Il s’ écarta pour me laisser pénétrer dans la maison et après m’avoir débarrassé de
mon pardessus et de mon chapeau, il me fit monter au premier et m’indiqua une
pièce (la bibliothèque de Gantvoort) où il me laissa seul.
[Chuquet & Paillard 1987: 151]

In other words, English and French differ in the way they relate events in time. English
seems to have a strong tendency to view events on a temporal axis (i.e., to present them
in a chronological sequence by means of coordination), while French is presented as
typically linking events more tightly together, thus making more intensive use of sub-
ordinating devices (Chuquet & Paillard 1987: 151).

3.1.2  English vs. Dutch

The English-Dutch contrastive literature is much less abundant than that dealing with
English and French. There is, however, some evidence that English and Dutch have
preferred ‘sentencing styles’ – to use Hannay’s (1997) words – that differ quite strikingly.
As suggested by Hannay & Mackenzie (1996: 102), the syntactic devices used in the

.  This broad definition of ‘coordination’ blurs the functional distinction between explicit co-
ordination and (asyndetic) parataxis (see e.g., Ramm & Solfjeld, this volume). The decision to
adopt such a broad definition of ‘coordination’ has been primarily motivated by its close adher-
ence to the definition assumed in the English-French and English-Dutch contrastive literature.
 Christelle Cosme

two languages differ in terms of availability, frequency and distribution, and function:
“Some devices are not readily available in Dutch; others are more flexible in English
than in Dutch; yet others just operate in different ways than their Dutch counterparts.”
Generally speaking, the contrastive literature presents English as being syntacti-
cally more complex than Dutch. Hannay and Mackenzie, for instance, claim that:
English has a tendency in argued text to use more sentences than Dutch which are
long (containing say over 30 words) and syntactically complex. It is for instance
not at all uncommon in English for sentences to have more than one combined
clause; indeed, sometimes a combined clause itself functions as the host clause for
another combined clause.  (Hannay & Mackenzie 1996: 118–119)

Sentence length is a first indicator of syntactic complexity. According to Hannay &

Mackenzie (1996: 44), “[c]omparison of similar English and Dutch texts has shown
that sentences in English, at least in sustained argued prose, tend to be on average
somewhat longer than their Dutch counterparts.”2 The greater syntactic complexity
of English, it is claimed, also manifests itself in a more intensive use of subordinate
clauses. In her investigation of Dutch novels translated into English, Vanderauwera
(1985: 78) notes that Dutch favours short and incomplete sentences and typically uses
coordinate rather than subordinate clauses. In the English translation, the author ob-
serves, textual organization is frequently adjusted to meet the English norms, as is
clearly the case in example (5).
(5) Hij gaat naar boven, naar de huiskamer, ziet dat het nu helemaal donker
is, staart nietsziend naar buiten, voelt zich eenzaam, weet nu al in welk
restaurant hij straks zal gaan eten. Hij trekt zijn jas aan en gaat naar een bar
in de benedenstad om kranten te kopen, iets te drinken en misschien op de
pinballmachine te spelen.
Upstairs he goes to the sitting-room; noticing that it is already dark, he gazes
outside not seeing a thing. Now he feels lonely, but already he knows at what
restaurant he will be eating presently. When he has put on his coat he steps out
to make his way to the bar downtown where he will buy newspapers, have a
drink and probably end up playing the pin-table.
 [Vanderauwera 1985: 101–102]

.  Hannay & Mackenzie (1996: 44) see two underlying reasons for this phenomenon. First,
Dutch is more tolerant than English of incomplete sentences. As a result, “what would in Dutch
be expressed by an incomplete sentence appears in English as a full sentence” or “the informa-
tion presented in Dutch as a separate but incomplete sentence is in English integrated into the
preceding sentence, making the latter correspondingly longer” (Hannay & Mackenzie 1996: 44).
Another reason for the higher sentence length of English texts is the frequent use English makes
of participial constructions, for which there is no real equivalent in Dutch.
 Clause linking in English, French and Dutch 

In relation to such typical textual adjustments, Vanderauwera (1985: 101) points out
that “[t]he simple and jerky juxtaposition of syntactically identical and independent
clauses suggestive of almost cinematic observation […] is matched here by more syn-
tactically varied and “better” linked narrative portions (by use of such means as sub-
ordinate or nonfinite clauses).” It thus appears that while the English text types are
characterized by a combining style, the Dutch text types clearly demonstrate a more
chopping style (Hannay 1997: 248).

3.2  Research hypothesis

The review has brought out major contrasts in the strategies pursued by English, French
and Dutch to link clauses. The three languages appear to be located on a continuum
ranging from incrementality to hierarchy (see Fabricius-Hansen 1996: 558 for defini-
tions of these two concepts). The two ways of organizing discourse (i.e., incrementality
vs. hierarchy) are accurately reflected, at the linguistic level, in the two clause linking
devices, i.e., in the distinction between coordination and subordination. An incremen-
tally organized text makes use of a relatively large number of coordinating construc-
tions and, respectively, a relatively low number of subordinating devices; in a hierar-
chically organized text, by comparison, there is a clear predominance of subordinating
structures and, conversely, a relatively low number of coordinating constructions.3
The hypothesis – henceforth referred to as the ‘Dependency Hypothesis’ – thus
reads as follows and may be graphically represented as in Figure 1:

INCREMENTALITY HIERARCHY
Parataxis Hypotaxis
(coordination) (subordination)

DUTCH ENGLISH FRENCH

Figure 1.  Dependency Hypothesis.

.  The distinction between incremental style and hierarchical style coincides with Chafe’s
(1982) distinction between fragmentation and integration. A fragmented style, Chafe (1982:
38) argues, is characterized by a more intensive use of coordinating structures, whereas an inte-
grated style typically makes greater use of subordinating constructions (e.g., participial clauses,
complement clauses or relative clauses).
 Christelle Cosme

It should be noted that the scale represented in Figure 1 is not to be taken in absolute
terms. Rather, incrementality and hierarchy are to be considered as relative concepts.
In other words, the hypothesis posits that French, in comparison to English, favours a
more hierarchical style and that Dutch, in comparison to English, has a typically more
incremental writing style.
The different positions of the three languages on the dependency cline represented
in Figure 1 have a clear impact on the translation process. It may indeed be postulated
that shifts in clause linking will take place. The following shifts may be predicted from
the Dependency Hypothesis:

a. Majority of shifts from coordination (in the broad sense, including mere juxta-
position by means of punctuation marks) to subordination in translations from
Dutch to English, from English to French and from Dutch to French;
b. Majority of reverse shifts (i.e., from subordination to coordination) in translations
from French to English, from English to Dutch and from French to Dutch.

These predictions will be tested against translation corpus data for three of the six
possible translation directions, namely (i) from English to French, (ii) from French to
English and (iii) from English to Dutch.
Before engaging in the corpus analysis proper, a few words about Vinay & Darbel-
net’s (1995: 15) distinction between so-called ‘options’ and ‘servitudes’ are in order, as
the distinction is highly relevant to the research presented here. It emerges that most
of the shifts under study are cases of options rather than servitudes. In other words,
the translator could most of the time have maintained a congruent construction in the
target language, but chose not do so probably because s/he was aware of the stylistic
norms and preferences of the target language. Thus, though translating the English
‘and’-coordination of example (6) with ‘et’-coordination in French would have yielded
a grammatically correct sentence in French, a translation by means of a subordinate
construction is stylistically more appropriate.
(6) In the course of years several ships had stranded and they now helped to
prop up the bank, […].
Plusieurs navires, qui s’étaient échoués au cours des années, contribuaient
maintenant à étayer la rive, […].
[pleci_news]

Though most of the shifts under investigation are optional, a number of them are the
result of language servitudes or language-systemic constraints. Thus, to translate the
English sentence given in (7), the translator had no choice but to operate a shift. A
word-for-word translation would have been grammatically unacceptable here, as ad-
verbial –ing constructions do not have any direct equivalent in Dutch.
(7) Ralph spoke to himself, sounding the bass strings of delight.
Ralph praatte in zichzelf en bracht de bassnaar van verrukking in trilling.
[pleci_news]
 Clause linking in English, French and Dutch 

4.  Corpus study

This section presents the findings of a large-scale corpus study, conducted in an attempt
to test and refine the Dependency Hypothesis on the basis of solid empirical data. The
reader is referred to Cosme (2004, 2006a) for an overview of preliminary findings con-
cerning the contrast in clause linking between English, French and Dutch.

4.1  Data and methodology

The corpus analysis conducted here is two-fold. To start with, a number of preliminary
observations are made on the basis of (under ‘corpus’) data (see Section 4.2.1), i.e.,
original texts in two or more languages that are matched by criteria such as time of
composition, register, etc. (see e.g., Johansson & Hasselgård 1999 and Granger 2003 for
a definition of comparable corpora). Concretely, the comparable data come from the
TRILLED corpus (TRILingual Louvain EDitorials corpus). Currently being compiled
by researchers at the Centre for English Corpus Linguistics (University of Louvain,
Belgium), the corpus is made up of original newspaper editorials in English, French
and Dutch.4 At its present state, it contains ca. 1 million words of English editorials, ca.
750,000 words of French editorials and ca. 500,000 words of Dutch editorials. For the
purpose of the present research, we have extracted a random sample of 300 sentences
in each language. A careful analysis of these 300 sentences per language, it is believed,
should provide useful insights into the frequency of the two clause linking devices (i.e.,
coordination and subordination) in the three languages.
In a subsequent step, use is made of a trilingual (under ‘corpus’) corpus called PLE-
CI (Poitiers-Louvain Échange de Corpus Informatisés) in order to retrieve the translation
shifts of interest (see Section 4.2.2). The corpus, which is currently being compiled by
the Centre for English Corpus Linguistics (University of Louvain, Belgium) and the
University of Poitiers (France), is made up of fiction and newspaper articles. At its pres-
ent state, it comprises ca. 600,000 words of English texts translated into French, ca.
700,000 words of French texts translated into English and ca. 200,000 words of English
texts and their respective Dutch translations. Because of the difficulties connected with
the automatic retrieval of the shifts under investigation, it was decided to use only a
sample of PLECI, namely ca. 50,000 words from the English ⇒ French component, ca.
50,000 words from the opposite direction (i.e., French ⇒ English) and ca. 25,000 words
from the English ⇒ Dutch component. Note that while fiction and newspaper language

.  Comparability in the TRILLED corpus is ensured by selecting editorials from quality news-
papers in the three languages. English editorials are taken from The Guardian, The Independent,
The Times, The Observer, The Sunday Telegraph, The Daily Telegraph and The Economist; French
editorials from Le Monde, Le Figaro and Libération; and Dutch editorials from De Volkskrant,
NRC Handelsblad, Trouw, Het Parool, Utrechts Nieuwsblad and Haagsche Courant. Moreover, the
editorials were collected during the same period of time.
 Christelle Cosme

are both equally represented in the English-French sub-corpus and in the French-Eng-
lish sub-corpus, the same does not apply to the English-Dutch sub-corpus, which is – at
its present state – restricted to fiction.5 In order to reinforce the English ⇒ Dutch part
of the study, we have also searched the Oslo Multilingual Corpus (OMC).6
The present study thus combines the respective strengths of comparable and transla-
tion data. Johansson and Hasselgård (1999: 146) were among the first to emphasize the
benefit of combining comparable and translation corpora, thus taking advantage of the
specific merits of both types. The view held in this article is in accordance with Johans-
son & Hasselgård’s (1999: 146) principle. The basic assumption is that comparable and
translation data are not mutually exclusive. Rather the contrary: they can supplement one
another. The two types of data will therefore be necessary to gain a deeper understanding
of the relationship between the three languages under investigation in our study.
While lexis lends itself perfectly to corpus investigations, the same does not always
hold for syntactic phenomena, the chief reason being that syntactic structures are usu-
ally more difficult to extract from a corpus than simple words (Gilquin 2002: 183).
Because of the difficulty connected with the automatic retrieval of macro-linguistic
patterns, most contrastive studies dealing with syntax or discourse are still to a large
extent dependent on intuition, or resort to corpus examples only to illustrate or sup-
port the claims being made. It must, however, be acknowledged that, though it is true
that lexis has the lion’s share in present-day contrastive linguistics, much effort has
been made in recent years to bridge the gap between lexical contrastive studies, on the
one hand, and syntactic or discourse studies, on the other.7
Though it must be admitted that the use of fully automatic methods is generally
impossible in corpus-based syntactic or discourse-oriented contrastive research, the

.  The size of the English-Dutch sub-corpus has been deliberately kept to 25,000 words so
as to facilitate comparisons with the 25,000-word English-French fiction sub-corpus and the
25,000-word French-English fiction sub-corpus. The decision to include two genres (i.e., fiction
and newspaper articles) in the English-French and French-English sub-corpora has been made
on the grounds that, at a later stage, we also intend to investigate genre differences (see Cosme
2006a for additional information).

.  We hereby thank Stig Johansson and Hilde Hasselgård for allowing us to search the English-
Dutch sub-corpus of the Oslo Multilingual Corpus (http://www.hf.uio.no/ilos/OMC/) compiled
within the framework of the Språk i kontrast (Languages in Contrast) project at the University of
Oslo (http://www.hf.uio.no/forskningsprosjekter/sprik/english/).
7.  In the Scandinavian countries, for instance, a number of contrastive studies based on the
ENPC (English-Norwegian Parallel Corpus) or on the ESPC (English-Swedish Parallel Corpus)
have dealt extensively with macro-linguistic phenomena. The SPRIK project (Språk i kontrast
/ Languages in Contrast; see http://www.hf.uio.no/forskningsprosjekter/sprik/english/), for in-
stance, has given rise to a series of corpus-based syntactic and discourse-oriented contrastive
studies (e.g., Fabricius-Hansen 1998).
 Clause linking in English, French and Dutch 

view held in this article is that it is possible to automate the process to some extent.
Computerized methods are thus adopted whenever technically feasible. Thus, in order
to ease the retrieval of the structures under study, it was decided to provide both the
comparable and the translation corpus data with problem-oriented linguistic annota-
tions using an automated method. Providing greater detail on the annotation scheme
and procedure falls beyond the scope of the present article. The reader is referred to
Cosme (2003) and Cosme (2006b) for more information on the annotation of compa-
rable and translation data, respectively.

4.2  Corpus findings and hypothesis testing

4.2.1  Preliminary observations
A comparison of English, French and Dutch newspaper editorials provides some use-
ful insights into the frequency of the two clause linking strategies, i.e., coordination
and subordination, across the three languages. The present section addresses the issue
of whether English, French and Dutch employ coordination and subordination differ-
ently. The objective is to test the hypothesis that Dutch uses coordination more often
than English and French and that French, with its intensive use of subordination, is the
most syntactically complex of the three languages.

4.2.1.1  Subordination.  A look at the overall frequency of subordinate clauses – both

hypotactic and embedded ones – in the three sub-corpora already reveals something
about the different nature of Dutch. While English and French contain a roughly simi-
lar number of sub-clauses (464 and 415, respectively), Dutch uses significantly fewer
subordinate clauses (255 only). Thus, while the mean number of sub-clauses per sen-
tence (SUB/SEN) in English is essentially identical to that in French (1.55 and 1.38
respectively), it appears to be markedly lower in Dutch (0.85) (see Table 1). An analysis
of variance (ANOVA) shows that there is considerable variation in terms of SUB/SEN
across the three languages (F(2,897) = 23.29, p ≤ 0.0001).

Table 1.  Mean number of subordinate clauses per sentence (SUB/SEN)

ENGLISH FRENCH DUTCH
No. of subclauses (SUB) 464 415 255
No. of sentences (SEN) 300 300 300
Mean no. of subclauses / 1.55 1.38 0.85
sentences (SUB/SEN)

Closer inspection of the languages on a two-by-two basis – by means of a t-test

procedure – reveals, however, that the difference in SUB/SEN is significant between
English and Dutch (t(535) = – 7.19, p = 0.0001) and between French and Dutch
 Christelle Cosme

(t(505) = – 5.14, p = 0.0001) only, hence not between English and French (t(592) =
1.38, p = 0.17).8
The corpus sentences may also be classified according to the number of subordi-
nate clauses they contain. For the purpose of the present research, a three-fold division
has been developed. Sentences containing no sub-clause are referred to as ‘zero-sub
sentences’; those that contain one single subordinate clause are labelled ‘mono-sub
sentences’ and the term ‘multi-sub sentences’ is used for sentences containing two or
more subordinate clauses. The exact breakdown of zero-sub vs. mono-sub vs. multi-
sub sentences in all three sub-corpora is summarized in Figure 2.

160
133 135
140

120 114
102 103
100 ENGLISH
82 83 84
80 FRENCH
64
DUTCH
60

40

20

0
Zero-sub Mono-sub Multi-sub

Figure 2.  Zero-sub vs. mono-sub vs. multi-sub sentences.

It emerges from Figure 2 that Dutch exhibits a larger number of zero-sub and mono-
sub sentences, and conversely, a much lower proportion of multi-sub sentences than
English and French. The difference between English and French, on the one hand, and
Dutch, on the other, as to their use of zero-sub vs. mono-sub vs. multi-sub sentences
proves highly significant (χ² = 40.82, df = 4, p ≤ 0.001). As for the difference between
English and French, it is again only slightly (i.e., not significantly) marked as regards
their respective use of multi-sub and zero-sub sentences, English using slightly more
multi-sub sentences, and respectively slightly fewer zero-sub sentences, than French.
This preliminary finding only partially confirms the Dependency Hypothesis.
While the corpus study clearly shows that Dutch employs fewer subordinating devices

.  Tests measuring the heterogeneity of variance have been conducted for each language pair.
The value of p being always below 0.20 (considered the threshold in such tests), variances have
to be considered unequal in all three sub-corpora. The results of the t-tests were thus considered
accordingly.
 Clause linking in English, French and Dutch 

than English and French (thereby positioning Dutch towards the ‘least hierarchical’,
and hence ‘most incremental’, end of the spectrum (see Figure 1)), it makes no strong
claims about the English-French contrast. French does not seem to be more hierar-
chical than English in any respect. On the contrary, as witnessed by its higher mean
number of sub-clauses per sentence (see Table 1) and its larger number of multi-sub
sentences (see Figure 2), English even seems to be using slightly – though not sig-
nificantly – more subordinating constructions than French. The role of the translation
corpus analysis (Section 4.2.2) will be to refine this preliminary finding.

4.2.1.2  Coordination.  A frequency-based analysis of the coordinating construc-

tions – both explicit/syndetic and implicit/asyndetic ones – in the three sub-corpora
reveals that French is markedly less fond of paratactic structures than English and
Dutch. Though an analysis of variance (ANOVA) of the mean number of coordinat-
ing constructions per sentence (amounting to 0.09 in French, 0.16 in English and 0.14
in Dutch) reveals no significant differences (F(2,897) = 2.32, p ≤ 0.10), the discrep-
ancy between French, on the one hand, and English and Dutch, on the other, is worth
pointing to: While the English and Dutch sub-corpora contain, respectively, 50 and 42
instances of coordination, the French sub-corpus scores 28 only.
A more delicate analysis of the syntactic patterns displayed by the coordinating
constructions indicates, however, that the lower frequency of coordination in French
is to a large extent attributable to so-called ‘VP-conjunction’, i.e., constructions in
which the conjoined units are not full clauses but predicates sharing a subject or predi-
cations sharing a subject and auxiliaries (see example (8)). As appears from Table 2,
which displays the raw frequencies of coordination of full clauses (COORFULL; see
example (9)) vs. VP-conjunction (VP-CONJ) in the three sub-corpora, French stands
out by using a higher percentage of COORFULL, and conversely, a lower proportion
of VP-CONJ than English and Dutch.
(8) Palestinian police have arrested three people and ___ are searching for two others
as part of their investigation into the bombing that killed three Americans.
[trilled]
(9) Traditionally, activity increases in the evenings during Ramadan after Muslims
break their day-long fast, and shops often stay open until early morning.
[trilled]

Table 2.  Coordination of full clauses vs. VP-conjunction

ENGLISH FRENCH DUTCH
# % # % # %
COORFULL 26   52.0 19   67.9 26   61.9
VP-CONJ 24   48.0   9   32.1 16   38.1
Total 50 100.0 28 100.0 42 100.0
 Christelle Cosme

An analysis of variance (ANOVA) comparing the mean numbers of coordination of full

clauses and VP-conjunction across the three sub-corpora shows that, while the differenc-
es for COORFULL (0.63 for French and 0.87 for English and Dutch) are not significant
(F (2,897) = 0.53, p ≤ 0.59), the differences for VP-CONJ (0.03 for French, 0.07 for English
and 0.05 for Dutch) nearly reach significance (F (2,897) = 2.67, p ≤ 0.07). Thus, as clearly
emerges from Figure 3, the French ‘under-representation’ of coordinating constructions –
as compared to English and Dutch – is especially noticeable as regards VP-conjunction.

60

50

40
24
16 VP-CONJ
30
COORFULL
9
20
26 26
10 19

0
ENGLISH FRENCH DUTCH

Figure 3.  Coordination of full clauses vs. VP-conjunction.

As is the case for the analysis of subordination (Section 4.2.1.1), the analysis of coordina-
tion thus only partially confirms the Dependency Hypothesis. While French clearly seems
to be using fewer coordinating structures than English and Dutch (thereby tending to-
wards the ‘least incremental’, and hence ‘most hierarchical’, end of the spectrum (see Fig-
ure 1)), there is no evidence to suggest that Dutch resorts more often to coordination than
English, as the English and Dutch sub-corpora behave in a surprisingly similar fashion in
that respect. The analysis also enables us to further refine the hypothesis, by postulating
that the French ‘underuse’ of coordinating constructions – as compared to English and
Dutch – is largely due to VP-conjunction (as opposed to coordination of full clauses).

4.2.2  Translation shifts

The aim of the present section is to supplement the comparable corpus analysis by
examining another type of data, namely translation corpus data.

4.2.2.1  English ⇔ French.  A cursory glance at the shifts involved in translations

from English to French and from French to English seems to corroborate the predic-
tions. Figure 4 shows that in English ⇒ French translations there is a majority of shifts
from coordination to subordination, while in French ⇒ English translations there is
an overwhelming majority of reverse shifts, i.e., from subordination to coordination.
 Clause linking in English, French and Dutch 

160

140 135

120

100 90
COORD => SUB
80
65 SUB => COORD
60
60

40

20

0
EN => FR FR => EN

Figure 4.  Translation shifts in English ⇒ French and French ⇒ English.

In what follows, we will focus exclusively on the two most represented shifts, viz. (a)
shifts from coordination to subordination in English ⇒ French and (b) shifts from
subordination to coordination in French ⇒ English.
A number of recurrent patterns emerge from a careful analysis of the translation
corpus data. It appears that the set of French subordinating constructions most fre-
quently corresponding to English coordination (in the broad sense) is fairly limited.
Table 3 gives an overview of the most regular patterns of correspondence. Note that
the term ‘correspondence’ is meant here to refer not only to French target structures
used to translate English coordination, but also to French source structures giving rise
to English coordination.

Table 3.  French subordinating structures corresponding to English coordination

EN ⇒ FR FR ⇒ EN
Target equivalents Source equivalents TOTAL

Relative clauses 19 49   68 (30%)

Participle clauses 20 34   54 (24%)
Present 10 14   24
Past 6 15   21
‘Gérondifs’ (‘en’ + present
participle) 4 5    9
Finite adverbial clauses 23 25   48 (22%)
pour + infinitive 9 6   15 (7%)
Others∗ 19 20   39 (17%)
TOTAL 90 135 225

* The category ‘Others’ includes non-finite adverbial clauses (e.g., introduced by avant de (‘before’)), verb-
less clauses, nominal clauses, etc.
 Christelle Cosme

It emerges from Table 3 that coordination in English most frequently corresponds

to a relative clause in French. Relative clauses, it appears, account for nearly a third (30%
exactly) of all the shifts mentioned in Table 3. What is particularly striking in these shifts
is that in 88.2% of the cases (i.e., 60 cases out of 68), it is the second conjoined clause in
English that corresponds to a relative clause in French (see Ballard 1995: 266). Example
(10) is thus representative. Cases where it is the first conjoined clause in English that cor-
responds to a relative clause in French (see example (11)) thus seem to form exceptions.
(10) Elles étaient filmées par les services officiels, qui envoyaient ensuite les reportages
aux petites stations américaines […].
These were then filmed by the official press services, and the footage was
distributed to small US TV stations […].
[pleci_news]
(11) Le plus âgé, qui pour la première fois de sa vie n’a pu accompagner son
troupeau dans sa transhumance annuelle, ne parvient pas à comprendre
pourquoi je suis venu de si loin.
For the first time in his life the father of the family was unable to accompany
his flock on its annual migration this year. He cannot understand why I have
journeyed so far.
[pleci_news]
(12) Il sortit de son portefeuille un instantané jauni qu’il tendit à son visiteur.
He drew a yellow snapshot out of his note-case and handed it over.
[pleci_news]

Careful analysis of the corpus data also reveals that the functions most generally ful-
filled by the relative pronoun in such shifts are those of subject (in 28 cases out of 68,
i.e., 41% of the cases) (see example (10)) and – to a lesser extent – direct object (in 11
cases out of 68, i.e 16% of the cases) (see example (12)).
Another frequent type of correspondence is that between French participle clauses
and English coordination. Participle clauses, it appears, account for about a quarter
(24% exactly) of all the shifts mentioned in Table 3. As indicated in Table 3, French
participle clauses fall into three types, viz. (a) present participle clauses (see example
(13)), (b) so-called gérondifs, formed with ‘en’ plus present participle (see example
(14)) and (c) past participle clauses (see example (15)).
(13) Ralph stood, one hand against a grey trunk, and screwed up his eyes
against the shimmering water.
Ralph se tenait appuyé contre un tronc gris, plissant les yeux pour regarder la
surface miroitante de l’eau.
[pleci_news]
(14) […] but she only shook her head and muttered about contamination.
[…] mais elle s’est contentée de secouer la tête en marmonnant quelque chose à
propos de la contamination.
[pleci_news]
 Clause linking in English, French and Dutch 

(15) Etablie par des anciens de la Harvard Business School, elle s’ attacha
avec succès à acclimater les méthodes pratiquées aux Etats-Unis dans un
environnement européen.
It was started by Harvard Business School graduates and succeeded in adapting
US methods to suit a European environment, with English-language teaching.
[pleci_news]

Ballard (1995: 259–260) argues that cases of correspondence between English co-
ordination and French present participle clauses occur predominantly when the se-
mantic relationship holding between the two coordinated clauses is consequential.
If the principle applies in certain cases, it is not as generalizable as Ballard’s (1995:
259–260) claim suggests. The English sentence of example (13), for instance, does
not involve any cause-and-effect relationship, and yet its French translation is a pres-
ent participle clause. By contrast, the correspondence between English coordination
and French so-called gérondif occurs mostly, Hoarau (1997: 56ff) argues, when the
meaning is one of simultaneity. Examination of the corpus data shows that this is
nearly always the case. Example (14) is thus representative in that respect, as the
shaking event and the muttering event are presumed to take place simultaneously.
Hoarau (1997: 56ff) adds that this type of correspondence is more likely to obtain
when English coordination joins a verb denoting an activity (e.g., ‘to shake’ in (14))
and a verb of speaking (e.g., ‘to mutter’ in (14)). This claim, however, cannot be sub-
stantiated, for lack of sufficient corpus data, example (14) being the only occurrence
found in the translation corpus. The correspondence between English coordination
and past participle clauses in French (see example (15)) also seems to follow a set
pattern. As is the case in example (15) above, the French construction corresponding
to English coordination is most of the time (in 62% of the cases exactly) an initial
past participle clause premodifying the subject (see Ballard 1995 and Hoarau 1997
for additional examples).
English coordination also frequently corresponds to a finite adverbial clause in
French. It appears from Table 3 that such correspondences account for 22% of all the
shifts mentioned in the table. While relative clauses and participle clauses are not per
se more specific (i.e., more explicit) than coordination, the same does not apply to
finite adverbial clauses, which generally imply a higher degree of explicitness than co-
ordination. While cases where English ‘and’ is turned into a finite adverbial clause in
French are quite common in the corpus data (see example (16)), cases where a finite
adverbial clause in French is rendered by means of coordination (in the broad sense,
including implicit linking by means of a punctuation mark) in English are equally
frequent (see example (17)).

(16) Il avait disparu pour se dissoudre dans la lumière jaune, pâle et indécise tandis
que la porte se refermait
He was gone, dissolving in the pale, uncertain, yellow light, and the door
was closing.
[pleci_news]
 Christelle Cosme

(17) Cette analyse et cette politique devaient conduire à de profondes divergences

avec les Etats-Unis, comme on le vérifia dès la rencontre, le 5 juillet 1958,
entre le général de Gaulle et le secrétaire d’Etat américain John Foster Dulles.
His analysis and the policies it inspired resulted in profound disagreement with
the US. This became immediately apparent, in July 1958, when De Gaulle met
the then US secretary of state, John Foster Dulles.
[pleci_news]

Corpus evidence suggests that such shifts mostly occur when the relationship between
the clauses is of a temporal nature (see example (16)).
Another interesting – though less frequent – type of correspondence is that between
English coordination and the construction ‘pour’ + infinitive in French. In French, it is
not quite clear whether the construction expresses purpose or mere chronological se-
quence. There is thus semantic indeterminacy (Chuquet & Paillard 1987: 152). When
the subject is endowed with intentionality, as is the case in example (18) for instance, it
seems as though the meaning can more readily be interpreted as being one of purpose.
Conversely, when the subject cannot be endowed with intentionality (e.g., when it
refers to some inanimate being), the meaning is more likely to be one of chronological
sequence (see example (19)). This ambiguity of meaning of the construction ‘pour’ +
infinitive in a way makes it bear some similarity to coordination.

(18) E
 n 1991, près de deux millions d’individus ont fait passer de la nourriture, des
télévisions, des radios et de l’essence en Pologne pour ramener des vêtements
polonais […] et des biens de consommation.
In 1991 nearly 2m people ferried food, televisions, radios and petrol to
Poland and brought back Polish clothes […] and consumer goods.
[pleci_news]
(19) Sur la droite, un ruisseau, qu’enjambait un pont de pierre, traversait ce champ
pour se perdre dans la mer.
To the right of the cottages a stream flowed beneath a low stone bridge, crossed
the meadow and lost itself in the sea.
[pleci_news]

It should be noted, however, that the above-mentioned typical cases of correspon-

dence are to be interpreted as tendencies rather than strict rules. A number of ‘counter-
tendencies’, i.e., (a) shifts from subordination to coordination in English ⇒ French
translations and, vice versa, (b) shifts from coordination to subordination in French ⇒
English translations, may be observed. These shifts, it appears, mainly involve corre-
spondence between French coordination (in the broad sense) and English –ing clauses
(see example (20)).
(20) -Je ne suis pas une rapporteuse, dit Patricia. Elle me défia du regard.
‘I’m not a tell-tale,’ Patricia said, staring at me defiantly.
[pleci_news]
 Clause linking in English, French and Dutch 

4.2.2.2  English ⇒ Dutch.  A cursory glance at the shifts involved in translations

from English to Dutch seems to confirm the predictions. Figure 5 shows that the ma-
jority (69% exactly) of shifts involved in that specific translation direction are shifts
from subordination to coordination.

COORD => SUB

31% SUB => COORD

69%

Figure 5.  Translation shifts in English ⇒ Dutch.

Close examination of the shifts involved in the English-Dutch sub-corpus of the Oslo
Multilingual Corpus (OMC) further confirms this finding, also showing a predomi-
nance (67%, i.e., 159 cases out of 239) of shifts from subordination to coordination.
A particularly regular pattern stands out from the corpus data, namely the transla-
tion of –ing clauses in English by means of coordination in Dutch. This translation strat-
egy – of which examples (21) and (22) are representative illustrations – accounts for
about three quarters (73% exactly) of all the shifts from subordination to coordination
occurring in the English ⇒ Dutch component of PLECI.
(21) T
 wee bergketens lopen zo ongeveer van noord naar zuid door de republiek en
vormen een aantal dalen en plateaus.
Two mountain chains traverse the republic roughly from north to south,
forming between them a number of valleys and plateaux.
[pleci_news]
(22) ‘Dit,’ zei ze, ‘is mijn dochter Sara.’ Ze sprak de naam zo uit dat de eerste
lettergreep rijmde op ‘kar’.
She said, ‘This is my daughter Sara,’ pronouncing the name so that the first
syllable rhymed with ‘car’.
[pleci_news]

It should be noted that such shifts are mostly cases of servitudes (i.e., constraints), in-
sofar as –ing clauses do not have any direct equivalent in Dutch.
As has been shown to be the case for the English-French pair, the shifts observed
in English ⇒ Dutch translations are to be interpreted as tendencies. In other words,
it is not always the case that translating an English text into Dutch necessarily im-
plies turning all subordinating constructions into coordinating ones. Sometimes, the
 Christelle Cosme

opposite tendency holds. The most typical ‘counter-tendency’ is reflected in example

(23), where coordination in English is turned into an infinitival clause of purpose in
Dutch.
(23) We zouden graag de schildpaden uit hun poel willen halen om ermee te spelen.
We would like to get the turtles out of their pool and play with them.
[omc]

5.  Clause linking as a gradient phenomenon

Besides shedding light on the contrast in clause linking between English, French and
Dutch (Section 4), the corpus study also touches upon a more theoretical issue, namely
the issue of clause linking as a continuum (rather than as a strict binary system). Im-
portant aspects of the relationship between ‘and’-coordination and subordination have
emerged.
Table 4 summarizes the assumed traditional dichotomy made between coordina-
tion and subordination.

Table 4.  Clause linking as a binary system

Coordination Subordination
1. Syntactic criteria
Dependency – +
Embedding – +
2. Semantic criteria
Backgrounding – +
Explicitness – +

As pointed out by Cristofaro (2003: 15), the criteria used to distinguish between co-
ordination and subordination are essentially syntactic in nature. Subordination is said
to be readily identifiable by a number of properties, such as (a) dependency, i.e., the
impossibility for a subordinate clause to occur in isolation, and (b) embedding, i.e.,
the intrinsic property of a subordinate clause to function as a constituent of the matrix
clause. These two features are traditionally thought to differentiate subordination from
coordinating constructions, insofar as the latter involve clauses that are both indepen-
dent and non-embedded. The difference between coordination and subordination can
also be accounted for in semantic terms. First, it is generally assumed that subordina-
tion is more explicit than ‘and’-coordination, the latter leaving the inference to the
reader (Quirk et al. 1985: 1040). Another fundamental difference between coordina-
tion and subordination is said to lie in the way in which the states of affairs expressed
by the connected clauses are perceived and conceptualized and the status they have in
the discourse context. Broadly speaking, “the information in a subordinate clause is
 Clause linking in English, French and Dutch 

often placed in the background with respect to the superordinate clause” (Quirk et al.
1985: 919). By contrast, the contents of coordinated clauses are commonly held to have
equal informational weight.
On closer examination, however, the traditional dichotomy between the two
clause linking devices – as expressed in Table 4 – turns out to be not as clear-cut as
generally assumed. We agree with Quirk et al.’s (1985: 919) claim that “coordination
is related by gradience to subordination.” First, dependency (i.e., the impossibility for
a clause to occur in isolation), which is taken to be distinctive of subordination, is not
a clearly delineated phenomenon, but is rather a matter of degree (Givón 2001: 327).
The degree of dependency of a clause may, for instance, vary according to the type of
coordination involved. The clauses conjoined in (24) are syntactically independent of
one another, each having its own subject and verb. The second clause of (25), by com-
parison, exhibits a higher degree of dependency, because it depends on the first clause
for the expression of one argument, namely the subject, and lacks an illocutionary
force of its own. Likewise, a non-finite subordinate construction like that found in
(27) exhibits a higher degree of dependency than a finite subordinate construction
like that given in (26), since it lacks tense and an illocutionary force of its own.
(24) The winter had come at last, and snow lay thick on the ground.
[Quirk et al. 1985: 946]
(25) Peter ate the fruit and drank the beer. [Quirk et al. 1985: 948]
(26) I can’t go out with you because I am studying this evening.
[Quirk et al. 1985: 992]
(27) Her aunt having left the room, I asked Ann for some personal help.
[Quirk et al. 1985: 993]
(28) I noticed that he spoke English with an Australian accent.
[Quirk et al. 1985: 1049]
(29) We have no electricity, because there is a power failure.  [Quirk et al. 1985: 1072]

Embedding is not a clear-cut criterion either. A nominal clause (see example (28)), for
instance, is more deeply embedded in the matrix clause than an adverbial clause (see
example (29)). Thus, systemic grammar (e.g., Matthiessen & Thompson 1988; see also
Givón’s (2001) scale of inter-clausal dependency) makes a sharp distinction between
true embedding and hypotaxis. While the nominal clause in (28) belongs to the first
category, on the grounds that it functions as an obligatory constituent of the main
clause, i.e., as direct object, the adverbial clause in (29) falls into the second, because it
is only there to provide additional information on the state of affairs described in the
matrix clause and is therefore more loosely attached to the matrix clause (i.e., less em-
bedded). Lastly, there are also cases where the semantic criteria (see Table 4) fail to be
distinctive of subordination. Consider example (30), for instance, in which a situation
 Christelle Cosme

that is part of the narrative sequence (i.e., of the foreground) is found in a subordinate
adverbial clause (see bold).

(30) Only after he stopped smiling and shrieking did he go to Stephanie and hug her.
That hug was also interrupted by additional shrieks. Quite a lot of noise from a
normally silent chimpanzee! After spending about fifteen minutes with Stephanie,
Nim went over to WER, Josh, and Jeannie, and hugged each of them in turn.
(Thompson 1987: 443)

There are thus fundamental weaknesses in the traditional binary opposition be-
tween coordination and subordination. It emerges that the boundary between the two
clause linkage types is blurred and that the various criteria conventionally applied to
draw the line between them are not as clear-cut as they might seem at first sight. Rath-
er, coordination and subordination should be considered as gradual phenomena. Cris-
tofaro (2003) summarizes this view – known as the ‘Continuum Approach’ – along the
following lines:

[…] clause linkage types should not be described in terms of the binary opposition
between coordination and subordination. Rather, they should be defined in terms
of a set of mutually independent and freely combinable features, which form a
more or less articulated continuum. Each clause linkage type may be more or
less coordinate-like or subordinate-like depending on the parameter taken into
account. (Cristofaro 2003: 22–23)

It is therefore much more reasonable to assume a continuum of clause linkage ranging

from prototypical coordination to prototypical subordination, with various intermedi-
ate degrees in between.
A corpus-based contrastive study of the type presented in this article lends sup-
port to the hypothesis that the borderline between the two clause linking strategies is
less clear-cut than it first seems. Drawing on various functional-typological theories
(e.g., Matthiessen & Thompson 1988; Lehmann 1988; Givón 2001), we posit a scale of
clause linking, like the one represented in Figure 6.
At one end of the spectrum, we find prototypical cases of coordination (such as (1)
in Figure 6), i.e., constructions that are [–dependent] and [–embedded]. The opposite
pole of the continuum features prototypical cases of subordination (like (5) and (6) in
Figure 6) that are clearly [+dependent] and [+embedded]. Note that (6) is more sub-
ordinate-like than (5), as non-finite constructions are traditionally assumed to be both
more embedded and more dependent than finite ones (see above). In view of the higher
degree of subordinateness of non-finite clauses, it is rather surprising to see the relatively
large proportion of cases where there is correspondence between a non-finite clause in
one language (e.g., a gérondif in French or an –ing clause in English) and a coordinat-
ing construction in the other language (Section 4.2.2.1). However, it is less of a surprise
if we consider the typical semantic indeterminacy of non-finite adverbial clauses. Like
coordinating constructions, non-finite adverbial clauses require more interpretation on
 Clause linking in English, French and Dutch 

[-dependent] [+dependent]
[-embedded] [+embedded]
PROTOTYPICAL PROTOTYPICAL
COORDINATION SUBORDINATION
coordination of
VP-conjunction hypotaxis embedding
full clauses
(1) (2)
finite non-finite finite non-finite
(3) (4) (5) (6)

(1) The winter had come at last, and snow lay thick on the ground.
(2) Peter ate the fruit and drank the beer.
(3) Before I could sit down, she offered me a cup of tea.
(4) Driving home after work, I accidentally went through a red light.
(5) I noticed that he spoke English with an Australian accent.
(6) He likes everyone to relax.
(Quirk et al. 1985)

Figure 6.  Clause linking as a continuum.

the part of the reader than is the case with finite adverbial clauses, which traditionally in-
volve more explicit linkage. This semantic likeness with coordination, incidentally, earned
English –ing adverbial clauses the name ‘ing-coordination’ (Johansson & Lysvåg 1986:
135). VP-conjunction (see example (2) in Figure 6) also falls between the two prototypical
poles. Though traditionally accounted for as coordination, it has been shown to exhibit
a higher degree of dependency than truly prototypical coordination, i.e., coordination of
full clauses (e.g., Cristofaro 2003: 23). It thus comes as no surprise that VP-conjunction is
the type of coordinating construction most frequently involved in the shifts investigated.
Similarly, hypotaxis (see (3) and (4) in Figure 6) has to be distinguished from embedding
on the grounds that such constructions are less embedded and less dependent (hence, less
subordinate-like) than truly embedded constructions like (5) and (6) (e.g., Matthiessen
& Thompson 1988; Givón 2001). The scale of clause linking posited in Figure 6 turns out
to be highly relevant to the investigation of translation shifts. Such shifts indeed prove to
occur most commonly with constructions located in the middle of the continuum (from
VP-conjunction to finite or non-finite hypotaxis, or vice versa).

6.  Conclusion

The primary concern of the article has been to show that languages – even closely-related
ones like English, French and Dutch – exhibit preferred clause linking patterns and that
these preferences have a clear impact on translation. A corpus-based cross-linguistic
 Christelle Cosme

i­ nvestigation of clause linking proves to confirm the Dependency Hypothesis to a large

extent. Though the comparable corpus analysis shows no striking difference between
English and French as to their use of subordination, the translation corpus analysis sheds
light on a major difference. In line with the predictions, English ⇒ French translations
are characterized by a large proportion of shifts from coordination to subordination,
whereas reverse shifts are found to be predominant in the other translation direction. As
for the contrast between English and Dutch, the translation corpus analysis has enabled
us to move one step further from the comparable corpus analysis. English and Dutch do
not appear – on the basis of the comparable corpus data – to differ in the frequency with
which they resort to coordination. The translation corpus analysis suggests, however,
that Dutch has a stronger preference for coordination than English, as English ⇒ Dutch
translations typically involve shifts from subordination to coordination. This finding
thus confirms the greater incrementality of Dutch, as compared to English.
The corpus-based investigation has not only confirmed the predictions, but it has
also permitted further refining of the hypothesis. Close examination of the translation
shifts reveals a limited set of correspondences. Thus, the shifts involved in English ⇒
Dutch translations have been shown to be predominantly of one type, namely ren-
derings of English –ing clauses by means of coordination in Dutch. Likewise, careful
analysis of the shifts involved in English ⇒ French and French ⇒ English transla-
tions shows that coordination corresponds to no more than four types of sub-clauses,
namely relative clauses, participle clauses, finite adverbial clauses and – to a slightly
lesser extent – constructions with ‘pour’ + infinitive. This is in accordance with claims,
among others by Beaman (1984), that subordination is not a unitary construct. There-
fore, merely stating that one language uses more subordination than another language
is somewhat too simplistic. Further investigation ought to include a combined formal-
functional analysis of the different types of sub-clauses that are preferred in one lan-
guage over another.
In addition to providing cross-linguistic descriptions of the preferred clause link-
ing patterns of the three languages, the article has also achieved a more theoretical
objective. Clause linking, it has been shown, is best considered as a continuum (rather
than as a strict binary system). Some constructions (e.g., VP-conjunction, non-finite
clauses) have been said to bear similarities to both coordination and subordination.
It has been suggested that this theory, known as the ‘Continuum Approach’, forms an
ideal framework for the investigation of translation shifts.

References

Asher, Nicholas & Vieu, Laure. 2005. Subordinating and coordinating discourse relations. Lingua
115: 591–610.
Ballard, Michel. 1995. La traduction de la conjonction ‘and’ en français. In Relations discursives
et traduction, Michel Ballard (ed.), 221–293. Lille: Presses Universitaires de Lille.
 Clause linking in English, French and Dutch 

Beaman, Karen. 1984. Coordination and subordination revisited. Syntactic complexity in spo-
ken and written narrative discourse. In Coherence in Spoken and Written Discourse, Debo-
rah Tannen (ed.), 45–80. Norwood NJ: Ablex.
Chafe, Wallace L. 1982. Integration and involvement in speaking, writing, and oral literature.
In Spoken and Written Language: Exploring Orality and Literacy, Deborah Tannen (ed.),
35–53. Norwood NJ: Ablex.
Chuquet, Hélène & Paillard, Michel. (eds). 1987. Approche linguistique des problèmes de traduc-
tion anglais >< français. Paris: Gap / Ophrys.
Cosme, Christelle. 2003. Clause combining in English, French and Dutch. State of the Art and
Pilot Study: Towards Annotating Sentences and Clauses. Unpublished MA thesis, University
of Louvain.
Cosme, Christelle. 2004. Towards a corpus-based cross-linguistic study of clause combining:
Methodological framework and preliminary results. Belgian Journal of English Language
and Literatures (BELL), New Series 2: 199–224.
Cosme, Christelle. 2006a. Clause combining across languages: A corpus-based study of English-
French translation shifts. Languages in Contrast 6(1): 71–108.
Cosme, Christelle. 2006b. “Spelling things out” in translations: A corpus-based study of
­explicitation. Paper presented at the SLT-Congress (The Study of Language and Transla-
tion), Ghent (Belgium), 12–14 January 2006.
Cristofaro, Sonia. 2003. Subordination [Oxford Studies in Typology and Linguistic Theory].
­Oxford: Oxford University Press.
Doherty, Monika. 1998. Clauses or phrases – a principled account of when-clauses in trans-
lations between English and German. In Corpora and Cross-linguistic Research: Theory,
Method and Case Studies [Language and Computers – Studies in Practical Linguistics 24],
Stig Johansson & Signe Oksefjell (eds), 235–254. Amsterdam/Atlanta: Rodopi.
Doherty, Monika. 1999. Position and explicitness: Language specific conditions for the use of
adverbial clauses in translations between German and English. In Sprachspezifische Aspekte
der Informationsverteilung [Studia Grammatica 47], Monika Doherty (ed.), 112–148. Berlin:
Akademie Verlag.
Doherty, Monika. 2005. Language-specific conditions for discourse linking and appropriate-
ness. Linguistics 43(1): 1–15.
Fabricius-Hansen, Cathrine. 1996. Informational density: A problem for translation and trans-
lation theory. Information Structure: A Key Concept for Translation Theory, Special issue of
Linguistics 34(3): 521–565.
Fabricius-Hansen, Cathrine. 1998. Informational density and translation, with special Reference
to German-Norwegian-English. In Corpora and Cross-linguistic Research: Theory, Method
and Case Studies [Language and Computers – Studies in Practical Linguistics 24], Stig
Johansson & Signe Oksefjell (eds), 197–234. Amsterdam/Atlanta: Rodopi.
Fabricius-Hansen, Cathrine. 1999. Information packaging and translation: Aspects of transla-
tional sentence splitting (German – English/Norwegian). In Sprachspezifische Aspekte der
Informationsverteilung [Studia Grammatica 47], Monika Doherty (ed.), 175–214. Berlin:
Akademie Verlag.
Gilquin, Gaëtanelle. 2002. Automatic retrieval of syntactic structures: The quest for the Holy
Grail. International Journal of Corpus Linguistics 7(2): 183–214.
Givón, Talmy. (ed.). 2001. Syntax. An Introduction, Vol. II. Amsterdam : John Benjamins.
Granger, Sylviane. 2003. The corpus approach: A common way forward for contrastive
­Linguistics and Translation Studies. In Corpus-based Approaches to Contrastive Linguistics
 Christelle Cosme

and Translation Studies [Approaches to Translation Studies 20], Sylviane Granger, Jacques
Lerot & Stephanie Petch-Tyson (eds), 17–30. Amsterdam/Atlanta: Rodopi.
Halliday, Michael A.K. & Matthiessen, Christian M.I.M. (eds). 2004. An Introduction to Func-
tional Grammar, Third Edition. London: Edward Arnold.
Hannay, Mike. 1997. Sentencing in Dutch and English. In Studies in English Language and teach-
ing, Jan Aarts, Inge de Mönnink & Herman Wekker (eds), 231–256. Amsterdam: Rodopi.
Hannay, Mike & Mackenzie, J. Lachlan. 1996. Effective Writing in English: A resource guide.
Groningen: Martinus Nijhoff.
Hasselgård, Hilde, Johansson, Stig, Behrens, Bergljot & Fabricius-Hansen, Cathrine (eds). 2002.
Information Structure in a Cross-linguistic Perspective [Language and Computers – Studies
in Practical Linguistics 39]. Amsterdam/New York: Rodopi.
Hoarau, Lucie. 1997. Etude contrastive de la coordination en français et en anglais. Paris: Ophrys.
Johansson, Stig & Hasselgård, Hilde. 1999. Corpora and cross-linguistic research in the Nordic
countries. In Contrastive linguistics and translation, Sylviane Granger, Ludo Beheydt & Jean-
Pierre Colson (eds), 145–162. Special issue of Le Langage et l’Homme 34(1). Leuven: Peeters.
Johansson, Stig & Lysvåg, Per. 1986. Understanding English Grammar, Vol. 1. Oslo: Norwegian
University Press (UniversitetsForlaget AS).
Lehmann, Christian. 1988. Towards a typology of clause linkage. In Clause combining in gram-
mar and discourse [Typological Studies in Language 18], John Haiman & Sandra A. Thomp-
son (eds), 181–225. Amsterdam/Philadelphia: John Benjamins.
Mann, William C. & Thompson, Sandra A. 1988. Rhetorical structure theory: Towards a func-
tional theory of text organization. Text 8(3): 243–281.
Martin, James R. 1992. English Text: System and Structure. Amsterdam/Philadelphia: John Ben-
jamins.
Matthiessen, Christian & Thompson, Sandra A. 1988. The structure of discourse and subordina-
tion. In Clause combining in grammar and discourse, John Haiman & Sandra A. Thompson
(eds), 275–329. Amsterdam: John Benjamins.
Moder, Carol Lynn & Martinovic-Zic, Aida (eds). 2004. Discourse across languages and cultures
[Studies in Language Companion Series 68]. Amsterdam/Philadelphia: John Benjamins.
Quirk, Randolph, Greenbaum, Sidney, Leech, Geoffrey & Svartvik, Jan. 1985. A Comprehensive
grammar of the English Language. London: Longman.
Solfjeld, Kåre. 1996. Sententiality and translation strategies German-Norwegian. Information
Structure: A Key Concept for Translation Theory, Special issue of Linguistics 34(3): 567–590.
Thompson, Sandra A. 1987. Subordination and narrative event structure. In coherence and
grounding in discourse, Russell S. Tomlin (ed.), 435–454. Amsterdam: John Benjamins.
Vanderauwera, Ria. 1985. Dutch Novels Translated into English: The transformation of a “minor-
ity” literature [Approaches to translation Studies 6]. Amsterdam: Rodopi.
Vinay, Jean-Paul & Darbelnet, Jean. 1995. Comparative stylistics of French and English: A
Methodology for translation [Benjamins Translation Library 11. Amsterdam/Philadelphia:
‘John Benjamins’ translated and edited by Juan C. Sager and Marie-Jo Hamel.
 Sentence splitting – and strategies to preserve
discourse structure in German-Norwegian
translations

Kåre Solfjeld
Østfold University College

Adjuncts at NP- and VP-level in German texts are often rendered as separate
sentences or clauses in coordinating structures in Norwegian translations.
Typically, syntactic subordination in the source version signals that the
information should not be interpreted as the main contribution of the sentence.
Hence, an important means of guiding the readers to a discourse structure
mirroring the original is lost in the target version. This study tries to unveil
the strategies used by translators to convey the discourse function reflected in
the original when syntactically subordinate source structures are translated by
paratactic target structures. In particular, the study focuses on the function of
coordination in the target version.

Keywords: sentence splitting, translation strategies, discourse structure,

subordination in SL vs. coordination in TL

1.  Introduction

Empirical studies show that translation from German non-fictional prose into Norwe-
gian often triggers sentence splitting. The syntactic complexity of the German source
sentences cannot be felicitously transferred to the Norwegian target version by ­analogous
structures, and so adjuncts at NP- and VP-level in the German version often give rise
to separate sentences or coordinate clause conjuncts in Norwegian (Fabricius-Hansen
1999, Solfjeld 2000, 2004, Ramm 2005). An illustration of this is (1), where the participle
forms of the German original are rendered in a separate sentence in Norwegian:
(1) … ging der Amtsrichter gemeinsam mit dem diensthabenden Polizeiarzt in die
Wohnung der Frau. Dort trafen sie schon zwei Polizisten, die vom verstörten, ohne
Unterlaß schluchzenden Dienstmädchen herbeigerufen worden waren. (dt4, 8–9)
‘… the court judge together with the police doctor on duty went into the
woman’s flat. There they already met two police officers who had been called by
the upset, [lit.:] without stop crying maid.’
 Kåre Solfjeld

… dro rettsdommeren sammen med vakthavende politilege avsted til fruens

leilighet. Der traff de to politifolk som allerede var blitt tilkalt av tjenestepiken.
Hun var aldeles forstyrret og gråt uten stans. (nt4, 8)
‘… the court judge together with the police doctor on duty went to the woman’s
flat. There they met two police officers who had already been called by the maid.
She was extremely upset and sobbed without stop.’

By this process syntactically integrated structures in the source versions have

­syntactically independent informational counterparts in the target version. Syntactic
subordination often signals some sort of informational ‘downgrading’ in the sense that
the information should not be read as part of the main story line (von Stutterheim
1997: 169, 245). When syntactically subordinated source structures are rendered as
separate sentences or conjuncts in the target version, a corresponding informational
downgrading may be hard to recover, as syntactic signals to that effect are lacking. A
central challenge for the translator seems to be to prevent interpretations where the
Norwegian sentence/conjunct corresponding to a syntactically embedded structure in
German is falsely understood as continuing the main line of thoughts or events, while
it is only meant to give additional or supplementary information of some kind.
The main purpose of this study is to look into the strategies Norwegian translators
use to prevent the information from being upgraded in a way not intended in the
German original. The study should be regarded as a follow-up of the work laid out in
Solfjeld (2008), presenting largely the same main claim that the discourse role of the
German adjuncts determines target structures. The present study is, however, based on
a wider set of data: 13 excerpts of German original texts and corresponding ­authorised
Norwegian translations. The material contains a few fictional texts, but most texts
are non-fictional, many of them biographies. Each translator is represented only once.
The present article is the result of cooperative work within the project ­Languages in
­Contrast (SPRIK), University of Oslo, where (among other topics) the relation between
syntactic and discourse functional subordination across languages has been a central
topic of research; cf. e.g., Behrens and Fabricius-Hansen (2005), ­Fabricius-Hansen et
al. (2005), Ramm and Fabricius-Hansen (2005), Fabricius-Hansen (1996, 2005), Ramm
(2005 and in this volume) and Solfjeld (2004).

2.  Main claims and data

A main challenge for the translator is to enable the reader to infer a discourse ­interpretation
that reflects the original. So it must be avoided that the downgraded ­information of the
original is wrongly seen as part of the Hauptstruktur (‘main ­structure’) of the text; that is,
it should keep its role as part of the Nebenstruktur (‘side structure’) – as adding supple-
mentary information of some kind – in the terms of the ‘quaestio’-model of Klein and
von Stutterheim (1992) and von Stutterheim (1997). Many of the texts on which this
 Strategies to preserve discourse structure 

study is based are narratives, where predicates relating a sequence of ­successive events
form the main structural line and where the predicate pushing the chain of events one
step further makes up the focal part of each sentence (von Stutterheim 1997: 106).
Some adjuncts, often state predicates, give information that does not directly con-
tribute to characterising the (focal) event of the sentence. They are not part of the main
story line of the narrative. Other adjuncts are more directly integrated into the event
predicate pushing the event chain of the narrative one step further. In the material on
which this study is based, there is a strong correlation between a position relatively far
to the left and background role and a position relatively far to the right and focal role.
This is reflected in most of the examples discussed below. The different informational
contributions of the adjuncts trigger different translation strategies or put different
constraints on the target structures (to be) used. In (2) below the background func-
tion of the original adjunct, along with the position of its clause counterpart in the
­Norwegian version, results in a target version where the connectives og (‘and’) and der-
for (‘therefore’, ‘for that reason’) are added. This seems to give a felicitous Norwegian
version. In (1) above og (‘and’)-coordination cannot be used.

(2) Anselmo Rodriguez, der mit dem Mann von Frau Balboa, dem Eigentümer der
größten Eisenwarenhandlung der Stadt, seit langem bekannt war, glaubte zuerst
an einen privaten Besuch. (dt4, 15)
‘Anselmo Rodriguez, who for long had been acquainted with the husband of
Mrs. Balboa, the owner of the biggest hardware store in the town, first believed
it to be a private visit.’
Anselmo Rodriguez var en gammel bekjent av fru Balboas mann, innehaveren
av byens store jernvareforretning, og han trodde derfor først det dreide seg om et
privat besøk. (nt4, 15)
‘Anselmo Rodriguez was an old acquaintance of the husband of Mrs. Balboa,
the owner of the biggest hardware store in the town, and he believed it therefore
first to be a private visit.’

A fruitful point of departure when discussing the typical translational patterns repre-
sented in the material, is found in works within Relevance theory (Blakemore 1987,
Carston 2002). They help to explain why different strategies are used under different
conditions and shed light on the interplay between the function of the original adjunct
and the choice of target structure, e.g., the order of the clauses/sentences in the target
version and the use of og (‘and’)-coordination vs. separate sentences.
The material comprises 274 cases of sentence splitting in the sense that a German
NP- or VP-adjunct gives rise to a separate sentence or a conjunct (as part of a coordi-
native structure) in the Norwegian target text. Adjunct is understood in a broad sense
as pre- and post-modifying structures at NP-level, including relative clauses, prepo-
sitional and participle phrases and what is often termed ‘appositions’, i.e., modifying
phrases of different sorts separated by comma. The adjuncts at NP-level largely seem
to be non-restrictive. Adjuncts at VP-level are primarily adverbials, often spatial or
 Kåre Solfjeld

temporal adverbials, realised as prepositional phrases or clauses. Adjuncts at VP-level

also include so-called free predicatives.
A basic assumption is that connectives are added in the translation process to ­ensure
a discourse interpretation reflecting the original. Among the connectives identified in the
material og (‘and’)-coordination stands out as unique in terms of ­frequency. 53 adjuncts
in the above sense split into coordination with the conjunction og (‘and’); cf. (2) above.
This means that 221 adjuncts split into independent sentences, i.e., sentences separated
by full stops or sometimes by comma as in (1) above; cf. the survey in Table 1.

Table 1.  Sentence splitting with adjuncts

Type of sentence splitting Number of cases
og (‘and’)-coordination   53
Sentences separated by full stop or comma 221
Total 274

On the whole few other connectives in the traditional sense of conjunctions, ­discourse
particles or adverbs creating some link with the preceding text are added in the trans-
lation process; cf. the survey in Table 2.

Table 2.  Connectives added in Norwegian translations

Connective added Number of cases
derfor (‘hence’, ‘so’) 7 (3 with og-coordination)
da (‘then’) 6 (2 with og-coordination)
dermed/-ved (‘therewith’) 4 (3 with og-coordination)
men (‘but’) 4
nemlig (‘namely’) 4
jo (approx. ‘as you know’) 3
så (‘then’) 1
likevel (‘however’) 1
altså (‘that is’) 1

In most of the cases listed here no explicit lexical counterpart can be identified for the
Norwegian connective in the German original. This supports the assumption that they
are added to ensure certain discourse structural effects. However, at the same time these
figures show that very often connectives in the traditional sense are not added. 198 of
the split sentences result in two separate sentences without any ­connective ­being added
at all.
The analysis presented in section 3 aims to test whether the material supports the
following basic claims:
– Adjuncts encoding background information have a tendency to be rendered as
clauses/sentences to the left of the sentences/clauses corresponding to the focal parts
of the original.
 Strategies to preserve discourse structure 

– Og (‘and’)-coordination is frequently used to obtain a downgrading function of the

clause encoding the same information as the original adjunct; cf. 3.1.
– Adjuncts being integrated into the focal parts of the original sentence have a strong
tendency to be rendered as clauses/sentences to the right of the sentences/clauses
­corresponding to (other elements of) the focal parts of the original. They have an
elaborating or specifying function and generally og (‘and’)-coordination is not used.
Occasionally other connective expressions are added, but the normal case seems to be
that no connectives are added at all; cf. 3.2.

3.  Analysis

3.1  P
 reserving discourse interpretation when the adjunct plays a background
role and is extracted to the left
One possible function for the adjuncts is to play some sort of background role giving
information that enhances the understanding of the sentence part giving the main in-
formational contribution of the sentence. Often these adjuncts are found in the open-
ing part of the German original. In the Norwegian target version the relative position
of the information generally mirrors the original. These are cases of information ex-
traction to the left in the terminology of Fabricius-Hansen (1999). The original ad-
junct gives rise to a sentence or clause preceding the predicate that corresponds (most
closely) to the main predicate of the original. In 81 out of the 95 cases of sentence
splittings that are extractions to the left, the source adjunct is found in the so-called
prefield of the original; cf. (2) above and also sentence pair (3):
(3) Auroras ältester Bruder, der als einziger der Familie noch in El Ferrol ansässig
war, schlug am Tag nach Bekanntwerden dieser Aufforderung seine Schwester
ins Gesicht. Er kam betrunken aus dem Königlichen Segelklub, wo seine Freunde
die Neuigkeit kommentierten und mit ihrem Spott auch ihn nicht verschonten.
(dt4, 39)
‘Aurora’s elder brother, who was the only one in the family who still lived in
El Ferrol, hit on the day after the announcement of this request his sister in
the face. He came drunk from the Royal Sailing Club where his friends had
commented on the news and ridiculed him as well.’
Auroras bror var den eneste i familien som ennå bodde i El Ferrol. Dagen etter
at oppfordringen var annonsert, slo han søsteren i ansiktet. Han kom full fra den
Kongelige Seilforeningen, hvor vennene hadde kommentert nyheten og gjort narr
av både ham og henne. (nt4, 36)
‘Aurora’s elder brother was the only one in the family who still lived in El Ferrol.
On the day after the request was announced, he hit his sister in the face. He
came drunk from the Royal Sailing Club where his friends had commented on
the news and ridiculed both him and her.’
 Kåre Solfjeld

Since it can be assumed that the main predicate very often carries the essential in-
formation of the sentence, the main challenge for the translators will be to rephrase
the target version in a way that keeps this focus intact, in the sense that the first
sentence/conjunct must be recognisable as giving background information to the
second clause/conjunct, which makes up the focal element(s). In narrative texts
the first sentence/clause must be identified as supplying background information
for the next step in the chain of events, which is encoded in the second sentence/
clause; cf. (3) above and (4). Semantic constraints seem to play a role in these cases.
The fact that the extracted information is conveyed by state predicates ensures the
background reading (more about this later). When information is extracted in the
adjunct to the left, coordination is a frequent target structure, as illustrated in (2)
above and also (4):

(4) Micha, wie Charly seit dem heißen Dezember dabei, erklärt die Unterschiede zu
den “Instandbesetzern, die meistens viel mehr Häuser haben als wir, mindestens
hundert”, … (dt5, 103)
‘Micha, like Charly there since the hot December, explains the differences from
the “Instandbesetzern, who normally have many more houses than we, at least a
hundred”, …’
I likhet med Charly har Micha vært med siden desember 1980, og han forklarer
oss forskjellen mellom oss selv og “Istandbesetterne […], som har mange flere hus
enn vi, minst hundre”: … (nt5, 68)
‘Like Charly Micha has been there since December 1980, and he explains us
the difference between themselves and “Istandbesetterne, who have many more
houses than we, at least a hundred”, …’

Out of the 95 cases of sentence splittings that are extractions to the left, 43 are rendered
as coordinate constructions, and in the remaining cases coordination is very often a
possible alternative.
Relevance theory (Blakemore 1987; Carston 2002) gives interesting clues as to why
translators often use coordination when splitting out information to the left. A coor-
dinative structure is processed as a unit, signalling that it is the conjoined proposition
that carries optimal relevance, and not each individual conjunct. Lang (1984) presents
the same basic ideas, assuming that the conjunction has a procedural meaning that
results in the subsumption of the conjuncts under a “common integrator”. Processing a
coordinative structure as a unit precludes the possibility that one of the conjuncts may
relate to the context separately (Carston 2002: 243), and so no inferences with regard to
discourse structure exclusively based on the relation between the first constituent and
the preceding text can be made. This may in some cases be important, as a discourse
relation established between the proposition of the first constituent individually and
the preceding text may affect how the next constituent is interpreted in relation to the
discourse interpretation built up so far.
 Strategies to preserve discourse structure 

At the same time, processing conjuncts in one step licenses certain pragmatic in-
ferences to the effect that there is a (con)sequential relation between what is conveyed
in the propositions of the two conjuncts. Often a temporal sequence can be inferred
and/or a weak causal or enablement relation between the first and the second conjunct,
at least in a narrative text; cf. the concept of consequentiality in Sandström (1993: 141).
This mechanism is obviously exploited by translators. Since the first conjunct of the
coordinative structure is interpreted as leading up to the second, the first conjunct is
prevented from assuming a position in the discourse hierarchy on an equal footing
with the second conjunct. In this way coordination guides readers more effectively to
interpretations where the second conjunct is seen as informational focus parallel to the
main predicate of the original. Text pair (5) illustrates this observation:
(5) a. Aurora hatte wenig Lust, das Elternhaus zu verlassen. Aber es blieb ihr
nicht anderes übrig, als ein Zimmer im Hause des Arztes zu beziehen;
undenkbar, dass sie sich der Anordnung widersetzte. Außerdem freute sich
Ochoas Frau, deren Ehe kinderlos geblieben war, auf weibliche Gesellschaft.
Aurora Rodríguez bestand aber darauf, täglich die Bibliothek ihres Vaters
aufzusuchen. (dt4, 40)
‘Aurora did not feel much like leaving her parents’ home. But she had no
choice but to move into a room in the house of the doctor; impossible
that she should resist the arrangement. Furthermore Ochoa’s wife,
whose marriage was childless, was happy about female company. Aurora
Rodriguez, however, insisted on visiting the library of her father daily.’
Aurora hadde liten lyst til å forlate barndomshjemmet. Men hun hadde ikke
annet valg enn å flytte inn i et rom i legens hus, det var utenkelig at hun
skulle motsette seg dette. Ochoas ekteskap var barnløst, og hans kone gledet
seg over å få kvinnelig selskap. Men Aurora Rodriguez insisterte på i det
minste å få besøke farens bibliotek hver dag. (nt4, 37)
‘Aurora did not want to leave her childhood home. But she had no
other choice than to move into a room in the house of the doctor; it was
impossible that she should resist the arrangement. Ochoa’s marriage was
childless, and his wife was happy about female company. Aurora Rodriguez,
however, insisted on at least visiting the library of her father daily.’
b. Aurora hadde liten lyst til å forlate barndomshjemmet Men hun hadde ikke
annet valg enn å flytte inn i et rom i legens hus, det var utenkelig at hun
skulle motsette seg dette. Ochoas ekteskap var barnløst. Hans kone gledet seg
over å få kvinnelig selskap. Men Aurora Rodriguez insisterte på i det minste å
få besøke farens bibliotek hver dag.
‘Aurora did not want to leave her childhood home. But she had no
other choice but to move into a room in the house of the doctor; it was
impossible that she should resist the arrangement. Ochoa’s marriage was
childless. His wife was happy about female company. Aurora Rodriguez,
however, insisted on at least visiting the library of her father daily.’
 Kåre Solfjeld

By choosing a coordinating structure the translator rules out misinterpretations or at

least garden-path-effects that might arise from the alternative sentence sequence; cf.
my paraphrase in (5b). Hence, coordination can be seen to have discourse structural
effects which guide the reader to the right interpretation. The full stop alternative seems
to be more open. Since a separate sentence may relate to the context independently, the
information that Ochoa’s marriage was childless can – in a first processing step – be
read as explaining why it was impossible for Aurora to protest against the arrangement
of moving into the doctor’s home. In the next processing step the following, separate
sentence can be interpreted as discourse functionally parallel to the preceding one.
The information that Ochoa’s wife was happy about female companionship can be read
as adding another reason why she could not contest the arrangement. Hence, the full
stop alternative can easily create a kind of list reading, and, in turn, blur the role of the
information that she was happy about gaining a female companion as the main focus.
By choosing a coordinating structure the translator prevents the possibility that the
first conjunct may relate to the context separately. There is no processing step which
allows Ochoa’s childless marriage to be seen as an independent reason why Aurora
could not protest against the arrangement. Coordination ensures that the two con-
juncts are processed as a single unit, and as part of the process of linking the conjuncts
the readers are licensed to infer a consequential relation between them (Blakemore
1987: 117, Carston 2002: 243). Hence, a coordinating structure guides the readers
more effectively to an interpretation where the information that Ochoa’s marriage was
childless is read as background for the fact that she was happy about female compan-
ionship. In this way the downgraded role of the information given in the syntactically
subordinate structure of the original is kept intact. It is the constituent corresponding
to the main predicate which adds to the argument that she had no other choice but to
move into the doctor’s home.
The potential for alternative discourse structural interpretations, not matching
the original, obviously has to do with the propositional content of the constituents.
Predicates split off to the left in the way illustrated by (2), (3), (4) and (5), are of-
ten state predicates supplying some background knowledge which makes the chain
of events easier to understand or more plausible. This is a typical pattern, at least in
the material on which this study is based: A state/activity predicate – not pushing the
chain of events further – functions as background for the predicate which it precedes.
Very often sentence splittings with information extraction to the left are rendered as
sentence sequences separated by full stops, and the backgrounding function of the
state predicate is easily derivable, as in (4) (above). The backgrounding function of the
information that Charly has been there since December, is so clear that coordination
is only optional; cf. the alternative (4b):

(4) b. I likhet med Charly har Micha vært med siden desember 1980. Han forklarer
oss forskjellen mellom oss selv og “Istandbesetterne […], som har mange flere
hus enn vi, minst hundre”: …
 Strategies to preserve discourse structure 

‘Like Charly Micha has been there since December 1980. He explains us
the difference between themselves and “Istandbesetterne, who have many
more houses than we, at least a hundred” …’

The recurring translation strategy of extracting the information of adjuncts to

the left and rephrasing it as state predicates seems to fit with the observation in
Sandström (1993: 168) that state predicates encoding facts that are directly relevant
to the interpretation of events are evaluated at a point on the time line (so-called
E-state) prior to the incorporation of the event for which it is relevant. It also seems
to comply with the observation made in Ramm and Fabricius-Hansen (2005) on
the examples of the discourse relation Background in the so-called RST-framework:
The so-called satellite constituent giving information with the function of increas-
ing the reader’s ability to comprehend what is conveyed in the nucleus generally
precedes this nucleus. The material used in the present study contains many cases
where the backgrounding function of a state predicate extracted to the left is self ex-
planatory, whereas the same state predicate in a position to the right of the predicate
which corresponds to the main predicate of the original (information extraction
to the right, Fabricius-Hansen 1999) obviously blurs the background function.
This is illustrated in sentence pair (6) below. In the Norwegian translated version
the function of the state predicate is to convey qualities of the woman which clearly
lead up to what comes next, whereas this is less clear if the order is reversed. The re-
versed order would rather seem to induce an interpretation where the two sentences
should be seen as a more parallel listing of the qualities of the woman. Consider (6b)
below.

(6) a. Der Traum zeigt deutlich, in welcher Zwangslage sie sich befindet. Ohne ihre
befriedigenden Berufserlebnisse fühlt sich die von Natur aus dynamische,
ehrgeizige Frau einsam und isoliert. (dt11, 19)
‘The dream clearly shows what a dilemma she is in. Without her satisfying
professional experiences the by nature dynamic, ambitious woman feels
lonely and isolated.’
Drømmen viser hvilken tvangssituasjon hun befinner seg i. Av natur er hun
en dynamisk, ærgjerrig kvinne. Uten tilfredsstillende yrkesopplevelser føler
hun seg ensom og isolert. (nt11, 22)
‘The dream clearly shows what a dilemma she is in. By nature she is
a dynamic, ambitious woman. Without her satisfying professional
experiences she feels lonely and isolated.’
b. Drømmen viser hvilken tvangssituasjon hun befinner seg i. Uten
tilfredsstillende yrkesopplevelser føler hun seg ensom og isolert. Av natur er
hun en dynamisk, ærgjerrig kvinne.
‘The dream clearly shows what a dilemma she is in. Without her satisfying
professional experiences she feels lonely and isolated. By nature she is a
dynamic, ambitious woman.’
 Kåre Solfjeld

3.2  P
 reserving discourse interpretation when the adjunct is part of the
sentence focus and is extracted to the right
Our material suggests that strategies to preserve discourse structures are different when
the adjuncts giving rise to sentence splitting are integrated into the focus part of the
source sentence, i.e., integrated into the sentence part that gives the main informational
contribution of the sentence. In narrative texts, for instance, where the predicate intro-
ducing the next step in the event sequence makes up the focused part of the sentence,
argument expressions, which refer to individuals and objects participating in the event,
will be syntactically dependent on the verb – and so be integral parts of the event predi-
cates. Hence, adjuncts giving information about the individuals or objects participating
in the event situations can often also be seen as part of the focus of the sentence (von
Stutterheim 1997: 113). Our data show that adjuncts relatively far to the right in the
source sentence are often integrated elements of the focused part of the original sen-
tence. In the splitting process these adjuncts typically give rise to separate sentences
to the right of the target text predicate which corresponds most closely to the main
predicate of the source sentence. Hence, these are examples of what Fabricius-Hansen
(1999) has termed information extraction to the right. In the material used for the pres-
ent study there are altogether 179 cases of information extraction to the right. In 153 of
these the source adjunct is found in a position to the right of the main predicate of the
original sentences, i.e., in the position to the right of the non-finite verb(s) in a predicate
consisting of more verb forms, or to the right of a finite verb when this verb is the only
verb form. It should be added that in the latter case more refined criteria might have
distinguished between source adjuncts in the so-called middle field and source adjuncts
in so-called extraposition. It should also be noted that in complex sentences the adjunct
giving rise to sentences or clauses in the Norwegian version may be placed to the left
of the predicate of a subclause, as was the case in (1) above. The position of the source
adjuncts relative to the main predicate – i.e., in complex sentences exclusively relative
to the highest ranking matrix predicate – may give a rough idea. Still, it shows that sen-
tence splitting with source structures in end or near end position of the source sentence
(complex) mirrors the sentence splitting in initial position. The information encoded
in an adjunct to the right of the main predicate is found in a position to the right of
the corresponding predicate in the target version. Hence, the relative position of the
information elements are largely kept intact, which strengthens the picture of a linear
mapping of the information structure of the monosentential version onto the multi-
sentence or multi-clause version. The target structures are almost exclusively sequences
of separate sentences with a full stop between them, but occasionally also separated by
comma.
When information is extracted from the focal parts of the source sentence the ad-
juncts and their corresponding target sentences seem to have an elaborating function.
The source adjuncts specify or give further information about individuals or objects,
and often this information encodes essential features of the event situation. In this way
 Strategies to preserve discourse structure 

the adjuncts give important contributions to the focal part(s) of the source sentence. In
the target text these adjuncts are rendered as sentences elaborating (in a wide sense) on
some individual or object referent. Syntactically, coordination is a rare choice in these
cases, but not totally absent: 10 cases of information extraction to the right are ren-
dered as coordinative structures in the Norwegian version. Elaboration seems to block
the possibility of attaching the elaborating constituent by coordination, as it gives the
answer to a question implicitly posed in the preceding sentence on which it elaborates,
and for that reason must relate to the context separately (Carston 2002: 246ff). So, it can
be expected that elaborating sentence splittings to the right are normally incompatible
with coordinative structures (more about this below). In fact, when information is ex-
tracted to the right, not very many other connectives are built into the Norwegian target
versions either, although connectives like nemlig (‘in other words’), jo (‘as said’, ‘as you
know’), da (‘then’) and altså (‘that is’) occur in a few cases. The strategy used to preserve
the discourse structure of the original in these cases is simply to present the elaborating
information in a separate sentence directly after the individual or object referent on
which it elaborates, ensuring the correct reference relation through anaphoric links. As
with the splitting off of background information to the left, the crucial point seems to be
to convey efficiently the informationally subordinate role of the elaborating sentences
and signal efficiently when the main story line is resumed. In a narrative text it must
be clear which sentence provides a new contribution to the chain of successive events.
These points are illustrated in text pairs (7) and (8) below.
(7) Bei Professor Dements Traumentzugsexperimenten reagierten die
Versuchspersonen auf den Traumsentzug mehrerer Nächte mit Reizbarkeit,
Unentschlossenheit und Feindseligkeit. Nach traumlosen hundert Stunden
wurden aus umgänglichen Menschen plötzlich aggressive Typen, die unter
Verfolgungswahn und Erinnerungslücken litten. Sobald man die Testträumer
ungestört träumen ließ, waren sie wieder ausgeglichen und normal. (dt11, 8)
‘In professor Dement’s dream denial tests the test persons reacted to several
nights of dream denial with hypersensitivity, apathy and hostility. After hundred
dreamless hours friendly people became aggressive types, who suffered from
paranoia and memory losses. As soon as the test dreamers were allowed to
dream undisturbed, they were in balance and normal again.’
Professor Dement hindret sine forsøkspersoner i å drømme. Etter noen netter
reagerte de med irritabilitet, ubesluttsomhet og aggresjoner. Etter noen
hundre timer ble ellers omgjengelige personer plutselig aggressive. De fikk
forfølgelsesforestillinger og hull i hukommelsen. Når forsøkspersonene fikk sove
uforstyrret igjen, ble de avbalanserte og normale som før. (nt11, 9)
‘Professor Dement prevented his test persons from dreaming. After some nights
they reacted with hypersensitivity, apathy and hostility. After some hundred
hours normally friendly persons became aggressive. They suffered from
paranoia and memory losses. As soon as the test persons were allowed to dream
undisturbed, they were in balance and normal again.’
 Kåre Solfjeld

In (7) the reactions of a group of test persons are rendered as a chain of successive events,
where the changing behaviour of the persons – encoded in the predicates including ad-
juncts characterising the persons at the different stages – make up the focused part of the
sentences. So, the relative clause die unter Verfolgungswahn und Erinnerungslücken litten
(‘who suffered from paranoia and memory losses’) seems to make up an important part
of the informational focus of the sentence in which it is integrated. Together with the ad-
jective aggressive (‘aggressive’) the relative clause ­provides the new relevant information
at that stage in the development. In the ­Norwegian version the information of the rela-
tive clause is rendered in a separate sentence. The anaphoric link de (‘they’) establishes a
connection to the referent personer (‘persons’) in the ­preceding sentence, simply adding
to the description of the test persons and with no linguistic expressions pointing to any
other discourse role. When in the next sentence the chain of events is pushed one step
further, the elaborating function of the Norwegian ­independent ­sentence is clearly con-
veyed. In line with the idea of the so-called right frontier c­ onstraint (Asher 1993; Asher &
Vieu 2005) easily identifiable elaboration relations keep the track open in the sense that
the superordinate sentence in the discourse hierarchy remains available as an attachment
point. Hence, the chain of successive events is equally ­effectively conveyed in the Nor-
wegian version as in the German one, where each step in the chain of events is mirrored
by each sentence in a more iconic way. The temporal adverbials in the surrounding sen-
tences seem to ­support the elaboration reading of the Norwegian independent sentence.
In (8) the prepositional phrase in verblaßenden Farben (‘in fading colours’) is part of a
­descriptive passage ­embedded in a narrative text part. Without going into further detail,
it can be seen that the splitting off to the right of the information of this prepositional
phrase causes no interpretational difficulties. The referents malingen, fargene (‘the paint’,
‘the colours’) establish a clear anaphoric link to gudevogn (‘god cart’). So the sentence
simply elaborates further on this cart and contains no linguistic expressions pointing to
another discourse function. Hence, the Norwegian version mirrors the German original
where the same information is kept together in one sentence. It does not disturb the
discourse structure of the passage, where the description part is continued in the next
sentence and the event sequence resumed in the last sentence.
(8) Rückfahrt […]. – Skriperumpudur. Den Namen dieses Kaffs werde ich nie
vergessen. Dort, am Straßenrand, verkam in Wind und Wetter ein hölzerner
Götterwagen in verblaßenden Farben, er kann in keiner Prozession mehr
mitgeführt werden. – Des unsicheren Lichts wegen knipste ich das einst stolze
Fahrzeug mit mit drei Kameras verschiedener Objektive, … (dt13, 192)
‘Travelling back […]. – Skriperumpudur. The name of this hole I shall never
forget. There, by the side of the road, fell apart through wind and weather a
wooden god cart in fading colours, it cannot be used in any procession any
more. – Because of the uncertain light I photographed the once proud vehicle
with three cameras with different objectives, …’
Så var det på tide å kjøre tilbake […]. – Skriperumpudur. Denne fattige, lille
landsbyens navn kommer jeg aldri til å glemme. Ved veisiden sto det en gudevogn
 Strategies to preserve discourse structure 

av tre og ble ett opp av vær og vind. Malingen, fargene var falmet eller avskallet;
den kan ikke lenger føres med i noen prosesjon. – På grunn av det dårlige lyset i
skumringstimen knipset jeg det en gang så stolte kjøretøyet med tre kamera med
ulike objektiver, … (nt13, 192)
‘Then it was time to travel back […]. – Skriperumpudur. The name of this
small poor village I shall never forget. There, by the side of the road stood a
god cart of wood and was eaten by wind and weather. The paint, the colours
were fading or peeling; it can no longer be used in a procession. – Because of
the weak evening light I photographed the once proud vehicle with three
cameras with different objectives …’

Our data suggest that clear anaphoric links contribute to convey the elaborating functions
efficiently, and normally no confusion arises with regard to discourse structure. However,
the material contains a few sentence pairs where special connective devices are built into
the Norwegian version to convey effectively the elaborating status of the independent Nor-
wegian sentence. This reflects that the information encoded in the source adjunct of the
original sentence does not provide contributions to the discourse structures other than
elaborating further on (elements in) the sentence in which it is integrated. As with the left-
extracted backgrounding cases explicit connective expressions seem to be added in those
cases where the propositional content of the extracted information may lead to interpreta-
tions deviating from the original or to certain garden-path-effects. Consider text pair (9):

(9) a. Die Konferenz endete mit einem Kompromiss. Die ursprüngliche SJVD-
Resolution erhielt zwei Zusätze, in denen die Herausbildung einer
neuen internationalen Organisation und die Schaffung einer neuen
Jugendinternationale zu Zielen erklärt wurden. Die neue Resolution sollte die
Grundlage für das “Büro internationaler Jugendorganisationen” bilden, das
spätestens sechs Monate nach der “Liller Konferenz” zu seiner ersten Sitzung
zusammentreten sollte. (dt8, 152–153)
‘The conference ended with a compromise. The original SJVD-resolution
had two supplements, in which the formation of a new international
organisation and the establishment of a new youth international were
declared as goals. The new resolution would form the basis of the “Bureau
of international youth organisations”, which should get together for its first
meeting not later than six months after the “Lille-conference”.’
Konferansen sluttet med et kompromiss. Den opprinnelige SJVD-resolusjonen
fikk to tillegg. Målene, het det nå, var å bygge opp en ny internasjonal
organisasjon og danne en ny ungdomsinternasjonale. Den nye resolusjonen
skulle være grunnlaget for “Det internasjonale byrå for revolusjonære
ungdomsorganisasjoner”, som skulle ha sitt første møte senest seks måneder
etter “Lille-konferansen”. (nt8, 111)
‘The conference ended with a compromise. The original SJVD-resolution
had two supplements. The goals, it was now stated, were to form a new
international organisation and to establish a new youth international.
 Kåre Solfjeld

The new resolution would form the basis of the CBureau of international
youth organisations”, which should get together for its first meeting not
later than six months after the “Lille-conference”.’
(9) b. Konferansen sluttet med et kompromiss. Den opprinnelige SJVD-
resolusjonen fikk to tillegg. Målene var å bygge opp en ny internasjonal
organisasjon og danne en ny ungdomsinternasjonale. Den nye resolusjonen
skulle være grunnlaget for “Det internasjonale byrå for revolusjonære
ungdomsorganisasjoner”, som skulle ha sitt første møte senest seks måneder
etter “Lille-konferansen”.
‘The conference ended with a compromise. The original SJVD-resolution had
two supplements. The goals were to form a new international organisation
and to establish a new youth international. The new resolution would form
the basis of the “Bureau of international youth organisations”, which should
get together for its first meeting not later than six months after the “Lille-
conference”.’
In text pair (9) the second sentence of the German original version includes a relative
clause, which is split off as a separate sentence in the Norwegian version. The ­syntactic
structure of the original clearly conveys that the relative clause encodes ­information about
the content of the referent Zusätze (‘supplements’), which the relative clause ­modifies.
In the Norwegian version the phrase het det nå (‘it was now stated’) is added by the
translator. This seems to be needed to ensure an interpretation ­mirroring the ­discourse
­structure of the original. Without het det nå (‘it was now stated’) the ­independent
Norwegian sentence corresponding to the relative clause might be ­interpreted as ­giving
some kind of reason for the compromise, as målene (‘the goals’) in principle may be
anaphorically linked with konferansen (‘the conference’) or possibly kompromiss
­(‘compromise’). By adding het det nå (‘it was now stated’) the readers are guided in the
direction of some written document or text. In this way tillegg (‘supplements’), as part of
resolusjon (‘resolution’), stands out as a natural anchoring for målene (‘the goals’).
The general picture is, however, that no connective expressions are added when
information is extracted to the right. Our data suggest that the role as integrated parts
of the focused elements is hard to combine with e.g., connectives such as nemlig, jo,
altså, which seem to presuppose a more independent backgrounding or commenting
function of the information which they modify. Hence, they often seem incompatible
with the ‘pure’ function of elaborating on some focused constituent(s), which is what
translators, for discourse structural reasons, aim for in most cases of information ex-
traction to the right. Our material includes cases where e.g., nemlig, jo, altså are added
in the translation process. For such cases see the discussion in Solfjeld (2008).

4.  Summary and discussion

The material confirms that preserving discourse structure is relevant for the choice
of target structure when adjuncts at NP- and VP-level in German originals are split
 Strategies to preserve discourse structure 

off in the translation process and in this way give rise to independent sentences or
conjuncts in the Norwegian target versions. If the extracted information has some
kind of ­backgrounding function, it is often moved to the left, preceding the informa-
tion which provides the contribution to the main line of events or thoughts of the
text (part). Often the ordering of the components seems to be enough to ensure an
­interpretation ­equalling the downgraded status of the original adjunct, particularly
if the extracted information is a state predicate and can be immediately ­recognised
as such. In cases where this is less clear, the backgrounding information is often
rendered as the first conjunct of a coordinative structure to prevent discourse func-
tional ­mismatches ­compared with the original. The (con)sequential inferences
licensed by the coordination ensure a downgrading effect of the first conjunct and a
corresponding upgrading of the second conjunct, which for that reason is ­preserved
as focal. Hence, to integrate adjunctive information with a backgrounding function
translators ­frequently use the connective og (‘and’). Other connectives are rare, but
do occasionally occur – for ­instance derfor (‘so’, ‘hence’) and dermed (‘therewith’).
Sometimes they are used in combination with og as illustrated in (2) above. These
connectives seem to have much the same downgrading effect on the preceding clause
as coordination.
When the adjuncts are integrated into the focal part(s) of the original, the strat-
egy is to keep the target sentence corresponding to the original adjunct in a position
where clear anaphoric links can be established. Normally, no connectives are added.
The position of the extracted information, including the anaphoric links, ensures the
intended interpretation, and the discourse function of the incoming sentences seems
equally clear in the target version as in the original. Sometimes, however, connective
expressions are added, and they seem to be added in those cases where they are needed
to prevent the extracted information from being falsely interpreted as a contribution
to the main story line.
The material shows that translators are generally sensitive to where connectives
should be added to prevent unintended discourse functional interpretations. However,
in some cases the Norwegian versions are less felicitous or at least more open or con-
fusing with regard to discourse structure than their German counterparts; cf. Ramm
(in this volume).
The material clearly shows that lexical items of different kinds may be add-
ed to obtain a discourse functional effect; cf. (9) above. Interestingly, most cases of
sentence splitting seem to work well without the adding of connective expressions
for which there are no explicit counterparts in the original. Evidently, the order in
which the ­constituents appear (cf. (6)), the use of different sorts of anaphoric expres-
sions and characteristics of the predicates as e.g., states or processes are all elements
which ­contribute to convey the discourse structures. These are aspects that have only
been briefly touched upon in the present study. The translation patterns in this study
­illustrate, however, that rhetorical structure makes up an important part of the linguis-
tic competence of the translators.
 Kåre Solfjeld

Interestingly, our data suggest a rather linear mapping of the sentence internal
i­ nformation structure of the original onto a sequence of sentences or clauses in the target
version. Many of our original sentences reflect the traditional claim that sentence open-
ings are the preferred position for background information linking the sentence with
the preceding context, whereas components further to the right carry the new, focused
information (Dietrich 1994; Nølke 1995). In this way the multisentence or multiclause
target versions encompass sequences where some sentences/conjuncts are split off
from a front position of the original. These sentences/conjuncts point to the right and
give background information for what comes next. Other sentences are split off from
focal parts of the original. These point to the left and elaborate on (components of) the
preceding sentence. The fact that many such sequences seem to be felicitous without
assisting connectives – the translators feel no need to add them – suggests that they are
often rather transparent. As the processing runs form left to right, a predicate without
any immediate relevance or connection to what has been said, is automatically inter-
preted as building up to what comes next. In particular state predicates in narratives
seem to come with this expectance of looking right. This fits well with the observation
that the ties to the left must be clearly stated when elaborating information is split off
to the right. Translators do, however, – as seen above – add connective devices to help
interpreting such sequences, when misinterpretations might occur.
The only frequently occurring connective built into the translation to preserve
discourse structure is og (‘and’). In this way, coordination seems to compensate for
the more restrictive use of syntactic subordination in Norwegian as compared to
­German. Coordination in Norwegian may also seem to be less restricted as to the
­relation holding between the conjuncts than corresponding und-conjuncts in ­German
and and-conjuncts in English (Ramm & Fabricius-Hansen 2005 and ­Fabricius-Hansen
et al. 2005). Interestingly, our data include some cases where ­information split off to
the right is rendered as a conjunct of a coordinative structure. In (10) the participle
form nachsichtig lächelnd (‘smiling indulgently’) is an event which accompanies the
event that Tom appears again, and together these event predicates make up the in-
formational focus of the German original. This seems to be typical of the informa-
tion splittings to the right where the extracted information is part of a coordinative
structure. The information extracted to the right is an integrated part of the main
informational contribution of the sentence, licensing that it is included in the second
conjunct of a coordinative structure, which often carries the focus of a coordina-
tive structure. As the first conjunct of a coordinative structure often seems to convey
background information, this translational strategy may mean that some element of a
syntactically complex focus in the German original is informationally downgraded to
a function of leading up to the second conjunct. This may for instance be the case in
(10), where the information that Tom appeared again, can be seen as more of a back-
ground information in the Norwegian version than in the German original.
(10) Jetzt traf der schwere Ball gegen Johns Kniekehlen. Er fiel um wie eine zu steil
gestellte Leiter, erst langsam und dann mit Wucht. Von der Hüfte und vom
 Strategies to preserve discourse structure 

Ellenbogen her breitete sich Schmerz aus. Tom stand wieder da, nachsichtig
lächelnd. Halblaut sagte er, ohne den Blick von John abzuwenden, etwas zu den
anderen, … (dt9, 14)
‘Now the heavy ball hit John’s knee. He fell over like a too steeply set up ladder,
first slowly and then with force. From the hip to the elbow the pain spread out.
Tom appeared again, smiling indulgently. Half aloud he said, without taking his
eyes away from John, something to the others, …’
Nå traff den harde ballen John i knehasene. Han falt om som en stige som stod for
steilt, først langsomt og så med tyngde. Smerten bredte seg fra hoften og albuen.
Tom stod der igjen og smilte overbærende. Halvhøyt sa han noe til de andre uten å
ta blikket fra John, … (nt9, 15)
‘Now the heavy ball hit John’s knee. He fell over like a ladder which was too
steeply set up, first slowly and then with force. From the hip to the elbow the
pain spread out. Tom appeared again and smiled indulgently. Half aloud he said
something to the others, without taking his eyes away from John, …’

Our data confirm that coordination is a frequently used strategy to keep syntactically
complex source structures together as units in the target version. To use it felicitously
this material at least suggests that the second conjunct must be focal and the first con-
junct must play the role of leading up to the second. This may be hard to combine with
the challenge of giving an equivalent target version. Linking constituents as ­conjuncts
in cases where it is hard to recover a consequence relation may give odd results. In
other cases a downgrading which does not mirror the original structure, may be in-
ferred. Furthermore, this leads to the very interesting question, discussed in Fabricius-
Hansen et al. (2005) and Ramm and Fabricius-Hansen (2005), of whether the use of
og-coordination in Norwegian really equals corresponding structures in German or
English – a question which certainly deserves more cross-linguistic r­ esearch.

Acknowledgments

This article is very much the result of joint work within the SPRIK-project, University of
Oslo. I would like to thank Cathrine Fabricius-Hansen, Wiebke Ramm and Bergljot Beh-
rens, Oslo, for invaluable cooperation and useful suggestions and comments. I would also
like to thank three anonymous reviewers for very constructive and helpful comments.

References

Asher, Nicholas. 1993: Reference to Abstract Objects in Discourse [Studies in Linguistics and
Philosophy 50]. Dordrecht: Kluwer.
Asher, Nicholas & Vieu, Laure 2005. Subordinating and coordinating discourse relations. Lin-
gua 115(4): 591–610.
 Kåre Solfjeld

Behrens, Bergljot & Fabricius-Hansen, Cathrine 2005. The relation accompanying circumstance
across languages. Conflict between expression and discourse subordination? [SPRIKreport
32]. URL: http://www.hf.uio.no/forskningsprosjekter/sprik/docs/pdf/bb/Sprik-Report32-
bb-cfh.pdf. Oslo: University of Oslo.
Blakemore, Diane 1987. Semantic Constraints on Relevance. Oxford: Basil Blackwell.
Carston, Robyn 2002. Thoughts and Utterances. The Pragmatics of Explicit Communication.
­Oxford: Blackwell.
Dietrich, Rainer 1994. Wettbewerb – aber wie? Skizze einer Theorie der freien Wortstellung. In
Was determiniert Wortstellungsvariation? Studien zu einem Interaktionsfeld von Gramma-
tik, Pragmatik und Sprachtypologie, Brigitta Haftka (ed.), 33–47. Opladen: Westdeutscher
Verlag.
Fabricius-Hansen, Cathrine 1996. Informational density: A problem for translation and transla-
tion theory. Linguistics 34: 521–565.
Fabricius-Hansen, Cathrine 1999. Information packaging and translation. Aspects of transla-
tional sentence splitting (German – English/Norwegian). In Sprachspezifische Aspekte der
Informationsverteilung, Monika Doherty (ed.), 175–213. Berlin: Akademie Verlag.
Fabricius-Hansen, Cathrine 2005. Elusive connectives. A case study on the explicitness dimen-
sion of discourse coherence. Linguistics 43(1): 17–48.
Klein, Wolfgang & von Stutterheim, Christiane 1992. Textstruktur und referentielle Bewegung.
Zeitschrift für Literaturwissenschaft und Linguistik 86: 67–92.
Lang, Ewald. 1984. The Semantics of Coordination [Studies in Language Companion Series 9].
Amsterdam: John Benjamins.
Nølke, Henning 1995. Utterance Focus. Elements of a modular theory. In Information Struc-
ture [Copenhagener Studies in Language 18], 75–114. Michael Herslund (ed.) København/
Frederiksberg: Samfundslitteratur.
Ramm, Wiebke 2005. Satzrelativsätze im Deutschen und ihre Übersetzung ins Norwegische.
Zum Zusammenhang zwischen Satzteilung und Diskursfunktion. In Estudios Filológi-
cos Alemanes. 9, Universidad de Sevilla: Revista del Grupo de Investigación de Filología
Alemana. 259–277. Also available as SPRIKreport Nr. 28. URL: http://www.hf.uio.no/
forskningsprosjekter/sprik/docs/pdf/wr/WiebkeRammReport28.pdf. Oslo: University of
Oslo.
Ramm, Wiebke & Fabricius-Hansen, Cathrine 2005. Coordination and discourse-structural
salience from a cross-linguistic perspective. In Salience in Discourse: Multidisciplinary
­Approaches to Discourse 2005, Manfred Stede, Christian Chiarcos, Michael Grabski & Luuk
Lagerwerf (eds). Münster: Stichting/Nodus, 119–128. Also available as SPRIKreport Nr. 30.
URL: http://www.hf.uio.no/forskningsprosjekter/sprik/docs/pdf/wr/WR-CFH-Report30.
pdf. Oslo: University of Oslo.
Sandström, Gørel 1993. When-clauses and the Temporal Interpretation of Narrative Discourse.
Umeå: Department of General linguistics, University of Umeå.
Solfjeld, Kåre 2000. Sententialität, Nominalität und Übersetzung. Eine empirische Untersuchung
deutscher Sachprosatexte und ihrer norwegischen Übersetzungen [Osloer Beiträge zur Ger-
manistik 26]. Frankfurt: Peter Lang.
Solfjeld, Kåre 2004. Informationsspaltung in Sachprosaübersetzungen Deutsch-Norwegisch. In
Kommunikasjon, Eva L. Björk & Sverre Vesterhus (eds), 111–130. Halden, Høgskolen i
Østfold.
Solfjeld, Kåre. 2008. Sentence splitting and discourse structure in translations (German-Norwe-
gian). Languages in Contrast 8.1, 21–46.
 Strategies to preserve discourse structure 

Stutterheim, Christiane von 1997. Einige Prinzipien des Textaufbaus. Empirische Untersuchungen
zur Produktion mündlicher Texte. Darmstadt: Niemeyer.

Source literature

Cremer, Peter 1982. Ali Cremer. U 333. Berlin – Frankfurt/M – Wien: Ullstein, 225–240 (dt1).
Norw. translation (bokmål) 1989. U 333. Vendepunktet. Translated by Tore Bjørn Stornæs-
Nilsen, Oslo: Faktum, 107–128 (nt1).
Däniken, Erich von 1985. Habe ich mich geirrt? München: Bertelsmann, 166–194 (dt2). Norw.
translation (bokmål) 1986. Har jeg tatt feil? Translated by Ådne Goplen, Oslo: Mortensen,
168–195 (nt2).
Franz, Uli 1987. Deng Xiaoping. Chinas Erneuerer. Stuttgart: Deutsche Verlags-Anstalt (dt3),
125–144. Norw. translation (bokmål) 1988. Deng Xiaoping. Kinas sterke mann. Translated
by Iver Tore Svenning. Oslo: Schibsted, 114–132 (nt3).
Hackl, Erich 1987. Auroras Anlass. Zürich: Diogenes, 7–42 (dt4). Norw. translation (bokmål)
1988. Auroras motiv. Translated by Lasse Tømte. Oslo: Cappelen, 7–39 (nt4).
Haller, Michael (ed.) 1981. Aussteigen oder Rebellieren. Authors of selected parts: Hans Halter
and Walter Tauber, Hamburg: Spiegel, 100–116 (dt5). Norw. translation (bokmål) 1982.
Isolasjon eller opprør. Translated by Truls Wyller. Oslo: Gyldendal, 66–80 (nt5).
Hermlin, Stephan 1979. Abendlicht. Berlin: Klaus Wagenbach, 58–87 (dt6). Norw. translation
(bokmål) 1980. Kveldslys. Translated by Carl Fredrik Engelstad. Oslo: Gyldendal, 53–78
(nt6).
Lorentz, Konrad 1963. Das sogenannte Böse. Zur Naturgeschichte der Aggression. Wien: Borotha-
Schoeler,19–33 (dt7). Norw. translation (bokmål) 1968. Den såkalte ondskap. Om aggresjon
hos mennesker og dyr. Translated by Brynjulf Valum. Oslo: Cappelen, 21–35 (nt7).
Lorenz, Einhart 1989. Willy Brandt in Norwegen. Kiel: Neuer Malik, 132–158 (dt8). Norw. trans-
lation (bokmål) 1989. Willy Brandt i Norge. Translated by Anne-Lise Risø. Oslo: Tiden,
97–115 (nt8).
Nadolny, Sten 1983. Die Entdeckung der Langsamkeit. München – Zürich: Piper, 9–17 (dt9).
Norw. translation (bokmål) 1985. Mannen som ilte langsomt. Translated by Ellen Harboe
Meisingset. Oslo: Cappelen, 11–17 (nt9).
Schreiber, Hermann 1978. Singles Allein leben. Besser als zu zweit? München: Bertelsmann, 115–
133 (dt10). Norw. translation (bokmål) 1980. Leve alene. Bedre enn å være to? Translated by
Leif Toklum. Oslo: Gyldendal, 89–103 (nt10).
Senger, Gerti 1985. Frauenträume Männerträume und ihre Bedeutung Niedernhausen/Ts:
Falken-Verlag, 8–29 (dt11). Norw. translation (bokmål): Kvinnedrømmer Mannsdrømmer.
Translated by Tormod Ropeid. Oslo: Teknologisk Forlag, 9–32 (nt11).
Prekop, Jirina 1988. Der kleine Tyrann. München: Kösel-Verlag, 49–67 (dt12). Norw. translation
(bokmål) 1989. Den lille tyrann. Translated by Peter Normann Waage and Eivind Tjøn-
neland. Oslo: Ex libris, 55–76 (nt12).
Wallraff, Günter 1985. Ganz unten. Köln: Kiepenheuer & Witsch, 84–101 (dt13). Norw. transla-
tion (bokmål) 1986. Aller nederst. Translated by Ola Johnsrud. Gjøvik: Pax, 73–87 (nt13).
 Upgrading of non-restrictive relative clauses in
translation
A change in discourse structure?1

Wiebke Ramm
University of Oslo

The paper investigates the upgrading of non-restrictive German relative clauses

to independent main clauses in Norwegian translations. The question is raised
whether the discourse function of the relative clause has any impact on the
interpretation of the upgraded translation. Previous research has revealed that
German non-restrictive relative clauses can serve different discourse functions: a
discontinuative/appositive or a continuative discourse function. These correlate
with different types of discourse relations. A corpus study supports the hypothesis
that upgrading is more problematic with respect to discontinuative/appositive
relative clauses. However, the study also reveals certain shortcomings of the
applied approaches. The upgrading of relative clauses makes the translations less
hierarchically organised than the original texts. This allows for some reflections on
the cross-linguistic assessment of coherence within and across sentence boundaries.

Keywords: Translation of relative clauses, subordination in syntax vs. discourse,

discourse organisation strategies, coherence.

1.  Introduction

In his contribution, Solfjeld (in this volume) shows how relatively complex structural
changes in a translation may nevertheless lead to similar interpretations of a source

.  The research presented here was developed as part of my PhD project funded by the Faculty
of Humanities, University of Oslo. I am also grateful to the project ‘SPRIK (Språk i kontrast/
Language(s) in contrast)’ funded by the Norwegian Research Council (project number NFR
158447/530), for supporting my participation at the DGfS-06 conference in Bielefeld. I would
like to thank Cathrine Fabricius-Hansen for her helpful comments on earlier versions of this
paper and Barbara Clare Dalton, Bergljot Behrens and Signe Oksefjell Ebeling for correcting my
English. Last but not least, I owe gratitude to the three reviewers, for their useful comments and
suggestions for improving the paper.
 Wiebke Ramm

language (SL) and a target language (TL) text. In this paper, we will take the opposite
case as a starting point and show that seemingly small adjustments in the sentence
structure may nonetheless have significant consequences for the discourse interpreta-
tion of the TL text for example as regards anaphora resolution. Examples of German
non-restrictive relative clauses upgraded to independent main clauses in Norwegian
translations will be taken from a German-Norwegian parallel corpus of popular sci-
ence articles. We will investigate how the interpretation of the translation differs from
the SL text and whether the discourse function of the relative clause (RC) in the origi-
nal text has an impact on the interpretation of the (upgraded) TL version.
Section 2 explains why non-restrictive RCs in German represent a problem when
translated into Norwegian and summarises two approaches addressing the discourse
function of (German) non-restrictive RCs. These approaches provide the framework
for the description of the discourse structures of German text fragments containing a
non-restrictive RC and for the comparison with their sentence-structurally upgraded
Norwegian translations. Section 3 presents a hypothesis about the consequences of
upgrading German RCs to independent sentences in translation and the predictions
that follow from it. We then present and evaluate the results of a corpus study testing
the hypothesis and its predictions. Two different discourse organisation strategies are
considered in relation to the RC examples and their translations in Section 4, and
some possible implications of the different strategies for the coherence of a text are
discussed. Finally, Section 5 summarises the findings, concluding with an outlook on
how the analysis of relatively small structural adjustments in translation – such as
those discussed in this paper – can contribute to a better understanding of a cross-
linguistically valid notion of coherence.

2.  Non-restrictive relative clauses in German

2.1  Non-restrictive relative clauses as a problem for translation

Due to typological differences between the two languages, it is not unusual for non-
restrictive German RCs to be translated as independent main clauses in Norwegian.2
German relative markers with nominal antecedents are pronouns with referential prop-
erties, and the case, number and gender agreement requirement helps to disambiguate
potential antecedents in the matrix clause. By contrast, the most frequent Norwegian

.  In a corpus study investigating sentence boundary adjustments in translations of popular

science texts between Norwegian and German, non-restrictive RCs turned out to be the most
frequent cause of sentence splitting (i.e., the translation of one SL clause or clause complex as
two or more independent sentences in the TL) for the translation direction German-Norwegian
(Ramm, in preparation).
 Upgrading of non-restrictive relative clauses in translation 

relative marker, som, is non-inflecting and usually considered as a subjunction (Faarlund

et al. 1997: 866). This lack of potential for disambiguation, sometimes in interaction with
other typologically motivated changes of word order in the matrix clause, often leaves
the upgrading of an RC to an independent main clause and the reintroduction of the
referent as the only option for the Norwegian translation, as in example (1a) vs. (1b).
(1) a. Von 1905 bis zum Ersten Weltkrieg existierten kaum außenpolitische
Berührungspunkte zwischen dem Deutschen Reich und Norwegen, dessen
Regierung nach der Unabhängigkeit eine permanente außenpolitische
Neutralität anstrebte. […] [BB]
Literal (linear) English gloss: ‘From 1905 until World War I existed hardly
any foreign-political points of contact between the German Reich and
Norway, whose government after the independence a permanent foreign-
political neutrality strived for.’
b. Fra 1905 til utbruddet av første verdenskrig fantes det få utenrikspolitiske
berøringspunkter mellom Det tyske riket og Norge. Etter uavhengigheten
ønsket den norske regjeringen utenrikspolitisk nøytralitet, […].[BBTN]
‘From 1905 until the outbreak of World War I existed few foreign-political
points of contact between the German Reich and Norway. After the indepen-
dence wished the Norwegian government foreign-political neutrality, […].’

RCs with the whole matrix clause as antecedent, so-called Satzrelativsätze (‘clause-­related
RCs’), are another type of RC frequently rendered as an independent sentence in Norwe-
gian translations. Norwegian equivalents of German relative adverbials − such as wobei
(lit. ‘where-by’), weshalb (‘wherefore’, ‘for what reason’), wodurch (lit. ‘where-through’) and
womit (lit. ‘where-with’) − are missing in Norwegian, or at least they are highly marked or
old-fashioned. Consequently, the only option available to the Norwegian translator is of-
ten to render the RC in the form of an independent sentence and to express the discourse
relation encoded by the relative adverbial in German by a corresponding connective (ex-
ample (2a) vs. (2b)). Alternatively, the discourse relation can be left implicit (example (3a)
vs. (3b)). The relative marker was (‘what/which’) is the only relative marker which has a
quasi-equivalent in Norwegian, noe (som) (‘something (which/that)’), which makes up-
grading dispensable in most of these cases.
(2) a. Das militärstrategische Primat wird auch daran erkennbar, dass
die Militärbehörden durch die Rüstungsprogramme über erhebliche
Kompetenzen in wirtschaftlichen Fragen verfügten, weshalb deutsche Firmen
auch mit diesen kooperieren mussten, obwohl die Wirtschaftspolitik formell
durch das Reichskommissariat gestaltet wurde. [BB]
‘The military-strategic primacy becomes also [daran (pron. adv., lit.
‘there-on’)] visible, that the military authorities through the arms programs
considerable competences in economic questions had, [weshalb] German
companies with these cooperate must, although the economic policy
formally through the Reichskommissariat formed/organised was.’
 Wiebke Ramm

b. Et forhold som reflekterer det militær-strategiske primatet, var at

rustningsprogrammene hadde gitt de militære myndighetene en betydelig
kompetanse i økonomiske spørsmål. Derfor måtte de tyske firmaene
samarbeide med dem, selv om den økonomiske politikken formelt sett ble
utformet av rikskommissariatet. [BBTN]
‘A fact that reflects the military-strategic primacy, was that the arms
programs had given the military authorities considerable competence in
economic questions. Therefore must the German companies cooperate
with them, although the economic policy formally seen was formed by the
Reichskommissariat.’
(3) a. Neben diesen engen wirtschaftlichen Kontakten bestanden rege kulturelle,
technische und wissenschaftliche Beziehungen, wobei für das norwegische
Bildungsbürgertum die deutsche Literatur und die deutsche Sprache den
Bezugspunkt darstellte. [BB]
‘Besides these close economic contacts existed active cultural, technical and
scientific relations, [wobei] for the Norwegian educated classes the German
literature and the German language the point of reference represented.’
b. Ved siden av disse tette handelsforbindelsene fantes det også viktige kulturelle,
tekniske og vitenskapelige kontakter. Tysk språk og litteratur stod sterkt i de
utdannede borgerlige kretser. [BBTN]
‘Besides these close trade connections existed also important cultural, tech-
nical and scientific contacts. German language and literature had a strong
position in the educated middle-class circles.’

The question is whether such modifications of clause linkage make a difference

for the discourse interpretation of the original vs. translated versions of the text, and
if they do, in what way?

2.2  The discourse function of non-restrictive relative clauses

German noun-related and clause-related non-restrictive RCs such as those presented
in the previous section are often described as weiterführende Relativsätze (‘continua-
tive RCs’), a subtype of non-restrictive RCs which can only appear in sentence-final
position, and, in studies on German RCs, their similarity to independent main clauses
is often emphasised (Brandt 1990: 46–52; Peyer 1997: 141–142; Laux 2002: 199–204).
In contrast to an adverbial or complement clause, a continuative RC is not syntacti-
cally embedded − it does not realise a syntactic function within the matrix clause −
and neither is it pragmatically integrated into the matrix clause (signalled by intona-
tion, for instance), having an information structure (focus-background structure) of
its own (Fabricius-Hansen 1992: 479; Holler 2005: 131; Holler in this volume). The
only obvious difference between a weiterführende Relativsatz and a main clause is its
syntactic dependency on the matrix clause, signalled by the clause-final position of the
finite verb in German. We are interested in specifying its discourse function, and in
 Upgrading of non-restrictive relative clauses in translation 

establishing the means by which this function is retained – or not retained – when it is
upgraded to an independent sentence in the Norwegian translation.
Laux (2002) and Holler (2005) investigate non-restrictive RCs in German – Laux
focussing on their temporal interpretation in narrative texts and Holler providing a
formal discourse-semantic account and computational-linguistic modelling in HPSG –
and develop similar criteria for the identification and description of weiterführende
Relativsätze. Both authors investigate the temporal structure of the matrix clause and
the RC in order to discover in what way the RC represents a continuation of the events
described in the matrix clause. Both use the distinction between Hauptstruktur (main
structure) and Nebenstruktur (side structure) as delineated by Klein and v. Stutter-
heim’s (1987, 1991) ‘quaestio’ approach in order to determine how the RC contributes
to answering the ‘quaestio’, the text question guiding the development of the respective
text fragment; and, finally, they both use the model of Segmented Discourse Represen-
tation Theory, SDRT, (Asher 1993; Asher & Lascarides 2003) to describe the discourse
relations holding between the matrix clause and the RC.
(4) a. Emil traf einen Bauern, der übrigens einen weiten Mantel trug. [Ihn fragte er
nach dem Weg.]
‘Emil met a farmer, who by the way a wide coat wore. [Him asked he for
the way.]’
b. Emil traf einen Bauern, den er dann nach dem Weg fragte. [So konnte er
ohne Probleme den Gasthof finden.]
‘Emil met a farmer, whom he then asked for the way. [So was he able with-
out problems the inn to find.]

Example (4a) instantiates what Laux (2002) calls a discontinuative (diskontinuativ)

and what Holler (2005) calls an appositive (appositiv) RC.3 The RC is not seen as part
of the main structure of the text, as it merely contains additional or background infor-
mation and does not push forward the development of the (narrative) text. ­Applying
Holler’s (2005) approach, an asymmetric, i.e., subordinating, discourse relation (Elabo-
ration) can be identified between the matrix clause and the RC. The distinction ­between
subordinating and coordinating discourse relations in the SDRT model (Asher &
Lascarides 2003; Asher & Vieu 2005) reflects an assumption common to many ap-
proaches to discourse structure − that discourse has a hierarchical structure. A central
criterion for distinguishing between the two types of discourse relation is the possible
attachment points for new information in a discourse structure: if two discourse units
are linked by a subordinating discourse relation (the prototypical example being Elabora-
tion), then both units provide possible attachment points to new information; whereas
with a coordinating discourse relation between two segments (as in the prototypical case
of Narration), the last but one unit is not available for the attachment of incoming new

.  Example (4a) and (4b) are adapted from Holler (2005: 211).
 Wiebke Ramm

information, i.e., it is not at the right end of the incrementally constructed discourse
representation. This is roughly what is called the right-frontier constraint (RFC) in SDRT
(Asher 1993: 270–271; Asher & Lascarides 2003: 10–12). Other important factors enter-
ing into play here are the temporal relations holding between the eventualities described
in the discourse units and how discourse topics are constructed and maintained.
Laux’s (2002) main criterion to distinguish between continuative (kontinuativ) and
discontinuative RCs is whether a Consequentiality relation in the sense of Sandström
(1993) – which is defined similarly to Narration in SDRT – can be identified between
the clauses. In contrast to Holler’s approach, not only is the discourse relation holding
between the matrix clause and the RC taken into account here, but so is the relation
holding between the RC and the following sentence. According to Laux’s definition
(2002: 189–190), a RC is continuative only when a Consequentiality relation holds be-
tween the matrix clause and the RC, and between the RC and the following sentence
(given in square brackets in the examples). Example (4a) is not continuative on this
premise, as there is neither a Consequentiality relation between the matrix clause and
the RC, nor between the RC and the following context.
The RC in (4b) would be classified as weiterführend in Holler’s and as kontinuativ
in Laux’s terminology. The RC has the same communicative weight as an independent
main clause and is part of the main structure of the text, since it makes an individual
contribution to answering the ‘quaestio’ of the text fragment (here something like
‘What happened?’). According to Holler’s definition, a symmetric i.e., coordinating,
discourse relation (Narration) holds between the matrix clause and the RC, while
Laux’s method leads to the assignment of a Consequentiality relation between the ma-
trix clause and the RC, and between the RC and the following sentence.
Example (4a) and (4b) illustrate a difference in the mapping between syntactic
structure and discourse organisation which is interesting from the perspective of
translation. While in (4a) syntactic subordination correlates with a subordinating (or
asymmetric) discourse relation, syntactic subordination in (4b) does not correspond
to subordination or asymmetry on the discourse level. Thus we have a mismatch between
syntactic structure and discourse organisation in this example (for similar observations
regarding the possible non-correspondence of subordination in syntax and discourse
see Delort in this volume). Although approaches operating with a concept of ‘sub-
ordination’ or ‘asymmetry’ of discourse relations such as SDRT or Rhetorical Struc-
ture Theory (RST) (Mann & Thompson 1988) often emphasise that there is no direct
mapping between a particular type of discourse relation and its lexico-grammatical
realisation, there seems to be a common understanding that syntactically subordinate
clauses are typically subordinated or less salient in discourse as well. Matthiessen and
Thompson (1988), for example, view hypotaxis as the grammaticalisation of a prop-
erty of the hierarchical structure of the discourse itself, and also Brandt (1990: 128)
concludes that – at least in the default case – the subordinate clause form of a continu-
ative RC is used to indicate that an information unit is less important than its matrix
clause. A similar view of the downgrading function that syntactic subordination has in
 Upgrading of non-restrictive relative clauses in translation 

discourse can be found in Carroll et al. (in this volume). The potential contribution of
subordinate vs. main clauses is also different as regards information-structure. Unless
explicitly marked to the contrary,4 a main clause always contains some new (in the
sense of ‘asserted’) information, whereas this is not necessarily the case for a subordi-
nate clause. In this way, syntactic upgrading may be accompanied by a change in the
information status of a discourse referent from being presupposed (given) to asserted
(new). However, while potentially relevant for the translations of the RC examples
discussed in this paper, this aspect of discourse interpretation will not be taken up in
the following discussion.
As mentioned above, what distinguishes a continuative RC from a ‘normal’ sub-
ordinate (or relative) clause is precisely its information-structural independence and
similarity to a main clause. We will now turn to the question of whether the different
functions a non-restrictive RC may have in discourse also play a role in how they are
translated into Norwegian, with particular regard to cases where a corresponding RC
in the Norwegian translation is not an option.

3.  Translational upgrading of relative clauses – a corpus study

3.1  Possible consequences of upgrading of relative clauses

Transforming a RC into an independent sentence is a translational operation that re-
quires relatively little syntactic restructuring in relation to the SL text. In many cases it
involves only smaller adjustments of the anaphoric links (such as replacing the relative
pronoun by a personal pronoun or a full NP), and the verbal predicate (for example,
the upgrading may require that a RC in the passive voice – typically starting with durch
(‘through’) + relative pronoun – is transformed into active voice in the translation).
Thus, if it holds that, in general, a translator tries to conserve as much of the ordering
and structuring of the SL text as possible (cf. Solfjeld in this volume), then it should be
expected that the syntactic upgrading of a RC will be one of the preferred translation
choices. Yet even seemingly minor modifications in the syntactic dependency struc-
ture of a sentence may have clear ramifications for its interpretation, depending on the
contribution the modified element makes to discourse development. One hypothesis
following from the discussion above is the following:
Hypothesis (H): The closer the RC is to an independent main clause and the
more it contributes to answering the quaestio, the less translational upgrading
will change the interpretation. This suggests that continuative RCs are less

.  For example, Fabricius-Hansen (2000: 73–74) discusses an example of a non-restrictive

German RC upgraded in the Norwegian translation, where givenness is marked by adding altså
(‘so’, ‘thus’) in order to block an interpretation as new information.
 Wiebke Ramm

problematic for discourse interpretation when syntactically upgraded in

translation than discontinuative/appositive RCs are.5

Example (5a) and (5b) – intra-lingually upgraded variants of (4a) and (4b) – illustrate
the intuition behind this hypothesis.
(5) a. Emil traf einen Bauern. Der/dieser trug übrigens einen weiten Mantel. [Ihn
fragte er nach dem Weg.]
‘Emil met a farmer. He/[dieser (dem.)] wore by the way a wide coat. [Him
asked he for the way.]’
b. Emil traf einen Bauern. Den/diesen fragte er dann nach dem Weg. [So konnte
er ohne Probleme den Gasthof finden.]
‘Emil met a farmer. Him/[diesen (dem.)] asked he then for the way. [So was
he able without problems the inn to find.]

In example (5a), where the discontinuative/appositive RC of (4a) has been upgraded,

the new independent sentence makes it more difficult to reconstruct the anaphoric
link between einen Bauern (‘a farmer’) in the first sentence and the masculine pronoun
ihn (‘him/it’) in the sentence following the upgraded RC. In fact, ihn (‘him/it’) in (5a)
is ambiguous, either referring to einen weiten Mantel (‘a wide coat’) – an interpreta-
tion supported by adjacency – or (the intended interpretation) to einen Bauern (‘a
farmer’) – an interpretation supported by parallelism of semantic roles. There is no
such ambiguity in the RC variant in (4a), due to the expectation of topic continuation
from the matrix clause in the first sentence of (4a) to the second sentence, i.e., due to
the syntactically downgraded nature of the RC. Example (5b), the upgraded version
of the continuative example (4b), however, does not cause any problems for reference
resolution, and the two versions seem to have very similar interpretations.
Hypothesis H allows for some interesting predictions regarding the practice of
how German RCs typically are translated into Norwegian:
Prediction (P1): Translators avoid upgrading discontinuative/appositive RCs in
Norwegian translation and use other translation strategies instead.
Prediction (P2): If discontinuative/appositive RCs nevertheless are upgraded in
translation (e.g., because no alternative translation options are available), this
leads to a less coherent text or to a different interpretation.

.  In the remainder of this paper we will use the terms ‘continuative RC’ and ‘discontinuative/
appositive RC’ as they are defined in Laux (2002) and Holler (2005). In theory, there could be ex-
amples which would be classified as continuative following Holler (2005) but not according to Laux
(2002) – namely, when a Consequentiality relation holds between the matrix clause and the RC,
but not to the following sentence (a constellation Laux (2002: 193) calls ‘blind end’ (Sackgasse) – in
practice, however, the correlation of the classification of the RC examples is relatively good, at least
in the examples discussed in this paper.
 Upgrading of non-restrictive relative clauses in translation 

To test these predictions, data from two parallel corpora of German original texts
translated into Norwegian were investigated: a corpus of three (popular-)scientific ar-
ticles on the subject of historical relations between Norway and Germany (consist-
ing of approximately 15000 words / 574 sentences), and the German-Norwegian part
of the Oslo Multilingual Corpus (OMC) containing fictional and non-fictional texts
(13 fictional and 17 non-fictional texts, about 40000 words in total at the time this
study was undertaken).6 As concerns the testability of the two predictions, it should be
pointed out that P1 and P2 have a different status: Whereas P1 should be testable on
the basis of a corpus study, namely by checking whether the relevant RCs are upgraded
or not, statements regarding P2 have to be more tentative, since a purely linguistic
analysis has its limits as regards judgements about the interpretation and cognitive
processing of a piece of discourse. The corpus data are discussed in the following two
sections: In Section 3.2, corpus examples of clause-related RCs upgraded in the Nor-
wegian translation are analysed and Section 3.3 addresses the translation of German
RCs with a nominal antecedent. Section 3.4. evaluates the results of the corpus study
with respect to the hypothesis and predictions formulated in this section.
A general observation as regards the translation of German RCs is that restric-
tive RCs appear to be never upgraded to independent sentences and almost always
correspond to RCs in the Norwegian versions of the texts. This does not come as a
surprise, since restrictive RCs denote properties (restricting the possible denotations
of the NP) and do not have so-called ‘root-clause’ properties. This implies that they
cannot be transformed into independent sentences without bringing about a semantic
change (Holler 2005: 30, Holler in this volume). But also many non-restrictive RCs
are translated by RCs in Norwegian, as long as grammatical contrasts between the two
languages do not block this option.

3.2  Translational upgrading of clause-related relative clauses

As regards clause-related RCs, some evidence confirming prediction P1 was found
in the OMC data which consist of a mixture of fictional (narrative) and non-fictional
texts,7 i.e., here the translators apparently avoid the upgrading of discontinuative/
appositive RCs: 42 out of 61 clause-related RCs found in the German SL texts were
discontinuative/appositive, and only 6 of these examples (14,3 %) were upgraded to
an independent sentence in the Norwegian translation. Paratactic clause coordination

.  See http://www.hf.uio.no/forskningsprosjekter/sprik/english/corpus/index.html for a descrip-

tion of the Oslo Multilingual corpus (OMC).

.  The data for clause-related RCs were retrieved from the OMC database by lexical search for
the individual relative adverbs. The relative marker was (‘what/which’) was excluded from the
study, because it was not possible to automatically filter out the instances of was not correspond-
ing to a relative marker.
 Wiebke Ramm

with og (‘and’), as in example (6b) below, is a pattern used relatively frequently as an

alternative translation strategy.8
(6) a. Meine Eltern untereinander sprachen deutsch, wovon ich nichts verstehen
durfte. Zu uns Kindern und zu allen Verwandten und Freunden sprachen sie
spanisch. [EC]
‘My parents with each other spoke German, [wovon (lit. ‘where-of ’)] I
nothing to understand was allowed to. To us children and to all relatives
and friends they spoke Spanish.’
b. Mine foreldre snakket tysk med hverandre, og det var ikke meningen at jeg
skulle forstå noe av det. Til oss barn og til alle slektninger og venner snakket
de spansk. [ECTN]
‘My parents spoke German with each other, and it was not intended that I
should understand anything of this. To us children and to all relatives and
friends they spoke Spanish.’

In the purely non-fictional corpus of scientific articles, however, evidence confirming

P1 was more tentative: 18 of 32 clause-related RCs found in the German SL texts were
discontinuative/appositive, but in several cases categorisation was problematic, cf. the
discussion below. 8 of these examples (44 %) were upgraded as an independent sen-
tence in the Norwegian translations.
One factor complicating the interpretation of these results is that it seems that
the different relative adverbs (wobei, weshalb, wodurch, etc.) behave differently. Their
individual meanings, often in combination with the occurrence of other connectives
(adverbs and particles), determine a continuative vs. discontinuative interpretation of
the RC in question. The relative adverb wobei (lit. ‘where-by’), for example, typically
introduces a clause-related RC describing some accompanying circumstance, which
in many cases leads to a classification as discontinuative/appositive as in example (3a),
repeated as (7a) below – 9 out of 11 wobei-examples in the corpus are discontinuative/
appositive. In her analysis of the German connective dabei (lit. ‘there-by’), which is
similar to wobei in that it consists of an anaphoric component combined with the
preposition bei (‘with, at’), Fabricius-Hansen (2005: 40) points out that “a main func-
tion of dabei is to prevent the narrative from ‘moving forward’”. Therefore dabei is
often added in German translations of English ing-progressive forms, a grammatical
resource which is not found in German. It is reasonable to assume that wobei has
a similar effect to dabei of blocking a continuative reading, not least because wobei
additionally triggers syntactic subordination (as do English ing-adjuncts), which is a
further indicator of a downgraded, discontinuative interpretation.

.  On the relative frequency and special use of clause coordination with og (‘and’) in Norwe-
gian texts in general, see Ramm and Fabricius-Hansen (2005), Fabricius-Hansen et al. (2005)
and Ramm (forthcoming).
 Upgrading of non-restrictive relative clauses in translation 

(7) a. Neben diesen engen wirtschaftlichen Kontakten bestanden rege kulturelle,

technische und wissenschaftliche Beziehungen,[I] wobei für das norwegische
Bildungsbürgertum die deutsche Literatur und die deutsche Sprache den
Bezugspunkt darstellte.[II] [Zahlreiche skandinavische Studenten studierten
in Deutschland, und zwischen den protestantischen Kirchen existierten enge
Bindungen.][III] [BB]
‘Besides these close economic contacts existed active cultural, technical
and scientific relations,[I] [wobei] for the Norwegian educated classes
the German literature and the German language the point of reference
represented.[II] [Numerous Scandinavian students studied in Germany, and
between the protestant churches existed tight connections.][III]’
b. Ved siden av disse tette handelsforbindelsene fantes det også viktige kulturelle,
tekniske og vitenskapelige kontakter.[I]Tysk språk og litteratur stod sterkt i de
utdannede borgerlige kretser.[II] [Utallige skandinaviske studenter studerte
i Tyskland, og det fantes tette bånd mellom de protestantiske kirkene.][III]
[BBTN]
‘Besides these close trade connections existed also important cultural,
technical and scientific contacts.[I] German language and literature had
a strong position in the educated middle-class circles.[II] [Innumerable
­Scandinavian students studied in Germany, and there were tight connec-
tions between the protestant churches.][III]’

Example (7) seems to give evidence for the validity of prediction P2 as the Norwegian
translation (7b) appears to be more ‘chopped up’ than the German (7a). One possible
explanation for this is that the relation of temporal and spatial overlap of the even-
tualities described in the matrix clause and the RC, signalled by wobei in (7a), is not
signalled by a corresponding connective in (7b) and thus has to be inferred from the
context only. Explicit marking of co-temporality and co-spatiality between the matrix
clause and the RC in (7a), together with the marking of syntactic dependency of the
RC, favours an interpretation of the matrix clause as being more salient than the RC,
which in turn assists the interpretation of the following sentence (7a[III]) as continu-
ing the matrix clause. Such an interpretation is much harder to get in (7b), since the
attachment of (7b[III]) in the discourse structure is not as obvious, i.e., it is less obvious
that the first sentence (7b[I]) is still on the ‘right frontier’ when processing (7b[III]), than
it is in (7a).
While the meaning of wobei itself clearly guides the interpretation of the RC
as ­appositive/discontinuative, it seems that adverbial connectives such as allerdings
(‘however’), freilich (‘admittedly’), letztlich (‘ultimately’) or bereits (‘already’), in combi-
nation with the relative marker and the semantics of the verbal predicate, often trigger
a continuative reading of the respective RC. In the scientific prose corpus, 9 out of 14
clause-related RCs introduced by the semantically underspecified relative marker was
(‘which/what’) contain an additional adverbial connective, and 6 of these examples are
continuative, as in (8a) below. This accords with Laux’s (2002: 260) observation that
 Wiebke Ramm

clear signals (e.g., lexical) are usually required to indicate the continuative discourse
function of a RC. Interestingly, in many cases the discourse relation expressed by an
adverbial connective in the German version is realised as a corresponding conjunc-
tion in the (upgraded) Norwegian translation, as in (8), where allerdings (‘however’) is
translated by men (‘but’) in the Norwegian version of the text.
(8) a. Vor allem nach der britisch-französischen Kapitulation auf der Konferenz
von München im Herbst 1938 mehrten sich die Anzeichen, dass das Deutsche
Reich eine wirtschaftliche und politische Interessensphäre in Skandinavien
anstrebte, was allerdings wegen der starken britischen ökonomischen
Präsenz in der Region nur wenig Erfolg versprechend war. [Dies zeigt, wie
weit die deutsche Außenwirtschaftspolitik sechs Jahre nach der so genannten
Machtergreifung noch davon entfernt war, eine Großraumwirtschaft zu
etablieren.] [BB]
‘First and foremost after the British-French capitulation at the conference
of Munich in autumn 1938 increased the indications, that the German
Reich an economic and political sphere of interest in Scandinavia strived
for, which however because of the strong British economic presence in the
region only little promising was. [This shows, how far away the German
foreign-economic politics six years after the so-called seizure of power was,
a Großraumwirtschaft to establish.]’
b. Framfor alt etter den britisk-franske kapitulasjonen under
Münchenkonferansen høsten 1938 ble det stadig tydeligere at Det tyske riket
siktet mot å opprette en økonomisk og politisk interessesfære i Skandinavia.
Men på grunn av den sterke britiske tilstedeværelsen i området hadde en
slik plan få utsikter til å kunne realiseres. [Dette viser hvor langt den tyske
utenrikspolitikken fortsatt befant seg fra å kunne etablere en økonomisk
kontinentalblokk, selv seks år etter nazistenes maktovertakelse.] [BBTN]
‘First and foremost after the British-French capitulation autumn 1938
became it more and more clear that the German Reich aimed to establish
an economic and political sphere of interest in Scandinavia. But because
of the strong British presence in the region had such a plan few chances to
be realised. [This shows how far away the German foreign policy still was
from to be able to establish an economic continental block, even six years
after the Nazi’s seizure of power.]’

A second factor that may blur the results of the corpus study is that, in many
cases, it is difficult to determine whether the RC is continuative or discontinuative – in
regard to both Laux’s and Holler’s approaches. A case on point is example (2a), from
a text discussing the role of Norway in German military strategy between 1914 and
1945. The example is repeated as (9a) below.
(9) a. Das militärstrategische Primat wird auch daran erkennbar, dass die
Militärbehörden durch die Rüstungsprogramme über erhebliche Kompetenzen
in wirtschaftlichen Fragen verfügten, weshalb deutsche Firmen auch mit
 Upgrading of non-restrictive relative clauses in translation 

diesen kooperieren mussten, obwohl die Wirtschaftspolitik formell durch das

Reichskommissariat gestaltet wurde. [Dadurch wurde nicht die zukünftige
Friedenswirtschaft eines ‘Großraumes’ vorprogrammiert, sondern Norwegen
in die aktuellen Bedürfnisse der deutschen Rüstungsproduktion eingebaut.]
[BB]
‘The military-strategic primacy becomes also [daran (pron. adv., lit. ‘there-
on’)] visible, that the military authorities through the arms programs
considerable competences in economic questions had, [weshalb] German
companies with them/these cooperate must, although the economic policy
formally through the Reichskommissariat formed/organised was. [Dadurch
(‘with this’) was not the future peace economy pre-programmed, but Norway
in(to) the actual needs of the German arms production built in.]’
b. Et forhold som reflekterer det militær-strategiske primatet, var at
rustningsprogrammene hadde gitt de militære myndighetene en betydelig
kompetanse i økonomiske spørsmål. Derfor måtte de tyske firmaene
samarbeide med dem, selv om den økonomiske politikken formelt sett ble
utformet av rikskommissariatet. [Følgelig ble ikke Norge programmert
for en plass innenfor en fremtidig kontinental økonomisk blokk, men
snarere utviklet i samsvar med de aktuelle behovene til den tyske
rustningsproduksjonen.] [BBTN]
‘A fact that reflects the military-strategic primacy, was that the arms
programs had given the military authorities considerable competence in
economic questions. Therefore must the German companies cooperate
with them, although the economic policy formally seen was formed by the
Reichskommissariat. [Consequently was not Norway programmed for a
place within a future continental economic block, but rather developed in
correspondence with the actual needs of the German arms production.]’

It is unclear whether the RC (together with the following concessive clause)

continues the events described in the matrix clause – which would correspond to a
coordinating discourse relation – or whether it comments on them – which would
rather support a subordinating discourse relation. Furthermore, this ambiguity is not
resolved in the following context, since the pronominal adverb dadurch (‘with this’) in
the sentence following the weshalb-clause is also ambiguous, i.e., it is open whether it
refers back to the whole clause complex (the intended interpretation?9) or only to the
clause-related RC (which would classify the RC as continuative according to Laux’s
definition). The følgelig (‘consequently’) in the Norwegian version (9b), however, has
its most likely antecedent in the sentence before, which is the upgraded counterpart
of the German weshalb-sentence. As a result, the connection between the economic

.  Due to space limits it is not possible to include more of the context before the example
­sentence. This context, however, would have made it clearer that taking dadurch as referring to
the whole preceding sentence, not only the RC, leads to a more plausible interpretation here.
 Wiebke Ramm

influence of the German military administration mentioned in the German matrix

clause, and the consequences this had on the role Norway played (economically, after
the First World War – which is what this text paragraph is about), are much harder
or almost impossible to infer in the Norwegian translation. What seems to make syn-
tactic upgrading of the RC additionally problematic in this example is the fact that
the RC in the German version is syntactically subordinate to a matrix clause which is
itself a subordinate clause. Translating the RC by an independent sentence thus ‘lifts’
the clause up two syntactic levels, possibly giving it too much prominence in discourse
structure when compared with the SL version.

3.3  Translational upgrading of relative clauses with a nominal antecedent

The examples of clause-related RCs illustrate that translational upgrading of a syntacti-
cally dependent clause to an independent sentence can change the discourse interpre-
tation and lead to problems, e.g., for the reconstruction of the referential structure of
the text passage. However, it is not clear to what extent this is due to the discontinuative
nature of the RC upgraded in the translation. This turned out to be even more difficult
to decide in the case of non-restrictive relative clauses with a nominal antecedent in
the matrix clause, since German RCs – particularly those in non-fictional texts of the
type seen in the scientific prose corpus – may be very complex. For instance, examples
with a RC depending on a matrix clause which is in itself a dependent clause (as in
(9a) above) seem to be even more frequent than is the case with clause-related RCs. A
noun-related non-restrictive RC may express very different types of things, depending
on the nature of the entities the RC is anaphorically linked to and the contribution of
the RC itself. This makes the application of either Holler’s or Laux’s criteria for distin-
guishing between discontinuative/appositive and continuative RCs almost unfeasible
in many instances. Whether the situation time in the matrix clause is completed, and
whether there is a shift of topic time between the matrix clause and the RC – two of
Holler’s diagnostic features for continuative RCs with a nominal antecedent (Holler
2005: 162) – is often difficult to resolve for complex examples in ‘wildlife’ texts, not
least as in some sentences it can even be difficult to identify the topic itself.
Example (10a) is a complex instance of a RC with a nominal antecedent, where
the RC contains additional information on a nominal discourse participant (Hitler);
but this description is not continued in the following sentence (although it could
have been), rather the claim in the matrix clause is taken up − namely that Rosen-
berg’s ‘Außenpolitische Amt’ did not have any influence on the ‘Auswärtige Amt’ and
Hitler.
(10) a. Das am 1. April 1933 unter Alfred Rosenberg gegründete Außenpolitische
Amt der NSDAP blieb in der Folge fast ohne Einfluss gegenüber dem
Auswärtigen Amt und Hitler,[I] der seit Ende 1933 punktuell in die
außenpolitischen Entscheidungsprozesse einzugreifen begann.[II] [Rosenberg
scheiterte wie auch Darré, der selbst ernannte völkische Parteiexperte für
 Upgrading of non-restrictive relative clauses in translation 

den Agrarsektor, damit, seiner Dienststelle in der sich herausbildenden

polykratischen Struktur des nationalsozialistischen Staates ein Profil und eine
Machtbasis zu verschaffen.[III]] [BB]
‘The on the 1. April 1933 under Alfred Rosenberg founded foreign-political
office of the NSDAP remained in the following time almost without
influence on the Foreign Office and Hitler,[I] who since the end of 1933
punctually into the foreign-political decision processes to intervene
started.[II] [Rosenberg failed as also Darré, the self-appointed national
(völkisch) expert for the agrarian sector, [damit (pron. adv.)] for his
department in the growing polycratic structure of the National Socialist
State a profile and a basis of power to provide.[III]]’
b. 1. april 1933 ble NSDAPs utenrikspolitiske kontor grunnlagt under
ledelse av Alfred Rosenberg, men den fikk svært liten innflytelse over
utenriksministeriet og Hitler.[I] Sistnevnte begynte fra slutten av 1933
av og til å gripe inn i de utenrikspolitiske avgjørelsene. [II] [På samme
måte som Darré, den selverklærte parti-ideologen for jordbruksspørsmål,
mislyktes Rosenberg i å profilere sitt kontor innenfor den flersidige,
kompliserte maktstrukturen som var iferd med å ta form innenfor den
nasjonalsosialistiske staten.[III]] [BBTN]
‘1. April 1933 was the foreign-political office of the NSDAP founded under
leadership of Alfred Rosenberg, but it gained very little influence over the
Foreign Office and Hitler.[I] The latter started from the end of 1933 oc-
casionally to intervene into the foreign-political decisions.[II] [In the same
way as Darré, the self-appointed party ideologist for agrarian questions,
failed Rosenberg to distinguish his office within the multi-sided compli-
cated power structure which was about to take form within the National
Socialist State.[III]]’

In terms of discourse relations as defined in SDRT (Asher and Lascarides 2003), the
RC would be an Elaboration of the matrix clause, which is the prototypical case of a
subordinating discourse relation. This means that the matrix clause as well as the RC
would be at the ‘right frontier’ of the discourse representation and thus could serve as
attachment points for information in the following discourse. In the German version,
syntactic subordination, here correlating with a subordinating discourse relation, helps
the reader to infer the appropriate attachment point for the following sentence (i.e., the
matrix clause), thus facilitating the processing of the complex information presented in
this text fragment. In the Norwegian translation in (10b), however, syntactic subordi-
nation as a discourse structuring signal is lost, and the discourse structure after (10b[II])
seems to be more open as to where following sentences could attach, when contrasted
to the German version. This leads to interpretation problems (garden path readings)
regarding (10b[III]), i.e., one first tries to interpret the sentence as a continuation of
(10b[II]) until one has to revise it upon encountering the subject ‘Rosenberg’. Placing
‘Rosenberg’ in sentence-initial position as in the German version would possibly facili-
tate the processing of this referent, but would not necessarily assist recognition of the
 Wiebke Ramm

hierarchical discourse structure. Thus, upgrading the RC does not block the right fron-
tier after (10b[I]) semantically − the same subordinating discourse relation can in the
end be inferred between (10b[I]) and (10b[II]). Nevertheless it causes difficulties for the
identification of discourse referents in the subsequent context, making the Norwegian
version appear less coherent than the German one, at least as far as reference structure
is concerned. So this example can be seen as evidence of the validity of P2, that trans-
lational upgrading of discontinuative RCs in some way changes the interpretation of
the texts.

3.4  Evaluation
As illustrated by the examples discussed so far, the upgrading of a RC almost always
changes the referential structure in the translation in some way. Often discourse ref-
erents have to be reintroduced or explicitated as a non-pronominal noun phrase as in
(10b), or the upgrading may force the disambiguation of referents which are (deliber-
ately) ambiguous in the source text, possibly leading to further interpretative problems
as in (9b). Furthermore, the discourse-structuring signal to process the matrix clause
and the RC together as one unit – also signalled by intonation if the text is read aloud –
is lost. This may cause ‘more trouble at the right frontier’, i.e., more garden path readings
of the context following the upgraded RC as in (10b), or it may lead to a change of the
discourse relations between the clauses involved, as occurs in (7).
Although it seems intuitively plausible that the continuative vs. discontinuative/
appositive discourse function of a non-restrictive RC should matter for its translation
into Norwegian, the predictions following from this hypothesis have proved difficult
to attest. This is probably due to the fact that the criteria for distinguishing between the
two discourse functions were often inoperable for the complex examples in this type
of text. One problem seems to be that approaches to discourse functions and discourse
relations often concentrate on texts of the narrative type, where the event structure
and its temporal development are decisive factors for the structuring and progression
of discourse. At least this holds for Klein and v. Stutterheim’s (1987, 1991) quaestio
approach and for Laux’s (2002) study of non-restrictive RCs. In fact, most of the clas-
sical examples illustrating the concept of coordinating and subordinating discourse
relations in SDRT in Asher and Lascarides (2003), are narrative. Many text passages in
the actual non-fictional corpus are, however, of non-narrative text types, where other
criteria, such as the description of (abstract) objects and their properties and relations
or the unfolding of argumentation, may guide the structuring of the discourse, as in
(11a) below.
(11) a. Unabhängig davon begannen im Herbst 1916 deutsche U-Boote in großem
Stil damit, den Dampferverkehr nach Russland um das Nordkap herum
anzugreifen,[I] wobei innerhalb von nur vier Monaten 143 norwegische
Schiffe versenkt wurden, ohne dass die Entente militärischen Schutz bieten
konnte.[II] [Diese brutale Form der Kriegsführung richtete sich nicht gegen
 Upgrading of non-restrictive relative clauses in translation 

Norwegen, obwohl […], sondern stellte einen deutschen Test dafür dar, ob ein
verschärfter U-Boot-Krieg gegen Handelsrouten der Entente Erfolge erzielen
konnte.[III]] [BB]
‘Independently from this started in autumn 1916 German submarines
in large scale [damit (‘with this’, pron. adv.)], the steamboat traffic to
Russia around the Northcap to attack,[I] [wobei] within four months 143
Norwegian ships were sunk, without that the entente military protection
to offer able was.[II] [This brutal form of warfare was aimed not against
Norway, although […], but represented a German test for whether an
intensified submarine war against trade routes of the entente success
achieve could.[III]]’
b. Uavhengig av dette innledet høsten 1916 tyske ubåter omfattende angrep mot
skipsfarten på Russland rundt Nordkapp.[I] I løpet av bare fire måneder ble
143 norske skip senket uten at ententen kunne tilby noen form for militær
beskyttelse.[II] [Denne brutale krigføringen var ikke rettet mot Norge, selv
om […], men var et tysk forsøk på å finne ut om en skjerpet ubåtkrig mot
ententens handelsruter kunne lykkes.[III]] [BBTN]
‘Independently from this initiated autumn 1916 German submarines
large-scale attacks against the shipping to Russia around the Northcap.[I]
In the course of only four months were 143 Norwegian ships sunk, without
that the entente could offer any form of military protection.[II] [This brutal
warfare was not aimed against Norway, although […], but was a German
attempt to find out whether an intensified submarine war against the en-
tente’s trade routes could succeed.[III]]’

The wobei-clause expands (‘elaborates’) on the German attack of the steamboat traffic to
Russia mentioned in the matrix clause, thus a subordinating discourse relation should
be assigned to the two clauses, following Holler’s (2005) approach. Further, no Conse-
quentiality relation can be identified according to Laux’s (2002) definition, neither does
the following sentence allow a Consequentiality interpretation. Nevertheless it clearly
relates to the clause-related RC (which cannot be omitted without the text becoming
incoherent). The demonstrative subject NP diese brutale Form der Kriegsführung (‘this
brutal form of warfare’) in (11a[III]) summarises the content of the preceding clause
complex and creates an attachment point for the rest of this sentence which interprets
the events described in the preceding clauses, (11a[I]) and (11a[II]), in this way pushing
forward the argumentation of the text. In the Norwegian translation (11b), the wobei-
clause is upgraded as an independent main clause without any explicit signalling of the
discourse relation. However, the Elaboration relation can be inferred from the context
without problem, and the interpretation of the Norwegian text diverges only slightly
from the German version in that the sentence following the upgraded RC, (11b[III]),
is interpreted as being related to (11b[II]) only, whereas the German original is more
ambiguous as to whether this sentence relates to the whole clause complex, ((11a[I])
and (11a[II])), or to the RC in (11a[II]) only. What this illustrates is that the RC in this ex-
ample seems in some way to be continuative, notwithstanding that Laux’s and ­Holler’s
 Wiebke Ramm

criteria of definition are left unsatisfied, which may be related to the criteria being tai-
lored towards the typical discourse progression in narrative texts. If this observation
is correct, namely that this RC would be continuative on a definition of continuativity
adapted to a non-narrative text type, this example would also support the validity of the
hypothesis presented in Section 3.1, viz. that translational upgrading of continuative
RCs does not significantly affect discourse interpretation.

4.  Discourse organisation strategies and coherence

The syntactic upgrading of German RCs in the Norwegian translations (for which
in some cases there is no alternative) raises some interesting questions regarding the
language-specificity of discourse organisation strategies. The loss of syntactic sub-
ordination renders the Norwegian translation less hierarchic (hypotactic) and more
paratactic than the German original in that there is a shift towards more syntactic
autonomy regarding the clause-linking pattern (Lehmann 1988: 189; see also Cosme
in this volume on the characteristic shifts of clause-linking patterns in translations
between Dutch, French and English). Splitting the SL clause complex into a sequence
of independent sentences at the same time increases the incrementality of the text, a
parameter of information packaging that accounts for the amount of new information
introduced per sentence (Fabricius-Hansen 1999). According to Fabricius-Hansen,
Norwegian translations of German scientific prose appear to follow the principles of
incremental discourse organisation (PIDO) stated as follows: (1) minimise the num-
ber of new discourse referents introduced per sentence, (2) minimise the information
stated per new or reactivated discourse referent in the sentences used to declare or
reactivate them in the discourse, and (3) minimise “the use of presuppositions trig-
gering expressions whose presuppositions are not justified by the preceding context”
(Fabricius-Hansen 1999: 184). The discourse organisation of the German original ver-
sions, however, is instead characterised by what Doherty (2006: 60) terms the strategy
of incremental parsimony (SIP). This implies that new information should be attached
to an appropriate point of attachment in the sentence under construction, if possible.
Thus, whereas for a text in Norwegian it is important not to ‘squeeze’ too much in-
formation into a single sentence, the guiding principle of German texts is to avoid
starting a new sentence if a piece of information can be attached to the sentence under
construction. According to Doherty (who investigates translations from English into
German), the two discourse organisation strategies PIDO and SIP compete with each
other and are regulated by Sperber and Wilson’s (1986) Principle of Relevance, aiming
at an optimal balance between processing effort and cognitive effect (Doherty 2006:
60). The trade-off between the two strategies may differ from language to language, the
availability of appropriate attachment points being one possible source of divergence –
as pertaining to the non-restrictive RCs discussed in this paper.
 Upgrading of non-restrictive relative clauses in translation 

The question is whether these discourse organisation strategies tell us any-

thing about the coherence and/or comprehensibility of the respective texts. Is a typical
­hierarchically organised German text consisting of complex clauses, where a lot of
complex information is expressed within the single sentence in a syntactically struc-
tured way, more coherent than a less hierarchically organised Norwegian version of
the text? And does this mean anything for the comprehensibility of a text? The exam-
ples discussed so far could give the impression that the less hierarchical organisation of
the Norwegian versions of the texts automatically makes them less coherent than the
German original texts, as illustrated by the problems with keeping track of discourse
referents and due to the greater effort required to integrate separate sentences into the
incrementally constructed discourse representation.
However, this is not always the case. The relative clause in (12a) below explains
(‘elaborates’) an ‘action-reaction scheme’ (referring to the German strategy in World
War II) mentioned in the matrix clause, which is a central topic of the whole para-
graph. In this case the following sentence takes up the topic of derartige historische In-
terpretationen (‘such historical interpretations’), i.e., what is said in the relative clause
and in the following contrastive während-adverbial clause.

(12) a. Hubatsch betrachtet die Besetzung des Landes von 1940 ebenfalls
innerhalb eines Aktions-Reaktions-Schemas,[I] bei dem das Deutsche
Reich aus strategischen Gründen den Briten zuvorkommen musste, um
die Rohstoffzufuhr aus Schweden sicherzustellen, während offensive
seestrategische Gesichtspunkte für die Kriegsführung gegen England keine
Rolle gespielt hätten. [II] [Erst in der Fischer-Kontroverse während der
späten 60er Jahre wurden derartige historische Interpretationen in der
Bundesrepublik außer Kraft gesetzt.][III] [BB]
‘Hubatsch looked at the occupation of the country of 1940 as well within
an action-reaction scheme,[I] [bei dem (‘where’)] the German Reich (nom.)
due to strategic reasons the British (dat.) would have to forestall, in order to
secure the commodity supplies from Sweden, while offensive sea-strategic
considerations played no role for the warfare against England. [II] [Not until
the Fischer controversy in the late sixties were such historic interpretations
in the Federal Republic suspended.][III]’
b. Hubatsch tolket også okkupasjonen av 1940 som en reaksjon.[I] I denne
tolkningen måtte Tyskland av strategiske grunner komme britene i forkjøpet
for å sikre råvareleveransene fra Sverige, mens offensive sjøstrategiske
synspunkter ikke hadde hatt noen betydning for krigføringen mot England.[II]
[Det var først under Fischer-debatten mot slutten av 1960-årene at slike
historiske tolkninger ble satt ut av kraft i Forbundsrepublikken.] [III] [BBTN]
‘Hubatsch interpreted as well the occupation of 1940 as a reaction. [I] In
this interpretation had Germany from strategic reasons to forestall the
British in order to secure the commodity supplies from Sweden, while
offensive sea-strategic considerations played no role for the warfare against
 Wiebke Ramm

England. [II] [It was only under the Fischer debate in the late sixties that
such historic interpretations were suspended in the Federal Republic.] [III]’

In this example, the upgrading of the RC in the Norwegian translation (12b) does not
lead to difficulties in interpretation. On the contrary, it seems that the nominal style
of the original German version (which is not untypical of this genre) – the RC being
syntactically only a NP modifier of Aktions-Reaktions-Schema – is quite ponderous:
Aktions-Reaktions-Schema is not taken up as a referent in the following sequence of
subordinate clauses (due to the nature of the RC construction chosen), until derartige
historische Interpretationen refers back to it in the following sentence (in non-initial
position). In the Norwegian translation, betrachtete […] innerhalb eines Aktions-
Reaktions-Schemas (‘looked at […] within an action-reaction scheme’) of the German
matrix clause is rendered as tolket […] som en reaksjon (‘interpreted […] as a reac-
tion), a translation which is licensed by the previous context. The RC is upgraded from
an informationally ‘heavy’ and complex NP modifier to an independent sentence by
anaphorically linking (12b[II]) to (12b[I]) with the sentence-initial adverbial i denne
tolkningen (‘in this interpretation’). Thus, the Norwegian version more explicitly keeps
track of the discourse referent ‘interpretation as a reaction’ than the German original
does and at same time is a smart translation solution for marking the indirect speech
indicated by the conjunctive mood of hätte (‘would have’) in the German version.
The example illustrates that the SIP strategy is not always applied appropriately
or may be overused, especially in texts with high information density such as scien-
tific articles.10 The tendency to use a nominal style attributed to German information-
oriented texts may lead to incongruence with or inadequate perspectivisations of the
states of affairs to be expressed, as in example (12a) where the bei dem-RC (structurally
a NP modifier) is a linguistic realisation that is rather difficult to recover in proportion
to the relative salience of its content.
As to the question of whether a hierarchical vs. non-hierarchical organisation
of discourse gives an indication of how coherent a text is, different factors seem to
interact and compete with each other and are constrained by the choices available
in the grammatical systems of the individual languages. One factor is the degree of
ease in establishing and maintaining (accommodating) referential relations in a text.
It can be argued (following Holler 2005: 211) that, with respect to this dimension of
coherence, non-restrictive RCs such as those in the German SL texts are superior to
their ­Norwegian translations by a corresponding sequence of independent sentences,
­because the syntactic link between the matrix clause and the RC restricts the search
space for the antecedent of the anaphor (the relative marker) introducing the RC. The
analysis of examples (9) and (10) may indicate that the search space for antecedents

.  One possible explanation for this is related to the space constraints authors often have to
meet.
 Upgrading of non-restrictive relative clauses in translation 

also plays a role for the resolution of anaphors in the sentence following a RC con-
struction (as in the German original versions), as opposed to anaphors in the sentence
following a sentence sequence (as in the Norwegian translations).
Further evidence that syntactic subordination as a RC has an impact on the ac-
cessibility of possible antecedents for anaphora in the following context comes from
research based on Centering Theory (Grosz et al. 1995). Miltsakaki (2005) analyses
English and Greek non-restrictive RCs and investigates whether a pronoun taking
up the topic (which typically means subject position in English) in the sentence fol-
lowing a clause complex containing a non-restrictive RC has its more likely anteced-
ent in the matrix clause or the RC. The study has revealed that processing a clause
complex containing a non-restrictive RC as a single unit together with the matrix
clause in most cases leads to a more coherent interpretation11 of the discourse seg-
ment than processing the matrix clause and the RC as a separate discourse unit each.
In the present study, the referential relations between entities in the matrix clause, the
RC and the following sentence are, in many cases, more complex (i.e., not restricted
to the topic only), therefore possibly not all of Miltsakaki’s results are comparable.
­Nevertheless, the present study also shows that syntactic subordination can operate to
signal a downplay in the salience of a discourse unit, but this is not always the case, as
the continuative RC examples demonstrate. In any case, subordination instructs the
reader to process the subordinate clause as one unit together with the matrix clause.
If a subordinate clause is (or must be) upgraded to a main clause in translation, then
the possibility of structurally signalling its relative prominence may be lost. If in ad-
dition this upgraded clause is separated from the translation of the matrix clause by a
full stop (and not by a semicolon or a comma, as in (6b) above), the signal to process
the two clauses as one unit is also lost which may lead to a more ‘chopped up’ text, as
in example (7b).
Assessing the coherence of a text is not only a matter of referent accessibility but
just as much a question of the recoverability of other types of content relations (con-
ceptual relations and discourse relations). Complex hierarchical information packag-
ing and high information density within a sentence do not imply that the information
can be easily unpacked by the reader, as illustrated by example (12a). Conceptual and
discourse relations are sometimes easier to retrieve when explicitly (lexically) sig-
nalled, but this usually makes a sentence longer and, at some point, the limits of what

.  In Centering Theory, coherence is measured in terms of the type of centering transition
computed between the topic update units. Centering transitions describe the type of topic con-
tinuation vs. topic shift in a discourse, where topic continuation is counted as the most coherent
transition (Miltsakaki 2005: 4). According to Miltsakaki (2005: 5), the question of whether each
tensed (main or subordinate) clause should count as a topic update unit (an utterance), or the
complex clause as a whole should be processed as a unit, was left open to empirical investigation
in the original version of Centering Theory – which is what Miltsakaki does in her study.
 Wiebke Ramm

can be expressed within a single sentence are reached. Moving pieces of information
into a new sentence, however, may cause new problems for keeping track of referents,
as seen above.

5.  Summary and conclusions

This paper has looked at the relation between subordination in syntax vs. discourse
from a translation perspective, taking upgrading of non-restrictive German RCs to
independent main clauses Norwegian – a frequent cause of sentence splitting in trans-
lations from German into Norwegian – as an example. Firstly, it has been shown that
contrasts between the grammatical systems of the two languages in regard to the re-
alisation of RCs are an important factor motivating this type of adjustment affecting
the (syntactic) hierarchical structuring of the SL vs. TL text and frequently also its
segmentation into sentences (discourse units separated by full stops).
We have taken a closer look at the discourse functions of German non-restrictive
RCs. On the basis of two approaches, viz. Laux (2002) and Holler (2005), we made
some predictions regarding the potential effects of translational upgrading of the dif-
ferent types of non-restrictive RCs upon the interpretation of the TL text. The two
accounts consider the temporal interpretation and the discourse relations holding
between the matrix clause and the RC, and Laux further takes into account the sen-
tence following the clause complex containing the RC. The main hypothesis was that
continuative RCs (weiterführende Relativsätze) would be less problematic when up-
graded as independent main clauses than discontinuative/appositive RCs because they
are more similar to main clauses than discontinuative/appositive RCs are. Continua-
tive RCs are non-restrictive RCs where, according to Holler’s (2005) approach, a sym-
metric/coordinating discourse relation (e.g., Narration) can be identified between the
matrix clause and the RC (applying the SDRT model of Asher & Lascarides 2003).
Alternatively, following Laux’ (2002) approach, a Consequentiality relation (as defined
in Sandström 1993) holds between the matrix clause and the RC as well as between
the sentence containing the RC and the following context. However, the hypothesis −
although intuitively plausible − proved difficult to test on the ‘wildlife’ data of two
German-Norwegian parallel corpora, one reason being the difficulties in determin-
ing the continuative vs. discontinuative/appositive nature of the RC in several of the
(non-narrative) text examples. As a possible explanation for these difficulties, it was
suggested that the criteria delineating the (dis)continuativity of a RC are based on
the typical discourse progression found in narrative texts, and that the application of
such criteria is not entirely appropriate for non-narrative text passages, such as those
in many of the corpus examples. With modified (i.e., text-type adjusted) criteria for
the definition of discourse continuity, the validity of the hypothesis seems to be better
supported by the corpus data.
 Upgrading of non-restrictive relative clauses in translation 

Translational upgrading of RCs changes the clause linking pattern from hypo-
taxis (clause subordination) to parataxis (clause coordination), cf. Lehmann (1988),
and in many cases – when an upgraded clause is separated from the preceding sen-
tence by full stop – also the incrementality of the text, i.e., it reduces the amount of
new information introduced or accommodated per sentence. Language-systematic
contrasts concerning the realisation of RCs can thus contribute to the preference of
Norwegian information-oriented texts (or at least of texts translated from German)
to be organised according to the PIDO principles of incremental discourse organisa-
tion (Fabricius-Hansen 1999), whereas it is rather SIP, the strategy of incremental par-
simony (Doherty 2006), which is followed in the German texts. The two discourse
organisation strategies potentially impact upon the establishment and maintenance
of discourse coherence in texts. Nominal referents for example may be linked ana-
phorically within or across sentence boundaries, a difference which influences their
recoverability. Inherent hierarchical or dependency relations between concepts may
be expressed by means of the sentence-internal syntactic structure (by adjunction or
syntactic dependency), they may be signalled lexically (e.g., by connectives), or they
may be contextually inferable between clauses or sentences. Finally, the communica-
tive weight (discourse prominence) that an utterance is intended to have, may be in-
dicated by sentence-internal structure (e.g., by syntactic subordination) or by lexical
means (across sentence boundaries).
Viewing coherence as a multi-dimensional phenomenon (cf. Stede in this volume
for a similar view), and taking into account the language-specific options for the realisa-
tion of the different dimensions, provides evidence for Doherty’s (2006: 60) claim that the
trade-off between the two discourse organisation strategies is language-specific. This im-
plies that coherence is a concept that is best measured against language-specific options
and preferences for discourse structuring within and across sentence boundaries. What
this means for text comprehensibility and text comprehension, however, is much more
opaque. The methodology used in this paper is not suitable to assess whether language-
specific preferences for hierarchical vs. non-hierarchical discourse organisation have any
implications for how texts are actually processed and understood. Whether (from the
perspective of native speakers) a hierarchically structured German text could be pro-
cessed and understood with equal ease as a less hierarchically structured Norwegian text
is only possible to test within the framework of a psycholinguistic methodology.

References

Asher, Nicholas. 1993. Reference to Abstract Objects in Discourse [Studies in Linguistics and
Philosophy 50]. Dordrecht: Kluwer.
Asher, Nicholas & Lascarides, Alex. 2003. Logics of Conversation [Studies in Natural Language
Processing]. Cambridge: CUP.
 Wiebke Ramm

Asher, Nicholas & Vieu, Laure. 2005. Subordinating and coordinating discourse relations. Lin-
gua 115: 591–610.
Brandt, Margareta. 1990. Weiterführende Nebensätze: Zu ihrer Syntax, Semantik und Pragmatik
[Lunder germanistische Forschungen 57]. Stockholm: Almqvist & Wiksell.
Doherty, Monika. 2006. Structural Propensities: Translating Nominal Word Groups from English
into German [Benjamins Translation Library 65]. Amsterdam: John Benjamins.
Faarlund, Jan Terje, Lie, Svein & Vannebo, Kjell Ivar. 1997. Norsk referansegrammatikk. Oslo:
Universitetsforlaget.
Fabricius-Hansen, Cathrine. 1992. Subordination. In Deutsche Syntax: Ansichten und Aussich-
ten, Ludger Hoffmann (ed.), 458–483. Berlin: de Gruyter.
Fabricius-Hansen, Cathrine. 1999. Information packaging and translation: Aspects of transla-
tional sentence splitting (German – English/Norwegian). In Sprachspezifische Aspekte der
Informationsverteilung, Monika Doherty (ed.), 175–213. Berlin: Akademie.
Fabricius-Hansen, Cathrine. 2000. Übersetzen mit Stil - ein unmögliches Ziel? In Übertragung,
Annäherung, Angleichung: Sieben Beiträge zu Theorie und Praxix des Übersetzens [Osloer
Beiträge zur Germanistik 25], Cathrine Fabricius-Hansen & Johannes Østbø (eds), 65–95.
Frankfurt: Peter Lang.
Fabricius-Hansen, Cathrine. 2005. Elusive connectives. A case study on the explicitness dimen-
sion of discourse coherence. Linguistics 43(1): 17–48.
Fabricius-Hansen, Cathrine, Ramm, Wiebke, Solfjeld, Kåre & Behrens, Bergljot. 2005. Coor-
dination, discourse relations, and information packaging: Cross-linguistic differences. In
Proceedings of the Symposium on the Exploration and Modelling of Meaning (SEM-05), Biar-
ritz, France, November 14–15 2005, Michel Aurnague, Myriam Bras, Anne Le Draoulec &
Laure Vieu (eds), 85–93.
Grosz, Barbara, Joshi, Aravind & Weinstein, Scott. 1995. Centering: A framework for modelling
local coherence in discourse. Computational Linguistics 21(2): 203–225.
Holler, Anke. 2005. Weiterführende Relativsätze: Empirische und theoretische Aspekte [Studia
Grammatica 60]. Berlin: Akademie.
Klein, Wolfgang & von Stutterheim, Christiane. 1987. Quaestio und referentielle Bewegung in
Erzählungen. Linguistische Berichte 109: 163–183.
Klein, Wolfgang & von Stutterheim, Christiane. 1991. Text structure and referential movement.
Sprache und Pragmatik 22: 1–32.
Laux, Britt Dalen. 2002. Die nicht-restriktive Relativsatzkonstruktion im Deutschen: Diskurs-
funktion und temporale Interpretation. PhD dissertation, NTNU Trondheim.
Lehmann, Christian. 1988. Towards a typology of clause linkage. In Clause Combining in Gram-
mar and Discourse, John Haiman & Sandra A. Thompson (eds), 181–225. Amsterdam:
John Benjamins.
Mann, William C. & Thompson, Sandra A. 1988. Rhetorical structure theory: Toward a func-
tional theory of text organization. Text 8: 243–281.
Matthiessen, Christian & Thompson, Sandra A. 1988. The structure of discourse and ‘subor-
dination’. In Clause Combining in Grammar and Discourse, John Haiman & Sandra A.
Thompson (eds), 275–329. Amsterdam: John Benjamins.
Miltsakaki, Eleni. 2005. A centering analysis of relative clauses in English and Greek. In Proceed-
ings of the 28th Penn Linguistics Colloquium, University of Pennsylvania, Philadelphia. Also
available as U. Penn Working Papers in Linguistics 11.1: 1–15. URL: http://www.cis.upenn.
edu/~elenimi/CenteringAnalysisOfRelativeClauses.pdf.
 Upgrading of non-restrictive relative clauses in translation 

Peyer, Ann. 1997. Satzverknüpfung – syntaktische und textpragmatische Aspekte [Reihe Germa-
nistische Linguistik 178]. Tübingen: Niemeyer.
Ramm, Wiebke. Forthcoming. Discourse-structural salience from a cross-linguistic perspective:
Coordination and its contribution to discourse (structure). In Salience. Multidisciplinary
Perspectives on its Function in Discourse [Trends in Linguistics. Studies and Monographs
(TiLSM)], Michael Grabski et al. (eds). Berlin: Mouton de Gruyter.
Ramm, Wiebke. In preparation. Satzgrenzenveränderungen in der Übersetzung: Zum Zusam-
menhang von Informationsportionierung und Diskursstruktur am Beispiel der Sprachen
Norwegisch und Deutsch. PhD dissertation, University of Oslo.
Ramm, Wiebke & Fabricius-Hansen, Cathrine. 2005. Coordination and discourse-structural sa-
lience from a cross-linguistic perspective. In Salience in Discourse: Multidisciplinary Approach-
es to Discourse 2005, Manfred Stede, Christian Chiarcos, Michael Grabski & Luuk Lagerwerf
(eds). Münster: Stichting/Nodus, 119–128. Also available as SPRIKreport Nr. 30. URL:http://
www.hf.uio.no/forskningsprosjekter/sprik/docs/pdf/wr/WR-CFH-Report30.pdf.
Sandström, Görel. 1993. When-clauses and the Temporal Interpretation of Narrative Discourse.
PhD dissertation, University of Umeå.
Sperber, Dan & Wilson, Deirdre. 1986. Relevance: Communication and Cognition. Oxford:
Blackwell.

Corpus texts referred to in the examples

[BB]: Barth, Boris 1999. Norwegen und der Norden in der deutschen Strategie 1914 bis 1945. In
Deutschland – Norwegen. Die lange Geschichte, Jarle Simensen (ed.), 165–186. Oslo: Tano
Aschehoug.
[BBTN] Translation: Hobson, Rolf. Norge og Norden i tysk strategi, 1914–1945. In Tyskland –
Norge. Den lange historien, Jarle Simensen (ed.), 145–163. Oslo: Tano Aschehoug.
[EC]: Canetti, Elias 1977. Die gerettete Zunge. München: Carl Hanser.
[ECTN] Translation: Qvale, Per 1982. Den reddede tungen. Oslo: Gyldendal.
 Subordination in narratives and
macro-structural planning
A comparative point of view

Mary Carroll, Antje Rossdeutscher, Monique Lambert

and Christiane von Stutterheim
University of Heidelberg/ University of Stuttgart/ University of Paris VIII/
University of Heidelberg

Based on a series of cross-linguistic studies covering Germanic (Dutch, English,

German) as well as Romance languages (French, Italian, Spanish), the present
paper addresses questions concerning the nature of the decisions made at the
level of macro-structural planning when producing a narrative. Focusing on
English, German, and French, it presents evidence for a hierarchy of factors
and associated constraints that are both grammatically based and perspective
driven. The comparison shows how some of the constraints at issue lead to cross-
linguistic differences in the way subordination as a downgrading operation is
implemented in the narratives.

Keywords: Macro-structural planning principles, decision hierarchy,

downgrading operations, crosslinguistic study.

1.  Introduction

The analyses presented in this paper are based on retellings of a silent animation film
(Quest) where speakers were asked to tell ‘what happened’ (20 per group). The type of
information organisation required in carrying out a task of this kind can be described at
different levels of analysis. There are questions involving information selection (deciding
what to say), thematic continuity (e.g., topic assignment), referential framing, which re-
lates to predicate-argument structures and how they are anchored with respect to times,
worlds, and spaces. In order for a sequence of propositions to be coherent, these referential
properties have to be related in consistent terms across utterances. The question is: are the
issues solved for each sentence anew or are there macro-structural principles that guide the
speaker at each relevant stage in the narrative? For example, is the decision to map an agent
of an action as subject of a main or subordinate clause made individually at each point in
 Mary Carroll et al.

the narrative, or are there principles that decide the issue on a default basis for the narrative
sequence as a whole? As mentioned above, the present paper is based on cross-linguistic
findings that show how the types of decisions that speakers preferably make correlate with
grammaticised means in the respective languages. The present objective is to show how
factors guiding decision making in information structure are hierarchically organised at
the level of macro-structural planning.
Given the assumption that sequencing principles form the underlying conceptual
structure in a narrative task (‘tell what happened’), the narratives are grouped in the
following according to the underlying temporal frames of reference and associated
perspectives. Although the present focus is placed on English, French, and German,
results are included in some cases for Dutch, Spanish, and Italian in order to illustrate
the overall typological trend (cf. von Stutterheim, Carroll & Klein 2003; Carroll &
Lambert 2003, 2006; Rossdeutscher & von Stutterheim 2006).
The following section presents a summary of the relevant findings, while section 3
gives an overview of the associated quantitative analyses on which these are based; sec-
tion 4 shows how patterns in information structure can be traced to a set of grammatically-
driven principles that guide decisions at the level of macro-structural planning in narra-
tive tasks and is followed by the first step towards a formal model of possible hierarchical
orders in section 5.

2.  The narrative task

Speakers were asked to retell the content of the silent film Quest which lasts approx.
9 minutes. It portrays the adventures of a clay figure, the only animate-like protagonist
in the film, who tries to find his way in a hostile world ‘peopled’ by natural forces (high
winds, flying rocks, huge sheets of paper) – which in some way or another get in his
way in his quest for salvation. For the analysis of information structure the constella-
tion is simple. There is one main character with a well defined status (animate-like)
with respect to the other major players (inanimate), so that mapping patterns in in-
formation structure can be compared accordingly. Although the speakers were asked
to tell ‘what happened’, the present tense was selected spontaneously across all the
languages studied (20 speakers per group).

2.1  Temporal frame with a protagonist-based perspective

Information structure for the narrative sequence may centre on a temporal frame of
reference that is organised on the basis of temporal shift. This constitutes the preferred
option in the German data and is marked by the following factors:

i. The main character is accorded a higher status compared to other agents. It is eli-
gible for mention as the subject of a main clause. The other agents (inanimate entities
 Subordination in narratives and macro-structural planning 

such as gusts of wind, falling rocks, etc.) are downgraded in that selection for men-
tion is restricted. Furthermore, when inanimate entities compete with the protagonist
for mention as subject of the clause (inanimate entity is agent, for example), they are
typically mapped into a passive construction as the logical or underlying subject. This
ensures maintenance of the status of the main character at the level of thematic conti-
nuity (see line 016 in example 1 below). In this sense the main character/protagonist is
assigned ‘topic’ status.
ii. In reference maintenance there is additional evidence for informational status ac-
corded to the protagonist since the means used involve zero anaphora (cf. lines 16, 17,
and 19, 20 in example 1 below).
iii. In the narrative sequence, the main structure of the text, events are presented
as constituting a change in state. This is typically brought about by an animate agent
who acts to bring about the change in state. The temporal relation is anaphoric in that
events are construed as having reached a point of completion (pc). Linkage is based
on the relation what happens after pc, whereby the post time of the preceding event
is taken as the topic time of the subsequent event. This is typically expressed by the
temporal shifter dann, ‘then’ (cf. in detail von Stutterheim, Carroll, Klein 2003; Ross-
deutscher & von Stutterheim 2006).

For German speakers, a protagonist-based perspective and temporal shift form the rel-
evant criterion in organising narrative progression. Before looking at the implications
with regard to decisions at the level of macro-structural planning, we will take a closer
look at examples of this pattern in information structure. Use of the passive to ensure
thematic continuity is illustrated in example 1 below (line 16). It occurs when another
agent comes into competition with the protagonist for mention as subject of a main
clause, as mentioned. The rate of occurrence for the passive in this kind of context is
high at 72.7 per cent (Murcia Serra 2001).

(Ex.1) de21
[12] das Männchen fällt runter in eine Papierwüste mit lauter einzelnen quadratischen
Blättern
the little-man falls down into a paper-desert with many
single square sheets
[13] und in diesem Raum weht ein starker Wind
and in this place blows a heavy wind
[14] und wirbelt die Papierblätter durch die Gegend
and whirls the paper-sheets through the area
[15] und das Männchen muss dauernd ausweichen
and the man must constantly avoid
[16] und wird dann schließlich von einem Blatt getroffen
and gets then finally by one sheet hit
 Mary Carroll et al.

[17] und fällt zu Boden

and falls to ground
[18] dann steht es wieder auf
then gets it again up
[19] und findet eine nasse Stelle im Papierboden
and finds a wet place in the paper-floor
[20] fängt an nach dem Wasser zu graben
begins after the water to dig

Dutch retellings show similar principles with regard both to perspective taking and the
temporal frame (same task). In other words, similar preferences in information structure
were observed in the two languages which share the ‘verb second’ constraint for word
order. In these languages the finite verb constitutes the second main constituent in main
clauses. Significantly, the verb can be preceded by only one main constituent, and this
need not be the subject. The ‘slots’ created by the verb second constraint are referred to
as the ‘Vorfeld’ and ‘Mittelfeld’. Both are relevant in topic assignment (Frey 2000, 2005).
We assume that there is an interdependency between the constituents that may typi-
cally occur in the ‘Vorfeld’ in V2 languages in the given task. For narratives (answering
the quaestio ‘what happened’) these are linguistic means that encode temporal relations
(such as dann, (‘then’), as well as the syntactic subject. Evidence for an interdependency
lies in the fact that in information selection in both Dutch and German (deciding what
to say), entities and events are filtered with respect to their ability to accommodate tem-
poral relations such as shift. Events that do not accommodate this relation are less likely
to be selected for mention (cf. in detail Carroll & Lambert 2003). In the narrative task
the protagonist is accorded ‘topic’ status on a global basis for the task as a whole, since
an intentional agent is more likely to be involved in events that reach a goal or point of
completion, thus accommodating temporal shift (typically expressed by dann (‘then’)).
The next example is taken from the English data and illustrates a temporal frame
based on anaphoric shift, and a ‘protagonist-based’ perspective with comparable use of
the passive in downgrading agents involving natural forces (cf. line 11). In contrast to
German, however, this is not the preferred pattern in information structure in retell-
ings in English, as will be illustrated below.
(Ex.2) ee03
[9] ok the man arrives in a paper world [10] and eh everywhere is covered with paper
[11] and he gets hit by the flying piece of paper [12] and then he walks / he hustles
around [13] and he finds a damp piece of paper [14] and he pushes the paper [15]
and then he falls through the paper

Although information organisation is similar at the temporal level, there are marked
differences compared to the German texts. In the English retellings maintenance of
the protagonist as subject across adjacent clauses does not constitute a criterion for
zero anaphora (VP-coordination). Pronouns are used and zero anaphora is restricted
to events that are very closely related in causal terms, regardless of the type of agent.
 Subordination in narratives and macro-structural planning 

This was also found in the larger set of retellings in English (see below) as well as in
French retellings (cf. Carroll & Lambert 2006 on causality and the contexts in which
zero anaphora is licensed in English and French). In German, by contrast, zero ana-
phora predominates and is closely bound to the event sequence and the temporal
means used to establish it (this will be taken up again below). While the protagonist-
based perspective is representative of a large group of coherent German narratives, the
English example belongs to a small set (3/20 speakers).

2.2  Temporal frame deictic with narrator-perspective

For the majority of English speakers (17/20) the relevant indicators in information
structure can be summarised as follows:

iv. All agents, both animate and inanimate, are selected for mention and accorded
the same status in that they can be mapped as the subject of a main clause; use of
downgrading procedures for this category is thus very low, compared to German (see
below).
v. In the temporal frame events are linked to a topic time which is externally an-
chored and deictic (‘now you see’; or ‘then you see’). All types of situations can be
hooked up to this temporal anchor, both bounded and unbounded, and ongoing events
(expressed by the progressive ‘–ing’) may form an integral part of the sequence; there
are strict constraints in switching from potentially bounded events (simple tense) to
ongoing events within an event sequence; significantly, the temporal relation expressed
by ‘then’, where this occurs, relates to the left boundary of the preceding event, and not
to the right boundary; it entails precedence but not necessarily completion of the prior
event; this principle is crucial since it facilitates the integration of ongoing events into
the story line (cf. Carroll & Lambert 2006).
vi. Causal relations (‘so’, ‘because’) predominate in advancing the story line.
vii. The means used in reference maintenance to the main character are pronouns.
Zero anaphora occurs when the core relation in advancing the story line (causal) holds
between the events in question (as in French retellings).

In this sense use of zero anaphora in information structure may be viewed as linked to
the type of relation at issue as much as to the entity to which reference is maintained.
In information organisation in German topic assignment and anaphoric shift coincide,
while in English and French causal relations and zero anaphora coincide.
Starting with narratives with the deictic anchor ‘then you see’, line 44 gives an ex-
ample of natural forces acting as the agent of an action. In contrast to the protagonist-
based perspective, it is mapped into a main clause.
(Ex.3) ee13
[34] then you see the sand man [35] falling from the sky onto this surface [36] and
­apparently falling through another world from the world of sand [37] and he gets up
 Mary Carroll et al.

[38] and he walks around [39] trying to figure out where he is [40] and it’s windy [41]
and those papers are blowing [42] there is a little sort of tornado going by him [43] as he
is walking [44] a piece of paper flies in his face [45] and he falls down [46] and he gets up
[47] and he takes the paper away from his face [48] and he / in front of him sees water
drip onto the surface onto the part of the surface [49] which is wet [50] and he goes over
to it [52] and he kneels / gets down on his hands and his knees in front of it [53] and he
feels the paper [54] he feels that it’s wet [55] so he reaches his hands up to wait for the
next drop of water [58] so where its wet on the paper surface [59] he just starts to dig [60]
like he was digging in the sand [61] and he only digs for a little while [62] and because the
paper is wet [63] and he ends up falling through the paper.

Occurrence of natural forces as subject of a main clause is the predominant pattern

(61.1%) – when these forces act as agents and compete with the protagonist for mention
as subject. In the remaining cases a passive construction or subordination is used. The
following example illustrates use of the deictic anchor ‘now you see’. As in the first
example for English above, zero anaphora is confined to events that are closely linked,
as in 48/49.

(Ex.4) ee02
[34] okay eh now you see [35] that the man has fallen onto another like really bleak
landscape [36] and it looks [37] as if it as all made out of paper [38] and there are like
sheets of paper [39] lying on the floor [40] and he falls down onto this ground [41] and
eh gradually he stands up again [42] and he can hear a wind [43] blowing a little bit [44]
and these pieces of paper keep flying past him [45] and he can still hear the sound of eh
water [46] falling [47] and dropping onto the ground [48] so he gets up [49] and goes to-
wards the sound [50] and as he does this [51] you see like [52] it’s a paper [53] flying past
him [54] and they’re quite big [55] they’re like the size of him [56] so one knocks him over.

In contrast to protagonist-based narratives where temporal shift is defined over

events with right boundaries, temporal sequencing, in this strict sense, is often un-
derspecified. The relation between one event and the next is supported by causal
relations (so, as in lines 48 and 56). In contrast to the use of dann (‘then’) in German,
the temporal shifter then relates to the precedence of the prior event (its left bound-
ary), and does not entail that this event has reached a point of completion before
the next one begins, as mentioned above. This factor, along with the presence of the
deictic anchor ‘you see’, facilitates the integration of unbounded ongoing events at
any point in the story line. Transitions from a bounded to an ongoing event are me-
diated by reference to a state or by the narrator (‘you see’): a pile of stones pushes him
up out of the ground and you see him looking around; unmediated sequences where a
bounded event is immediately followed by an ongoing event are not licensed; com-
pare – x pushes him up, he is looking around to the example above (cf. in detail Car-
roll & Lambert 2006). The narrator as deictic anchor ensures adequate integration of
ongoing events into the sequence, as in lines 49, 50, 51, and 52 in the example above,
 Subordination in narratives and macro-structural planning 

where a transition of this kind is mediated by ‘you see’, combined with the marking
of simultaneity in (50).

2.3  Protagonist- and narrator-based perspective

On the whole, speakers of French follow a consistent frame of reference which can
be described as incorporating both dimensions to form a specific cluster. English and
French share the feature ‘subject prominent’ in information structure:

i. All agents, both animate and inanimate are selected for mention and mapped as
subject of a clause, as in English.
ii. However, events involving natural forces are downgraded and related entities are
mapped as subject of a subordinate clause when advancing the story line; this means
that non-intentional agents are downgraded, in contrast to English.
iii. The change in status accorded to different event types (switch from main clause
to a subordinate clause) leads to disruptions in the temporal chain of bounded events
given in main clauses, since anaphoric linkage, which relates to the right boundary of
an event in an adjacent clause, is interrupted.
iv. Predicates relating to intentions, attitudes and perception are accorded a predom-
inant status in that they allow for the introduction of a causal chain, thus bridging
events that differ in informational status.
v. Events are also linked externally as components of an interpretative frame; al-
though interpretation occurs in all languages, a ‘narrator perspective’ is present to a
significantly high degree in French retellings.

Ex. 5 illustrates a typical sequence where many events are unbounded, showing how
this temporal property is often underspecified. Lines 63 to 68 express situations of
perception or reflection by the protagonist. Furthermore the text illustrates the rela-
tive predominance of causal relations over temporal relations (donc, (‘so’) i.e., causal)
as stated above.
(Ex. 5) ef08
[61] il s’aperçoit
he realises
[62] que c’est mouillé partout
that it is wet all around
[63] donc là aussi il est un petit peu perplexe, on peut dire
so again he is a little bit astonished, one might say
[64] d’abord il essaie de récupérer l’eau par terre
first he tries to collect the water from the ground
[65] il s’aperçoit
he realises
[66] que c’est pas possible
that this is not possible
 Mary Carroll et al.

[67] donc il essaie plutôt de l’attraper vers le haut

so he rather tries to catch it from above
[68] comme s’il est en attente d’une offrande ou quelque chose comme ça
as if he were waiting for an offering or something like that
[69] et il essaie à nouveau de retrouver cette goutte d’eau
and he tries again to find this drop of water
[70] et il s’engouffre à nouveau
and he is engulfed again
[71] oui, il creuse en fait
yes, he digs basically
[72] pour essayer de trouver
in order to find out
[73] où va cette eau
where this water goes to
[74] et il s’enfonce à nouveau dans un gouffre
and he sinks into the abyss again

The following example again illustrates this overall pattern, showing how events with
non-intentional agents are mapped into subordinate clauses, although the events in
question form part of the story line (e.g., 62, 63, 64).
(Ex.6) ef02
[57] et c’est à ce moment là qu’il trempe ses mains
and in this moment he immerses his hands
[58] puisqu’il s’est dit que
because he said to himself that
[59] comme il ne pleut il ne pleuvait pas
since it is not raining/was not raining
[60] l’humidité ne pouvait venir que de ce sol
the wetness could not come from somewhere else than the ground
[61] et donc il trempe ses mains dans cette feuille humide
and so he immerses his hands into the wet paper
[62] qui se froisse
which crumbles
[63] et se déchire
and tears
[64] et l’emporte en profondeur
and takes him into the depths

The analysis also shows that, although both dimensions are present, speakers may dif-
fer with respect to the relative weight attributed to the two worlds (narrator, protago-
nist). Some speakers refer to the narrator on a marginal level in introductions (6 out
of 18), taking the protagonist as the global topic, while 12 out of 18 speakers rely on an
interpretative frame, with inclusion of the narrator as witness on a consistent basis. In
addition to the narrator as an interpreting subject and prominent source for implicit or
 Subordination in narratives and macro-structural planning 

explicit causal relations in the latter group, the main character (the clay figure) can take
on this role as well, given its salient characteristics as a human-like being. This gives
a ‘two-layered’ form of macro-structural organisation with both the narrator as well
as the protagonist as interpreting subject, both with the same potential in establishing
coherence between events.
In the following we will give a brief quantitative overview over the relevant fea-
tures in the French data with respect to expression of attitude or stance (il est perplexe;
‘he is perplexed’; il s’aperçoit, ‘he realises’) and the type of relation used in linking states
of affairs (causal, temporal): In French, verbs expressing attitude or stance occur more
frequently, as compared to German (French 40.2%; German 18.1%). Causal relations
(donc (‘so’), for example) predominate in French (12.6%), in contrast to temporal shift
(3.8%) as expressed by et puis (‘then’). In German, by contrast, temporal shift is the
most frequent means expressing linkage (28.2%), compared to causal relations (3.8%),
as mentioned above.

3.  D
 owngrading operations in French, German, and English:
Quantitative overview

This section provides a quantitative overview of the results presented above. We have
illustrated how the use of downgrading, in the form of subordinate clauses and pas-
sive constructions, is driven by the temporal frame and thematic continuity. In French
and English it was shown how all agents are eligible for mention as subject of a clause.
They are downgraded in French by means of subordination (subject of a subordinate
clause), even though the events in question serve to advance the story line. Downgrad-
ing does not occur on a systematic basis in English, reflecting differences that can be
linked to the grammaticised means available to express temporal relations.
The following figures on subordination in the narratives, covering both perspec-
tives, reflect the scale on which this form of downgrading is implemented in the three
languages. As indicated above, overall occurrence is highest in French: English 742/3886
utterances (19.1%); German 366/2740 utterances (13.3%); French 850/2297utterances
(37.0%). The preferences shown do not constitute an isolated phenomenon but fall in
line with other decisions made in information structure. In the following summary
we will indicate how they form part of a set of principles, starting with information
selection.

3.1  Downgrading at the level of information selection

In German, where the story line is advanced by means of the temporal relation based
on anaphoric shift (relating to events with a right boundary), speakers do not men-
tion natural forces to the same extent, as illustrated above. The following overview
illustrates differences in information selection across the different languages. In order
 Mary Carroll et al.

to illustrate the overall preferences, it takes into consideration another member of the
V2 group (Dutch), as well as other members of languages that are ‘subject prominent’
in information structure (Italian and Spanish). The following figures cover mentions
in all clause types.
References to inanimate entities (natural forces) as agent or experiencer in an
event (e.g., sheets of paper are swirling around; rocks push up out of the ground; a sheet
of paper knocks him over), amount to 34.5% in English, 35.8% in French, 37.1% in
Italian, 30.9% in Spanish. But there is a significant drop in both German 24.5% and
Dutch 23.6%. The difference between speakers of German and Dutch and the other
languages is statistically significant: t-test English – German p = 0.006 highly significant;
Spanish – German p = 0.04 significant; while English – Spanish p = 0.25 is not signifi-
cant; English – French p = 0.21 is also not significant. These figures cover mentions
as subject of both main and subordinate clauses and are based on average values, in
percent, for 20 speakers per group (cf. in detail Carroll & Lambert 2003). So speakers
of German and Dutch not only downgrade by use of the passive, they also ‘defocus’
at a higher level, so to speak, in that in many cases some entities and events are not
selected for mention at all.
The criterion for selection is relevance for the protagonist, as will be shown in the
examples below. By contrast, in English and French speakers mention minor events,
even when they are not of direct relevance for the protagonist – events that pose a
threat but do not actually affect the protagonist. The first example illustrates the rel-
evance principle and speakers of all the languages studied mention the following situ-
ation in which a natural force is the agent of the action.
English: It’s very windy with a lot of paper and a very big sheet hits him in the face
and knocks him down: Mention of this event (given in percent) amounts to 90.0 in
English and 60.0 in German. Mention in German is markedly lower however, if there
are no immediate consequences for the protagonist, as when the rock in the next event
misses him, for example:
English: The first thing we see is a rock dropping from the sky directly towards his
head. Mention in English is high at 80.0, compared to 40.0 in German. Another ex-
ample is given with the event where a saw slightly damages the clay man’s foot .
English: As he was lying there this saw came right by and slightly damaged part of
his toe. The rate in this case is 70.0 in English compared to 30.0 in German. Relevance
with respect to the protagonist is not at issue in information selection in English (or
French and Spanish), in contrast to German.

3.2  Mapping inanimate entities as subject of a main clause or not

The next set of figures gives the extent to which events involving natural forces are mapped
into a main clause within the narrative sequence in the different languages. Again, figures
for Italian, Spanish, as well as Dutch are included in order to underline overall trends.
Mention of natural forces as subject of a main clause are as follows: English 26.0, Spanish
 Subordination in narratives and macro-structural planning 

24.1, while the values in French and German are lower with French 8.0, German 14.0.
The figures for German and French reflect downgrading by means of the passive, sub-
ordination, or omission (cf. Carroll & Lambert 2003). (The figures are percentages and
references to the remaining option, the protagonist or the narrator amount to: English
74.0, French 92.0, German 86.0.

3.3  Reference introduction

Focusing and downgrading patterns can be observed with the means used in the in-
troduction of natural forces. In English where natural forces are eligible for mention as
subject of a main clause throughout the narrative, this status correlates with the way in
which they are typically introduced. The referents are introduced as the main content
of a clause. In other words, there is a clause which serves to introduce the entity at
issue (presentationals). This contrasts with the second option in which referents are
embedded in clauses involving actions of the protagonist. Taking, for example, the first
mention of new entities in the case of ‘sheets of paper’, the preferred means in English
are as follows:
(Ex. 7) There are sheets of paper swirling around (main content of clause)

This contrasts with German (as well as French) where first mentions are more fre-
quently embedded:
(Ex. 8) Das Männchen fällt nun auf eine neue Ebene (…), die mit Blättern (…) also
keine Baumblätter sondern Schreibblätter große Blätter überdeckt ist.
The little-man falls now onto a new level (…), which with leaves (…), not tree-
leaves but sheets of paper big sheets covered is.
(Ex. 9) Dieses Wesen scheint in die nächste Ebene gefallen zu sein, wo da der
Fußboden aus lauter einzelnen Papierstücken besteht.
This creature seems into the next level to have fallen, where the floor out of lots
of single sheets of paper consists.

As illustrated in the examples, the entities at issue are often introduced in German nar-
ratives as a property of the place where the protagonist lands. Overall figures for these
options are as follows: In English, introductions as the main content of a clause, given
in percentages, is 83.1 while the embedded form amounts to 16.9; (Spanish: main con-
tent 70.9, embedded 29.1).
In contrast to English, these options in reference introduction occur with compa-
rable frequency in French and German: In French, introductions as main content of
the clause amount to 54.6, while the embedded form is 45.4; in German, introduction
as the main content of the clause accounts for 58.6 of the cases and the embedded form
for 41.4. (cf. Carroll & Lambert 2003).
Up to now we have isolated some of the main downgrading operations, showing
how they are implemented in consistent terms in the different languages.
 Mary Carroll et al.

3.4  N
 arrator-perspective and downgrading at the level of the
narrative as a whole
With this perspective, speakers explicitly refer to the narrator’s world by means of
then you see (example 3 above, line 34), or now you see. The events in the story world
referred to in lines 35 and 36 in ex. 3 above are mapped into dependent clauses. The
subsequent story world descriptions, as a whole, are subordinate in the sense of modal
subordination, where all events in the narrative are presented as witnessed (cf. Roberts
1989). Subordination can take the form of a complement clause introduced by that.
This can be compared with other possible patterns of subordination at this level,
taking, for example, the following retelling of a German speaker, one of the very few
who produces a series of narrative descriptions with syntactic subordinate clauses, i.e.,
relative clauses. The pattern is one of the few in German in which the perspective is
narrator based.
(Ex.10) (de013)
[1] am Anfang sieht man so ne Art Pappmachemännchen im Sand
at-the beginning see you a kind of paper-man in the sand
[2] liegen vor ‘ner Flasche
lying in-front-of a bottle
[3] das dann aufwacht
who then wakes up
[4] sich anfängt zu bewegen
starts to move
[5] die Flasche in die Hand nimmt
the bottle in the hand takes

This is found in English as well, though relative clauses are not used to the same extent.
All speakers starting out in this way end up sooner or later with modal subordination.
In theory one could have narratives where descriptions of the narrator’s world are
downgraded or subordinated with respect to the story line, as in und er hebt den Arm,
den man vorher nicht gesehen hat (‘he raises an arm which one did not see before’).
There is no evidence, however, in the present database of a pattern of this kind, at least
not on a systematic basis.
In sum, one of the main types of subordination in the English narratives arises
from the status accorded to the narrator. In French subordination of this kind is also
found for the narrative as a whole, in addition to its use in downgrading certain types
of events when advancing the story line.

4.   vidence for macro-structural principles underlying

E
information s­ tructure

The cross-linguistic comparison between English, French, and German allows isola-
tion of a number of core parameters that are relevant for macro-structural planning.
 Subordination in narratives and macro-structural planning 

Starting with the role of temporal-aspectual means, English and French are similar
typologically in that they are subject prominent in information structure (and both
have fixed word order), they differ, however, with respect to the grammatical means
available to encode aspectual relations (present tense). Similarities in information
structure can be observed in that all agents, whether animate or inanimate, are se-
lected for mention. The languages differ, however, with respect to the way in which
these events are integrated into the story line, taking both degree and type of event
integration. This finding points to the relevance of the progressive in English and the
associated deictic anchor (‘now you see’; ‘then you see’) in accommodating events of
different types. Both languages mention events that are not fine tuned with respect
to temporal shift. Both languages rely on causal relations in advancing the story line.
However, French speakers downgrade events in which the protagonist is not mapped
as subject, and make prominent use of the narrator in an interpretative role in estab-
lishing coherence. In this sense there is what has been termed a ‘narrator perspective’.
Speakers of English can rely on a deictic ‘now you see; then you see’ in integrating
agents of all kinds. We can thus identify core parameters that are well defined given
the nature of the structural means available – parameters that can be investigated at the
level of macro-structural planning for information structure as a whole. The presence
of an aspectual marker to code the distinction ‘ongoing’ allows speakers to establish a
temporal frame which delivers a default setting with respect to decisions concerning
event integration.
However, the comparison with French shows that a temporal frame of this kind
may, but need not be the sole determining factor in information selection. This ques-
tion was tested with a comparison across different languages that are ‘subject promi-
nent’ in information structure: in the cross-linguistic analyses English, French, Italian,
Spanish all show a similar pattern or default case in information selection – there
are no restrictions. This shows that in English and French information selection (e.g.,
events) is not dictated by the nature of the temporal frame. However, event integra-
tion clearly is. Prominence is accorded to the protagonist in the narrative sequence in
French (not downgraded), in contrast to English, a principle in information structure
that is found in all the languages in the present study that do not make use of aspectual
distinctions such as the progressive in the temporal frame. If events that advance the
story line (inanimate agents) are systematically downgraded, as in French, an external
point of reference is implemented (narrator as witness). We assume that this point of
reference serves to bridge differences (events with the protagonist as subject in main
clauses vs. natural forces as subject in subordinate clauses) and allows the speaker to
establish coherence via other relations (causal, intentional, etc).
The fact that the speakers of English choose between two frames is useful in the
analysis of determining factors since the small number of speakers who implement
the relation temporal shift in the temporal frame evidence patterns in information
integration (downgrading) that differ markedly from the main group (deictic frame).
This provides evidence for the assumption that the temporal frame of reference
­constitutes the most basic component at the level of macro-structural planning in a
 Mary Carroll et al.

narrative task (tell ‘what happened and why’). If speakers of English decide to recount
the narrative using the simple tense form (with little or no use of the progressive),
events which are not fine tuned for temporal shift are filtered out and not selected
for mention, as in German and Dutch (in as far as one can rely on results for 3 narra-
tives). As mentioned, this finding points to the relevance of the temporal frame and
associated sequencing principle in delivering a default for decisions regarding the sta-
tus assigned to entities in topic assignment, reference maintenance, and downgrading
(passive, subordination). All of these factors correlate with the temporal frame used
in sequencing events: (i) causal relations and a deictic anchor allowing incorporation
of ongoing events (for the main group in English) versus (ii) temporal shift (relating
to events with a right boundary) in frames that do not include the progressive but rely
on the simple tense form. Significantly, there is no external anchor or reference point
(i.e., no witness as reference point or deictic anchor) in this latter option in English
so that sequencing principles in the temporal frame proceed on the basis of internal
means only.
Information structure in German, and Dutch, is fine tuned so as to accommodate
the relation temporal shift on an internal basis so to speak. In German and Dutch the
syntactic subject, and the information it encodes, shares its potential status (encode
topic information) with other categories which map into the ‘Vorfeld’ or pre-verbal
‘slot’. These are temporal relations, e.g., the times introduced by temporal adverbials
such as dann which are associated with change in state events and encode temporal
shift. We assume that information selection and topic assignment is defined in these
terms, and with this the decision as to what type of entity can typically be mapped as
subject of a main clause: there is a focus on candidates that can deliver on temporal
shift. Entities that do not qualify are either filtered out in information selection, or
when selected, are downgraded on a systematic basis (passive), as revealed in the anal-
yses for Dutch and German. The cross-linguistic comparison shows that this pattern in
macroplanning is not excluded in English, for example. However, the extent to which it
is found in both V2-languages provides evidence for the role of grammaticised means
in driving this option, as given with the ‘V2’ constraint for word order. In keeping with
these overall patterns, zero anaphora in reference maintenance correlates with events
involving temporal shift in Dutch and German, while zero anaphora is restricted to
events that are closely related in causal terms in English and French.

5.  Towards a model of macro-structural planning principles

Based on the empirical findings for macro-structural principles in the preceding sec-
tions the present section presents the first step towards a formal model of principles
that guide macro-structural planning. However, this initial step is confined to the
­subset presented above, i.e., language-specific principles that are driven by grammati-
cal means and operate at the macro-structural level when advancing the story line.
 Subordination in narratives and macro-structural planning 

Principles relating to the narrator’s world and how the two dimensions in story telling
are combined will be addressed in a future paper.

5.1  Comparability of the speaker’s knowledge base

We assume that speakers build up a knowledge base relating to the events seen in the
film, and that knowledge at this level is comparable across the speakers. Although this
assumption is clearly a simplification, it allows a common point of departure in the
cross-linguistic comparison. All speakers saw the same film and, other things being
equal, we can ask what factors are then imposed by language structure at the level of
macro-structural planning in re-narrating the content of the film. In order to give this
idea more substance let’s assume that all speakers, irrespective of their languages, have
access to the same set of events in their knowledge base. For the sake of argument let
us assume that this set contains every event ever selected by any speaker. In particular
it contains ‘minor events’ without immediate relevance for the protagonist. Events that
remain unmentioned in the individual narratives are modelled as successfully filtered
by global principles when organising information for expression. The comparison of
production principles will entail a comparison of principles that govern this kind of fil-
tering. In the present framework these principles will be stated in very general terms.

5.2  Producing as deciding

Assuming that the speakers have access to the same set of events we further assume
that global planning principles apply event by event. Each event of the set will at some
point form the current event in the sequence of events under planning. For each event
the speaker must decide whether it should be selected for mention and how to en-
code it. Assume for a starting point that the speaker has to decide no more than three
things: (i) whether to select the event or not (ii) which verbal predicate to select in
encoding the event (iii) which participant to select for subject. This is written as the
following set of three decisions:
(1) {<select (e), not select(e)>, <predicate(e)>, <subject(e)>}

Binary decisions are represented by ordered pairs in (1), constructive decisions are
one-tuples. The set (1) is an unordered set of decisions. However, models of macro-
structural planning will be ordered sets of decisions. This section deals in particular
with possible hierarchical orders and explores the extent to which the relevant set of
decisions are ordered in different ways in different languages. Here is a first example:
The option of selecting the current event for mention can determine the selection of
the subject of the related clause in a straightforward manner: The speaker can select as
subject the participant that is highest on the scale for ‘agentivity’. As the findings show,
the majority of English speakers proceed in this way. Another order is given when
the decision to select and encode an event depends on whether a certain entity can
be mapped as subject. In other words, if the current event does not allow the speaker
 Mary Carroll et al.

to map information into a clause with the ‘predetermined’ entity as subject it will be
omitted. Again this provides a criterion that can operate globally since it will apply for
every current event under description. We have seen that German speakers follow this
principle. These interdependencies between selection for mention and subject selec-
tion in English (E) and German (D) can be presented as follows:
(1’) E <select(e), not select(e)> G <subject(e)>
  <subject(e)>   <select(e), not select(e)>

The question arises as to why we have converse ranking in German and English. The
answer cannot be delivered on the basis of (1) since there are more obligatory binary
options that have to be decided for the description than the three in (1). We will go
into the other obligatory options in the next subsections thereby following three main
assumptions:

i. If a language demands an obligatory binary decision on the linguistic form

of a sentence predication these decisions will be made at the outset for the
whole unit of macroplanning.
ii. Languages differ with respect to the set of obligatory options on the decision
agenda, given constraints rooted in the grammatical system.
iii. This leads to different decision hierarchies for the individual languages.

5.3  Structurally-determined options in German

In German the pre-verbal position can be filled by different constituent types. As the
analyses show, the decision includes whether or not to fill it with the syntactic subject,
since this may, but need not, be placed in this position. The subject may also be placed
sentence-internally, thereby mapping another constituent into the pre-verbal position.

5.4  Structurally-determined options in English

There is no option and thus no obligation in deciding where to place the subject in
the English narratives. Word order constraints at the syntactic level determine its pre-
­verbal position – constraints which also hold for French. In contrast to both German
and French, English is a special case in that the question of temporal aspect, i.e., use
of the simple tense or the progressive, arises with every event and associated sentence
predication. There is no obligation to mark temporal perspective by grammatical
means either in German or French, at least not in the present tense (the preferred
tense in the present task in all three languages).
The (unordered) set of decisions on the agenda differs in German (G), French (F),
and English (E):
(2) G {<select(e), not select(e)>, <pred(e)>, <subj(e)>,
  < [subj |V| … ] , […|V| subj]>}
 Subordination in narratives and macro-structural planning 

(3) F {<select(e), not select(e)>, <pred(e)>, <subj(e)>}

(4) E {<select(e), not select(e)>, <pred(e)>, <subj(e)>, <prog, not prog>}

Following (3) the set of decisions for French with respect to our model is smaller than
in German and English. The consequences of this finding will be treated following our
presentation of the constraints on planning principles in German and English.

5.5  Implementation of these principles in context

In German the syntactic subject, and the information it encodes, shares its potential
status (encode topic information) with other categories which map into pre-verbal
position, the ‘Vorfeld’ or pre-verbal ‘slot’. These are temporal relations introduced by
anaphoric temporal adverbials which are associated with change in state events and
encode temporal shift (cf. Rossdeutscher & Stutterheim, 2006, on dann). Thus (3) can
be given more substance for the set of decisions on the agenda in practice. There is the
option of either placing the time relation or discourse referent (the latter as subject)
sentence-initially or sentence-internally, respectively. Thus (3) can be rewritten as (5).
(5) G {<select(e), not select(e)>, <pred(e)>, <subj(e)>,
  <[subj(e) |V| t-rel], [t-rel |V| subj(e)]>}

‘t-rel’ represents relational descriptions of temporal discourse referents of the ana-

phoric type. The temporal discourse referent has an antecedent in the context of the
narrative sequence.
In English decisions with an obligatory status relate to the temporal domain. The
main impact of the global decision on verbal aspect relates to the temporal frame.
The common feature of the narratives of the speakers which include use of both the
progressive tense as well as the simple tense in the narrative sequence is (i) imple-
mentation of a temporal deictic anchor and (ii) a narrator-oriented perspective. In
Rossdeutscher & Carroll (2006), this form of temporal structure is defined as follows:
the times which the event descriptions relate to (also referred to as ‘topic times’ in
Klein 1994) are subintervals of evaluation times of the propositional attitude which
the narrator has about the events he has witnessed. Those speakers who do not choose
this anchor and implement the simple tense only in the narrative sequence follow a
principle of anaphoric shift in sequencing events, i.e., it is non-deictic. Thus in practice
the aspectual option involves (i) implicit or explicit reference to some witness time,
t-witness, around which the states in the story world obtain and within which events
are located, or (ii) implicit or explicit reference to some relational time t-rel, in which
the states obtain and within which the events are located. Leaving states aside, we write
‘e in t‘ when the event is located in the period t. The global decision to adopt a deictic
anchor is written as ‘e in t-witness‘ and the global decision for anaphoric shift as ‘e in
t-rel’. The grammatical option <prog, not prog> in (4) is substituted by the more ­general
decision in perspective taking that correlates significantly with the decision regarding
aspect. Thus (4) can be rewritten more substantially as (6):
 Mary Carroll et al.

(6) E {<select(e), not select(e)>, <pred(e)>, <subj(e)>,

   <e in t-witness, e in t-rel>}

5.6  Decision hierarchies

Taking the set of decisions which the speaker has to solve, we will now show how rank-
ing is established within these sets. In German, eligibility for mention in pre-verbal
position constitutes a core aspect with respect to information structure: Constituents
on the left periphery play a role in linking the new predication to the preceding context
(cf. Lambrecht 1994; Frey 2000, 2005). If the decision is made to ‘fill’ this position by a
stable or recurring discourse referent, one form of coherence is fulfilled at the level of
macroplanning. The conditions that allow use of zero anaphora are complex (cf. Klein
1993; Reinhart 1980) but one obvious factor, of course, is that the utterances at issue
must share discourse referents (or else referential linkage must be marked in more
explicit terms). The default condition for use of zero anaphora in German instantiates
a global decision of this kind that at the same time outranks event selection. Event
selection is outranked by subject selection also, if the subject (protagonist) is placed
sentence internally.
In order to display the ranking we will introduce an irreflexive and transitive order-
ing ‘>’ that obtains between subsets of the decision sets. The ordering is displayed in
addition in different lines. At the level of macroplanning for the German narratives, the
decision as to which individual referent forms the subject belongs to the subset of deci-
sions that outrank the decision as to whether or not an event is selected for mention.
Nothing will be determined at present about the relative ranking between subject selec-
tion and subject position: the decision regarding selection has the same ranking as the
decision on where to place the referent. We also assume that predicate choice and event
selection are of the same ranking. The choice of predicate cannot outrank the decision
regarding sentence-internal or pre-verbal position but is definitely outranked by subject
selection. This is the case since German speakers choose passive descriptions even if a
natural force constitutes the discourse referent with the highest status on the agentivity
scale for the event at issue. The ranking so far can be displayed as (7).
(7) {{<subj(e)>, <[subj(e) |V| t-rel], [t-rel |V| subj(e)]>} >
{<select(e), not select(e)>, <pred(e)>}}

The ranking in (7) predicts our findings for German where events that do not support
the selection of the protagonist as subject tend to be filtered out.
In English we are concerned with two different rankings: For the few speakers
who do not select a deictic temporal anchor for the temporal frame, subject selection
and non-deictic anchoring outrank event selection and predicate selection. Again the
relative ranking between the former two must be left open at this point. Assuming
here that the choice of the event-dependent predicate is independent of the option
between deictic vs. anaphoric anchoring and corresponding tense usage, we include
 Subordination in narratives and macro-structural planning 

the decision in the subset of outranked decisions, giving (8). (The decision to imple-
ment anaphoric shift (and thus not to use the other sequencing principle) is presented
as ‘<--- , e in t-rel>’; a decision to implement a deictic temporal anchor is presented as
‘<e in t-witness, --->’ (see (9) below).
(8) E {{<subj(e)>, <---, e in t-rel>} >
   {<pred(e)>, <select(e), not select(e)>}}

The parallels between the English and the German narratives (based on temporal shift)
are mirrored in (7) and (8) in that subject selection outranks event selection.
For the larger group of English speakers we get a reverse ranking with regard to
event selection and subject selection, see (9). (Compare also (1’))
(9) E {{<select(e), not select(e)>, <e in t-witness, --->} >
   {<pred(e)>, <subj(e)>}}

The principle of setting eligibility for mention as subject in terms of agentivity (high-
est candidate), for the given event, is on a par with the decision involving a deictic
temporal anchor.
In (9) there are no constraints on the selection of events, which mirrors our find-
ings that events are eligible for mention irrespective of the role the protagonist may
play. Event selection is not constrained by subject selection. The question is: which set
of criteria is relevant in deciding whether to select an event or not? This will be one of
the issues dealt with in the next subsection.

5.7  Strengthening the decision hierarchies

Up to now we have been dealing with arguments for the hypothesis that the differ-
ences in information structure in German and English are the consequence of gram-
matically-driven hierarchies within a set of decisions. We will now turn to the ques-
tion as to whether we can strengthen the hierarchies and discuss which principles may
be involved. For example, could some ranking on the basis of ‘protagonist-is-subject’
and ‘temporal frame-is-deictic’ outrank the decision as to whether an event should be
selected or not? That is to say: Is a hierarchy of the following form to be expected?
(m stands for the protagonist, the clay man)
(10) E {{<subj(e) = m>, <e in t-witness, --->} >
   {<pred(e)>, <select(e), not select(e)>}}

To give a brief answer: None of the English narratives can be reconstructed on the
basis of global planning principles of this kind. Why is this so? We believe that a com-
bination of this kind is excluded in global planning by general principles. Let’s as-
sume for a moment that (10) could form a possible decision hierarchy (even if this is
contradicted by the facts). We could then also assume that either the global decision
‘protagonist m is subject’ outranks the global decision for a deictic perspective or else
the latter decision outranks the former.
 Mary Carroll et al.

i. If the speaker opts in the first place for the construction of a narrative sequence
that is protagonist-centred, events can be linked by anaphoric means such as then.
In this case the temporal relation between two events – selected on the basis of the fact
that the protagonist can be chosen as subject, is always specified by anaphoric connec-
tors. Any period succeeding the last mentioned event qualifies as a then-interval. Tem-
poral anaphoric anchors provide temporal succession whether or not the described
events are inherently, e.g., causally, connected. And the events chosen on protagonist-
centred grounds are not causally connected in general.
ii. If, on the other hand, the speaker decides to establish a deictic perspective in the
first place, he is committed to establishing the story line with predications in which
the topic times constitute an extended time span (e.g., ‘now’) of the witness time of
the narrator. Sets of events can be described as witnessed during a time span of this
kind if they are continuous events, for example. In Rossdeutscher and von Stutterheim
(2006) we define two events as continuous if the consequent event overlaps in part
with the state or result state of the antecedent. There are in principle three ways of
accomplishing a narrative task involving a sequence of events that qualify as continu-
ous: The speaker could produce a state by state description, each one corresponding
to one witness time (‘now’) only. However, witness time can involve an extended ‘now’
and include temporal updates of the topic times of events. The events at issue can be
conceptualised as continuous. Furthermore, the speaker can decide not to omit any of
the events in his knowledge base and select each one. Continuity can be accomplished
in this way if the events in the knowledge base are continuous. But this strategy would
not go far. Given the constraints in (10) it cannot work at all: the events do not gener-
ally qualify for a description with the protagonist as subject. There is a third solution:
A sequence of events can be conceptualised as continuous if conceptualised as causally
connected or as part of a common plan on the part of the protagonist. But this third
option of predicating a causal chain of events cannot be easily resolved by relying on
protagonist-centred descriptions either. In establishing a causal chain the speaker may
want to mention events that interfere with the current states and plans of the protago-
nist. Events of this kind need not have the protagonist as a participant and may have
an antagonist as a likely subject.

We sum up the results for macroplanning principles following deconstruction of the

non-existent (10):
If <subj(e) = m> would outrank <e in t-witness, ---> the task would be hard to
solve (if solvable at all) and the speaker would be better off with the anaphoric option
<----, e in t-rel> in (8), excluded in (10). In this case the speaker is free to choose the
times for which he predicates properties and changes relating to the protagonist. If <e
in t-witness, ---> outranks <subj(e)> the extended ‘now’ of the witness time can be
filled by describing a continuous sequence of events. Restricting the selection further
so as to reduce the set to events that allow for ‘protagonist-is-subject’ would be too
rigid.
 Subordination in narratives and macro-structural planning 

The majority of English speakers opt for a deictic perspective which outranks
selection; selection is restricted by filling the witness time with causally connected
events. The subject is selected with regard to each particular, taking the candidate with
the highest ranking on the agentivity scale as subject. (We display this as ‘subj(e) = x’.)
This option is modelled as the appropriate strengthening of (9) yielding (11):
(11) E {{<e in t-witness, ---->} >
   {<select(e), not select(e)>} >
   {<pred(e)>, <subj(e) = x>}}

The decision hierarchy for the minority of English speakers displayed in (8) can be also
strengthened in accordance with the general principles treated in this subsection. As
already discussed, the highest ranking for ‘protagonist-is-subject’ leads to anaphoric
anchoring.
There is another possible ranking – the one where the option for anaphoric an-
choring outranks any other decision in the decision set, subject selection in particular.
We claim that the option for anaphoric linkage leads to restrictions in subject selection.
Note first that it allows restriction of the set of events to be chosen. Any time succeed-
ing the last topic time may be chosen as the predication time of the event. But any time
is not a permissible feature in this context. If all options are open the speaker has a high
burden at every point in the chain with regard to selection of the ‘next’ event (whether
to omit it or not). Restricting the set by restriction via subject selection is a possible
option. If the speaker chooses this solution the protagonist is the only individual dis-
course referent at issue. Which of the two rankings is actually appropriate cannot be
decided. The latter procedure would allow for the theoretically interesting situation
that production principles in English are the consequences of a binary decision on the
way to link the times to which the event descriptions relate. We end this subsection by
displaying this solution in (12). (11) and (12) are the two decision hierarchies in our
model for macroplanning principles in English so far.
(12) E {{<---- , e in t-rel>} >
   {<subj(e) = m>} >
   {<select(e), not select(e)>, <pred(e)>}}

5.8  German
Are there any consequences for German, given the claims made for English above? It
should be clear that the restriction of event selection via restrictions for the subject of
the clause would be a consequence of anaphoric temporal anchoring. And the option
<subj(e) = m> entails anaphoric anchoring as its consequence. Whatever the speaker
decides with respect to the question as to where to place the subject is in accordance
with a possible strengthening of (7), repeated here as (13).
(13) G {{<subj(e) = m>, <[subj(e) = m |V| t-rel], [t-rel |V| sub(e) = m]>} >
   {<select(e), not select(e)>, <pred(m)>}}
 Mary Carroll et al.

The constraints given with ‘verb second’ involves more than mere constraints on sub-
ject selection, however. As mentioned above, the dual role in information structure in
German of both the syntactic subject and constituents that encode temporal relations
(temporal shift with dann, ‘then’, channeled by the verb second constraint), may focus
attention in information selection on a specific subset of entities in events that indicate
a point of completion and thus set the preconditions for temporal shift.

5.9  French
French differs from German and English in that there are no further elements in the
decision set besides the basic ones to which our comparison was confined. There are
no requirements which have to be considered in solving the narration task concern-
ing aspect or subject position. Subject position is clear and aspectual distinctions are
not marked, as the analyses reveal. There is no evidence of any grammatically-driven
constraints on event selection (other than the absence of restrictions found for all
­subject-prominent languages in the cross-linguistic studies). Event selection outranks
any other decision and the speaker must solve this decision problem in an appropriate
way, given the narrative task. The question is solved via subject selection: the protago-
nist qualifies as subject in main clauses when advancing the story line. (Other entities
do not, even though they also advance the story line.) In contrast to German, sentence
structure in French (SVO) does not give rise to the case where temporal adverbs enter
the stage of global sentence planning. In French the story line, as the analysis shows,
is established via relations such as ‘means/goal’, ‘reactions’ and ‘plans’, often ascribed
by the narrator. The intensional dimension of the story is in focus and the preferred
properties predicated, in relation to the protagonist, are attitudinal properties. But as
we have already discussed in connection with the English narratives, a causal chain
will not provide a basis for omitting reference to antagonists, the natural forces which
interfere with the current beliefs, intentions of the protagonist, especially where they
appear as causal elements. In both SVO languages, the events in question are eligible
for mention and are not filtered out in contrast to the V2 languages (German and
Dutch). So word order constraints in French and English do not drive a restriction at
this level, in contrast to the V2 languages. In French omission would mean filtering
out events that can contribute to progression at a causal level. They are downgraded at
a temporal level (subordinated) but not in causal terms. In English a deictic temporal
anchor allows integration of any event that fills the extended ‘now you see’ or ‘then
you see’ of the witness time. There are no grounds at this level in English to filter out a
certain set of events. So in both languages eligibility for mention may be guided by the
fact that they qualify as candidates in forming a causal chain.
The discussion of planning principles in French ends with the decision hierarchy
resulting from our considerations. In the notation, <subj = m> states that subject se-
lection is not event dependent. The restriction to intensional properties is presented
in <pred(attitude(m)>.
 Subordination in narratives and macro-structural planning 

(14) F {{<subj = m>} >

   {<select(e), not select(e)>, <pred(attitude(m)>}}

The ranking predicts the restriction given for event sequences.

6.  Conclusions

The present analysis provides evidence for downgrading procedures and their de-
terminants as going beyond the functional description of subordination as a down-
grading device that operates across well established divisions such as foreground and
background (Hopper 1979, 1995; Tomlin 1985; Cristofaro 2005). It shows how deter-
mining factors can be linked to grammaticised means that are language specific – a
verb second constraint at the level of word order, the presence or absence of aspectual
distinctions such as the progressive. It also illustrates other forms of downgrading that
are not random, or subject to questions such as ‘style’, but form part of a set of general
principles that hold at different levels in information structure. Analyses of the related
factors and associated constraints, that pair with relevant typological contrasts, point
to a hierarchically ordered set of principles at the level of macro-structural planning
that are grammatically driven.

References

Carroll, Mary & Lambert, Monique. 2003. Information structure in narratives and the role of
grammaticised knowledge: A study of adult French and German learners of English. In
Information Structure and the Dynamics of Language Acquisition, Dimroth, Christine &
Marianne Starren (eds), 267–287. Amsterdam: John Benjamins.
Carroll, Mary & Lambert, Monique. 2006. Reorganizing principles of information structure in
advanced L2s: French and German learners of English. In Educating for Advanced Foreign
Language Capacities, Heidi Byrnes, Heather Weger-Guntharp & Katherine Sprang (eds),
54–73. Washington DC: Georgetown University Press.
Cristofaro, Sonia. 2005. Subordination. Oxford: OUP.
Frey, Werner. 2000. Über die syntaktische Position des Satztopiks im Deutschen. In Issues on
Topics. ZAS Papers in Linguistics 20: 137–172.
Frey, Werner. 2005. Pragmatic properties of certain German and English left peripheral con-
structions. Linguistics 30: 5–55.
Hopper, Paul. J. 1979. Aspect and foregrounding in discourse. Syntax and Semantics 12: Dis-
course and Syntax. Talmy Givon (ed.), 213–241. New York: Academic Press.
Hopper, Paul J. 1995. The category ‘Event’ in natural discourse and logic. In Discourse Gram-
mar and Typology, Werner Abraham, Talmy Givon & Sandra Thompson (eds), 139–153.
Amsterdam: John Benjamins.
Klein, Wolfgang. 1993. Ellipse. In Syntax. Ein internationales Handbuch zeitgenössischer For-
schung, Joachim Jacobs (ed.), 763–798. Berlin: de Gruyter.
 Mary Carroll et al.

Klein, Wolfgang. 1994. Time in Language. London: Routledge.

Lambrecht, Knud. 1994. Information Structure and Sentence Form. Cambridge: CUP.
Murcia Serra, Jorge. 2001. Grammatische Relationen im Deutschen und Spanischen. Frankfurt:
Peter Lang.
Reinhart, Tanya. 1980. Conditions for text coherence. Poetics Today 1: 161–180.
Roberts, Craige. 1989. Modal subordination and pronominal anaphora in discourse. Linguistics
and Philosophy 12: 683–721.
Rossdeutscher, Antje & Carroll, Mary. 2006. How to cope with narration tasks in German and Eng-
lish. University of Heidelberg. URL: http://www.ims.uni-stuttgart.de/~antje/narration-task.
pdf.
Rossdeutscher, Antje & von Stutterheim, Christiane. 2006. Semantische und pragmatische Prin-
zipien der Positionierung von ‘dann’. Linguistische Berichte 206: 29–60.
Stutterheim, Christiane von, Carroll, Mary & Klein, Wolfgang. 2003. Two ways of construing
complex temporal structures. In Deictic Conceptualization of Space, Time, and Person.
Friedrich Lenz (ed.), 97–133. Berlin: de Gruyter.
Tomlin, Russell S. 1985. Foreground-background information and the syntax of subordination.
Text 5(1–2): 85–122.
part iii

Monolingual studies
 German dependent clauses from a
constraint-based perspective

Anke Holler
University of GÖttingen

In this article the main focus is on German non-canonical clause linkage.

Inspecting five instances of non-canonically linked clauses (i.e., weil-verb second
clauses, continuative wh-relative clauses, verb second relative clauses, free
dass-clauses, and dependent verb second clauses) it is argued that subordination
should be treated as a multidimensional phenomenon. By means of the
aforementioned clauses, the article empirically investigates several factors
underlying the compositeness of subordination. To account for the presented
clause linkage facts a non-derivational analysis in the framework of Head-driven
Phrase Structure Grammar is developed employing constraints and sorts partially
ordered in multiple-inheritance hierarchies as conceptual devices to model
several degrees of clause linkage.

Keywords: non-canonical clause linkage, subordination, HPSG

1.  Introduction

Traditional, structural as well as generative approaches to grammar usually act on the

assumption that clauses that can stand alone as complete sentences differ grammatical-
ly from the ones that are dependent on a prior clause. This becomes manifest in the
­distinction of main and subordinate clauses. Related to this opposition is a stipulated
contrast between two syntactic linkage relations: hypotaxis and parataxis. On the other
hand, several studies (cf. e.g., Haiman & Thompson 1984; van Valin 1984; Lehmann
1988; König & van der Auwera 1988; Fabricius-Hansen 1992; Reis 1997; Peyer 1997)
have shown that a dichotomic view on clause types and syntactic linkage in complex
sentences oversimplifies the linguistic facts. In this article I will revisit this problem.
I will firstly review relevant aspects of the debate by discussing pertinent empirical
data and influential theoretical approaches. On the basis of clause linkage data from
­German I will take up the position that the opposition of two syntactic linkage relations
should be abandoned entirely, and that clauses should be grammatically distinguished in
terms of the degree they are integrated into a potential host clause ­depending on their
­syntactic form, their interpretation, and their functional usage. Further, I will ­elaborate
on the proposal that typical non-integrated clauses are best analysed as ­being orphan
 Anke Holler

c­ onstituents that are linked to the prior clause only discourse-structurally. I will ­provide
a constraint-based analysis in the framework of Head-driven Phrase Structure Gram-
mar (Pollard & Sag 1994) arguing that a ­sign-based linguistic theory ­employing sort-
ed feature structures and multiple-inheritance sort hierarchies as ­conceptual ­devices
seems to be well suited to account for the presented clause linkage facts.
The article is organised as follows. In the next section some more general issues
concerning the nature of clause linkage and the distinction between main and sub-
ordinate clauses are addressed. Section 3 recalls data of German dependent clauses
that challenge any approach implementing a twofold differentiation between main and
subordinate clause types. Section 4 sketches a constraint-based analysis that organizes
clause types according to the way they are grammatically linked to the prior clause.
Section 5 concludes the article.

2.  Clause linkage and subordination

This section discusses two essential issues concerning the nature of clause linkage and
subordination, viz. (i) the question whether there exist more than one relation linking
a putative subordinate clause to its matrix clause, and (ii) by which grammatical means
a subordinate clause can be identified at all.

2.1  Is there more than one subordinating linking relation?

Complex sentences are usually analysed as being composed of clauses joined in a
­hypotactic or paratactic relation. Roughly speaking, a clause is viewed as hypotactically
linked to a prior clause if it is subordinated to this clause; it is viewed as paratacti-
cally linked to a prior clause if it is coordinated with it.1 At the syntactic level subor-
dination can be expressed by specific connectives such as subordinating conjunctions
like dass (‘that’), (relative or interrogative) pronouns and relative adverbs. A coordinate
­relationship is realised by coordinating conjunctions like und (‘and’) and oder (‘or’) or
by ­punctuation marks. Whereas it seems to be general consensus that clauses ­being
in a ­coordinate relationship are sequentially added together, it is much more under
­debate what ­subordination exactly means. One way to look at subordination is to define
­subordinate clauses as clauses that are embedded as a constituent of a matrix sentence.
­According to this view a subordinate clause functions as a part of another clause, for
instance ­being an argument or an attribute. Another approach is to say that subordination is
based on ­dependence, plus assuming that subordinate clauses are characterised by internal
formal properties such as a certain position of the finite verb (e.g., a clause final posi-
tion in German) or the presence of a subordinating conjunction. At first glance defining

.  The notions of parataxis and coordination are not identical, cf. Pasch et al. (2003).
 German dependent clauses 

subordination involving either embeddedness or dependence seems to be equivalent. In

fact it is not. Being connected with a prior clause does not automatically result in having
a syntactic function in that clause. For instance, this can be seen in examples like (1). The
dependent clause, although looking like a relative clause, is not syntactically embedded
in the prior clause, cf. Holler (2005).
(1) Emma gewann die Schachpartie, was Oskar ärgerte.
Emma won the chess.match which Oskar annoyed
‘Emma won the chess match, which annoyed Oskar.’
In addition, typologists studying non-Indo-European languages have persuasive-
ly ­argued that clause chaining is evidence that dependence and embeddedness are
­distinct parameters in defining syntactic relations in clause linkage, see e.g., van Va-
lin (1984), Haiman & Thompson (1984), Longacre (1996). Thus, seeking for language
universals also rejects identifying embeddedness and dependence.
On the basis of these results it is striking that clauses must generally be evalu-
ated according to at least two separate criteria: (i) whether they are linked to another
clause (dependence), and (ii) whether they fulfil a syntactic function in another clause
(embeddedness).2

2.2  How to identify a subordinate clause?

Even if embeddedness and dependence are recognized as isolable and independent param-
eters of the relationship that two adjacent clauses can have, it is not yet clarified which set
of properties characterises subordinate clauses. That this is in fact a complex issue shows
work done in the eighties and nineties on various Germanic languages, e.g., by Lehmann
(1988), König & van der Auwera (1988), Fabricius-Hansen (1992), Santorini (1992) and
Peyer (1997). It has been demonstrated in the course of these studies that a clear-cut sepa-
ration of the main clauses from the subordinate ones seems to be impossible.
Despite this comprehensive work, it is still often taken for granted that German
clause types could be clearly distinguished by surface-related syntactic means such as
the placement of the finite verb. This point of view has been particularly revitalised in
construction-oriented and/or constraint-based approaches to Germanic languages, which
usually act on the assumption that the finite verb is generally fronted in main clauses,
while it has to occur clause-finally in subordinate clauses. For instance, this ­position
has been adopted in Uszkoreit (1987), Kathol (1995), Netter (1998), or ­Ginzburg & Sag
(2000), mostly consulting examples similar to (2) as evidence.
(2) a. Oskar ist vom Stuhl gefallen.
Oskar has from.the chair fallen
‘Oskar has fallen from a chair.’

.  This view complies well with the approaches proposed for instance by Lehmann (1988) and
Pasch et al. (2003) in order to account for phenomena of German clause combining.
 Anke Holler

b. Emma bezweifelt, dass Oskar vom Stuhl gefallen ist.

Emma doubts that Oskar from.the chair fallen is
‘Emma doubts that Oskar is fallen from a chair.’

These data form the basis of stipulating that the position of the finite verb is ‘hard-
wired’ to the representation of a certain clause type. In other words, the mentioned
constraint-based approaches follow the idea that a fronted finite verb generally marks
main (root) clauses, whereas its final position generally signals a subordinate clause.
This perspective involves the additional assumption that the set of main clauses com-
prises all clauses and only those clauses that can be uttered independently. See Kathol
(1995) for an explicit proposal elaborating on this idea. Splitting clause types into main
and subordinate depending on the position of the finite verb and interconnecting this
with the (in-)dependence of the respective clauses yields an approach that classifies
dependent verb second clauses such as (3a) as main clauses, but independent verb final
clauses such as (3b) as subordinate clauses, predicting – contrary to the facts – that the
verb second clause is uttered independently but not the verb final one.
(3) a. Emma glaubt, Studierende schlafen lange.
Emma believes students sleep long
‘Emma believes that students sleep long.’
b. Ob er noch kommt?
Whether he till omes
‘I wonder whether he will still come?’

Although approaches acting on a dedicated relation between the placement of the fi-
nite verb and the classification of a clause as main or subordinate are obviously flawed,
most linguists probably share the intuition that the position of the finite verb may
somehow contribute to the main-subordinate distinction. This is not contradictory if
we assume that a clausal typology is not based on a single phenomenon, but is rather
associated with a number of different aspects of grammar. From this point of view it is
conceivable that a phenomenon being normally considered as typical for main clauses
(such as verb second) may occur in dependent clauses (cf. example (3a)),3 or may not
be realised at all in an independent clause (cf. example (3b)). If we look this way on
a clausal typology, it comes as no surprise that the competent user of language only
seems to be able to identify a clause as main or subordinate by consulting a bunch
of grammatical properties. This means that the terms ‘main clause’ and ‘subordinate
clause’ must be taken as labels for a set of several independent linguistic properties.
In canonical cases, each of these properties can be identified; in non-canonical cases,

.  Germanic verb second is a so-called main clause phenomenon. It has been studied e.g.,
by Vikner (1995), Santorini (1992), Wechsler (1991), Reis (1997), Meinunger (2004), Truck-
enbrodt (2006) and many others. Emonds (1970), Hooper & Thompson (1973) as well as
Green (1976) are concerned with related phenomena in English, i.e., constructions involv-
ing root phenomena. For reasons of limited space, the reader is referred to the original work.
 German dependent clauses 

however, only a subset of the formal properties that are associated with the one or the
other clause type is realised. This is the reason why clauses may differ in degree they
are viewed as main or subordinate, as I will further discuss below.
Taking subordination as a multidimensional phenomenon in the aforemen-
tioned way raises the question of the types of factors underlying the compositeness
of subordination. Following Haiman & Thompson (1984), Lehmann (1988) as well as
Fabricius-Hansen (1992) and Reis (1997), I propose that these properties pertain to
syntactic form, interpretation, and functional usage. On the basis of empirical data of
German complex clause constructions I will discuss the essential factors in the next
section. The data by themselves are not new, but the consequences drawn from them
may contribute to the theoretical discussion on the analysis of clausal subordination.

3.  The empirical challenge

In this section pertinent complex clause constructions, primarily drawn from German,
are investigated with respect to a set of selected grammatical and pragmatic criteria that
are widely accepted as being relevant for the distinction of main and subordinate clauses,
cf. among others Haiman & Thompson (1984), Fabricius-Hansen (1992), Reis (1997).
The following criteria are employed: (i) intonation and prosodic features, (ii) internal
form of the clause, (iii) external syntax and topological placement, (iv) interpretational
aspects as well as (v) information structure and association with illocutionary force.

3.1  Weil-verb second clauses

The so-called weil-verb second clauses4 (henceforth weil-V2 clauses) as exemplified by
(4a)5 are an alternative type of standard causal adverbial clauses such as (4b), which has
been substantiated by work of Wegener (1993), Keller (1995), Pasch (1997), ­Uhmann
(1998) among others. At the syntactic surface, weil-V2 clauses are characterized by an
introducing conjunction weil (‘because’) and by a finite verb in second position.6
(4) a. Peter kommt zu spät, weil er hat keinen Parkplatz gefunden.
Peter comes too late because he has no parking.space found
‘Peter is late because he could not find a parking space.’

.  Weil-verb second clauses are mainly attested for colloquial German, but can be observed in
­written German as well, cf. Wegener (1993) and Uhmann (1998). Both authors have convinc-
ingly ­argued that verb second in weil-clauses is not an irregularity of performance but a system-
atic part of speakers’ language competence.

.  The example is taken from Uhmann (1998).

.  Uhmann (1998) citing Altmann (1997) points out that there are cases where the finite verb
occupies an initial position, e.g., Ich brauche jetzt unbedingt neue Kleider, weil – möchtest du
­eigentlich noch so mit mir ausgehen? (‘I absolutely need new clothes, because – do you really want
go out with me that way?’)
 Anke Holler

b. Peter kommt zu spät, weil er keinen Parkplatz

Peter comes too late because he no parking.space
gefunden hat.
found has
‘Peter is late because he could not find a parking space.’
Example (5) indicates that weil-V2 clauses are topologically restricted to a final posi-
tion, which means that they neither stand in front of the clause they are combined
with (cf. (5a)) nor within it (cf. (5b)). This is clearly contrary to standard weil-adverbial
clauses as (6) shows.7
(5) a. *Weil er hat keinen Parkplatz gefunden, kommt Peter zu spät.
because he has no parking.space found comes Peter too late
b. *Peter kommt, weil er hat keinen Parkplatz gefunden, zu spät.
Peter comes because he has no parking.space found too late
(6) a. Weil er keinen Parkplatz gefunden hat, kommt Peter
because he no parking.space found has comes Peter
zu spät.
too late
‘Peter is late since he could not find a parking space’
b. Peter kommt, weil er keinen Parkplatz gefunden hat,
Peter comes because he no parking.space found has
zu spät.
too late
‘Peter is late since he could not find a parking space’
In contrast to canonical adverbial clauses weil-V2 clauses do not have a syntactic func-
tion in the clause they are combined with. Evidence for this comes from the fact that
it is impossible to refer to a weil-V2 clause using a correlative or to attach them by a
supplement, as (7) demonstrates.8
(7) a. Peter kommt (*deswegen) zu spät, weil er hat keinen Parkplatz
Peter comes therefore too late because he has no parking.space
gefunden.
found
‘Peter is late because he could not find a parking space.’

.  The described topological restrictions may be violated if the weil-V2 clause has parenthetical
function. This does not blur the argument concerning the mentioned topological restrictions,
but rather substantiates the view that weil-V2 clauses show a root-like behaviour: ?Peter ist – weil
er hat keinen Parkplatz gefunden – schon wieder zu spät. (‘Peter is, because he could not find a
parking space, late again.’)

.  There is further evidence that weil-V2 clauses do not function as a component part (Satzglied)
of the clause they are linked to. Contrary to causal weil clauses, weil-V2 clauses cannot be transferred
into an adverbial phrase as the following example illustrates: (i) can only paraphrase (ii), but not (iii).
 German dependent clauses 

b. *Peter kommt zu spät, und.zwar weil er hat keinen

Peter comes too late namely because he has no
Parkplatz gefunden.
parking.space found.

In addition, weil-V2 clauses behave exceptionally with respect to their internal syntax.
Uhmann (1998:111) cites corpus data showing that weil-V2 clauses allow left dislocation
and topicalisation, which are typical root phenomena.
Apart from the mentioned syntactic peculiarities, weil-V2 clauses deviate seman-
tically and pragmatically from their adverbial counterparts. It is a well-known fact that
weil clauses in general may be interpreted either causally or epistemically (cf. Sweetser
1990; Keller 1995; Blühdorn forthcoming). The preferred reading of weil-V2 clauses is
however an epistemic one. According to this interpretation, the speaker gives reasons
for the assumption expressed in the host clause. There are in fact cases that can only
get the epistemic interpretation. In example (8) for instance, the speaker concludes
that it hailed from the fact that her car has dints. Obviously, the damaged car cannot
be a reason for hailing, but the damage is a plausible reason for the speaker’s assump-
tion that it hailed. For further discussion of the epistemic reading of weil-V2 clauses
I refer the interested reader to the literature, e.g., Wegener (1993), Keller (1995), and
Blühdorn (forthcoming).
(8) Es hat gehagelt, weil mein Auto hat Dellen.
it has hailed because my car has dints
‘My car is damaged because it has hailed.’

Another semantic characteristic of weil-V2 clauses is that they may behave differently
from causal weil clauses with respect to scope of negation and quantifier binding. As
Wegener (1993) has shown, the content of a weil-V2 clause is not negated if the host
clause contains a negative particle: Whereas the content of the weil clause in (9a) is not
in the scope of the matrix negation, (9b) can be interpreted such that it is denied that
Peter went home because of a head ache.9
(9) a. Peter ist nicht nach.Hause gefahren, weil er hatte Kopfweh.
Peter is not home driven because he had a.head.ache
‘Peter did not drive home because he had a head ache.’
b. Peter ist nicht nach.Hause gefahren, weil er hatte,
Peter is not home driven because he had

(i) Wegen seines Versprechens kommt Max. (‘Max comes because of his promise to do it.’)
(ii) Max kommt, weil er es versprochen hat. (‘Max comes because he has promised it.’)
(iii) Max kommt. Weil – er hat es versprochen. (‘Max comes. Because he has promised it.’)

.  Note that negation generally does not have scope over epistemically interpreted weil clauses.
The verb final weil clause in example (9b) is actually ambigious between a causal and an epis-
temic reading.
 Anke Holler

Kopfweh (sondern …).

a.head.ache but
‘Peter did not drive home because he had a head ache, (but …).’

Example (10) also taken from Wegener (1993) demonstrates that a quantifier in the
host clause does not scope over a weil-V2 clause: (10a) justifies why the speaker ­believes
that some guests will come, while (10b) means in a preferred reading that some guests
will come because of the sunny weather.10

(10) a. Einige Gäste werden kommen, weil heute scheint die Sonne.
some guests will come because today shines the sun
‘Some guests will come, because today the sun is shining.’
b. Einige Gäste werden kommen, weil heute die Sonne scheint.
some guests will come because today the sun shines
‘Some guests will come, because the sun is shining today.’

With respect to the pragmatic properties of weil-V2 clauses, several authors emphasise
that weil-V2 clauses have an illocutionary force that is independent of that of the host
clause, see e.g., Wegener (1993), Keller (1995), Uhmann (1998). In other words, a weil-
V2 clause is not part of the speech act expressed by the prior clause. This is ­illustrated
in example (11). The weil-V2 clause in (11a) expresses a statement and is clearly not
part of the question. This contrasts with canonical causal weil clauses as shown in
(11b).

(11) a. Kommt Peter? Weil er hat es versprochen.

comes Peter because he has it promised
‘Is Peter coming? Because he promised to.’
b. Kommt Peter, weil er es versprochen hat?
comes Peter because he it promised has
‘Is Peter coming because he promised to?’

Data like (11) furthermore provide evidence that the intonational unit of a weil-V2
clause is separated from that of the host clause. This is also observed in Uhmann
(1998:121). For instance, she points out that the sequence in (12a) is only acceptable if
separated in two intonation units marked by a rising boundary tone at the final boundary
of the first intonation phrase. In contrast, (12b) indicates that the same sequence is
unacceptable if phrased as one single intonation unit. This is, however, not the case for
causal weil clauses as (12c) illustrates.

(12) a. Ist Peter zu spät gekommen? Weil er hat keinen

is Peter too late came because he has no

.  Example (10b) is in fact ambiguous. There is a second reading of the verb final weil clause
which is similar to the interpretation of example (10a).
 German dependent clauses 

Parkplatz gefunden.
found parking.space
‘Is Peter coming too late? Because he has not found a parking space.’
b. * Ist Peter zu spät gekommen, weil er hat keinen
is Peter too late came because he has no
Parkplatz gefunden.
parking.space found
c. Ist Peter zu spät gekommen, weil er keinen Parkplatz
is Peter too late came because he no parking.space
gefunden hat?
found has
‘Did Peter come too late because he has not found a parking space?’

As a matter of course, the aforementioned intonational differences of causal weil ­clauses

and weil-V2 clauses are closely related to their interpretation. For instance, the ques-
tion operator scopes over the weil clause only in (12c), but not in (12a). ­Consequently,
the weil-V2 clause in (12a) states a reason for asking the question, which is a speak-
er ­attitude. In (12c), however, the reason for Peter’s being late is asked about. For a
­profound ­discussion of these data see Uhmann (1998: 120 et seqq.).

3.2  Continuative wh-relative clauses

There is a subclass of non-restrictive relative clauses that behave all about the same as
weil-V2 clauses. I call this class continuative wh-relative clauses or short wh-relatives
because at the syntactic surface these clauses can be recognised by a possibly complex
left-peripheral wh-expression as exemplified in (1), repeated here as (13).11
(13) Emma gewann die Schachpartie, was Oskar ärgerte.
Emma won the chess.match which Oskar annoyed
‘Emma won the chess match, which annoyed Oskar.’

At first glance, wh-relatives look like canonical subordinate clauses since the finite verb
is placed in final position. On closer inspection, however, wh-relatives show a number
of properties that are usually not considered as being characteristic of subordinate
clauses. First, wh-relatives are prosodically detached from their host clause as they
establish a separate intonation unit. They have to be preceded by a pause, and the prior
clause ends with falling intonation.
(14) Emma kaufte einen TEURen Schrank.(\), was ÄRGerlich ist.
Emma bought an expensive cupboard which annoying is
‘Emma bought an expensive cupboard, which is annoying.’

.  For a comprehensive empirical and theoretical analysis of the class of continuative relative
clauses see Holler (2005).
 Anke Holler

Secondly, the described prosodic behaviour forms the basis for the pragmatic autonomy of
the wh-relatives. Their focus domain is independent of that of the prior clause, cf. (15).
(15) Was ist passiert? (‘What happened?’)
#[Emma kaufte einen teuren Schrank, was ÄRGerlich ist.]F
Emma bought an expensive cupboard which annoying is
‘Emma bought an expensive cupboard, which is annoying.’
The construction in (15) cannot be uttered as an ‘all-focus’ answer to the question
What did happen?, which is usually taken as an argument that the wh-relative clause
is not integrated into the information structure of the host. That the whole con-
struction ­cannot be uttered as an ‘all-focus’ sentence is attributed to the fact that
every clause of the construction needs a focus accent, and thus projects a separate
focus-background-structure.12
From examples like (16) comes evidence that wh-relatives have an autonomous
illocutionary force. On the one hand, they are not part of the content that is asserted
or asked for, cf. (16a). On the other hand, they may be used to accomplish a speech
act, cf. (16b).
(16) a. * Hat Emma einen Schrank gekauft, was Oskar erstaunte?
has Emma a cupboard bought which Oskar astonished
b. Hat Emma einen neuen Schrank gekauft? Was Oskar
has Emma a new cupboard bought which Oskar
nämlich sehr erstaunen würde, weil sie doch so
namely very astonish would because she PART so
geizig ist.
mean is
‘Did Emma bought a new cupboard? This would astonish Oskar because
she is very mean.’
Example (17) demonstrates that epistemic expressions, performative indicators, ­modal
particles can appear in wh-relatives. This is further evidence for the illocutionary inde-
pendence of wh-relatives.
(17) Die Firma handelt mit Waffen, weshalb ich {sicher/
the company deals with weapons that’s why I {certainly/
hiermit/wohl} kündige.
hereby/well} hand.in.my.notice
‘The company deals with weapons, and that’s why I {certainly/ hereby/well}
hand in my notice.’
Thirdly, wh-relatives do not have a syntactic function in their host clause because
they neither receive a theta-role from the matrix predicate nor may function as an
adverbial adjunct. They rather express a commentary or provide additional back-

.  For a similar argumentation see Brandt (1990).

 German dependent clauses 

ground ­information. Thus, wh-relatives can be fairly freely added to a clause. Exam-
ple (18) also illustrates that wh-relatives are not a component part of the prior clause
since they cannot be linked to it using a supplement expression like und zwar.
(18) *Oskar verlor die Schachpartie, und.zwar was ihn ärgerte.
Oskar lost the chess.match namely which him annoyed
Fourthly, wh-relatives show a scope behaviour that is characteristic of non-integrated
clauses as they disallow variable binding from outside, and a negative particle placed
in the host clause does not scope over the wh-relative.
(19) a. Oskari hat die Schachpartie nicht gewonnen, was ihni ärgerte.
Oskar has the chess.match not won which him annoyed
‘Oskar did not win the chess match, which annoyed him.’
b. * Kein Spieleri gewann die Schachpartie, was ihni ärgerte.
no player won the chess.match which him annoyed
Fifthly, wh-relatives are topologically restricted occuring only at the very end of a com-
plex sentence, as demonstrated in (20) and (21).13
(20) a. Oskar verlor die Schachpartie, was ihn ärgerte.
Oskar lost the chess.match which him annoyed
‘Oskar lost the chess match, which annoyed him.’
b. * Was ihn ärgerte, verlor Oskar die Schachpartie.
which him annoyed lost Oskar the schess.match
(21) a. * Oskar verlor die Schachpartie, was ihn ärgerte, die er
Oskar lost the chess.match which him annoyed that he
unbedingt gewinnen wollte.
absolutely won wanted.to
b. Oskar verlor die Schachpartie, die er unbedingt gewinnen
Oskar lost the chess.match that he absolutely won
wollte, was ihn ärgerte.
wanted.to which him annoyed
‘Oskar lost the chess match that he absolutely wanted to win, which
annoyed him.’

On the basis of the presented empirical properties of wh-relatives we can conclude that
continuative wh-relative clauses seem to be in accordance with weil-V2 clauses with
respect to all principal criteria investigated.

3.3  Verb second relative clauses

Gärtner (2001) investigates a further peculiar subclass of relative clauses in ­German,
which he dubs integrated-verb second relative clauses (henceforth V2 ­relatives).­

.  Note that there are cases where continuative wh-relatives appear inside the host clause.
However, this is possible only if they are pronounced parenthetically.
 Anke Holler

Using criteria along the lines pursued in the present article, he shows that V2 relatives
such as (22) are clauses that “form a hybrid between hypotaxis and parataxis” (Gärtner
2001:29).

(22) Das Blatt hat eine Seite, die ist ganz schwarz.
the sheet has one side that is completely black
‘The sheet has one side that is completely black.’

Gärtner argues that V2 relatives show to some extent grammatical properties that are
normally taken as evidence for embeddedness and are hence attributed to subordinate
clauses: (i) V2 relatives have attributive function and get a restrictive interpretation.
This can be seen in example (23) indicating that the content of the relative clause is
picked up by the indefinite pronoun eins (‘one’) if this pronoun substitutes the nominal
phrase. (ii) V2 relatives can be prosodically integrated into the prior clause, cf. (24),
and (iii) V2 relatives are able to constitute a single informational unit with the host
clause, which is according to Brandt (1990) definable as a single partition into focus
and background. This demonstrates the ‘all-focus’ sentence in (25).

(23) Anna hat ein Buch, das liest sie nicht, und Emma hat
Anna has a book that reads she not and Emma has
auch eins.
also one
‘Anna has a book that she does not read and Emma also has one.’
(24) Anna hat ein Buch.(/), das liest sie nicht.
Anna has a book that reads she not
‘Anna has a book that she does not read.’
(25) Was ist passiert? (‘What happened?’)
[Anna hat ein Buch, das liest sie nicht]F
Anna has a book that reads she not
‘Anna has a book that she does not read.’

In terms of the aforementioned facts V2 relatives should be treated on a par with

­restrictive relative clauses, but things are more intricate as Gärtner (2001) argues.
He provides evidence that V2 relatives have properties that diverge from canoni-
cal ­(restrictive) relative clauses: (i) The finite verb occurs in second position; (ii) V2
­relatives must remain clause final as (26) suggests since V2 relatives cannot precede
ordinary verb final restrictive relative clauses; (iii) a quantifier cannot bind a variable in
a V2 relative, cf. (27); (iv) V2 relatives cannot be in scope of a negation operator realised
in the ­matrix clause, cf. (28); and (v) V2 relatives seem to be illocutionary independent,
because (29a) is ungrammatical indicating that the relative clause cannot be subsumed
under the question operator. Additionally, epistemic expressions and modal particles
may occur in V2 relatives, cf. (29b). Gärtner (2001) concludes from these data that V2
relatives must have an “assertional nature”.
 German dependent clauses 

(26) a. Anna hat Bücher, die sie sehr schätzt, die liest
Anna has books that she very appreciates that not
sie nicht.
reads she
‘Anna has books that she very appreciates that she does not read.’
b. *Anna hat Bücher, die liest sie nicht, die sie sehr schätzt.
Anna has books that reads she not that she very appreciates
(27) a. *Keine Studentini hat ein Buch, das liest siei nicht.
no student has a book that reads she not
(28) Anna findet ein Buch nicht, das hat Grass geschrieben.
Anna finds a book not that has Grass written
‘Anna does not find a book that Grass has written.’
(29) a. *Hat Anna ein Buch, das liest sie nicht?
has Anna a book that reads she not
b. Anna hat ein Buch, das liest sie {sicher/wohl} nicht.
Anna has a book that reads she {certainly/well} not
‘Anna has a book that she {certainly/well} does not read.’

The presented facts, thus, suggest that V2 relatives seem to oscillate grammatically
between main and subordinate clauses.

3.4  Free dass-clauses

A further class of clauses having properties residing between main and subordinate
clauses are the so-called free dass-clauses, which have been thoroughly described by
Reis (1997).14 Example (30) illustrates this clausal class.
(30) Max ist größenwahnsinnig, dass er jetzt noch.einen Porsche kauft.
Max is megalomaniac that he now another Porsche buys
‘Max is megalomaniac to buy another Porsche now.’

Free dass-clauses demonstrate that the presence of a complementizer is another pa-

rameter traditionally taken to indicate subordination that cannot be generally valid.
Apart from the clause final position of the finite verb, free dass-clauses behave only
with respect to their prosody and information structure like subordinate clauses. Ac-
cording to Reis (1997), it is possible to prosodically integrate a free dass-clause into its
host clause, which results in one single informational unit. In other words, free dass-
clauses do not necessarily form a separate phonological phrase and focus domain as is

.  I follow the terminology of Reis (1997). The term ‘free dass- clause’ is sometimes also used
for independent sentences like Dass Du mir nicht wegläufst. (‘Don’t run away.’), which I do not
discuss in this paper.
 Anke Holler

illustrated in example (31) showing that free dass-clauses can in principle be uttered
as ‘all-focus’ sentences.15
(31) Was ist passiert? (‘What happened?’)
[Max ist GRÖßenwahnsinnig, dass er jetzt noch.einen Porsche kauft.]F
Max is megalomaniac that he now another Porsche buys
‘Max is megalomaniac to buy another Porsche now.’
On the other hand, free dass clauses show properties that are not characteristic of
subordinate clauses. Contrary to dass-complement clauses, free dass-clauses are not
subcategorized by the host clause’s predicate. Hence, free dass-clauses do not allow
correlatives or supplements as (32) exemplifies.
(32) a. Maria ist (*es) blöd, dass sie kommt.
Maria is (it) kind.of.stupid that she comes
‘Maria is kind of stupid to come.’
b. *Maria ist blöd, und.zwar dass sie kommt.
Maria is kind.of.stupid namely that she comes
Free dass-clauses are also more restricted than dass-complement clauses with respect
to their topological behaviour. As noticed by Reis (1997), they are obligatorily extra-
posed, cf. (33).
(33) a. Du musst verrückt sein, dass du kommst.
you must crazy be that you come
‘You must be crazy that you come.’
b. *Dass du kommst, musst du verrückt sein.
that you come must you crazy be
c. Was ist denn gerade los, dass er so schreit?
what is part just.now the.matter that he like.that screams
‘What is wrong just now, that he screams like that?’
d. *Was ist denn, dass er so schreit, gerade los?
what is part that he like.that screams just.now the.matter
The fixed order is also evidence that free dass-clauses do not behave like subordi-
nate adverbial adjunct clauses, either, which for instance can occupy the clause initial
­position, as (34) indicates.
(34) Weil er so schreit, beschweren sich die Nachbarn.
because he likes.that screams complain refl the neighbours
‘Because he screams like that the neighbours complain.’
In addition, free dass-clauses occur before rather than after adverbial clauses intro-
duced by weil (‘because’) as Reis (1997) shows by means of examples like (35).

.  Note that it is generally also possible to decompose the construction into two informational
units: Max ist GRÖßenwahnsinnig, dass er jetzt noch einen PORsche kauft.
 German dependent clauses 

(35) a. Du bist blöd, dass du Ernas Nerzmantel bezahlst, weil

you are kind.of.stupid that you Erna’s mink.coat pays.for because
sie nämlich selbst genügend Geld auf dem Konto hat.
she namely herself enough money on the account has
‘You are kind of stupid to pay for Erna‘s mink coat because she has
enough money on her account.’
b. *Du bist blöd, weil sie nämlich selbst genügend
you are kind.of.stupid because she namely herself enough
Geld auf dem Konto hat dass du Ernas
money on the account has that you Erna’s
Nerzmantel bezahlst.
mink.coat pays.for
Moreover, Reis (1997) argues that free dass-clauses cannot appear as sentence ­fragments
in question-answer pairs, cf. (36a), which is again in contrast to canonical subordinate
adverbial clauses introduced by e.g., weil, cf. (36b).
(36) a. Wieso/Warum ist Fritz blöd? (‚Why is Fritz stupid?‘)
*Dass er Ernas Nerzmantel bezahlt.
that he Erna‘s mink.coat pays.for
b. Wieso/Warum ist Fritz blöd? (‘Why is Fritz stupid?‘)
Weil er Ernas Nerzmantel bezahlt.
because he Erna‘s mink.coat pays.for
‘Because he’s paying for Erna’s mink coat.’
The semantico-pragmatic properties of free dass-clauses are not completely under-
stood, yet. But it can be observed that free dass-clauses are not generally interpreted
in the scope of negation or negative predicates contained in the host clause, cf. (37),
and that a quantifier in the host clause may not bind a variable occurring in the free
dass-clause, cf. (38).16, 17
(37) *Max ist nicht größenwahnsinnig, dass er jetzt noch.einen Porsche kauft.
Max is not megalomaniac that he now another Porsche buys
(38) *Fast jeder Politikeri ist größenwahnsinnig, dass eri jetzt
almost every politician is megalomaniac that he now
noch.einen Porsche kauft.
another Porsche buys

.  These empirical facts contrast to Reis’s original results. She claims by means of examples like
Jederi war blöd, dass eri darauf eingegangen ist. (‘Everybody was stupid to accept it.’) that free dass-
clauses can contain a pronoun that is bound by a quantifier in the matrix clause. Having consulted
several speakers of German, I do not think that this example is acceptable in the intended reading.
.  Examples (37) and (38) improve a lot if the particle so (‘so’) is added in front of the host
clause’s adjective, but, in my opinion, this modification results in a new construction type.
 Anke Holler

Reis (1997) further noticed that free dass-clauses give reasons for a presumption or an
­assessment expressed in the prior clause based on the fact that they denote, and that
free dass-clauses are to some extent illocutionary independent of the preceding clause.
For ­instance, in example (39a) the content of the free dass-clause cannot be part of the
question expressed in the host clause. Uttering (39a) the speaker wonders whether
Max is megalomaniac and additionally gives a reason for her question/wonder, viz.
that Max bought another Porsche. Examples (39b) and (39c) illustrate contrary to this
that a matrix clause’s question operator may scope over a complement dass-clause or a
subordinate adverbial clause.
(39) a. Ist Max größenwahnsinnig, dass er jetzt noch.einen Porsche kauft?
is Max megalomaniac that he now another Porsche buys
‘Is Max is megalomaniac to buy another Porsche now?’
b. Sagt Max, dass er jetzt noch.einen Porsche kauft?
says Max that he now another Porsche buys
‘Does Max say that he is buying another Porsche now?’
c. Ist Max größenwahnsinnig, weil er jetzt noch.einen Porsche kauft?
is Max megalomaniac because he now another Porsche buys
‘Is Max megalomaniac because he is buying another Porsche now?’
Moreover, it is generally possible to add expressions typically indicating illocutionary
force (such as epistemic expressions, modal particles or performative indicators) to a
free dass-clause as (40) shows.
(40) Max ist größenwahnsinnig, dass er {sicher/wohl/hiermit}
Max is megalomaniac that he {certainly/well/hereby}
noch.einen Porsche kauft.
another Porsche buys
‘Max is megalomaniac to buy {certainly/well/hereby} another Porsche now?’
In a sense, free dass-clauses remind of V2 relative clauses in terms of their illocution-
ary force potential. To account for V2 relatives Gärtner (2002) assigns them an “asser-
tional proto-force”, thereby claiming that V2 relatives share this kind of illocutionary
force with similar “embedded root” constructions. Although Gärtner (2002), who has
not studied free dass-clauses, establishes the hypothesis that assertional proto-force is
triggered in embedded V2 declaratives only, I suppose that free dass-clauses are anoth-
er clausal class having assertional proto-force since they show a certain independent
­assertionality being “parasitic” on the force of their matrix clause. If this assumption
is correct, proto-force construal is not reducible to V2 phenomena anymore. I have to
leave this interesting issue for further research.

3.5  Dependent verb second clauses

German dependent verb second clauses (henceforth dependent V2 clauses) such as
(41) have attracted attention in a number of studies.
 German dependent clauses 

(41) Oskar denkt, er gewinnt das Spiel.

Oskar thinks he wins the match
‘Oskar think that he will win the match.’

This clausal class has been recently discussed for instance by Vikner (1995), Reis
(1997), Auer (1998), Steinbach (2007), Meinunger (2004), and Truckenbrodt (2006).
The mentioned work reports unanimously that dependent V2 clauses similarly show
properties of subordinate and main clauses. In the following, I briefly summarize the
fundamental results.
First of all, dependent V2 can be observed only in the complements of a certain
set of predicates.18 Since the present article does not aim at classifying predicate groups
that allow or do not allow for dependent V2 clauses, I simplify the matter a little bit
by assuming that dependent clauses with V2 order are restricted to the class of bridge
verbs.19 There is an ongoing debate as to whether the dependent V2 clause indeed syn-
tactically functions as an object clause satisfying an argument of the matrix predicate.
The main evidence speaking for its complement status is obvious: The matrix predicate
remains otherwise unsaturated.20 In addition, a rising tone at the end of the matrix
predicate signals that it awaits completion, as (42) exemplifies.
(42) Sie hatte geglaubt.(/), er SCHLIEfe.
she had believe he slept[SUBJ]
‘She had believed that he would sleep.’

.  Some nouns also allow for dependent V2 clauses, such as die Befürchtung, Max verspäte
sich (‘The apprehension that Max is[SUBJ] late’). Lexically, these nouns are mostly related to verbs
showing verb second as well. I have nothing to say on this issue in the present article.
.  In their classical paper on the applicability of root transformations Hooper & Thompson
(1973) already distinguished verb classes with respect to their ability to embed root phenomena.
Classifications for German verbs combining with dependent V2 clauses that have been developed
more recently are recapitulated in Meinunger (2004).
.  Extraction data like (i) are usually adduced as further evidence for the argument status
of V2 dependent clauses. Reis (1997), however, invalidates this argument by claiming that (i)
is not a case of extraction, but results from parenthetical insertion of the matrix clause glaubt
er (‘believes he’) into the dependent V2 clause. Even if one does not follow this argumentation,
data like (ii) indicate that extraction out of dependent V2 clauses is not allowed unconditionally.
I thank one of the anonymous reviewers for pointing me to the latter example.
(i) Wo glaubt er wohnt man billig?
where believe he lives one inexpensively
‘Where do you believe one can live inexpensively?’
(ii) *Wen glaubt er, Maria wird einladen?
whom believe he Maria will invite
 Anke Holler

Reis (1997), however, seriously questions that dependent V2 clauses realise an argu-
ment of the matrix predicate in the usual way. She concludes this from the fact that the
grammatical behaviour of dependent V2 clauses clearly deviates from the behaviour of
verb final dass-clauses, whose complement status is proven beyond doubt. One of the
various arguments Reis (1997) discusses is a topological one: Dependent V2 clauses
are restricted to a final position within the matrix clause.21 Consequently, they cannot
appear in subject or topic position. This is demonstrated by example (43), which takes
advantage of binding facts. Unlike a verb final dass-clause appearing in initial position,
a topicalised dependent V2 clause may not contain a pronoun bound by a quantifier
in the matrix clause.

(43) a. Jederi möchte gern glauben, eri sei unheimlich beliebt.

everyone want.to gladly believe he is[SUBJ] extremely popular
‘Everyone would like to believe that he is extremely popular.’
b. *Eri sei unheimlich beliebt, möchte jederi gern glauben.
he is[SUBJ] extremely popular want.to everyone gladly believe
‘Everyone would like to believe that he is extremely popular.’
c. dass eri unheimlich beliebt ist, möchte jederi gern glauben
that he extremely popular is want.to everyone gladly believe
‘That he is extremely popular, everyone would like to believe.’

Besides the aforementioned topological difference, dependent V2 clauses and verb ­final
dass-clauses behave differently with respect to correlatives and und zwar-supplements, as
the contrast in (44) illustrates.

(44) a. Hans hat (*es) geglaubt, Peter geht dahin zu Fuß.

Hans has it believed Peter goes there on foot
‘Hans believed Peter goes on foot there.’
b. Hans hat (es) geglaubt, dass Peter dahin zu Fuß geht.
Hans has it believed that Peter there on foot goes
‘Hans believed that Peter goes on foot there.’

Whether dependent V2 clauses may be in scope of a negated matrix predicate is still an

open empirical issue. Acceptable examples like (45) suggest that dependent V2 clauses
may appear as complements of negated verbs.

(45) Sie behauptet ja nicht, er sei Kommunist, [sondern ...].

she claims part not he is[SUBJ] communist but
‘She does not claim that he is a communist, but …’

.  An exception to this generalisation are V2 clauses that are dependent on nouns, e.g., Die
Befürchtung, Max verspäte sich, war völlig unberechtigt. (‘The apprehension that Max is[SUBJ] late
was completely unsubstantiated.’).
 German dependent clauses 

On the other hand, data like (46) substantiate the widely accepted view that V2 is
blocked under a negated matrix predicate.
(46) *Er glaubt nicht, Maria möchte das Theorem beweisen.
he believes not Maria wants.to the theorem prove
‘He doesn't believe Maria wants to prove the theorem.’
Conducting some corpus research, Auer (1998) observed that only a small group of verbs
(primarily verba dicendi) may license V2 in a dependent clause if they are n ­ egated. Mei-
nunger (2004), mainly referring to corpus work done by Butulussi (1991), also ­presents
a list of apparent counter examples to the conventional observation that V2 is ruled out
under negation. Discussing the problem at length he in a way affirms Auer’s original
result and additionally states that (i) only subjunctive mood can save a dependent V2
clause under negative elements, and that (ii) all of the listed negated matrix clauses “must
not be interpreted as an assertion, but must be interpreted as a command to give up a
certain belief.” (Meinunger 2004: 213). In the light of these very restrictive conditions,
advocating the view that negation blocks V2 still seems to be justified.22
Although the facts discussed so far concerning order, supplements and negation
point to a root-like behaviour of dependent V2 clauses, it should be noted that depen-
dent clauses with V2 order also show typical properties of subordinate clauses. As I have
mentioned earlier, their phonological shape as well as the thematic relationship between
the matrix predicate and the dependent V2 clause are characteristic for a subordinate
status. Binding facts provide even more positive evidence: Since binding is ­usually im-
possible across a sentence boundary, examples like (47) indicate that the dependent V2
clause must be structurally analysed as being deeper than the ­matrix clause; otherwise
the pronoun could not be c-commanded by the quantifier and, hence, not be bound.
(47) Fast jederi glaubt, eri sei der Beste.
almost everyone believes he is[SUBJ] the best
‘Almost everyone believes that he is the best one.’

In addition to admitting variable binding from outside, dependent V2 clauses may

appear as complements to predicates that are questioned, which is also in accordance
with ordinary verb final dass-complement clauses, cf. (48).
(48) a. Behauptet sie, er sei Kommunist?
claims she he is[SUBJ] communist
‘Does she claim that he is a communist?’
b. Behauptet sie, dass er Kommunist sei?
claims she that he communist is[SUBJ]
‘Does she claim that he is a communist?’

.  The matter is even more complicated since dependent V2 clauses cannot generally be
­combined with negative predicates like bezweifeln (‘doubt’) either, cf. Auer (1998), Steinbach
(2007). For reasons of limited space I will not go into this issue here.
 Anke Holler

The observation that the interrogative operator may take scope over a dependent V2
clause meets Reis’s claim that dependent V2 clauses may be information-structurally
integrated into their matrix clause, meaning that the matrix clause and the dependent
V2 clause are part of one and the same focus domain. In the face of the ­aforementioned
phonological facts, this information-structural behaviour comes as no surprise.
Much more sensation creates the interaction of verb second in German with the
assignment of illocutionary force. Details apart, many researches have argued that the
movement of the finite verb to the second position has its reason in the assignment
of assertional illocutionary force, cf. e.g., Wechsler (1991), Lohnstein (2000), Gärtner
(2002), Meinunger (2004), Truckenbrodt (2006). This theoretical assumption is based
on the well-known empirical observation that dependent V2 clauses may be used to
assert a certain proposition, cf. (49).
(49) Emma glaubt, Max wählt rechts.
Emma believes Max votes.for right
‘Emma believes that Max votes for right-wing candidates.’

It is widely accepted that two assertions can be made uttering (49): (i) Emma believes
something, and (ii) Max votes for right-wing candidates. Thus, any account of dependent
V2 clauses must explain this assertional force (potential).
Summing up, the presented facts reveal that dependent V2 clauses show mixed
properties in terms of the main-subordinate-clause distinction. Before turning to a
formal analysis accounting for the grammatical behaviour of the discussed five clause
types, let me give a short overview of the outlined facts.

3.6  Summary of the empirical results

The clause types discussed in section 3.1 to 3.5 are various instances of dependent
clauses that relate to their prior clause in a non-canonical way.
Table 1 gives an overall picture of these facts in comparison to canonically
linked clauses, i.e., standard complement clauses such as verb final dass complement
clauses or adjunct clauses such as ordinary adverbial clauses or restrictive relative
clauses.23
Table 1 makes perfectly clear that the behaviour of non-canonical clauses cannot be
reduced to the position of the finite verb. This result rejects all existing constraint-based

.  It is not intended to exhaustively represent all existing German clause types. Table 1 rather
summarises the subset of data discussed in the present article according to the criteria that
have been applied. Apart from the listed properties the table hence does not make any specific
assumption concerning the grammatical behaviour of canonical subordinate clauses. In particu-
lar, it says nothing about the relation between complement clauses on the one hand and adjunct
clauses on the other hand.
 German dependent clauses 

Table 1. Comparison of empirical properties of canonical and selected non-canonical

dependent clauses in German
Non-canonical clause types

Canonical Free dass-cl. Wh-rel. cl.

clause types Dep. V2 cl. V2 rel. cl. Weil-V2 cl.

Phonologically integrable + + + –
Information structurally integrable + + + –
Syntactic function + + – –
Variable binding from outside + + – –
Within scope of negation + – – –
Topologically free + – – –
Dependent illocutionary force + – – –

approaches relying on the V2 property as a means for distinguishing main (indepen-

dent) and subordinate (dependent) clauses. For instance, influential ­approaches such as
Uszkoreit (1987), Kathol (1995), Netter (1998) and Ginzburg & Sag (2000) are explicitly
based on this flawed assumption.
In addition, it is striking that the relation between a dependent clause and its
host clause may be differently tight, which confirms earlier generalisations saying
that clause integration in German is not a clear-cut phenomenon but underlies a cer-
tain gradation. See, for instance, König & van der Auwera (1988), Lehmann (1988),
­Fabricius-Hansen (1992), and Reis (1997).

4.  A constraint-based approach to non-canonical clause linkage

To account for the presented facts several theoretical proposals have been developed.
One consists in exploiting prototype theory by making a distinction between proto-
typical and marginal cases of subordination (e.g., Fabricius-Hansen 1992). Another
approach is to organise clause types in an elaborate taxonomy. Reis (1997), for instance,
defines a set of grammatical and pragmatic criteria upon which she distinguishes three
major classes of clauses: integrated clauses, i.e., all clauses being licensed by the matrix
predicate and having hence a syntactic function in the matrix clause, and unintegrated
clauses, which do not fulfil this criterion. Reis (1997) further subdivides the latter class
into a subclass of ‘relatively’ unintegrated clauses consisting of free dass-clauses and
dependent V2 clauses, and a second subclass of ‘absolutely’ unintegrated clauses com-
prising, for instance, continuous wh-relative clauses. A third solution, proposed inter
alia by Lehmann (1988) and Peyer (1997), is to evaluate clause types with regard to
a number of independent scalar parameters, and thus to analyse clause linkage as a
 Anke Holler

semanto-syntactic continuum. While being sympathetic with the idea to regard sub-
ordination as a composite term of various linguistic properties, I believe that none of
the sketched approaches offers a satisfactory solution for the described non-canonical
clause linkage phenomena in German. Therefore I will next develop a sort-driven
constraint-based analysis couched in the framework of Head-driven Phrase Structure
Grammar (Pollard & Sag 1994) that accounts for the data presented in this article.

4.1  Why is HPSG eligible for modelling degrees of clause linkage?

The architecture of Head-driven Phrase Structure Grammar (henceforth HPSG) qual-
ifies very well to capture a graded clause linkage. Firstly, all linguistic objects (i.e.,
words, subsentential phrases and sentences) are described as internally structured
complex signs consisting of phonological, syntactic, semantic, discourse, and phrase-
structural information, thereby positing multiple levels of structures. The linguistic
information is represented using hierarchically ordered attribute-value pairs. A sign
is assumed to have at least the attributes Phon(ology), Syn(tax)-Sem(antics), and
(in the case of phrases) D(augh)t(e)rs. The Synsem value distinguishes Loc(al)
and Nonloc(al) ­features. Whereas Nonlocal information is central in the analy-
sis of unbounded ­dependencies, Loc information has in turn three attributes of its
own called Cat(egory), Cont(ent), and Con(te)x(t). The Cat value includes
categorical information of a word and its ­required grammatical arguments. The
Cont value constitutes the word’s semantic contribution, and the Conx value con-
tains context-dependent semantico-pragmatic information. This feature architec-
ture on a par with a non-derivational perspective on grammar allows a simultane-
ous access to information stored at different structural levels.24 ­Secondly, a grammar
cast in the framework of HPSG is formulated as a declarative system of constraints,
which restrict the set of well-formed linguistic structures. Thirdly, the ­linguistic
structures employed in HPSG are sorted, which means that they are labelled with a
sort symbol. The finite set of sort symbols is partially ordered resulting in a multiple-
inheritance-hierarchy. Combining the latter two conceptual devices enables to model
­effects of constraint ranking, and, thus, to formulate restrictions that can be used to repre-
sent degrees of clause linkage. In addition, the sign-based setup of  HPSG allows extending
the application of general grammatical constraints to more and more ­‘peripheral’ clause

.  Although there are essential differences in the nature of structural levels posited in HPSG
on the one hand and in approaches influenced by Chomsky’s Government and Binding Theory
and its descendants on the other hand, we can regard Phon and Dtrs as rough analogs of the
GB levels phonetic form and S-structure; Cat plays a role roughly analogous to that of GB’s
D-structure, and Cont is most closely analogous to the GB level of logical form. However, the
Conx value, which contains linguistic information that bears on certain context-dependent as-
pects of semantic interpretation, does not map directly to a GB level.
 German dependent clauses 

phrase

independent dependent

integrated non-integrated

fully-integr weakly-integr weakly-non-integr fully-non-integr

Figure 1.  Partition of phrase in terms of the dimension Dependence.

linkage phenomena without being forced to posit a bifurcation ­between ‘core’ and ‘periph-
eral’ constructions. In the following I will sketch an analysis based on the hypothesis that
dependent clauses differ in the way they are combined with the preceding clause.

4.2  The proposed analysis

Following Green (1996), I firstly assume that objects of sort phrase may be partitioned
in terms of a Dependence dimension into independent and dependent objects, as de-
picted in figure 1.
While the sort independent comprises all independently uttered sentences,25 the sort
dependent describes all objects that are somehow attached to their linguistic ­surrounding
applying to all (canonical and non-canonical) clausal types ­presented in table 1. ­Building
on the empirical facts presented above, the sort dependent is ­further partitioned by two
subsorts, which are called integrated and non-integrated. The sorts fully-integr(ated) and
weakly-integr(ated) are in turn subsorts of the sort integrated, ­whereas the sorts fully-
non-integr(ated) and weakly-non-integr(ated) are subsorts of the sort ­non-integrated. By
formulating restrictions on these sorts we can easily model the empirical result that de-
pendent clausal objects differ in the degree they are integrated into a prior clause.26 It is
thereby assumed that the sort fully-integrated describes ­canonical dependent clauses and the
sort weakly-integrated dependent V2 clauses. The sort weakly-non-integrated represents

.  The class of objects of this sort also includes independent verb final clauses such as Ob er
noch kommt? (‘I wonder whether he will still come?’), which cannot be investigated in this arti-
cle. See for instance Oppenrieder (1991) for discussing this phenomenon.
.  Unfortunately, it cannot be discussed in this article to which extent this distinction can be
used for constituents other than clauses. At least, there is evidence from German and English
that nominal left-peripheral elements also need to be classified regarding their degree of (non-)
integrateness into a clause, cf. Shaer & Frey (2004).
 Anke Holler

free dass-clauses as well as V2 relative clauses. Last but not least, the sort fully-non-inte-
grated accounts for continuative wh-relative clauses and weil-V2 clauses.

4.2.1  Clauses of sort integrated

To begin with, clauses of sort integrated share the property of being related to the
prior clause by a selection or modification relation as they are either subcategorized
by the matrix predicate (complement clauses, dependent V2 clauses) or function as an
adjunct to it (adverbial clauses). In other words, these clauses are ­embedded into their
preceding clause. Assuming that canonical subordinate clauses of sort f­ ully-integrated
are analysed in a standard way (see e.g., Pollard & Sag 1994), I shall say nothing in
particular about them. I will rather concentrate on the analysis of non-canonical
types of dependent clauses starting with clausal phrases of sort weakly-integrated, i.e.,
dependent V2 clauses. Apart from the V2 property27, they differ in one major aspect
from canonical complement clauses: the assertional (proto-) force assigned to them.
This is subject to the constraint shown in figure 2 which says that all clauses of sort
weakly-integrated must contain an assert relation in their B(ac)kgr(ound) set.28
C AT | H EAD v2
weakly-integrated Æ SYNSEM | L OC
CONX | BKGR {assert_rel}

Figure 2.  Constraint restricting clauses of sort weakly-integrated (i.e., dependent V2 clauses).

On the other hand, constraint inheritance guarantees that dependent V2 clauses behave
like ordinary complement clauses in receiving a theta-role from the matrix predicate
since the sort weakly-integrated is defined as subsort of sort integrated. That dependent
V2 clauses are outside of the scope of negation is modelled as a lexical property of
negative particles requiring that they do not take scope over clausal phrases of sort
weakly-integrated. The peculiar topological behaviour of dependent V2 clauses is for-
malised by linear precedence rules. I will come back to both issues later.

4.2.2  Clauses of sort non-integrated

Since the empirical results show that clauses of sort non-integrated do not fulfil a syntactic
function in the prior clause and are not in its binding domain, they are neither a complement

.  For the sake of simplicity I say nothing specific about the way the V2 property is mod-
elled in HPSG. For a sort-driven proposal to represent verb final and verb second clauses see,
for instance, Holler (2005). The analysis presented here is, however, compatible with any other
­account to verb placement in German.

.  The Background feature is defined for objects of sort context. This sort represents the val-
ue of the Context feature modelling discourse and pragmatic information. The value of Back-
ground is a set of parametrized states of affairs. See Green (2000) for a proposal for ­structuring
the Background set to appropriately treat illocutionary force.
 German dependent clauses 

nor an adjunct of the matrix predicate. Thus, I assume at first that clauses of sort weakly-
non-integrated modify the complete CP representing the prior clause. Correspondingly,
this is realised by restricting the Mod(ified) value of objects of this sort to CPs, as the
constraint in figure 3 shows. This constraint additionally states that assertional (proto-)
force must be assigned to clauses of sort weakly-non-integrated, which is again achieved by
manipulating the Background set.
verb
H EAD
C AT M OD CP
weakly-non-integrated Æ SYNSEM | L OC
S UBCAT 〈〉
CONX | BKGR { assert_rel }

Figure 3.  Constraint restricting clauses of sort weakly-non-integrated (i.e., V2 relative clauses,
free dass-clauses).

To account for clauses of sort fully-non-integrated, I adapt an approach to ­English periph-

eral adverbials by Haegeman (1991). Haegeman argues that peripheral ­adverbials can best
be treated as being outside the syntactic representation of the ­sentences that they modify.
Elaborating on a proposal for non-restrictive relative ­clauses made by ­Safir (1986), she
pursues the hypothesis that peripheral adverbials are syntactically ­unattached “orphan
constituents”. Under this view orphan constituents are neither ­attached at D- or S-struc-
ture nor at logical form. Nevertheless, syntax may ­constrain the interpretation of orphan
constituents. For instance, Haegeman (1991) suggests that ­peripheral ­adverbials have to
be coindexed with a sentential modifiee. She further claims that the interpretation of or-
phan constituents must be dealt with on the basis of a theory of ­utterance ­interpretation. I
propose to analyse clauses of sort ­fully-non-integrated as orphan ­constituents in the terms
of Haegeman (1991). The ­sign-based architecture of HPSG described in section 4.1 inte-
grating ­grammatical, pragmatic and phrase-structural ­information as well as HPSG’s rep-
resentational ­formalism employing sorted feature structure ­descriptions are best suited
for implementing such an orphan approach. It ­allows to built constituents that are not
­syntactically but rather (discourse-)interpretationally linked. Such constituents can be
conceived of suprasentential units combining sentential constituents that, although syn-
tactically unattached, are ­interdependent on the level of utterance ­interpretation. Under
this view, the highest unit of linguistic description is not the sentence (i.e., a CP or an ob-
ject of sort clause), but a minimal ­discourse unit paratactically linking sentential phrases.
Units of this kind have been stipulated before, for instance by Banfield (1973), Fabb (1990)
or Gärtner (2001). Due to the HPSG setup the hypothesis outlined here, however, differs
from earlier ­approaches in not requiring that there is a syntactic phrase marker immediately
­dominating both sentential ­constituents. It is an asset of the HPSG formalism that it allows
within one object of sort sign, i.e., a linguistic object, establishing a discourse-structural
relation without being forced to ­simultaneously ­establish a syntactic ­relation.
The just described fundamental idea can be formally realised in three steps. In the
first step, phrases are distinguished in the dimension Headedness whether they have
 Anke Holler

phrase

hd-phrase non-hd-phrase

hd-nexus-phrase hd-fill-phrase host-orphan-phrase […]

Figure 4.  Part of the partial sort hierarchy in the dimension Headedness.

a head or not (following Ginzburg & Sag 2000), and for phrases of sort non-h(eade)d-
phrase a subsort called host-orphan-phrase is stipulated.29 See figure 4.
In the second step, the constraint depicted in figure 5 is formulated. It defines
­fully-non-integrated clauses30 as part of a phrase of sort host-orphan-phrase and ­restricts this
phrase in an appropriate way. The constraint under discussion specifies that a fully-non-
integrated clause figures as an orphan daughter. It additionally states that orphan constituents
do not syntactically modify another phrase because the orphan daughter’s Mod feature is
specified by the value none, and the host daughter’s Subcat list is required to be saturated
(empty). Thus, there is no clause to which the orphaned clause may attach ­syntactically.
However, there is a discourse-structural relation that is established between the host and the
orphan constituent. This is expressed by structure sharing the Content value of the host
daughter with an element of the Background set of the orphan daughter. The semantics of
the host hence acts as the contextual background of the orphan constituent.31

.  I leave it open which further subsorts of non-headed-phrase should be defined.

.  With Uhmann (1998) I assume that two lexical entries for weil (‘because’) must be ­stipulated.
In Holler (2005) I have argued that continuative wh-relative clauses are projected from a peculiar
non-restrictive relativizer. Note that these two elements trigger the value f­ ully-non-integrated for
weil-V2 clauses and continuative wh-relative clauses, respectively.
.  One reviewer raised the question of how the proposed discourse-based analysis may differ-
entiate between example pairs like (i) Peter kommt zu spät, weil er hat keinen Parkplatz gefunden.
(‘Peter is late because he could not find a parking space.’) vs. Peter kommt zu spät. Er hat nämlich
keinen Parkplatz gefunden. (‘Peter is late. He namely could not find a parking space.’), and (ii)
Emma gewann die Schachpartie, was Oskar ärgerte. (‘Emma won the chess match, which annoy-
ed Oskar.’) vs. Emma gewann die Schachpartie. Das ärgerte Oskar. (‘Emma won the chess match.
This annoyed Oskar.’) because each second clause of the four example constructions in (i) and
in (ii) relates discourse-functionally to the prior clause in a similar way. This, however, is no
objection against the proposed analysis for one important reason: in both example pairs, the sort
dependent is assigned only to the latter clause of the complex clause construction, i.e., the first
construction, but not to the latter clause of the sequence of clauses, i.e., the second construc-
tion. Hence, only the first construction in (i) and (ii) may figure as an object of sort host-orphan
phrase, and is thus described as a sign. The sequence of clauses, however, must be described by
 German dependent clauses 

In the third step, a constraint on objects of sort fully-non-integrated is assumed

positing that all clauses of this sort contain an assert relation in their Background set,
and thus may have illocutionary force, cf. figure 6.

HEAD verb
CAT
HOST-DTR | SS | LOC SUBCAT 〈〉

CONT 1
host-orphan-phrase Æ
fully-non-integrated
ORPH-DTR CAT | HEAD| MOD none
SS | LOC
CONX | BKGR { 1 , assert_rel}

Figure 5.  Constraint defining clauses of sort fully-non-integrated (i.e., weil-V2 clauses, con-
tinuative wh-relative clauses) as orphan constituents.

CAT | HEAD verb

fully-non-integrated Æ SYNSEM | LOC
CONX | BKGR {assert_rel}

Figure 6.  Constraint restricting clauses of sort fully-non-integrated (i.e., weil-V2 clauses,
­continuative wh-relative clauses).

If we now reinspect the facts concerning the scope of negation, we can generalise that
a negative particle may only scope over clauses of sort fully-integrated, as can ­easily be
specified in the lexicon.
Last but not least let me comment on a further advantage of the sort-driven ­approach
as pursued here. Employing linear precedence rules, it allows accounting for the ob-
served topological facts in a plausible way. See for instance the following set of rules:

i. Objects of sort weakly-non-integrated must precede objects of fully-non-integrated.

ii. Objects of sort fully-integrated must precede objects of weakly-integrated.
iii. Objects of sort integrated must precede objects of non-integrated.

two separate sign objects. Formally, this different treatment is achieved by the lexical specifica-
tion of the introducing elements (weil (‘because’) in weil-V2 clauses and a wh-relativizer in con-
tinuative wh-relatives) in each dependent clause of the complex clause construction in (i) and
(ii). Since these elements are not present in the sequence of clauses, in these cases each clause is
separately assigned the sort independent. Furthermore, that a dependent clause may be linked
to a preceding clause by one and the same discourse function as underlying two subsequent
independent clauses is not exceptional and hence says nothing about the dependence status.
Compare for instance Maria weint. Max hat ihr Spielzeug weggenommen. (‘Maria screams. Max
has taken away her toys.’) with Maria weint, weil Max ihr Spielzeug weggenommen hat. (‘Maria
screams because Max has taken away her toys.’)
 Anke Holler

In a nutshell, the proposed analysis has three essential ingredients: (i) Clauses are dif-
ferentiated in terms of their dependence on a putative matrix clause. (ii) Dependent
clauses that link non-canonically to their prior clause are partitioned into three sets,
each labelled by a specific sort. (iii) By constraining these sorts the empirical properties
of the discussed clause types are modelled non-derivationally, thereby realising effects
of graded clause linkage.

5.  Conclusion

Taking German as the main example, I have argued that dependent clauses must be
­distinguished regarding their degree of integration into a putative matrix clause. This
has been empirically shown by means of five instances of non-canonical clause types:
weil-verb second clauses, continuative wh-relative clauses, verb second relative clauses,
free dass-clauses, and dependent verb second clauses. Accounting for the presented data,
a constraint-based analysis has been developed that formulates restrictions on ­partially
ordered sorts, which are assigned to certain clause types. In accordance with the facts,
the analysis does not depend on the position of the finite verb, and hence ­rejects earlier
constraint-based approaches that implement a strict dichotomic ­distinction ­between
main and subordinate clauses relying only on the V2 property. In particular, the
­presented analysis allows distinguishing the (in)dependence relation from other gram-
matical factors contributing to the multidimensional phenomenon of subordination.

Acknowledgements

The paper is a considerably revised version of Holler (2005). I am grateful to the audi-
ence of the HPSG 2005 conference in Lisbon and the International CoGETI work-
shop on constraint-based grammar 2005 in Bremen for inspiring discussions. I also
thank Cathrine Fabricius-Hansen and Wiebke Ramm for giving me the opportunity
to present the core ideas of this paper at the workshop “‘ Subordination’ vs. ‘Coordina-
tion’ in Sentence and Text” at the 2006 DGfS meeting in Bielefeld and the audience
there for helpful comments. I would like to thank furthermore Hardarik Blühdorn,
Danièle ­Godard and three anonymous reviewers for critical remarks and stimulating
­suggestions on an earlier version of this paper. All errors and shortcomings are mine.

References

Altmann, Hans. 1997. Verbstellungsprobleme bei subordinierten Sätzen in der deutschen

Sprache. In Syntax im Fokus. Festschrift für Heinz Vater, Christa Dürscheid, Karl Heinz
Ramers & Monika Schwarz (eds), 69–84. Tübingen: Niemeyer.
Auer, Peter. 1998. Zwischen Parataxe und Hypotaxe: ‘abhängige Hauptsätze’ im gesprochenen
und geschriebenen Deutsch. Interaction and Linguistic Structures 2.
 German dependent clauses 

Banfield, Ann. 1973. Narrative Style and the grammar of direct and indirect speech. Foundations
of Language 10: 1–39.
Blühdorn, Hardarik. Forthcoming. Epistemische Lesarten von Satzkonnektoren – Wie sie
­zustande kommen und wie man sie erkennt. To appear in Semantik vs. Pragmatik oder
Semantik und Pragmatik?, I. Pohl (ed.). Tübinjen: Niemeyer.
Brandt, Margareta. 1990. Weiterführende Nebensätze. Stockholm: Almquist & Wiksell International.
Butulussi, Eleni. 1991. Studien zur Valenz kognitiver Verben im Deutschen und Neugriechischen.
Tübingen: Niemeyer.
Emonds, Joseph E. 1970. Root and Structure Preserving Transformations. PhD dissertation,
MIT.
Fabb, Nigel. 1990. The difference between English restrictive and non-restrictive relative clauses.
Journal of Linguistics 26: 57–78.
Fabricius-Hansen, Cathrine. 1992. Subordination. In Deutsche Syntax. Ansichten und Aussich-
ten, Jahrbuch des IDS 1991, L. Hoffmann (ed.), 458–483. Berlin: Walther de Gruyter.
Gärtner, Hans-Martin. 2001. Are there V2 –.relative clauses in German? Journal of Comparative
Germanic Linguistics 3(2): 97–141.
Gärtner, Hans-Martin. 2002. On the force of V2 declaratives. Theoretical Linguistics 28: 33–42.
Ginzburg, Jonathan & Sag, Ivan A. 2000. Interrogative Investigations: The Form, Meaning, and
Use of English Interrogatives. Stanford CA: CSLI.
Green, Georgia M. 1976. Main clause phenomena in subordinate clauses. Language 52: 382–397.
Green,  Georgia M. 1996. Distinguishing main and subordinate clause: The root of the prob-
lem. Unpubl. Ms., University of Illinois. (URL: http://www.linguistics.uiuc.edu/g-green/
marseille.ps)
Green, Georgia M. 2000. The nature of pragmatic information. In Grammatical Interfaces in
HPSG, R. Cann, C. Grover & P. Miller (eds), 113–138. Stanford CA: CSLI.
Haegeman, Liliane. 1991. Parenthetical adverbials: The radical orphanage approach. In Aspects
of Modern English Linguistics, S. Chiba, A. Ogawa, Y. Fuiwara, N. Yamada, O. Koma &
T. Yagi (eds), 232–254. Tokyo: Kaitakusha.
Haiman, John & Thompson, Sandra A. 1984. ‘Subordination’ in Universal Grammar. In Proceedings
of the Tenth Annual Meeting of the Berkeley Linguistics Society, C. Brugman et al. (eds),
520–523. Berkley CA: Berkeley Linguistics Society.
Holler, Anke. 2005. Weiterführende Relativsätze. Empirische und theoretische Aspekte. Berlin:
Akademie Verlag.
Hooper, Joan & Thompson, Sandra A. 1973. On the applicability of root transformations.
­Linguistic Inquiry 4: 465–497.
Kathol, Andreas. 1995. Linearization-Based German Syntax. PhD dissertation, Ohio State
­University.
Keller, Rudi. 1995. The epistemic weil. In Subjectivity and Subjectivisation. Linguistic Perspec-
tives, D. Stein & S. Wright (eds), 16–30. Cambridge: CUP.
König, Ekkehard & van der Auwera, Johan. 1988. Clause integration in German and Dutch con-
ditionals, consessive conditionals, and concessives. In Clause Combining in Grammar and
Discourse, J. Haiman & S.A. Thompson (eds), 101–133. Amsterdam: John Benjamins.
Lehmann, Christian. 1988. Towards a typology of clause linkage. In Clause Combining in
­Grammar and Discourse, J. Haiman & S.A. Thompson (eds), 181–225. Amsterdam: John
Benjamins.
Lohnstein, Horst. 2000. Satzmodus – kompositionell. Zur Parametrisierung der Modusphrase.
Berlin: Akademieverlag.
Longacre, Robert. 1996. The Grammar of Discourse. New York NY: Plenum.
 Anke Holler

Meinunger, André. 2004. Verb position, verbal mood and the anchoring (potential) of sentences.
In Syntax and Semantics of the Left Periphery, H. Lohnstein & S. Trissler (eds), 313–341.
Berlin: Mouton de Gruyter.
Netter, Klaus. 1998. Functional Categories in an HPSG for German. PhD dissertation, German
Research Center for Artificial Intelligence (DFKI) & University of Saarland.
Oppenrieder, Wolfgang. 1991. Von Subjekten, Sätzen und Subjektsätzen. Untersuchungen zur
Syntax des Deutschen. Tübingen: Niemeyer.
Pasch, Renate. 1997. Weil mit Hauptsatz – Kuckucksei im denn-Nest. Deutsche Sprache 25(3):
252–271.
Pasch, Renate, Brauße, Ursula, Breindl, Eva & Waßner, Ulrich H. 2003. Handbuch der deutschen
Konnektoren. Linguistische Grundlagen der Beschreibung und syntaktische Merkmale der
deutschen Verknüpfer (Konjunktionen, Satzadverbien und Partikeln). Berlin: de Gruyter.
Peyer, Ann. 1997. Satzverknüpfung: syntaktische und textpragmatische Aspekte. Tübingen:
­Niemeyer.
Pollard, Carl & Sag, Ivan. 1994. Head-Driven Phrase Structure Grammar. Stanford: CA CSLI.
Reis, Marga. 1997. Zum syntaktischen Status unselbständiger Verbzweit-Sätze. In Syntax im
Fokus. Festschrift für Heinz Vater, C. Dürscheid, K.H. Ramers & M. Schwarz (eds), ­121–144.
Tübingen: Niemeyer.
Safir, Ken. 1986. Relative clauses in a theory of binding and levels. Linguistic Inquiry 17(4):
663–689.
Santorini, Beatrice. 1992. Variation and change in Yiddish subordinate clause word order.
­Natural Language and Linguistic Theory 10: 595–640.
Shaer, Ben & Frey, Werner. 2004. Integrated and non-integrated leftperipheral elements in
­German and English. ZAS Papers in Linguistics 35(2), 465–502.
Steinbach, Markus. 2007. Integrated parentheticals and assertional complements. In parentheti-
cals, N. Dehé & Y. Kavalova (eds), 53–87. Amsterdam: Benjamins.
Sweetser, Eve. 1990. From Etymology to Pragmatics. Metaphorical and Cultural Aspects of Seman-
tic Structure. Cambridge: CUP.
Truckenbrodt, Hubert. 2006. On the semantic motivation of syntactic verb movement to C in
German. Theoretical Linguistics 32–3: 257–306.
Uhmann, Susanne. 1998. Verbstellungsvariation in weil-Sätzen. Zeitschrift für Sprachwissen-
schaft 17(1): 92–139.
Uszkoreit, Hans. 1987. Word Order and Constituent Structure in German. Chicago IL: Chicago
University Press.
Van Valin, Robert. 1984. A typology of syntactic relations in clause linkage. In Proceedings of the
Tenth Annual Meeting of the Berkeley Linguistics Society, C. Brugman et al. (eds), 542–558.
Berkley CA: Berkeley Linguistics Society.
Vikner, Sten. 1995. Verb-Movement and Expletive Subject in the Germanic ­Languages. New
york: Oxford University Press.
Wechsler, Stephen. 1991. Verb second and illocutionary Force. In Views on Phrase Structure, K.
Leffel & D. Bouchard (eds), 177–191. Dordrecht: Kluwer.
Wegener, Heide. 1993. Weil – das hat schon seinen Grund. Zur Verbstellung in Kausalsätzen mit
weil im gegenwärtigen Deutsch. Deutsche Sprache 21: 289–305.
 To the right of the clause
Right dislocation vs. afterthought*

Maria Averintseva-Klisch
University of Tübingen

In my paper I will introduce two different, although apparently similar

constructions at the right sentential edge in German: right dislocation (RD) and
afterthought (AT). I show that RD is a discourse-structuring device: it marks the
topic for the following discourse segment. AT, on the contrary, is a local repair
strategy. I assume that a syntactically independent repair like AT and a syntactically
adjoined discourse-structuring device like RD are cross-linguistically two options of
the right sentential edge, and test this assumption for French and Russian.

Keywords: Right dislocation, discourse topic, afterthought, discourse relation

1.  Introduction

‘German right dislocation’ is since Altmann (1981) a term for a construction consist-
ing of an NP1 at the right edge of the clause (i.e., after the formal completion of the
clause) and a coreferent intraclausal pro-form, as in (1):
(1) Hast Du ihn schon gesehen, (ich meine) den Karl?
have you himi already seen, (I mean) the Karli
(Schindler 1995: 44)

*  This paper emerged as part of my doctoral research conducted first at the DFG-funded
graduate school “Economy and Complexity in Language” at the Humboldt University Berlin,
and then at the University of Tübingen. I would like to thank my supervisor Claudia Maienborn
(Tübingen), Manfred Consten (Jena), Mareile Knees (Jena) und Fabienne Salfner (Berlin) for
numerous valuable comments on my work, the organizers and participants of the AG “ ‘Sub-
ordination’ versus ‘coordination’ in sentence and text – from a cross-linguistic perspective” for
stimulating discussions, and Cathrine Fabricius-Hansen as well as three anonymous reviewers
for their helpful remarks on the first draft of this paper.
.  Altmann (1981) also mentions PP- and CP-right dislocations; these are not considered in
this paper. See, however, Averintseva-Klisch & Salfner (2007) for PP-RD.
 Maria Averintseva-Klisch

Traditional analyses of the German right dislocation assume that this is a construction
of spoken German that serves to resolve a potentially unclear pronominal reference
(Altmann 1981; Auer 1991; Schindler 1995; Selting 1994; Uhmann 1993; 1997). The
speaker of (1) notices that the use of the pronoun ihn (‘him’) is unclear in the context
(it might be that there are either several referents that are equally suitable as an ante-
cedent, or no referent is activated highly enough for the reference of the pronoun to be
resolved effortlessly by the addressee). This observation causes the speaker to provide
as an “afterthought” information that should make it easier to identify the intended
referent.
However, there are abundant cases where a right dislocation cannot serve the ref-
erence clarification, as the reference is pretty clear, cf. (2):
(2) a. (“Der Taifun!” rief Lukas dem Kapitän zu.
The typhoon.mask called Lukas the captain to
“Da ist er!”) Ja, da war er, der Taifun.
There is he Yes there was hei the typhooni
(Michael Ende, Jim Knopf und die Wilde 13: 190)
b. (Den Tag, den vergess‘ ich nicht,)
(That day, d-pron forget I not)
der war viel zu schön, der Tag.
d-proni was much too wonderful the dayi.
(Altmann 1981:129)

Here the right dislocation marks the referent of the NP (the typhoon in (2a) and the
day in (2b)) as being especially important for the discourse. To be more exact, the
referent is marked as the discourse topic in the sense of aboutness (see section 2.2.).
I argue that (1) and (2) are instances of two functionally and formally different con-
structions that have been subsumed under the label of right dislocation. I name these
constructions ‘right dislocation proper’ (RD), cf. (2), and ‘afterthought’ (AT), cf. (1),
following the distinction made for other languages, e.g., in Ziv (1994) and Ward &
Birner (1996) for English and Fretheim (1995) for Norwegian.
In this paper, I will first dwell upon the functional and formal differences between
RD and AT in German. In section 2, I will briefly introduce prosodic and syntactic fea-
tures of RD and AT (section 2.1.), and then turn to their respective discourse functions
(section 2.2.). I will propose that AT is a syntactically independent ‘orphan’ (Haegeman
1991), that gets introduced into the discourse via a discourse relation Afterthought for-
mulated for this case (section 2.3.). As for RD, it is syntactically integrated into its host
sentence, and serves to mark the discourse topic for the following discourse segment.
In section 3, I will address the general issue of the use of the right clausal edge.2 I will

.  I use the expression “right clausal edge” here and later to refer to constructions coming after
the syntactic completion of the clause irrespectively of the fact whether they really pertain syn-
 To the right of the clause 

argue with data from German, Russian and French that RD (i.e., a syntactically inte-
grated global discourse-structuring device) and AT (i.e., a syntactically non-integrated
local repair) are two cross-linguistically possible options of the use of the right clausal
edge. Finally, in section 4, I will sum up and draw some conclusions.

2.  RD vs. AT: Form and function

In this section, I introduce the prosodic and syntactic differences between RD and AT.
Many of the facts presented here have been already pointed out in the previous re-
search, e.g., in Altmann (1981), but have mostly been done up with as exceptions from
a certain pattern of a so-called “right dislo­cation”. However, distinguishing between
RD and AT allows to dispence with most exceptions, and to describe distinct patterns
for RD and AT instead. The formal differences between RD and AT suggest that RD is
prosodically and syntactically part of its host sentence, while AT is not. After describ-
ing these differences I will specify the discourse functions of RD and AT which have
been mentioned above.

2.1.  RD vs. AT: Prosodic and syntactic differences

RD is prosodically integrated into its host sentence (3a), i.e., it continues the tone
movement of the host sentence and thus does not build a prosodic unit (an intona-
tional phrase in terms of Selkirk (1978)) of its own, whereas AT builds an intonational
phrase (optionally divided from the clause by a pause) with a tone movement and a
clause-like accent of its own (cf. Uhmann 1997), (3b)3:
(3) a. [Ich MAG siei nicht, die Brigittei]. RD
b. [Ich MAG siei nicht], | [die BriGITtei]. AT
Ilike her not, the Brigitte.
‘I don’t like her, Brigitte.’
(|: pause; [ ]: intonational phrase; CAPITALS: primary accent)

Crucially, prosody is only one of several criteria for distinguishing between RD and
AT. Prosodic differences go along with syntactic differences, which are briefly listed

tactically to the clause (being base-generated or moved right IP-adjuncts) or whether they are
syntactically autonomous. That is why I do not use the term “right periphery” here, as this term
implies that a right-peripheral construction belongs syntactically to its host sentence. “Right
clausal edge” in this sense covers right peripheral items as well as autonomous afterthought ad-
ditions, which are syntactically not part of the host sentence, as I argue below.

.  See Fretheim (1995) for a similar analysis for Norwegian: he shows that in Norwegian,
as in German, prosodically integrated structures are RDs, and prosodically non-integrated
ones ATs.
 Maria Averintseva-Klisch

below; they all suggest that RD belongs in a much more straightforward way to its host
sentence than AT.
I. Strict morphological agreement (in case, gender and number) between the
clause-internal pro-form and the NP is necessary for RD, while at least gender agree-
ment is only optional for AT, cf. (4) vs. (5):
(4) (Ach ja, mein Nachbar! Er hat gestern wieder einen Wutanfall
bekommen, nur weil die Kinder im Garten gespielt haben. )
‘(Oh yes, my neighbour! He went completely berserk yesterday, only
because the children were playing in the garden.’)
Ich kann ihn nur bedauern, den Mann /        RD
I can himakk.mask only regret the manakk.mask /
*die Giftspritzei / *der Manni.
*the spitfireakk.fem / *the mannom.mask
(5) a. Esi ist ein bisschen gleichförmig, | deine Melodiei.4
it.neutr is a bit homogeneous your melody.fem AT
(Attested oral data)
b. Und dann passierte das Unglücki, (ich meine)
And then happened [the misfortune]neutr (I mean) AT

dieser schreckliche Autounfalli / ich meine
[this terrible traffic.accident]mask.nom / I mean

diesen schrecklichen Autounfalli.5
[this terrible traffic.accident]
mask.akk

II. A subordinate clause between the clause-internal pro-form and the NP is not
possible for RD and possible for AT, cf. (6):
(6) Ach, diese Münchner Stadtväter!
a. Es könnte ihneni ja wirklich aufgefallen sein, RD
it could themi yes really standing.out be

.  It is not quite clear, whether es and the AT-NP deine Melodie really corefer, or whether
the neutral weak personal pronoun es here is a so-called “anticipatory es”, which refers in a
very abstract way (Smith 2002: 95). In the latter case coreference in the strict sense of the word
does not take place. For RD in any case coreference between the RD-NP and the pro-form is
required. Thus, irrespective of the fact whether grammatical incongruence corresponds to non-
coreference here or not, (5a) is only possible as AT, and not as RD.

.  If an AT is introduced with the addition ich meine (‘I mean’), the case of the AT-NP can
vary between nominative and accusative. Altmann (1981) argues against the assumption that
the verb meinen (‘to mean’) constituting the part of the addition is really able to govern the ac-
cusative case, so that the case variation cannot be explained with the case marking through the
verb.
 To the right of the clause 

daß dazu das Geld fehlt, *den Münchner Stadtväterni.

that for.it the money lacks *the of.Munich city.fathersi

b. Es könnte ihneni ja wirklich aufgefallen sein, daß dazu
it could themi yes really standing.out be that for.it

das Geld fehlt,| ich meine (damit) die M. Stadtv.i. AT
the money lacks I mean (with.it) the of.M. city.fathersi
(Altmann 1981: 115)
‘Oh, these city fathers of Munich! Theyi really could have noticed, that the
money for this purpose is lacking, (I mean) these city fathers of Munich /
*these city fathers of Munichi!’

Furthermore, optional additions (ich meine (‘I mean’), also (‘that is’), tatsächlich (‘re-
ally’) etc.) between the clause-internal pro-form and the NP are possible for AT and
not for RD, cf. (7) vs. (8):6
(7) (“Der Taifun!” rief Lukas dem Kapitän zu. „Da ist er!“)
‘“The typhoon!” Lukas called to the captain. “Here it comes!”’ RD
Ja, da war er, (*ich meine /*also/*tatsächlich) der Taifun.
yes, there was iti (*I mean /*that.is/*really) the typhooni
(8) (Lisa und Melanie haben sich gestritten.)
‘Lisa and Melanie quarrelled.’
Dann ist sie weggelaufen, | (ich meine / also) Lisa. AT
then is shei run.away (I mean / that.is) Lisai

III. The NP is not tied to the right-peripheral position in the case of AT, but can
have a fairly free position in its host sentence, while RD is only possible at the right
periphery, cf. (9) vs. (10):
(9) a. Ich habe ihn gestern nur mit Mühe wiedererkannt, | ich
I have him yesterday only with effort recognized, I
meine den Peter. AT
mean the Peter
b. Ich habe ihn, | ich meine den Peter, | gestern nur mit Mühe wiedererkannt.
c. Ich habe ihn gestern | ich meine den Peter, | nur mit Mühe wiedererkannt.
‘I hardly recognized him yesterday, I mean Peter.’

.  As the examples show, the (im)possibility of additions with RD and AT is not due to the
meaning and function of the addition, as one might be tempted to believe in the case of ich
meine / also (‘I mean’ / ‘that is’), which are additions explicitly assisting the reference-clarifying
function of AT. Also such additions like natürlich (‘of course’), tatsächlich (‘really’) etc., which
are insensitive to the functional difference between RD and AT, are bad with RD and perfectly
acceptable with AT. Thus, this difference seems to be a syntactic one.
 Maria Averintseva-Klisch

(10) Den Tag, den vergess’ ich nicht,

the dayi d-proni forget I not
a. deri war viel zu schön, der Tagi.
d-proni was much too beautiful the dayi
(Altmann 1981: 129)
b. *deri, der Tagi, war viel zu schön. RD
*d-proni the dayi was much too beautiful
‘That day, I will never forget it, it was so beautiful, the day.’

To summarize: there is ample evidence that RD belongs prosodically and syntac-

tically in a much more straightforward way to its host sentence than AT. Prosodically,
RD is a part of its host sentence’s tone contour. Con­sidered syntactically, morphological
agreement of the RD-NP with the clause-internal pro-form suggests that NP is part of
the clause, as morpho­lo­gical agreement is a sentence-bound phenomenon.7 Moreover,
RD occu­pies a fixed position in the host sentence (at its right periphery), and does not
allow insertions (neither subordinate clause insertion nor optional additions of any kind)
between the host sentence and the RD-NP. The former constraint, i.e., not allowing sub-
ordinate clause insertion, is since Ross (1967) known in the literature as “Right Roof Con-
straint”: Ross assumes rightward movement to be possible, but never across a sentence
boundary. For reasons of length and thematic contiguity of this paper I am not going to
discuss the issue whether RD is a product of syntactic movement or base-generated at the
right periphery.8 I use the term ‘right dislocation’ without implying any syntactic analysis,
but rather as it is the term generally accepted in the literature for this construction, cf. e.g.,
Ziv (1994), Fretheim (1995), Ward & Birner (1996), Lambrecht (2001).
However, RD in German seems to be island-sensitive, cf. (11), which might be
considered an argument for the movement analysis:

(11) Über Meyer1 werden öfters Gerüchte verbreitet. Mal soll er1 Beziehungen zur
Regierung haben, mal werden seine1 Erfolge mit Kontakten zur Mafia erklärt.
‘There are lots of rumours about Meyer. One hears that he is bound to have
connections to the government, or even that his success is due to his contacts to
the mafia.’

.  Consten (2004: 91) shows that intersentential anaphoric resumption is generally possible
without gender agreement, whereas it is impossible for intrasentential anaphora.

.  I am not aware of any detailed syntactic analysis of German RD, but see Ross (1986/1967),
Kayne (1994), and Culicover / Jackendoff (2005), to name but a few, for different possibilities of
syntactic analysis of RD in general. Ross assumes that RD is a result of a rightward movement,
whereas for Kayne RD emerges through the movement of the remaining clause to the left of the
RD-NP. Villalba (2000) modifies this analysis for RD in Catalan slightly, still assuming that RD
involves the leftward movement of its host clause. In contrast, Culicover / Jackendoff (2005)
advocate the base generation of the RD to the right of its host sentence.
 To the right of the clause 

a. *Ich habe schon oft [NP das Gerücht, dass ihm1 die Mafia
I have already often [NP the rumour that him1 the mafia
geholfen hat] gehört, dem Meyer1.
helped has] heard the Meyer1
b. Ich habe schon oft [NP das Gerücht ta] gehört, [dass ihm1
I have already often [NP the rumour ta] heard [that him
die Mafia geholfen hat]a, dem Meyer1.
the mafia helped has]a the Meyer
‘I have already often heard the rumour, that the mafia has assisted
him, this Meyer.’

Complex NPs being movement islands in German might explain the ill-formedness of
(11a) as opposed to (11b). In this paper, I solely assume that RD is syntactically part
of its host sentence, i.e., it belongs to the ‘Nachfeld’ (“afterfield”) in terms of Zifonun,
Hoffmann & Stecker (1997). For the purposes of this paper I assume that RD is a right
IP-adjunct.9 As far as the structural position of the clause-internal pro-form is con-
cerned, RD does not impose any additional placement constraints on it (to the struc-
tural position of unstressed personal pronouns in general see Abraham 2007: 185 ff.)10
AT, on the contrary, can vary its position in its host sentence. Furthermore, AT
does not strictly require morphological agreement between the NP and the clause-
internal pronoun, and it allows various insertions between the host sentence and AT-
NP. Besides, AT is not island-sensitive, cf. (12):
(12) Alles spricht jetzt nur noch über die Freilassung von Meyer1 und Müller2, und es
gibt die wildesten Gerüchte.
‘Everybody is talking about the discharge of Meyer1 and Müller2, and the
strangest rumours go around.’
Ich habe schon oft [NP das Gerücht, dass ihm1 die Mafia
I have already often [NP the rumour that him1 the mafia
geholfen hat] gehört, | ich meine dem Meyer1.
helped has] heard I mean the Meyer1

All in all, AT appears to be syntactically fairly independent of its host sentence. That is
why I propose to analyse AT as an ‘orphan’ in terms of Haegeman (1991), Shaer (2003)
and Shaer & Frey (2004). ‘Orphans’ are linguistic units that are syntactically totally

.  As for rightward adjunction there are no syntactic tests allowing to discern the exact func-
tional layer that serves as the adjunction site, I keep the notation “IP” to refer to the functional
domain above VP. This does not mean that I theoretically disagree with the split IP assumption,
but solely that there is virtually no possibility to account for the adjunction site of the RD-NP
in a more precise way.
.  In this respect German RD is expected to differ from clitic RD like in Catalan, Italian or
Greek, where clitics are subject to structural constraints different from those on full pronouns.
 Maria Averintseva-Klisch

independent from their host sentence and get integrated into it only at the level of the
discourse, as other utterances do. One illustration: according to Shaer (2003), certain
adverbials like with his X-ray vision in (13) are ‘orphans’ in this sense:

(13) With his X-ray vision, John located the files.

(Shaer 2003: 458)

Shaer shows that the PP with his X-ray vision in (13) is syntactically independent. For
instance, it does not show any effects that a syntactic movement would produce (e.g., it
violates island constraint), so that a movement-based analysis of such adverbials fails.
Furthermore, this PP can occur parenthetically at different positions in the clause, as
in (14), which is expected if it does not have any syntactically determined position in
its host sentence:

(14) a. With his X-ray vision, John located the files.

b. John, with his X-ray vision, located the files.
c. John located the files, with his X-ray vision. etc.

Besides, an ‘orphan’ has to build a prosodic unit of its own. As I have argued above,
the same characteristics apply to AT: it does not show any features typical for ele-
ments syntactically dependent on its host sentence. Thus, AT allows morphological
non-agreement, and does not obey the island constraint. Prosodically, it also has to
build a unit of its own. In other words, AT appears to be a syntactic ‘orphan’. In terms
of Zifonun, Hoffmann & Stecker (1997) AT is a construction of the ‘rechtes Außenfeld’
(“right outer field”) that embraces syntactically independent additions to the clause.
The assumption that AT is syntactically non-integrated can directly account for its
appearance after an explicit marking of the sentence boundary, the latter being either
marked graphically with an appropriate punctuation mark, e.g., a full stop, in the case
of written language, cf. (15a), or, in spoken language, by an intervention of another
speaker (so-called ‘alien-initiated repairs’ in terms of Uhmann (1993)), like in (15b):

(15) a. (Der Koch war schon an Bord, der Matrose ebenfalls.)

‘The cook was already on board, the sailor too.’
Er aß die Fliegen. Der Koch, nicht der Matrose.
He ate the flies. The cook not the sailor
(Martel Yann, Schiffbruch mit Tiger: 364)
b. A: Ich weiß nicht, wann sie kommt. B: wer?
A: I know not when shei comes. B: who?
A: na die Anna.
A: interj the Annai. (Attested oral data)

After showing how the formal differences between RD and AT correspond to func-
tional differences, I will turn to the issue of how the discourse integration of AT takes
place.
 To the right of the clause 

2.2  RD vs. AT: Functional differences

My proposal is that RD is used to mark the discourse topic. I understand discourse
topic informally as the discourse referent that is stably activated in the discourse repre-
sentation during the reception of a particular discourse segment,11 or, in other words,
the referent about which the current discourse segment is, cf. (16), where the old shoe-
maker is the discourse topic:12
(16) A broad ray of light fell into the garret, and showed the workman with an
unfinished shoe upon his lap, pausing in his labour. [...] He had put up a hand
between his eyes and the light, and the very bones of it seemed transparent. So he
sat, with a steadfastly vacant gaze, pausing in his work. [...]
(Charles Dickens, The Tale of Two Cities, Ch. 6)

Each particular language has preferred options of referring to discourse topics. Thus,
personal pronouns are generally considered to be the cross-linguistically preferred
anaphorical means of reference to discourse topics (see e.g., Bosch et al. 2003). Also in
(16) the discourse topic referent is constantly referred to with the personal pronoun
he. Besides, the so-called Left Dislocation in English, the construction consisting of an
NP to the left of a clause and a coreferent pro-form inside it as in (17), has been argued
to set the current discourse topic (e.g., Frey 2004):13

.  I understand discourse segment intuitively as a relatively small span of a discourse (mini-
mally one utterance) that is characterized through a fairly tight thematic contiguity. In written
language a discourse segment mostly corresponds to a paragraph.
.  The problem of the status of discourse topic has been extensively and controversely dis-
cussed in literature (see Büring (2003); Asher (2004a) and (2004b); Kehler (2004); Oberlander
(2004); Stede (2004) and Zeevat (2004), to name just a few). However, in spite of theoretical con-
troversies and unclarities concerning the exact definition of the discourse topic, the existence of
some kind of entity that is most salient at a given stage of the discourse and that is relevant for
establishing coherence seems to be uncontroversial. This is the intuition shared by the papers in
the recent issue of Theoretical Linguistics dedicated to discourse topics, although the authors use
different terms for the same intuition of “the thing” that “chunks of text are about” (Asher 2004b:
255). My understanding of the discourse topic as being valid on the local level of a discourse
segment corresponds more or less to the concept of ‘local topic within discourse segments’ in
Kehler (2004).
.  Generally a comparison of the left and right dislocation seems promising, even if it is out
of reasons of space impossible in this paper. In short, both dislocation constructions are topic-
related devices, although there are differences. In general, LD seems cross-linguistically to be
functionally more tightly connected with its host sentence than RD. Due to the linear order of
the discourse, LD is bound to set the topic for the discourse segment including its host sentence,
whereas RD is more important for the segment following its host sentence. In German, a spe-
cial variant of left dislocation, the so-called ‘Hanging topic’ like in (a), has been argued to be a
discourse-topic-marking device, cf. Frey (2004):
 Maria Averintseva-Klisch

(17) Annai, I like heri a lot. Shei [...]

Similarly, RD in German serves to mark the discourse topic, but is has the peculiarity
that the topic is marked not only to be the current one, but also as the topic for the
discourse segment following the RD. In (18), Madame Dutitre is set as “what is being
talked about” for the segment following the host sentence of the RD:
(18) Und als der König seine Frau verloren hatte, bedauerte ihn die Dutitre: “Ach ja,
für Ihnen is et ooch nich so leicht [...].”
‘And when the king lost his wife, Dutitre pitied him: “Dear me, I should say, for
you things aren’t that easy either [...]” ’
Siei war ein Original, die Madame Dutitrei.
shei was an original the Madame Dutitrei
Sie verstand nie, warum man über ihre Aussprüche lachte. Sie war eben echt und
lebte, wie alle wirklich originalen Menschen, aus dem Unbewussten. Kein falscher
Ton kam deshalb bei ihr auf.
‘Shei never understood why everybody always laughed at heri remarks. Shei
was genuine and lived unconsciously, as all unique people do. Shei never came
across as being artificial.’
(Siegfried Fischer-Fabian, Berlin-Evergreen: 125)

Thus, RD in German is a global discourse strategy in the sense that it helps to structure
a whole discourse segment by marking its topic. More specifically, RD can either pro-
mote a discourse-old referent to the discourse topic, or to signal maintenance of the old
discourse topic (especially after a change of the narration perspective).14 Importantly, it
is a forward-looking strategy, as it structures the segment following the host sentence.
RD being situated at the right periphery of the sentence predestines it to influence the
subsequent segment. This means that in the default case the referent of the RD-NP is the
discourse topic for the following segment, as in (19a). Interestingly, RD is also possible at
the very end of a discourse like in (19b). This only seems to contradict my claim: the cru-
cial point is that no other discourse referent is available as topic as long as the discourse
continues, cf. (19c), which is a pragmatically unsuitable continuation in the context of

(a) Peter, I like him a lot.

See also Zybatow & Junghanns (1997) for Russian and Lambrecht (1981) for French counter-
parts to Hanging topic, as well as Averintseva-Klisch (2006) for similarities and differences be-
tween Hanging topic and RD.
.  Thus in (18) the discourse topic is maintained over the change of the perspective intro-
duced by the quotation; as this maintenance cannot be taken for granted, it is explicitly signalled
through the RD.
 To the right of the clause 

the RD (marked with “#”). This means that even in (19b) the referent of the RD-NP
remains the only discourse topic that is hypothetically possible:
(19) a. Siei war ein Original, die Madame Dutitrei.
shei was an original the Madame Dutitrei
Sie verstand nie, warum man über ihre Aussprüche lachte.
‘Shei never understood why everybody laughed at heri remarks.’
b. Ja, siei war ein Original, die Madame Dutitrei.
yes shei was an original the Madame Dutitrei
‘Yes, she was unique, that Madame Dutitre.’ (context: e.g., as the last
sentence of the chapter about Madame Dutitre.)
c. Monsieur und Madame Dutitre waren damals ziemlich berühmt, er für seine
Pferde, sie für ihre Sprüche. Sie war ein Original, die Madame Dutitre. #Er
hatte die besten Pferde Berlins in seinen Ställen.
‘Monsieur1 and Madame2 Dutitre were rather famous at that time, she2 in
particular for her2 bon-mots, and he1 for his1 race horses. She2 was unique,
that Madame D.2. #He1 had the best race horses of Berlin in his stables.’

As for AT, it is a repair strategy used to resolve a potentially unclear reference in the
host sentence, as in (20):

(20) (Sie [Die Mutter] hat den Wohnzimmerschrank aber auch nicht leiden
können,[...], aber mein Vater hat sich auf keine billigen Sachen mehr eingelassen,)
‘Mother hated the wardrobemask, [...], but my father didn’t want to have any
more cheap things around.’
er ist ihr auch zu dunkel gewesen, der Wohnzimmerschrank [...]
hei is for.her also too dark been the wardrobei
(Birgit Vanderbeke, Das Muschelessen)

Contrary to RD, AT is a local strategy, as it is used to repair its host sentence and does
not have any impact on the global discourse structure. In this sense it is backward-
looking, as it refers back to its host sentence and does not influence the following dis-
course segment. In the next section, I will introduce my proposal for the integration of
syntactically independent AT into its host sentence.

2.3  Syntax-discourse asymmetry with AT

Being detached syntactically, an orphan gets attached to its host sentence at the level
of the discourse (cf. Haegeman 1991). This attachment occurs for orphans in a regular
way used for discourse units (cf. Shaer 2003). I assume with Asher & Lascarides (2003)
that discourse units get attached to each other via discourse relations. In Averintseva-
Klisch (forthc.), I have argued that a special discourse relation Afterthought has to be
formulated for the attachment of ATs. What is relevant for the present issue is that
 Maria Averintseva-Klisch

Afterthought is argued to be a subordinating discourse relation in terms of Asher &

Vieu (2005). Asher and Vieu put forward four tests to distinguish between subordinat-
ing and coordinating discourse relations, cf. (21):
(21) Given are two constituents, α and β, a relation R (α, β), and a possible extension
with a constituent γ; the nature of R is to be tested:
1. Attachment Test: If it is possible to attach γ to α, then R is subordinating; if
attachment is possible only to β, then R is coordinating.
2. Continuation Test: if γ continues β in its relation to α, then R is
subordinating; if it is impossible, then R is coordinating.
3. Anaphora Test: if for any γ attached to β no pronominal element in γ can
be bound by referents in α, then R is coordinating; if some can, then R is
subordinating.
4. ‘Prototype’ Test: if R can co-occur with Narration (as prototypical
coordinating relation), then R is coordinating; if R can co-occur with
Elaboration (prototypical subordinating relation), then R is subordinating.
(Summed up from Asher & Vieu 2005)

Test 4 is not applicable to Afterthought, as the semantics of Afterthought is not compat-

ible with Elaboration. In short: firstly, Afterthought differs crucially from Elaboration
in its impact on the truth conditions of the whole sentence: AT first makes the estab-
lishing of the truth conditions for an utterance possible; due to the lack of referential
clarity , it is not possible before the adding of the AT takes place. Secondly, Elaboration
and Afterthought are different kinds of discourse relations. Asher & Lascarides (2003)
distinguish between ‘content-level’ discourse relations and ‘cognitive-level’ discourse
relations. For the former, it is only the content of the utterances building up a discourse
that matters; this is the case with Elaboration. For the latter not only the content of the
utterances, but also the intentions of the speaker and the addressee are important for
defining their semantics. This seems for me to be the case with Afterthought (see also
Averintseva-Klisch (forthc.)). Because of these differences between Elaboration and
Afterthought Test 4 cannot be applied to Afterthought. The applicable tests all yield the
same results: Afterthought is subordinating, cf. (22)-(24):

(22) Attachment Test:

a. Dann ist sie weggelaufen, (α) α
b. (ich meine) die Serena. (β) ↓ Explanation
c. Das macht sie immer wenn sie wütend ist. (γ) γ
‘Then she ran away (α), (I mean) Serena (β). That‘s what she always does
when she is angry (γ).’

According to the Attachment-Test, the relation R (α, β) is bound to be subordinating,

because γ gets attached to α (and not to β) via discourse relation Explanation: the last
constituent explains the first one.
 To the right of the clause 

(23) Continuation Test

a. Dann ist sie weggelaufen, (α) α
  ↓
b. (ich meine) die Serena. (β)   β Afterthought
  ↓
c. Also die Kleine mit blonden Zöpfchen. (γ) γ
‘Then she ran away (α), (I mean) Serena (β). That is the little one with
blonde pigtails (γ).’

Here, it is possible to proceed with a constituent that continues β in its Afterthought

relation to α; thus, R again fulfils the conditions for subordination.

(24) Anaphora Test

, (α) attachment site
a. Dann ist sie weggelaufen
b. (ich meine) die Serena. (β)
c. Das war nicht besonders schlau. (γ)... das ...
c’. Das macht sie immer wenn sie wütend ist. (γ) ... das ...

‘Then she ran away (α)1, (I mean) Serena (β). That1 was not very clever (γ). /
That1’s what she always does when she is angry (γ).’

It is possible to resume the event token (24c) as well as the event type (cf. Asher 1993)
of the constituent α (24c’) with the pronoun das (‘that’) in γ. To sum up: the tests 1–3
all show that Afterthought is a subordinating discourse relation.
According to the analysis presented here, corrections like (25) are a subtype of
afterthoughts:

(25) A: John failed his exams. B: No, he didn’t, he got 60%. A: I meant John Smith.
(Asher & Lascarides 2003:305)

Asher and Lascarides (2003) do not consider in detail cases like (25), but they seem
tacitly to handle them in a similar way as the instances of the discourse relation
Correction, cf. (26):

(26) A: John distributed the copies. B: No, it was Sue who distributed the copies.
(Asher & Lascarides 2003: 470)

However, for Correction the constituents involved are per definition required to be in-
consistent with each other (Asher & Lascarides 2003: 469). This is not the case in (25),
where the constituent With “John” I mean John Smith is in no way inconsistent with the
constituent John failed his exams. Besides, with respect to their function corrections like
(25) are exactly like ‘alien-initiated repairs’ (Uhmann 1993), where the hearer explicitly
signals his inability to resolve the pro-form, like in (27) (cf. also example (15b)):
 Maria Averintseva-Klisch

(27) A: I don’t like her at all. B: Whom? A: Anna I mean.

That is why I propose to subsume corrections like (25) under afterthoughts. Reflect-
ing the character of AT as described above, the discourse relation Afterthought can be
informally stated as in (28):
(28) Afterthought is a subordinating discourse relation, which holds whenever the
speaker of the host sentence and the AT supplies the AT with the intention
of clearing the reference of a discourse referent x that has been introduced in
the host sentence by establishing a relation x=z, where z is a discourse referent
introduced in the AT, and the reference of z in the discourse representation is
assumed to be unambiguous.

A point that I would like to make here is the following: syntax and discourse attachment
do not mirror each other in the case of AT: syntactically, AT is an orphan, i.e., not at-
tached at all. At the level of the discourse structure, however, it gets attached via a subor-
dinating discourse relation. This supports the widely assumed hypothesis that generally
the symmetry of syntactic relations and discourse relations between two discourse units
is not necessarily required (see e.g., Blühdorn in this volume and Holler in this volume),
although in some cases syntactic and discourse relations might go hand in hand.
As for RD, I assume that it does not constitute a separate discourse unit, but is a
part of the unit containing the host sentence. In Averintseva-Klisch (2006) I show how
exactly RD contributes to the semantics of its host sentence.
Summing up this section one might say that the right clausal edge in German
holds two options: it is either used for syntactically integrated RD, or for syntacti-
cally non-integrated AT repairing some intraclausal reference.15 The former is a global
discourse-structuring device, in particular concerning the immediately following dis-
course segment. The latter, on the contrary, is a local (i.e., related to the host sentence)
repair strategy. In the next section I will argue that these are cross-linguistically two
options the right clausal edge might have.

3.  Right periphery in the discourse: an outlook

The issue of interest now is whether the usage of the right clausal edge for backward-look-
ing local repairs and for forward-looking discourse topic marking is a peculiarity of Ger-
man or whether these two options (local repair and global discourse structuring) are used

.  Strictly speaking, RD and AT hold two different positions: while I assume that RD being
part of the sentence is located at its right periphery, AT comes after the sentence boundary;
being syntactically fully independent, it cannot be analysed as a right-peripheral construction.
Zifonun, Hoffmann & Stecker (1997) introduced the term rechtes Außenfeld (‘right outer field’)
for non-integrated phrases as the right edge of the clause (as compared to Nachfeld (‘afterfield’)
for syntactically integrated ones).
 To the right of the clause 

cross-linguistically. The following is to be understood as an outlook for further research. My

hypothesis is that local repair and discourse topic marking are cross-linguistically available
due to the general properties of the right clausal edge. This position is on the one hand the
last possibility to add locally to the clause or to comment on some information within it.
On the other hand, by virtue of its placement it relates the clause to the following discourse.
As the first step to test this hypothesis I will in this paper very briefly compare German to
French and Russian with respect to the use of RD and AT. I will begin with AT.

3.1  Afterthought in Russian and French

AT is expected to be cross-linguistically generally available (primarily as an option of
the spoken language). Not being syntactically attached at all, ATs do not depend on the
syntactic characteristics of a language, and in fact nothing should prevent the possibil-
ity of adding repairs after the actual end of the clause. As expected, ATs are possible in
Russian, cf. (29), as well as in French, cf. (30):
(29) Ego žena revnujet ego k Tereze Lido.
‘His wife is jealous of Teresa Lido.’
Ona puskajetsja za nim v pogonju, | ego žena.
Shei starts for him in pursuit his wifei.
‘Shei pursues him, his wifei’ (Attested oral data)
Here, the speaker retells the contents of a movie, and the sentence preceding the clause
with the pronominal reference ona (‘she’) introduces two female referents, “his wife”
and “Teresa Lido”. The speaker uses a personal pronoun to refer to the wife, but then
believes it might be unclear to the hearer whom she actually means, and she adds an ex-
plicit reference resolution device. Similarly in (30), where the reference of the pronoun
il (‘he’) is unclear in the context, it is resolved explicitly with an afterthought NP:
(30) (context: Jean and Jacques are eating.)
Il a déjà mangé la soupe, | Jean.
he has already eaten the soup Jean
As in German, the AT-NPs in Russian are those constructions involving NPs to the
right of the clause that have been investigated most often.16 Now, I will turn to the less
clear and more interesting issue of RD proper in Russian and French.

3.2  Right dislocation in Russian and French

RDs in Russian have to my knowledge barely been considered in the literature. The
only analysis I am aware of distinguishing between a repair and another kind of

.  Lapteva (1976) assumes that adding a NP having a coreferent pro-form inside the clause
after a syntactically complete clause has a function of “explanation”. Cf. also Zemskaja (1973)
and Švedova et al. (1982) who are talking about the function of “specification of the pronoun”
for the added NP.
 Maria Averintseva-Klisch

right-peripheral NP is given in Lapteva (1976). She assumes that besides AT there is

another kind of added NPs with a coreferent pro-form, which seems “not to have any
function at all” (Lapteva 1976: 267). In (31), however, one could hardly assume that
the right-peripheral NP does not have any function; it is clearly used to enhance the
intended meaning of the passage in that the wall is marked as the discourse topic:

(31) Meier Wolf vsju žizn’ kopil den’gi, čtoby uvidet’ Stenu Plača.
‘MW saved money his whole life long to see the Wailing Wall’
On videl ejë teper’, ėty stenu.
he saw heri now this wall.femi
Ona naxoditsja za linijej železnoj dorogi [...]. Ėto prostaja požarnaja stena [...]
‘Now he has seen iti, this walli. Iti is situated behind the railways [...]. Iti is a
simple fire protection wall [...]’
(Aleksandr Galič, Matrosskaja Tišina)

Characteristically, the reference of the pronoun in RD is often quite clear in the con-
text, as in (31) and (32). That is why I argue that both are cases of RD proper.

(32) A: A Petrova ušla čto-li?

A: and Petrovai is.gone interr.pron
B: Net, ona po-mojemu ne byla Petrova
B: no shei I.believe not was Petrovai
‘A: Is Petrova already gone? B: No, I believe she has not been here at all, Petrova.’
(Lapteva 1976: 267)

In (32) the referent of Petrova is clearly the discourse topic, so that (31) and (32) are
similar and in both the right-peripheral NP is an RD in the sense of a right-peripheral
discourse topic marking construction.17
RD is also attested for French (Lambrecht 1981, 1987), e.g., (33):18

(33) a. Il a mangé la soupe, Jean.

hei has eaten the soup Jeani
b. Il est beau, ce tableau!
hei is beautiful, this picture.maski (Lambrecht 1981: 80)

Formally, RD is prosodically and syntactically integrated in Russian as well as in French,

(see Lapteva (1976) resp. Lambrecht (1981)). For example, morphological agreement
is required for Russian RD (Zemskaja 1973), cf. (34) as well as for French RD (35):

.  In (32) the discourse topic is similarly to (18) marked as maintained over the change of a
perspective (here the change of the speaker).
.  The French data presented in section 3 are taken from Lambrecht (1981), (1987) and Ashby
(1988); the glosses are mine.
 To the right of the clause 

(34) Mne eje podarili nedavno, ėtu knigu /

To-me her.acc presented.3.ps.plur recently this.acc book.acc /
*ėta kniga.
*this.nom book.nom
‘I recently got it as a present, this book.’ (Zemskaja 1973: 166)

(35) Il-faut y aller quand il-fait chaud à la plage / *la

one.should therei go when it.is warm to the beachi / *the
plage.
beach.nomi (Lambrecht 1981: 78)

In (34), the default nominative case is not available for the RD-NP if the intraclausal
pro-form is in the accusative. In (35) RD-phrase has to preserve the morphological
marking which an argument phrase would have in its canonical position (Lambrecht
1981: 79).19 This suggests that RD in Russian and French is syntactically integrated in
the same way as RD in German.
As for the function of RD, I argue for Russian as well as for French that in these
languages RD serves in a way similar to German RD to mark the discourse topic.
Observations made in the previous research on French RD are compatible with my
analysis of RD as a discourse topic marking device. Lambrecht argues that RD is used
to refer to the “previously established topic” (Lambrecht 1987: 237). Ashby (1988)
claims that at least sometimes the referent of the RD “continue[s] to be talked about in
succeeding sentences” (Ashby 1988: 216),20 which would correspond to its being the
discourse topic for the following segment, cf. (36):

(36) Et puis ils étaient méchants, les Allemands. A la fin, quand

and then theyi were angry the Germansi at the end when
ils ont vu que la situation était perdue [...]
theyi have seen that the situation was lost [...]
(Ashby 1988: 214)

Here the speaker tells about the end of World War II in France and introduces the
new discourse topic, the Germans, with the help of the RD. Generally, Ashby (1988)
differentiates between 6 pragmatic functions of French RD: turn closing, filler,

.  In (35) the dislocated constituent is a PP and not an NP. However, Averintseva-Klisch &
Salfner (2007) argue that PPs like in (35) that are prepositional objects behave in a RD exactly
like NPs as far as their function is concerned. Here it is only due to the subcategorization frame
of the verb aller that requires a PP, that a PP instead of an NP is dislocated.

.  It is slightly surprising, that Ashby does not explicitly state referring to the discourse topic
as a function of RD, although he observes that 93% of RDs in his corpus refer to discourse topics
(Ashby 1988: 216, Table 4).
 Maria Averintseva-Klisch

clarification, topic shift, contrast and epithet. However, these, with possible exception
of clarification, appear to me to be secondary functions accompanying the discourse
topic marking. Thus, turn closing, which Ashby claims to be the most important func-
tion of RD, seems to me to be a special case of the discourse topic marking: the dis-
course topic gets “passed on” to the next speaker, so to speak: the first speaker wants to
get sure that his communication partner continues to talk about the same entity as he
did. So, in (37) “his son” is the discourse topic for both speakers:
(37) A: Il est en dernière année, son fils.
A: he is in last year his son.
B: Ah mais, oui, c’est en dernière année.
B: Ah but yes it.is in last year.
(Ashby 1988: 222)

As for the cases analysed as clarification in Ashby, these cases I expect to be ATs. This
is however, an issue I cannot decide upon, as Ashby intro­duces only one example of
clarification, and it does not have enough context for me to be able to decide on this
issue.
In other words, an informal cross-linguistic comparison supports the claim that
RD marks the discourse topic. An important additional evidence for this claim coming
from French are the pronominal RDs like in (38a), cf. Ashby (1988); these also occur
in Norwegian (Fretheim 1995, 2001), cf. (38b):21
(38) a. Madame X, elle est née ici, elle.
Mrs. Xi shei is born here shei
(Ashby 1988: 204)
b. Scott heter Glenn til etternavn, han.
Scotti is.named Glenn as surname hei
(Fretheim 2001: 62)

Pronominal RDs seem to me to be the clearest case of a discourse topic marking func-
tion of the RD. Firstly, adding a pronominal reference definitely cannot be a refer-
ence clarification. Secondly, personal pronouns are traditionally assumed to be the
preferred means for referring to discourse topics (e.g., Zifonun, Hoffmann & Stecker
(1997), Bosch et al. (2003), Consten & Schwarz-Friesel (forthc.)), so it is to expect that
also right dislocated personal pronouns refer to discourse topics.
This cross-linguistic affinity of the discourse topics to the right periphery might
seem unexpected, as especially in the conversation analysis the right clausal edge is usu-
ally seen as the position used for turn taking or for delaying of turn taking (cf. e.g., Auer
(1991)). In accordance with this analysis information coming at the right periphery is not
important, so that its potential loss through a turn-taking is not problematic. However,

.  Note that my analysis with respect to the function of RD differs from the actual proposals
in Ashby (1988) and Fretheim (2001).
 To the right of the clause 

it seems to me that the repetition of information that is already given and is thus seem-
ingly “unnecessary” must be something more than just a strategy to retain turn-taking.
Rather with this repetition the importance of the corresponding referent for the dis-
course is made clear, i.e., the referent is explicitly set as the current discourse topic.

3.3  Discussion: RD vs. AT cross-linguistically

A comparison of German, Russian and French with respect to their use of RD and AT
shows that these languages have AT and RD as the options of the use of the right clausal
edge. These two uses of the right clausal edge are due to its twofold character. On the one
hand, for reasons of linearity the right periphery constitutes a point of intersection with
the following discourse. As such it is predestined to host NPs referring to the discourse
topics that pertain in the following discourse segment. On the other hand, the position
immediately after a clause is the last possibility to add something locally to this clause;
that is why afterthought NPs are possible. Moreover, in all three languages observed RD
is syntactically integrated. A similar observation was made in Lambrecht (2001: 1068),
who stated that RD is cross-linguistically “more tightly connected with the predicate-
argument structure of the clause” than left-dislocated elements. In the languages un-
der discussion AT is syntactically non-integrated. That is to be expected, as a speaker
adds AT as a repair device when he has already completed the clause and only after
that notices that the clause might be unclear. The question that remains is: Why is RD
cross-­linguistically prosodically and syntactically integrated? Why a forward-looking
discourse topic marking prefers a realization through a syntactically integrated NP at
the right periphery is a challenging question for further research. It would be especially
interesting to compare in a systematic way RD in languages featuring clitic RD (like e.g.,
Catalan or Greek) with RD in languages having only full pronouns like German. It seems
that clitic RD differs from non-clitic RD in that multiple dislocation is non-restrictedly
allowed (Vallduvì 1992: 85 for Catalan). As for Left dislocation, Grohmann (2003) pro-
poses different syntactic analyses for non-clitic and clitic left dislocation. For Romance lan-
guages it is often assumed that RD is either derived from left dislocation (e.g., Frascarelli &
Hinterhölzl (2007) for Italian) or that both constructions are symmetrical (Beninca &
Poletto (2003) for Italian and Vallduvì (1992) for Catalan), which would imply also dif-
ferent analyses for clitic and non-clitic RD. If the syntax of the clitic and non-clitic RD
also differs, then one would expect functional differences, too. Still, it seems that also clitic
RD is a topic-marking construction, cf. Villalba (2000: 20), Lambrecht (2001: 1072), so
that the generalization made above, viz. that the right periphery is a position designated
for hosting expressions referring to discourse topics, still holds.

4.  Summing up and conclusions

In my paper, I first introduced two prima facie similar, but formally and functionally
different constructions of German, right dislocation and afterthought. I argued that
 Maria Averintseva-Klisch

whilst AT is a local reference repair strategy, RD is an important discourse-structuring

device used to mark the discourse topic for the segment following RD. At the level
of syntax, RD is part of its host sentence, presumably a right IP-adjunct. AT, on the
contrary, is syntactically independent from its host sentence, and gets integrated into
it only at the level of the discourse via a special discourse relation. Thus, AT might be
seen as an illustration of the general asymmetry of syntactic and discourse relations:
AT is added via a subordinating discourse relation, although there is no subordination
on the level of syntax, AT being syntactically not attached at all.
I assume that local repair and global discourse-structuring devices are cross-
linguistically two options of the use of the right clausal edge, and put forward the
question whether AT is always syntactically detached, whereas RD is always syn-
tactically integrated. To follow up this point, I compared German to Russian and
French with respect to the use and characteristics of RD and AT.
I found that all three languages feature AT as an ‘orphan’ and RD as a syntactically
integrated construction used to mark the discourse topic for the following segment.
The former is due to the character of AT as a local repair.
As for RD, this discourse-topic marking device at the right clausal periphery
seems to be available cross-linguistically in a similar way. Discourse topic, contrary
to sentence topic, is not a sentence-bound notion and thus also not a syntactic, but
a pragmatic category. Accordingly, there cannot be a prototypical position for dis-
course topic inside the clause. However, it seems that right periphery, being a syntacti-
cally unnecessary and thus entirely pragmatically ruled position, is cross-linguistically
preferred for the placement of NPs referring to discourse topics. This leads to the hy-
pothesis that there is a cross-linguistic interdependence of the discourse function and
the syntax. I assume that RD is cross-linguistically part of its host sentence, situated at
the right periphery. Respectively, AT and other kinds of repair constructions are cross-
linguistically ‘orphans’, i.e., syntactically independent units generated after a sentence
is completed. Testing of this assumption for more languages with different syntactic
character would be a challenging task for further research.

References

Abraham, Werner. 2007. Topic, focus and default vs. contrastive accent. In On Information
Structure, Meaning and Form: Generalizations across Languages, Kerstin Schwabe &
Susanne Winkler (eds), 183–203. Amsterdam: John Benjamins.
Altmann, Hans. 1981. Formen der “Herausstellung” im Deutschen: Rechtsversetzung, Linksverset-
zung, Freies Thema und verwandte Konstruktionen [Linguistische Arbeiten 106]. Tübingen:
Niemeyer.
Ashby, William J. 1988. The syntax, pragmatics and sociolinguistics of left- and right-dislocations
in French. Lingua 75: 203–229.
Asher, Nicholas. 1993. Reference to Abstract Objects in Discourse. Dordrecht: Kluwer.
 To the right of the clause 

Asher, Nicholas. 2004a. Discourse topic. Theoretical Linguistics 30: 163–201.

Asher, Nicholas. 2004b. Troubles with topics: Comments on Kehler, Oberlander, Stede and
­Zeevat. Theoretical Linguistics 30: 255–262.
Asher, Nicholas & Lascarides, Alex. 2003. Logics of Conversation. Cambridge: CUP.
Asher, Nicholas & Vieu, Laure. 2005. Subordinating and coordinating discourse relations.
­Lingua 115: 591–610.
Auer, Peter. 1991. Vom Ende deutscher Sätze. Zeitschrift für Germanistische Linguistik 19:
139-157.
Averintseva-Klisch, Maria. 2006. ‘Separate performative’ account of German right dislocation.
In Proceedings of the Sinn und Bedeutung 10 [ZAS Papers in Linguistics 44]. Christian
Ebert & Cornelia Endriss (eds), 15–28. URL: http://www.zas.gwz-berlin.de/papers/zaspil/
articles/zp44/Averintseva.pdf.
Averintseva-Klisch, Maria. Forthcoming. German right dislocation and afterthought in discourse.
In Constraints in Discourse, Anton Benz & Peter Kühnlein (eds). 213–235. Amsterdam: John
Benjamins.
Averintseva-Klisch, Maria & Salfner, Fabienne. 2007. German PP-dislocations to the left and to
the right: against a symmetric approach. In Proceedings of Sinn und Bedeutung 11, Estela
Puig-Waldmüller (ed.), 46–62. Barcelona: Universitat Pompeu Fabra. URL: http://mutis.
upf.es/glif/pub/sub11.
Beninca, Paola & Poletto, Cecilia. 2003. Topic, focus and V2: Defining the CP sublayers. In The
Structure of CP and IP, Luigi Rizzi (ed), 52–75. Oxford: OUP.
Bosch, Peter, et al. 2003. Demonstrative pronouns and personal pronouns. German der vs. er,
Proceedings of the EACL2003, Workshop on The Computational Treatment of Anaphora.
Budapest.
Büring, Daniel. 2003. On D-trees, beans, and B-accents. Linguistics and Philosophy 26(5): 511–545.
Consten, Manfred. 2004. Anaphorisch oder deiktisch? Zu einem integrativen Modell domänenge-
bundener Referenz. Tübingen: Niemeyer.
Consten, Manfred & Schwarz-Friesel, Monika. 2007. Anapher. In Wortarten des Deutschen,
Ludger Hoffmann (ed.). 265–292. Berlin: de Gruyter.
Culicover, Peter & Jackendoff, Ray. 2005. Simpler Syntax. Oxford: OUP.
Frascarelli, Mara & Hinterhölzl, Roland. 2007. Types of topic in German and Italian. In On
­Information Structure, Meaning and Form: Generalizations across Languages, Kerstin
Schwabe & Susanne Winkler (eds), 87–116. Amsterdam: John Benjamins.
Fretheim, Thorstein. 1995. Why Norwegian right-dislocated phrases are not afterthoughts. Nor-
dic Journal of Linguistics 18: 31–54.
Fretheim, Thorstein. 2001. The interaction of right-dislocated pronominals and intonational
phrasing in Norwegian. In Nordic Prosody, Wim A. van Dommelen & Thorstein Fretheim
(eds), 61–75. Frankfurt: Peter Lang.
Frey, Werner. 2004. Notes on the syntax and pragmatics of the German left dislocation. In The
Syntax and Semantics of the Left Periphery, Horst Lohnstein & Susanne Trissler (eds),
­203–233. Berlin: de Gruyter.
Grohmann, Kleanthes K. 2003. Prolific Domains. On the Anti-Locality of Movement Dependencies.
Amsterdam: John Benjamins.
Haegeman, Liliane. 1991. Parenthetical adverbials: The radical orphanage approach. In Aspects
of modern English linguistics: Papers presented to Masatomo Ukaji on his 60th birthday, S.
Chiba et al. (eds), 232–254. Tokyo: Kaitakusha.
 Maria Averintseva-Klisch

Kayne, Richard. 1994. The Antisymmetry of Syntax. Cambridge MA: The MIT Press.
Kehler, Andrew. 2004. Discourse topics, sentence topics, and coherence. Theoretical Linguistics
30: 227–240.
Lambrecht, Knud. 1981. Topic, Antitopic and Verb Agreement in Non-Standard French [Pragmat-
ics and Beyond II-6]. Amsterdam: John Benjamins.
Lambrecht, Knud. 1987. On the status of SVO sentences in French discourse. In Coherence and
Grounding in Discourse, Russel S. Tomlin (ed.), 217–262. Amsterdam: John Benjamins.
Lambrecht, Knud. 2001. Dislocation. In Language Typology and Language Universals / 
Sprachtypologie und sprachliche Universalien. An International Handbook, Vol. 2, Martin
Haspelmath et al. (eds), 1050–1078. Berlin: de Gruyter.
Lapteva, Ol’ga A. 1976. Russkij razgovornyj sintaksis. Moskva: Nauka.
Oberlander, Jon. 2004. On the reduction of discourse topic. Theoretical Linguistics 30: 213–225.
Ross, John R. 1986. Infinite Syntax! Norwood: Ablex. [= Ross, John R. 1967. Constraints on
Variables in Syntax. PhD dissertation, MIT].
Schindler, Wolfgang. 1995. Zur Topologie von Herausstellungen und Zusätzen am rechten
­Satzrand. Sprache & Sprachen 16: 44–56.
Selkirk, Elisabeth. 1978. On prosodic structure and its relation to syntactic structure. In Nordic
prosody II, Thorstein Fretheim (ed.), 111–140. Trondheim 1981: TAPIR.
Selting, Margret. 1994. Konstruktionen am Satzrand als interaktive Ressource in natürlichen
Gesprächen. In Was determiniert Wortstellungsvariation?, Brigitta Haftka (ed.), 299-318.
Opladen: Westdeutscher Verlag.
Shaer, Benjamin. 2003. An ‘orphan’ analysis of long and short adjunct movement in English. In
WCCFL 22 Proceedings, G. Garding, G. & M. Tsujimira (eds), 450–463. Sommerville MA:
Cascadilla Press.
Shaer, Benjamin & Frey, Werner. 2004. ‘Integrated’ and ‘non-integrated’ left-peripheral elements
in German and English. ZAS Papers in Linguistics 35(2): 465–502.
Smith, Michael. 2002. The polysemy of German es, iconicity, and the notion of conceptual dis-
tance. Cognitive Linguistics 13(1): 67–112.
Stede, Manfred. 2004. Does discourse processing need discourse topics? Theoretical Linguistics
30: 241–253.
Švedova, Natalija et al. (eds). 1982. Russkaja Grammatika. Moskva: Nauka.
Uhmann, Susanne. 1993. Das Mittelfeld im Gespräch. In Wortstellung und Informationsstruktur,
Marga Reis (ed.), 313–354. Tübingen: Niemeyer.
Uhmann, Susanne. 1997. Grammatische Regeln und konversationelle Strategien. Fallstudien aus
Syntax und Phonologie. Tübingen: Niemeyer.
Vallduvi, Enric. 1992. The Informational Component. New York NY: Garland.
Villalba, Xavier. 1999. Symmetry and antisymmetry in syntax. Syntaxis 2: 1–25.
Ward, Gregory & Birner, Betty J. 1996. On the discourse function of rightward movement in
English. In Conceptual Structure, Discourse and Language, Adele Goldberg (ed.), 463–479.
Stanford CA: CSLI.
Zeevat, Henk. 2004. Asher on discourse topic. Theoretical Linguistics 30: 203–211.
Zemskaja, Elena A. (ed.). 1973. Russkaja razgovornaja reč. Moskva: Nauka.
Zifonun, Gisela, Hoffmann, Ludger & Stecker, Bruno. (eds). 1997. Grammatik der deutschen
Sprache. Vol. 1 & 2. Berlin: de Gruyter.
 To the right of the clause 

Ziv, Yael. 1994. Left and right dislocations. Discourse functions and anaphora. Journal of Prag-
matics 22: 1–17.
Zybatow, Gerhild & Junghanns, Uwe. 1998. Topiks im Russischen [Sprache und Pragmatik 47].
Lund: University of Lund.
 Exploring the role of clause subordination in
discourse structure
The Case of French avant que

Laurence Delort
Université Paris 7

The goal of this paper is to explore the role of clause subordination in discourse
structure. Through the study of the French subordinating conjunction avant que
(‘before’) and its interaction with discourse context, I will attempt to show that
clause subordination can affect temporal structure and also discourse structure,
by conveying either subordination or coordination between discourse units.

Keywords: clause subordination, subordinating and coordinating discourse

relations, discourse structure, avant que

1.  Introduction

This paper deals with the interaction between the French subordinating conjunction
avant que (‘before’) and discourse context, and with the effect of this interaction on
both temporal structure and discourse structure.
In a discourse C1 avant que C2,1 the subordinate clause has an adverbial function:
it temporally anchors the eventuality conveyed in the main clause, (Kamp & Reyle
1993). This function allows the subordinate clause to be replaced by another temporal
adverbial, as illustrated in (1) and (2).
(1) Paul a trouvé la solution avant que Marie la lui donne.2
‘Paul found the solution before Marie gave it to him.’
(2) Paul a trouvé la solution {avant la tombée de la nuit / avant 20h00}.
‘Paul found the solution {before nightfall / before 8.00 pm}.’

.  Or C1 avant de V2, if the subject is identical in both clauses. Therefore, this paper is
c­ oncerned with one-sentence discourses rather than discourses made up of sentence sequences.
Notation: Ci is a clause, and Vi is an infinitive verb, which denotes an eventuality (also called
situation – state or event, (Asher 1993)) noted ei.
.  The fact that the subordinate clause has a factual or a counterfactual interpretation is not
my point here. Whatever the interpretation of the subordinate clause, this discourse has a cir-
cumstantial interpretation.
 Laurence Delort

Interestingly, this adverbial function is lost in some discourse contexts. Indeed, in

­discourses (3a) and (3b), the subordinate clause cannot be replaced by a temporal
adverbial, cf. (4a) and (4b) respectively.
(3) a. Paul a d’abord cherché la solution avant que Marie la lui donne.
‘Paul first sought the solution before Marie gave it to him.’
b. Paul a longtemps cherché la solution avant que Marie la lui donne.
‘Paul sought the solution for a long time before Marie gave it to him.’
(4) a. #Paul a d’abord cherché la solution avant la tombée de la nuit.
‘Paul first sought the solution before nightfall.’
b. #Paul a longtemps cherché la solution avant la tombée de la nuit.
‘Paul sought the solution for a long time before nightfall.’

From this observation, several questions arise: how can a temporal subordinate clause
lose its adverbial function? What does the contrast between discourses such as (1) and
(3) hide? Does the loss of adverbial function have an effect on temporal structure and
discourse structure? I will try to provide some answers in this paper.
The paper is organized as follows. First, I propose an analysis of the interaction
between the conjunction avant que and discourse context which gives rise to non-
adverbial uses of the subordinate clause, as exemplified by discourses (3a–b). Then, I
present some effects of this interaction on temporal structure (in particular, avant que
conveys a strongly constrained temporal relation, rather than a simple one), and on
discourse structure (in particular, avant que conveys coordination, rather than subor-
dination, between discourse units).

2.  Interaction between avant que and discourse context

Discourse (1) exhibits a ‘circumstance’ interpretation, that is, the subordinate clause
­introduced by avant que has a temporal adverbial function. As observed in the
­discourses in (3), the same subordinate clause can lose this adverbial function. Avant
que does not introduce an eventuality providing a temporal anchor for the eventuality
conveyed in the main clause. If a ‘circumstance’ interpretation is impossible or not
plausible, what does a discourse C1 avant que C2 express? My hypothesis is that
­interaction between the semantics of avant que and discourse context gives rise to two
interpretations:

– ‘continuation’: e1 and e2 are in a narrative sequence, e2 being the continuation of e1;

– ‘pre-condition’: e1 is the (necessary) condition of the realization of e2, i.e., realiza-
tion of e2 depends on e1.

In these two interpretations, as well as in a ‘circumstance’ interpretation, the tempo-

ral relation conveyed by avant que is the same: a temporal precedence between the
 Clause subordination in discourse structure 

­eventualities, i.e., e1 < e2. The central idea behind this work is that it is discourse con-
text, and sometimes also extra-linguistic knowledge, that allows avant que to appear in
­non-circumstantial discourses (and trigger coordination in discourse, see section 3). To
argue for this idea, the next two subsections investigate what kind of discourse context
supports ‘continuation’ (section 2.1) and ‘pre-condition’ (section 2.2).

2.1  Continuation
‘Continuation’ can be inferred thanks to linguistic cues capturing either a topic
­elaboration or a preparatory event.

2.1.1  Topic elaboration

There are some linguistic cues, such as verbs and adverbs, which unambiguously ­express
a continuation between two eventualities. These cues can be found in ­combination
with avant que: for instance, the adverb d’abord (‘first’) in (5), or the verb commencer
par (‘to start off by’) in (6).
(5) Permettez-moi d’abord de présenter quelques réflexions générales avant d’entrer
dans les détails. (A. de Tocqueville – Correspondance avec H. Reeve)3
‘Let me first present some general comments before going into the details.’
(6) Ils commencèrent par se dévisager, les uns les autres, avant de se parler.
(P. Loti – Mon frère Yves)
‘They started off by staring at one another, before speaking.’

When made explicit by such lexical items, ‘continuation’ is constrained by a com-

mon theme, called topic, shared by the eventualities, (cf. Danlos, 2005). That is, avant
que conveys ‘continuation’ only if the eventualities linked share a topic (supported by
­identical subjects). For instance, in (7), there is no thematic link between e1 ‘to try to
bring me out’ and e2 ‘to blow up’. In fact, e1 is continued by e3, introduced by ensuite
(‘then’). Avant que in (7) does not convey ‘continuation’, but ‘circumstance’.
(7) Eh bien, elle a d’abord essayé de m’entraîner hors du club avant que cela n’éclate.
Ensuite, elle a fait preuve d’un sang-froid […] que bien peu de femmes possèdent
en pareille occasion. (M. Droit – Le retour)
‘She first tried to bring me out of the club before it blew up. Then, she kept her
head, such as few women do in such a situation.’

The cues in (5) and (6) are cataphoric, because they call for a succeeding eventuality.
Sometimes, avant que is used in combination with anaphoric cues, calling for a preceding

.  Most of examples are taken from the French literature database Frantext (http://www.
frantext.fr/). If there is no reference to author, the discourses are constructed. All examples in
French are followed by their translation in English.
 Laurence Delort

eventuality, as illustrated in (8) and (9), with the verbs ajouter la suite (‘to add’) and the
verb conclure (‘to conclude’), respectively.
(8) Il répondit en reprenant une phrase de Virginie : “Faut pas se plaindre...” avant
d’ajouter la suite : “…y’a plus malheureux que nous !”. (R. Sabatier – David et
Olivier)
‘He answered by borrowing a sentence from Virginie: “ There is no need to
complain…” before adding: “There are people more unfortunate than us!”.’
(9) Pour ma part je me demandai surtout si je pouvais arrêter ma carrière – avant de
conclure que non. (M. Houellebecq – La possibilité d’une île)
‘I was wondering if I could stop my career – before concluding that I couldn’t.’

All these lexical cues for ‘continuation’ reflect the fact that e1 cannot occur after e2. The
eventualities occur in a certain temporal order, with respect to a topic, on which they
elaborate. So, these discourses C1 avant que C2 have nothing to do with a circumstan-
tial interpretation.
Lexical cues are necessary with avant que when a topic elaboration with ‘continu-
ation’ is used. If there are cues, e1 can be either instantaneous or durative, cf. (10a) and
(10b) respectively, and there is no ambiguity.
(10) a. Paul a d’abord donné la solution à Marie avant de la donner à Lisa.
‘Paul first gave the solution to Marie before giving it to Lisa.’
b. Paul a d’abord cherché la solution avant que Marie la lui donne.
‘Paul first sought the solution before Marie gave it to him.’

If there are no cues, when e1 is instantaneous, there is an ambiguity between ­‘continuation’

and ‘circumstance’, cf. (11a). When e1 is durative, there is no ambiguity, cf. (11b).
(11) a. Paul a donné la solution à Marie avant de la donner à Lisa.
‘Paul gave the solution to Marie before giving it to Lisa.’
b. Paul a cherché la solution avant que Marie la lui donne.
‘Paul sought the solution before Marie gave it to him.’

Discourse (11b) reflects another discourse configuration where avant que conveys
‘continuation’.

2.1.2  Preparatory Event

Some discourses display no thematic link between eventualities, i.e., there is no topic
elaboration. Nevertheless, they express ‘continuation’ with avant que in combination with
lexical semantics of verbs. In particular, there is ‘continuation’ when e1 is a ­durative
situation, as in (12) and (13).

(12) Il erra longtemps avant de s’asseoir sur un banc de pierre. (R. Sabatier – Les
noisettes sauvages)
‘He wandered a long time before sitting down on a stone bench.’
 Clause subordination in discourse structure 

(13) Quand je suis entré, elle a attendu quelques secondes avant de lever les yeux vers
moi. Je l’avais encore jamais trouvée aussi belle. (P. Djian – 37.2 Le matin)
‘When I entered, she waited a few seconds before looking at me. I had never
found her so beautiful.’

One can observe that none of the subordinate clauses can be replaced by a temporal
adverbial, cf. (14) and (15) respectively.

(14) #Il erra longtemps avant 20h00.

‘He wandered a long time before 8.00 pm.’
(15) #Elle a attendu quelques secondes avant 20h00.
‘She waited a few seconds before 8.00 pm.’

The eventuality expressed by the main clause can be interpreted as a preparatory phase
for e2: there is no thematic link between e1 and e2, but e1 leads to e2 (not necessarily
naturally). This preparatory phase is even clearer when one observes discourses such
as (16) where e1 is an interval of time between two events e0 ‘the noise echoes through
the whole house’ and e2 ‘he walks down’.

(16) La porte claque très fort et le bruit résonne dans toute la maison. Quelques
minutes s’écoulent avant qu’il redescende. (B. Clavel – La maison des autres)
‘The door bangs very loudly and the noise echoes through the whole house.
A few minutes go by before he walks down.’

The situation e1 can be iterative, as illustrated by (17).

(17) La porte battit derrière lui, oscilla plusieurs fois avant de s’immobiliser.
(M. Genevoix – Eva Charleboix)
‘The door banged behind him, hovered several times before coming to a
standstill.’

In this case, the subordinate clause can be replaced by a temporal adverbial in (18), but
it does not imply that (17) and (18) share the same meaning. In (17), e1 leads to e2: ‘to
hover several times’ leads to ‘to come to a standstill’. While in (18), e1 does not lead to
e2, e1 is only temporally anchored by the adverbial phrase.

(18) La porte oscilla plusieurs fois avant 20h00.

‘The door hovered several times before 8.00 pm.’

In discourses (12), (13), (16) and (17), lexical semantics of verbs (conveying that e1 is
a durative/iterative situation leading to e2) explicitly represents the semantic relation
‘continuation’ between the eventualities.
The next interpretation, ‘pre-condition’, arises from this interpretation: e1
is a preparatory (durative or iterative) event that is realized with the purpose of
­realizing e2.
 Laurence Delort

2.2  Pre-condition
‘Pre-condition’ means that e1 is the necessary condition of the realization of e2. It can
be either explicit, through linguistic clues, or inferred, through the semantics of even-
tualities (in particular e1). Again, avant que interacts with discourse context to trigger
a non-circumstantial interpretation.

2.2.1  Explicitness of ‘pre-condition’

‘Pre-condition’ can be expressed via lexical cues such as modal verbs, in addition to
expression of durative or iterative situations. These cues can be found either in C1, as
e.g. devoir (‘must’, ‘to have to’) in (19) and (20), or in C2, as e.g. pouvoir (‘to be able to’)
in (21) and (22).
(19) Arlequin dut tambouriner longtemps avant que la porte s’ouvrît. (M. Tournier –
Le medianoche amoureux)
‘Arlequin had to drum a long time before the door opened.’
(20) L’air siffla entre les lèvres de Bensoussan qui dut s’y reprendre à deux fois avant
d’évacuer un nom : - Toni. (A. Page – Tchao Pantin)
‘Bensoussan had to try twice before saying a name: - Toni.’
(21) J’ai attendu encore une ou deux secondes avant de pouvoir débloquer mes
mâchoires. (P. Djian – 37.2 Le matin)
‘I waited one or two seconds longer before I could unlock my jaw.’
(22) J’en étais tellement déconcertée et froissée qu’il se passa une minute avant que je
puisse réagir. (P. Labro – Des bateaux dans la nuit)
‘I was so stumped and piqued that an entire minute went by before I could react.’

In all these discourses, e1 must be realized in order to make e2 happen. Modal verbs
trigger this interpretation unambiguously. For instance, if a modal verb is inserted in
the ‘continuation’ discourse (16), cf. (23a), e1 is interpreted as a necessary ­condition for
his walking down, cf. (23b). See also the parallel between (13) and (21), and ­between
(16) and (22).
(23) a. Quelques minutes s’écoulent avant qu’il redescende.
‘A few minutes go by before he walks down.’
b. Quelques minutes s’écoulent avant qu’il puisse redescendre.
‘A few minutes go by before he can walk down.’

Some contexts do not display lexically specified modalities for conveying ‘pre-condition’.
The semantics of eventualities and extra-linguistic knowledge represent clues for the
inference of ‘pre-condition’.

2.2.2  Inference of ‘pre-condition’

Sometimes, modalities are not expressed but e1 is such that it is easily understood
as the necessary condition of the realization of an event, which is e2. Corpora show
 Clause subordination in discourse structure 

numerous ‘continuation’ discourses (where e1 is a preparatory event) parallel to ‘pre-

condition’ discourses. For instance, discourses (12) and (24) (see (19) with modality)
both display a durative situation. However, in (12), e1 leads (with no intent) to e2,
while in (24), e1 is realized in order to realize e2. The same observation can be made
on discourses (17) and (25) (see (20) with modality). Both display an iterative situation
but in (17), the repeated hovering is not realized in order to come to a standstill (it just
leads to it), while in (25), he tried twice in order to speak. Finally, discourses (13) and
(26) (see (21) with modality) show the same distinction between ‘continuation’, on the
one hand, when there is no intention, and ‘pre-condition’, on the other hand, when
there is intention.
(24) Lucie tambourina longtemps à la porte de sa chambre avant qu’il se décidât à
ouvrir. (J. Rouaud – Les champs d’honneur)
‘Lucie drummed a long time at the door before he decided to open.’
(25) Il s’y reprit à deux fois avant de parler, mais enfin il dit : - Pourquoi? (L. Guilloux –
Le pain des rêves)
‘He tried twice before speaking, but at last he said: - Why?’
(26) J’ai attendu que la bonne femme se décide à aller chercher son argent avant de
considérer le boulot comme terminé. (P. Djian – 37.2 Le matin)
‘I waited for the woman to take her money before considering the job done.’

Interaction between avant que and lexical semantics cues licenses ‘continuation’
or ‘pre-condition’. Differences with ‘circumstance’ pertain to semantics, as we have
seen, but also to temporal structure and discourse structure, as we will see in the
next section.

3.  Effects on temporal structure and discourse structure

Avant que appears in several discourse contexts and I put forward that this subor-
dinating conjunction licences different interpretations: ‘circumstance’, ‘continuation’
or ­‘pre-condition’. These three interpretations are summed up respectively by the
discourses (27) (=(1)), (28) (=(3)), and (29), which are examples constructed from
discourses taken from the Frantext database (see note 3), and illustrate the previous
analysis in section 2.

(27) Paul a trouvé la solution avant que Marie la lui donne.

‘Paul found the solution before Marie gave it to him.’
(28) a. Paul a d’abord cherché la solution avant que Marie la lui donne.
‘Paul first sought the solution before Marie gave it to him.
b. Paul a longtemps cherché la solution avant que Marie la lui donne.
‘Paul sought the solution for a long time before Marie gave it to him.’
 Laurence Delort

(29) a. Paul a dû longuement insisté avant que Marie lui donne la solution.
‘Paul had to insist strongly before Marie gave the solution to him.’
b. Paul a longuement insisté avant que Marie lui donne la solution.
‘Paul insisted strongly before Marie gave the solution to him.’

‘Circumstance’ and ‘continuation’ echo two discourse relations from Segmented

­Discourse Representation Theory (SDRT) as found in Asher and Lascarides (2003),
viz. Background and Narration, respectively. Since I did not find a detailed descrip-
tion of a possible corresponding discourse relation for ‘pre-condition’ in SDRT, or
in Mann and Thompson’s (1988) Rhetorical Structure Theory (RST), I put this in-
terpretation aside in this section. Nevertheless, one interesting aspect arises from
Background (for ‘circumstance’) and Narration (for ‘continuation’): they do not in-
volve the same effects on temporal structure and discourse structure. Does avant
que convey Background in (27) and Narration in (28)? If this proves to be the case,
it would mean that the subordinating conjunction avant que has an important role
at the discourse level.
After a general definition of Background and Narration in SDRT (section 3.1), I try
to show that avant que can be a marker of these discourse relations in contexts similar
to those in (27) and (28) (section 3.2).

3.1  Definition of Background and Narration in SDRT

3.1.1  Background in SDRT
Background is defined as follows in Asher and Lascarides (2003: 460): “this relation
holds whenever one constituent provides information about the surrounding state of
affairs in which the eventuality mentioned in the other constituent occurred”. That
is, the eventuality described in the first clause is the main event, and the eventuality
described in the second clause is the (e.g., temporal, spatial) circumstance of this main
event. Since temporal progression is broken, it is a subordinating discourse relation
(Vieu & Prévot, 2004). Discourse (30), taken from Asher and Lascarides (2003), illus-
trates Background between the two constituents π1 and π2 (noted as Background(π1,
π2)), representing the semantics of e1 and e2, respectively.
(30) Max opened the door. The room was pitch dark.

The semantic effect of Background is a temporal overlap between the eventualities, as

axiom (A1) expresses, namely that e2 and e1 overlap.
(A1) Background(π1, π2) → overlap(e2, e1)

3.1.2  Narration in SDRT

Narration is defined as follows in Asher and Lascarides (2003: 462): “this relation
holds if the constituents express eventualities that occur in the sequence in which
 Clause subordination in discourse structure 

they were described”. That is, the eventuality described in the first clause occurs, then,
the ­eventuality described in the second clause occurs. Since temporal progression is
­continuous, it is a coordinating discourse relation. Discourse (31), taken from Asher
and Lascarides (2003), illustrates Narration(π1, π2).
(31) Max came into the room. He sat down.

This discourse relation implies semantic effects on discourse interpretation. The first
effect is the temporal relation between the denoted events, cf. axiom (A2) proposed
in Bras et al. (2001). This axiom means that, when Narration holds between π1 and
π2, “post(e1) persists up to the beginning of e2, and pre(e2) starts when (or before) e1
ends”, i.e. there is a strong contiguity between the events expressed in the units linked
by Narration. That is, no relevant event can occur between the events. Hence, this tem-
poral relation has to be distinguished from the simple temporal relation eα < eβ.
(A2) Narration(π1, π2) → e1⊃⊂(post(e1)∩pre(e2))⊃⊂e2

The second effect of Narration is both semantic and structural: it reflects the need for a com-
mon topic between the events, and the insertion in the discourse structure of a constituent
corresponding to this topic, cf. axiom (A3). If Narration holds, then there should exist a unit
(the topic, noted π*) summarizing the units linked by Narration, and structurally dominat-
ing the complex unit (π’) made by Narration(π1, π2) via the discourse relation Topic.
(A3) Narration(π1, π2) → ∃π*(π*=π1∩π2)∧Topic(π*,Narration(π1, π2))

This topic constraint is a means for ensuring coherence in a narrative text. For ­instance,
the representation in SDRT of discourse (31) is given in Figure 1. It illustrates that, in
SDRT, a coordinating relation like Narration is drawn with a horizontal arrow, while
a subordinating relation like Topic (or Background) is drawn with a vertical arrow, cf.
Asher and Lascarides (2003: 146–147).

π*

Topic

π¢

π1 π2
Narration

Figure 1.  SDRT structure for discourse (31).

3.2  Is avant que a marker of Background and Narration?

3.2.1  Avant que and Background
Given the definition of Background in SDRT, one can easily see that avant que t­ riggers
this discourse relation in (27): π1 describes the main event (conveyed by the main
 Laurence Delort

clause), and π2 describes the temporal circumstance (conveyed by the subordi-

nate clause introduced by avant que). It is also the case in discourse (32), where the
­subordinating nature of Background is indubitable because of the attachment of the
subsequent constituents (π3 and π4) on π1 via Narration, with respect to the right
frontier constraint, (cf. Polanyi 1988).
(32) Il remit soigneusement sa casquette avant de sortir, salua de nouveau et ouvrit
maladroitement la porte. (R. Sabatier – David et Olivier)
‘He carefully put on his cap before leaving, said good-bye once again and
opened the door awkwardly.’

Moreover, the temporal effect of Background is compatible with the semantics of

­discourses conveying ‘circumstance’. For instance, in (27) and (32), e1 does not really
occur before e2, but e1 occurs during an event occurring before e2. In more formal
terms: there is not simply e1 < e2, but overlap(e, e1) and e < e2, i.e. e overlaps e1, and e
occurs before e2 (same temporal overlapping operator as in axiom (A1)). Nevertheless,
a detailed investigation remains to be done.
Because of the subordinating nature of Background, the fact that avant que is a
marker of Background exemplifies the hypothesis of the mapping between clause sub-
ordination and discourse subordination made by Matthiessen and Thompson (1988).
This issue is discussed in the next subsection.

3.2.2  Avant que and Narration

There is some linguistic evidence showing that the temporal precedence relation in
discourse (27), on the one hand, and discourse (28), on the other hand, is not the same.
First, the temporal relation can be modified in (27) but not in (28), cf. (33) and (34)
respectively.
(33) Paul a trouvé la solution trois heures avant que Marie la lui donne.
‘Paul found the solution three hours before Marie gave it to him.’
(34) a. #Paul a d’abord cherché la solution trois heures avant que Marie la lui donne.
‘Paul first sought the solution three hours before Marie gave it to him.’
b. #Paul a longtemps cherché la solution trois heures avant que Marie la lui
donne.
‘Paul sought the solution for a long time three hours before Marie gave it to
him.’

This adverbial modification with trois heures (as well as with juste, quelques secondes,
longtemps (‘just’, ‘a few seconds’, ‘a long time’, etc.), bearing on the temporal relation, works
with ‘circumstance’, but not with ‘continuation’. This reflects the fact that not only is it im-
possible to modify the temporal relation, but it is also impossible to ­extend the temporal
distance between the eventualities. That is, with ‘circumstance’, the ­distance between the
eventualities can be quantified, while with ‘continuation’, this distance is constrained such
that e2 occurs immediately after e1, i.e. there is no distance between e1 and e2.
 Clause subordination in discourse structure 

Inserting a third eventuality between e1 and e2 can test this constraint, cf. Bras et
al. (2001): in (35), e3 can occur between e1 and e2, whereas in (36), it cannot.
(35) Paul a trouvé la solution avant que Marie la lui donne. Entre-temps, il l’avait
donnée à Lisa.
‘Paul found the solutions before Marie gave it to him. Meanwhile, he had given
it to Lisa.’
(36) a. Paul a d’abord cherché la solution avant que Marie la lui donne. #Entre-temps,
il l’avait trouvée.
‘Paul first sought the solution before Marie gave it to him.’ #Meanwhile, he
had found it.’
b. Paul a longtemps cherché la solution avant que Marie la lui donne.
#Entre-temps, il l’avait trouvée.
‘Paul sought the solution for a long time before Marie gave it to him.’
#Meanwhile, he had found it.’

I did not come across discourses such as (36) in a corpus. However discourses like
(37) have been attested in corpora, where the third eventuality always corresponds to
an eventuality occurring during the time interval of e1. So, this is not a contradicting
observation.

(37) J’ai passé quinze ans à donner des cours et à écrire des livres avant de comprendre
que je n’étais pas fait pour cela, mais je ne peux vraiment me retirer, car,
entre-temps, j’ai acquis quelque renommée et les collègues me retiennent.
(J. Kristeva – Les samouraïs)
‘I spent fifteen years lecturing and writing books before understanding that I am
not made for this, but I cannot withdraw because, meanwhile, I had gained a
reputation and my colleagues retain me.’

It seems clear that the temporal relation between the end of e1 and the beginning of e2
is constrained. With ‘continuation’, the temporal relation is like the one proposed by
Bras et al. (2001) for describing the temporal effect of Narration, see axiom (A2). So,
with respect to temporal structure, in discourse (28), and in other discourses pertain-
ing to ‘continuation’, Narration is used.
The topic constraint, expressed in axiom (A3), is satisfied in topic elaboration, but not
in a preparatory event. These two subtypes of ‘continuation’ share the same ­temporal rela-
tion but not the topic constraint. Nevertheless, they imply the same structural ­effect: co-
ordination between discourse units. Finally, by conveying Narration, avant que is a chal-
lenge to the hypothesis of a mapping between clause combining and discourse ­structure.
To put it in a nutshell, avant que is a subordinating conjunction that can trigger
(at least) two discourse relations, involving different temporal and structural effects.
On the one hand, avant que can trigger Background, which implies a temporal overlap
relation between the main event and its temporal location, and subordination in dis-
course (i.e. narrative digression). On the other hand, avant que can trigger Narration,
 Laurence Delort

which implies a temporal precedence relation between two main events, and coordination
in discourse (i.e. narrative progression).

4.  Conclusion and perspectives

The goal of this paper was twofold. First, one aim was to shed light on the fact that
the subordinating conjunction avant que can convey several interpretations when it
interacts with discourse context. Second, it aimed at showing that avant que has an
important role in discourse structure, by triggering either subordination or coordination
between discourse constituents. It follows that there is no direct mapping between
clause combining and discourse structure: a subordinating conjunction can convey a
coordinating discourse relation.
If avant que is a cue-phrase of Narration, a comparison between avant que and
puis would be interesting, as the following observations show. When there is a top-
ic elaboration, avant que can be translated by and then, as observed in the bilingual
­database TransSearch,4 cf. (38).
(38) a. En deux heures à peine, l’espace aérien nord-américain a été plongé dans le
chaos le plus complet avant d’être complètement fermé.
b. In the space of a few hours, North American air space was thrown into
complete chaos and then shut down completely.

When ‘continuation’ occurs, and e1 is durative, one can find puis in place of avant que,
cf. (39).
(39) Il a attendu un moment puis m’a touché l’épaule et s’est relevé. - Je sors par les
cuisines, il a fait. (P. Djian – 37.2 Le matin)
‘He waited for a while and then touched my shoulder and got up.’

But puis cannot always be replaced by avant que without changing the original mean-
ing of the discourse. The ‘continuation’ interpretation of (40a) is totally lost with avant
que: (40b) rather conveys ‘circumstance’ despite the thematic link (topic) between
eventualities.
(40) a. Il frappa à la porte. Puis il entra.
‘He knocked at the door. Then he entered.’
b. Il frappa à la porte avant d’entrer.
‘He knocked at the door before entering.’

This confirms that avant que needs a particular discourse context to convey a dis-
course relation such as Narration. But further investigation is needed.

  TransSearch is a database of translations between English and French accessible from http://
www.tsrali.com/.
 Clause subordination in discourse structure 

Finally, this paper gives an account of several interpretations illustrated by unambig-

uous discourses. But corpora display many ambiguous discourses for which it is hard to
find a clear interpretation, especially to decide if a discourse represents ‘circumstance’ (or
Background), or ‘continuation’ (or Narration). ‘Pre-condition’ ­discourses are not ambigu-
ous because there are always (extra-) linguistic clues for inferring such an interpretation.
However, from discourses without lexical clues (such as d’abord, ­commencer par) or no
possible semantic inference, ambiguity arises. Future research will concentrate on other
clues for inferring the correct interpretation of avant que, such as taking into account
a wider discourse context or discourse attachment on the content of the subordinate
clause, as discourses (41) and (42) illustrate. In (41), e2 is part of a set of events, all elabo-
rating the topic expressed in the first sentence. In (42), the last sentence is to be attached
to the subordinate clause and not to the main clause (as in discourse (32), for instance).
(41) Heureusement, Mme Bernard arracha les autres à la torpeur en jouant la meneuse
de jeu. Elle entraîna les jeunes dans une partie de croquet, puis organisa une partie
de mikado avant de faire des tours de cartes et de montrer des jeux avec une simple
ficelle. (R. Sabatier – Les filles chantantes)
‘Fortunately, Mrs Bernard acted as a leader. She dragged the youths into a match
of croquet, then organised a game of pick-up-sticks before doing card tricks and
showing games with a simple piece of string.’
(42) Olivier trouva là une heureuse occasion de promenade. Il jeta un regard dans la
glace avant de sortir en sifflotant. Il faisait beau. (R. Sabatier – David et Olivier)
‘Olivier found an opportunity for a stroll. He peeked in the mirror before going
out whistling. The sun was shining.’

Acknowledgments

I would like to thank the reviewers for their many valuable comments, and André
­Bittar for checking my English.

References

Asher, Nicholas. 1993. Reference to Abstract Objects in Discourse. Dordrecht: Kluwer.

Asher, Nicholas & Lascarides, Alex. 2003. Logics of Conversation. Cambridge: CUP.
Bras, Myriam  Le Draoulec, Anne & Vieu, Laure. 2001. Temporal information and discourse relations
in narratives: The role of French connectives puis and un peu plus tard. In Proceedings of the 39th
Annual Meeting of ACL, 49–56, Toulouse, France.
Danlos, Laurence. 2005. Partition of an entity with aspectuo-temporal operators. In Proceedings of
the Third International Workshop on Generative Approaches to the Lexicon (GL’2005), Geneva,
Switzerland.
Kamp, Hans & Reyle, Uwe. 1993. From Discourse to Logic. Introduction to Model-theoretic ­Semantics
of Natural Language, Formal Logic and Discourse Representation Theory. D ­ ordrecht: Kluwer.
 Laurence Delort

Mann, William & Thompson, Sandra. 1988. Rhetorical structure theory: Toward a functional
theory of text organization. Text 8(3): 243–281.
Matthiessen, Christian & Thompson, Sandra. 1988. The structure of discourse and ‘subordination’.
In Clause Combining in Grammar and Discourse [Typological Studies in Language vol. 18],
John Haiman & Sandra Thompson (eds), 275–329. Amsterdam: John Benjamins.
Polanyi, Livia. 1988. A formal model of the structure of discourse. Journal of Pragmatics 12:
601–638.
Vieu, Laure  & Prévot, Laurent. 2004. Background in SDRT. In Proceedings of the conference
Traitement Automatique du Langage Naturel, 485–494. Fez, Morocco.
 Pseudo-imperatives and other cases of
conditional conjunction and conjunctive
disjunction*

Michael Franke
Universiteit van Amsterdam

Pseudo-imperatives, a special kind of sentential conjunctions and disjunctions,

display a surprisingly divergent preference bias. This paper aims to explain this
pragmatic preference puzzle based on the different discourse segmentation
behavior of conjunction and and disjunction or. To lend credence to the
suggested explanation, related non-standard uses of conjunction and disjunction
will be discussed in detail.

Keywords: pseudo-imperatives, discourse relations, non-standard uses of

conjunction/disjunction

1.  Pseudo-imperatives and a pragmatic puzzle

1.1  The problem of pragmatic asymmetry

There is an interesting pragmatic asymmetry in the following pattern (cf. van der
­Auwera, 1986):1
(1) a. Close the window and I will kiss you. (A and P+)
b. Close the window and I will kill you. (A and P–)
c. Close the window or I will kill you. (A or P–)
d. ?Close the window or I will kiss you. (A or P+)

*  A lot of people have helped me considerably in writing this text including the anonymous
reviewers. I’m thankful to all of them. Anke Lüdeling kindly provided me with access to the cor-
pora that I have used and I am particularly grateful for that, as well as for the stimulation I have
received from discussions with Robert van Rooij, Martin Stokhof, Paul Dekker and Magdalena
Schwager. Thanks to Tikitu de Jager for proofreading. Needless to say, all errors are my own.
.  I assume here for the sake of the example that (it is common knowledge between inter-
locutors that) the hearer wants to be kissed by the sender, but not killed. I will write P+ (P–) for
declarative clauses that denote (un‑)desirable propositions in this sense.
 Michael Franke

I will call the sentences in (1) pseudo-imperatives,2 or PIs for short, which I take to be a
label for a particular surface form: an imperative clause3 A is followed by either and or
or which is followed by a declarative clause P. If we assume that in a given context the
hearer has a clear preference whether the content expressed by the declarative clause P
of a pseudo-imperative should become true or not, each sentence (1a–1c) is a natural
argument for or against the act named by A: (1a) and (1c) are arguments for, (1b) is an
argument against. However, sentence (1d) is not a natural thing to say at all: for fixed
hearer-desirability of P it is simply infelicitous. In particular, the infelicitous (1d) is not
an argument against the act named in A, although this might be expected when looking
at the conditional sentences in (2) which correspond loosely to each sentence in (1).
(2) a. If you close the window, I will kiss you. (If A, then P+)
b. If you close the window, I will kill you. (If A, then P–)
c. If you don’t close the window, I will kill you. (If not-A, then P–)
d. If you don’t close the window, I will kiss you. (If not-A, then P+)

For the conditional sentences in (2) the picture is entirely symmetric: positive (nega-
tive) consequences of act A are an argument for (against) it, while positive (negative)
consequences of not-A are an argument against (for) A. This symmetry in (2) is broken
in (1) by the infelicity of (1d).
This paper takes this Problem of Pragmatic Asymmetry (PoPA) observed in pseudo-
imperatives as its starting point. The puzzle can be approached from two sides. Firstly, (i)
we should ask why there are instantiations of variable X so that (3a) becomes an argument
against A, while there are no instantiations of Y that turn (3b) into an argument against A.
(3) a. Do A and X will be the case/happen.
b. Do A or Y will be the case/happen.

Secondly, (ii) we should ask why instantiations of X in (3a) may be desirable or unde-
sirable to the hearer, while instantiations of Y in (3b) may not be hearer-desirable.4

.  This does not quite match Clark’s (1993) terminology where the term was chosen for its par-
ticular connotation. In this paper I do not want to hint at a possible difference between genuine
imperatives and mere pseudo-cases. Here “pseudo-imperative” is just a name for a particular
linguistic form.

.  Let me justify this categorization: It is indeed not entirely self-evident that the first con-
nect in sentences (1) is really an imperative clause. It might as well be an infinitive or a bare VP
(see Bolinger, 1979, for some early discussion). Although this paper deals with English, I take
the fact that other languages, such as German or modern Greek (cf. Han, 1998), have parallel
constructions where the first connect morpho-syntactically is undeniably an imperative clause
to be reason enough to reject the idea that in English these forms should be anything else than
imperative clauses (contra Russell, 2007).

.  Strictly speaking, these two questions really address two different problems, or at least two
different challenges from the same problem set. Nevertheless, there is an obvious connection.
 Pseudo-imperatives and other cases of conjunction/disjunction 

1.2  Some possible solution strategies

There are several ways in which the PoPA might be solved. Let me sketch some pos-
sibilities in order to briefly discuss previous accounts and to situate the current contri-
bution more clearly in the greater scheme of things.5

1.2.1  Pragmatics only

Perhaps the most appealing account for the PoPA would be entirely pragmatic in terms
of principles of how to say what and what not to say when, especially in order to influ-
ence someone else’s behavior or judgment.6 Here is a rough sketch of such a purely
pragmatic account. Suppose that we start with the minimal assumption that the pseu-
do-imperatives in (1) are semantically equivalent to the conditional sentences in (2)
one by one. Then, obviously, whatever pragmatic account for the infelicity of (1d) we
would like to give, we would have to take into consideration that, after all, (2d) is ac-
ceptable. So we need to acknowledge at least some difference between the disjunction
(1d) and the conditional (2d). In Franke (2005a) I suggested that the difference might
be found in topicality or aboutness. With some intuitive plausibility we could say that
conditionals (2c) and (2d) mention, refer to or talk about not-A. Disjunctions (1c) and
(1d), on the other hand, mention or talk about A instead while still giving conditional
information about what happens when A is not performed. Based on this intuition we
might endorse the following Mention-Principle:
Mention-Principle: Do not bring to attention an action that you do not want to
be performed (because mentioning choices just makes them salient and more
probable to be chosen), unless you immediately discredit what you mention (by
stating negative consequences of it, for instance).

It is not crucial to find the Mention-Principle convincing as long as it helps illustrate

what a purely pragmatic approach could look like. For it is palpable that only (1d)

For instance, suppose we have a satisfactory independent explanation for the preference bias in
(3b) as an answer to question (ii). Then an answer to question (i) may be given on top of that.
We could say that in order to be an argument against A, (3b) needs a desirable Y, because this
is how conditional information influences choice of action. But (3b) cannot get a positive Y by
assumption. Hence it cannot be an argument against A. But clearly not every such argument
that connects (i) and (ii) needs to be accepted. In this sense, the two questions should be kept
apart and each account of the PoPA should make clear whether the chicken or the egg part of
the puzzle is solved first and how we get from one to the other.

.  The present approach restricts itself to semantic and pragmatic considerations. More syn-
tactic considerations are left for another occasion.

.  Both van der Auwera (1986) and Clark (1993) propose a solution that fits into this
category. For discussion and criticism of these and other accounts see Franke (2005b) and
Schwager (2006).
 Michael Franke

v­ iolates this principle while all other sentences in (1) and (2) agree with it. So here we
have a rough sketch of a pure pragmatic solution to (aspect (ii) of) the PoPA.
One of the driving ideas of this paper is that a pure pragmatic explanation of the
PoPA as outlined here, however plausible on its own, would ignore a considerable
amount of (at least prima facie) relevant empirical data. It is therefore the central
concern of this paper to give further, mostly authentic and in this context hitherto
overlooked data in order to put PIs into a broader linguistic perspective. Doing so
will support a particular alternative approach, namely the Connector Hypothesis
which I will introduce in section 1.3 and whose plausibility I will discuss critically
thereafter.

1.2.2  Forceless imperative clauses

To appreciate the central hypothesis of this paper, it is advisory to contrast it with a su-
perficially similar, but different solution strategy. The solution strategy I have in mind
here is based on the following idea. Suppose that we can make plausible that impera-
tive clauses are associated with directive force in most, but not all linguistic contexts.
In particular, consider the following Force Hypothesis:7
Force Hypothesis: If an imperative clause is followed by conjunction and and
a declarative it will not be associated with directive force, but if followed by
disjunction or and a declarative clause it will.

Indeed, the Force Hypothesis gives us an answer to question (i) of the PoPA: PIs with
conjunction, call them conjunctive pseudo-imperatives (ConjPIs), are not associated
with directive force, but disjunctive pseudo-imperatives (DisjPIs) are. Hence, a ConjPI
(3a) can be an argument for and against the content of its imperative clause A, but a
DisjPI (3b) cannot be an argument against A, because here the imperative clause is
always associated with the directive force that A should be performed.
Obviously, (any serious careful formulation of) the Force Hypothesis needs in-
dependent support. What is needed is independent morpho-syntactic evidence that
imperative clauses lose their directive force exactly in those contexts where they can
be used to argue against the performance of the act they describe. Perhaps such inde-
pendent evidence can be found. Again, I do not want to argue that there is no plausible
continuation of this idea. At present, I want to suggest a similar, but different hypoth-
esis for which independent evidence can be given easily in terms of parallel examples
that do not involve imperatives at all. And once more, I argue that this approach has an

.  For any serious proposal along these lines more severe qualifications are necessary and, of
course, have been spelled out by followers of this strategy such as Han (1998) and Russell (2007).
To name just one example of such a refinement, we obviously have to require the absence of
speech-act triggers such as please (see Bolinger, 1979). I omit the details and refer the reader
again to the discussion in Franke (2005b) and Schwager (2006).
 Pseudo-imperatives and other cases of conjunction/disjunction 

advantage also over the Force Hypothesis in that it places our explanandum in a wider
context of relevant examples.

1.3  The Connector Hypothesis

Let me start by suggesting and evaluating critically an account of the PoPA based on
what I will call the Connector Hypothesis (CH). The idea behind the CH is that it is
not pragmatic principles of debate or the pragma-semantic properties of imperatives
that are responsible for the pragmatic asymmetry in PIs, but rather properties of the
sentential connectors and and or. I want to suggest that connectors and and or in PIs
have different discourse segmentation properties. To spell out the CH, I will therefore
first briefly introduce this difference in discourse segmentation behavior in section
1.3.1. Subsequently, in section 1.3.2, I will investigate authentic PIs in order to assess,
again on an intuitive basis only, their discourse segmentation behavior. This finally
leads to a concrete formulation of the CH in section 1.3.3.

1.3.1  Discourse segmentation

According to Mann and Thompson (1987), discourse interpretation is a three-step
procedure consisting of:
Segmentation: What are the units of discourse?
Relation: Which units of discourse relate to each other and what is the
relation between them?
Coordination: Which hierarchical structure holds between related units of
discourse; which ones, if any, are subordinate?

Especially in the context of sentential connections such as PIs whose connects are of
different clause types the segmentation step is not trivial and deserves extra attention.
For a simple motivating example, consider the sentences in (4).
(4) a. Drive past the gas station and then, after five minutes, you will see the
roundabout.
b. Drive past the gas station until you see the roundabout.

Intuitively, at speech-act level we have two discourse units in (4a), but only one in (4b):
(4a) gives advice about driving in a certain direction followed by the statement that, if
that direction is followed, the roundabout will be reached. In clear contrast, (4b) is just
a directive to drive in a certain direction for a specific amount of time. In other words,
the clause “you will see the roundabout” in (4b) is not realized as a speech-act in its
own right, because of the (semantically subordinating) connector until, while in (4a)
both connects of the (presumably semantically coordinating) connector and (then)
give rise to two separate speech-acts.
This simple observation suggests the following intuitive distinction: (an occur-
rence of) a sentential connection “X ⊗ Y”, where X and Y are sentences or clauses and
 Michael Franke

⊗ is a sentential connector such as until or and, is discourse separating if the connects

X and Y are treated as separate discourse units and are thus associated with separate
communicative events, or speech acts (in a particular discourse context); otherwise, if
(an occurrence of) “X ⊗ Y” is treated as one communicative event and both connects
together are associated with one speech act (in a particular discourse context), (this
occurrence of) “X ⊗ Y“ is discourse integrating.8

1.3.2  Pseudo-imperatives in the wild

The distinction between discourse separating and discourse integrating sentential con-
nections is based merely on introspection. At present, I am not interested in theoreti-
cal refinement. Rather, I suggest to apply naive intuition about discourse segmentation
behavior of connections to a selection of examples of PIs collected from various sourc-
es. To start with, here are three cases of ConjPIs:
(5) a. The creature sighed, ‘Come closer and I will tell you a great secret.’ I moved
closer, until I could see my own breath condense on its skeletal shoulder.
(http://elfwood.lysator.liu.se/libr/s/h/shalene/ random.html, 17.8.2006)
b. Please send me the issues you are experiencing, and I will update this page,
along with information when they are resolved.
(http://www.bloglet.com, 17.8.2006)
c. “And now rearrange yourself,” I said, “and in the meanwhile I will go and
wash the baby.”
   (http://www.worldwideschool.org/library/books/lit/
shortstories/ThroughRussia/Chap1.html, 17.8.2006)

The examples in (5) are all authentic examples corresponding more or less to the
artificial (1a): for instance, under the likely assumption that the addressee wants to
be told a great secret, the whole sentence(5a) urges him or her to come closer. The
other examples are similar in this respect. But now ask yourself: are the examples in
(5) discourse separating or discourse integrating connections? For instance in (5a),
does the creature give a directive followed by a (conditional) statement or does the
creature only assert a (conditional) statement? Maybe the following three readings
are most plausible:
(6) a. inform(“if you come closer, I will tell you a secret”)
b. direct(“send issues”) & inform(“if you send, I will update”)
c. direct(“rearrange yourself ”) & inform(“I will wash the baby”)

.  The ideas of discourse integrating connections and non-veridical discourse relations in Seg-
mented Discourse Representation Theory (Asher and Lascarides, 2003) are related but not iden-
tical: the former is a property of a linguistic form (token), the latter is a property of a semantic
object (token).
 Pseudo-imperatives and other cases of conjunction/disjunction 

Example (5a) could well just be an assertion of a conditional, (5b) gives a clear direc-
tive followed by a conditional statement, and (5c) gives a directive followed by a non-
conditional statement.
There is plenty of room for disagreement about the judgments in (6a). But this is
not crucial for the argument that I would like to make. There are other examples where
things are less controversial. Here is one:
(7) Say one more word and I’ll scream!
 (From the song ‘Show me’ from the musical My Fair Lady
by Frederik Loewe and Alan J. Lerner)

To my mind, example (7), which corresponds to (1b), can only be treated as a dis-
course integrating connection: a statement of a conditional as in (8a).
(8) a. inform(“if you say one more word, I’ll scream”)
b. direct(“say one more word”) & inform(“(if you do) I will scream”)
c. direct(“don’t say one more word”) & inform(“(if you do) I will scream”)

Clearly, (7) does not get reading (8b), and even (8c) seems implausible, because it is
unclear how an imperative clause can give rise to a directive with the exact opposite
content of what is mentioned.9
To remove all doubt that there are examples of discourse integrating conditional-
like ConjPIs consider finally the following example:
(9) ‘You see what kind of people they are,’ he said: his eyes moved around restlessly,
he did not seem to be speaking to anyone in particular. ‘You think there is
nothing and all the time the ground beneath your feet is ­rotten with tunnels.
Look around a place like this and you would swear there wasn’t a living soul in
miles. Then turn your back and they come crawling out of the ground. […]’
(J.M. Coetzee, Life and Times of ­Michael K, pp. 121–2)10

The last ConjPI in (9) certainly does not get reading (10b) or (10c), but simply (10a).
(10) a. inform(“if you turn your back, they come crawling out”)
b. direct(“turn you back”) & inform(“(if you do) they come crawling out”)
c. direct(“don’t turn you back”) & inform(“(if you do) they come crawling
out”)

.  In fact, both van der Auwera (1986) and Clark (1993) maintain that under certain circumstanc-
es an imperative clause can be interpreted like an ironic or sarcastic remark to mean the opposite of
what has been said literally. Still I find it implausible to argue for analysis (8c) on these grounds, be-
cause in order to be perceived as ironic or sarcastic it is necessary that the literal content is assessed
as if meant literally after all. But this is not intuitive for cases like (7) where it is not the case that the
imperative clause is first taken literally and then reinterpreted along pragmatic principles.
.  Page numbers refer to the Vintage 2004 paperback edition.
 Michael Franke

Example (9) is special, not only because it clearly does not involve an imperative
speech act, but also because it involves a different modality than the examples that
we have considered so far: other examples referred to the concrete immediate fu-
ture, but (9) states a generic relationship between events. We could speak of generic
ConjPIs here. The crucial observation is that generic ConjPIs are clear cases of dis-
course integrating ConjPIs with a conditional-like interpretation. The reason why
this is crucial is because it contrasts with DisjPIs. First of all, look at some genuine
examples:
(11) a. The relevant Minister for Finance and the Budget says, ‘Put the brakes on
or we will lose our European Union aid!’
(From corpus: Europarl (en) (EU-EN), 2049840)
b. Don’t bother to resist, or I’ll beat you
(From the song ‘The Beautiful People’ by Marylin Manson)
c. Bush Tells UN, Make War or I Will
(http://www.thenation.com/blogs/capitalgames?pid = 100, 17.8.2006)

In all examples in (11) the speaker urges the addressee to bring about some state of
affairs or perform some action and further enhances this urge by a threat. In example
(11a), for instance, the minister demands that the breaks be put on and threatens that
if the brakes are not put on something bad will happen:
(12) direct(“put brakes on”) & inform(“if brakes are not put on, we lose aid”)

In other words, sentence (11a) is a discourse separating connection. This is similar for
the other examples in (11) and, it seems, for most, if not all DisjPIs.
There is but one class of DisjPIs for which it is not clear whether they are discourse
separating or discourse integrating. These are sentences like (13) which we could call
generic DisjPIs because, like generic ConjPIs, they too refer to a generic conditional
relation between events.
(13) a. I’m telling you, working for a successful start-up is no fun. The atmosphere
is so tensed. Agree to everything he says, or your boss fires you
immediately.
b. Speak at least six languages or you are not a cosmopolitan.

I will come back to generic DisjPIs in section 4 where I discuss whether these cases are
a threat to the CH in its strong formulation given next.

1.3.3  The Connector Hypothesis: formulation & application

The PoPA can be accounted for if we assume that conjunction and and disjunction or,
as they occur in PIs, have a different discourse segmentation behavior:
Connector Hypothesis (CH): While ConjPIs can be discourse integrating con-
nections, in which case they get a pure conditional-like reading such as (6a), or
discourse separating connections, in which case they get a speech-act ­conjunction
 Pseudo-imperatives and other cases of conjunction/disjunction 

reading such as (6c), DisjPIs can be discourse separating connections only and
they always get a speech-act conjunction reading as in (12).

Based on the CH the following pragmatic account of the PoPA can be given. If a DisjPI
like (3b) is a discourse separating connection like (12), it will always contain a direc-
tive to perform act A. Hence, the whole disjunction cannot possibly be an argument
against A. This answers part (i) of the PoPA, the question why a DisjPI cannot be an
argument against the content of its imperative connect.
As for part (ii) of the PoPA, the presence of a directive to perform A also explains
why a (hearer-)desirable declarative connect Y is pragmatically infelicitous in (3b). If
Y is desirable then the conditional statement “If not-A, then Y” urges the addressee not
to bring about or perform A, but at the same time the addressee is told to do so by the
directive. This is clearly incongruous, whence the infelicity.
Like the Force Hypothesis, the CH also requires independent evidence in its sup-
port. In particular, there are three issues to be addressed in order to lend credence to
the CH: firstly, the claim that there are discourse integrating conditional-like read-
ings of English and needs support; this will be given in section 2. Secondly, we need
evidence for the stipulated discourse separating readings of English or. To this end,
section 3 gives corroborating examples of discourse separating or which, to the very
best of my knowledge, have not been addressed in the relevant literature in this form.
Finally, we have to justify the claim that DisjPIs can only be discourse separating. This
is the Achilles tendon of the CH and I will address this issue critically when I turn to a
closer investigation of generic DisjPIs in section 4.

2.  Conditional conjunction

2.1  The problem of conditional conjunction

It is not surprising that and can realize a speech-act conjunction as in (6b) and (6c).
Similarly it is not surprising that, in (6b), the declarative connect P of a ConjPI is
interpreted in the context of A, as it may be considered a modal subordination phe-
nomenon (Roberts, 1989) and has parallel examples in contexts different from ConjPIs
such as in (14).
(14) “I’m leaving one for all the others,” said Rabbit, “and telling them what it means,
and they’ll search too. I’m in a hurry, good-bye.” And he had run off.
 (A.A. Milne, The House at Pooh Corner, ‘In which
Eeyore finds the Wolery and Owl moves into it’)

In (14) Rabbit’s prognosis that the others will search too is interpreted in the context
of, i.e., conditional upon, the others receiving and understanding “Owl’s Mysterious
Missage” (the subject of conversation at that particular point in the story).
 Michael Franke

What is much more surprising is the behavior of discourse integrating conjunc-

tion with its conditional-like readings, as in (6a). However, it has been observed before
that conditional conjunctions, as we might want to call them, occur in a variety of con-
texts (cf. Culicover and Jackendoff, 1997; Schwager, 2006). Apart from ConjPIs, there
are three further contexts in which and is naturally interpreted to express a conditional
relationship between connects. First of all, we find conjunctions of declaratives (15),
either with the simple present tense in both connects as in (15a), where, in the present
case, we find a generic interpretation, or with a simple present tense in the first connect
and a will-future in the second (15b), where, in the present case, we find a conditional
interpretation relating to the immediate future course of events.
(15) a. I am not sure he is wholly of this world. One tries to imagine him running
a staging post for insurgents and one’s mind boggles.
(J.M. Coetzee, Life and Times of Michael K, p. 130)
b. ‘And if I climb the fence? What will you do if I climb the fence?’ ‘You climb
the fence and I’ll shoot you, I swear to God I won’t think twice, so don’t try.’
[…] ‘You climb the fence and I’ll shoot you dead, mister. No hard feelings.
I’m just telling you.’
(J.M. Coetzee, Life and Times of Michael K, p. 85)

Second, conjunctions of NPs with declaratives, often with a negative polarity item ex-
pressing minimality (16), such as one more or any, also give rise to conditional read-
ings (Culicover, 1970, 1972).
(16) Bob: I’m real disappointed in you, Modesto; pullin’ a gun on an old saddle
pal like that.
Chico: One more word and I will kill you!
Bob: One more word, huh? Let me see if I can think of one. How about
g-r-e-a-s-e-r? Greaser?
(Passage from the film One-Eyed Jacks (1961) by Marlon Brando)

The third and last case of conjunctions with a conditional reading are conjunctions of
sufficiency-modal constructions, such as only have to or it’s sufficient to, and a declara-
tive (17).
(17) You only have to install the MSDTC once in Cluster Administrator and
MSDTC will be configured on all nodes in the cluster. You do not have to install
MSDTC manually on each node.
(http://support.microsoft.com/kb/301600/, 21.8.2006)

In conclusion, there are indeed a number of examples for conditional conjunctions

beyond PIs. This then supports part of the CH.

2.2  Semantics of conditional conjunction

Of course, the observation that there are discourse integrating conditional readings of
and also raises an interesting question: how can one of our prototypical coordinators
 Pseudo-imperatives and other cases of conjunction/disjunction 

give rise to a conditional-like, i.e., a subordinate-clause-like, meaning? In particular,

what do all the contexts in which and can get conditional readings have in common
that together with a single reasonable semantic analysis of and the conditional read-
ings can be explained?
In order to address this problem, Schwager (2006), who assumes and applies the
thesis that imperatives denote necessity modals, proposes that the first connects con-
tribute a necessity modal, not only for ConjPIs but also for other examples of condi-
tional conjunctions. Necessity modals in turn denote, in simplified terms, a necessity
operator Nec(P,Q) that takes two propositions as arguments, a restrictor P and a do-
main Q, with the following obvious semantics: Nec(P,Q) = λ w . (w ∈ P → w ∈ Q).
Schwager now proposes that due to the (topical) intonational properties of con-
ditional conjunctions (deaccenting of first connect and lack of a low boundary tone
before and), the content of the first connect is not mapped onto the domain of the
necessity operator, but onto its restrictor while the content of the second connect is
mapped onto its domain, thus yielding a conditional reading.
However, Schwager herself notices some obvious shortcomings of this idea. First
of all, it is unclear why (18a) does not get the same conditional reading as (18b).11
(18) a. You have to drink one more beer and I’ll leave.
b. Drink one more beer and I’ll leave.

It is moreover not entirely obvious why sufficiency modals such as (17) should be
analyzed as contributing a necessity modal. Similarly, Schwager’s own example (73b)
of chapter 12.4, here (19), clearly contains a possibility modal.
(19) You can even call him at MIDnight and he won’t be angry.

In the light of this, the stipulation of a covert necessity operator in the first connect
of, for example, generic simple present tense clauses, as in (15a), is drawn into doubt
as well.
So how could we improve on the analysis? Let me sketch a very rough alternative
idea. Culicover and Jackendoff (1997) proposed to account for the conditional read-
ing of and in terms of a generic operator Gen(P)(Q). That a generic operator can be
tweaked to model conditional readings is clear: very roughly Gen(P)(Q) means the
same as the above necessity operator, namely that under all normal circumstances, if
the restrictor P is met, Q is true.12
That the content of the first connect of conditional conjunctions is then mapped
onto the restrictor of the generic operator can again be justified with reference to the

.  Notice that (18a) may get a different conditional reading that we are not interested in: “If
you have to drink one more beer, then I’ll leave.”.
.  Interestingly, in an entirely unrelated account, Krifka (1995), for instance, uses a generic
operator with a semantics that is basically equivalent to the semantic analysis of conditional
sentences as proposed, among others, by Kratzer (1991).
 Michael Franke

above-mentioned topical intonation properties. So suppose we had a satisfactory

­account for generic conditional conjunctions. Maybe it would not be too unlikely to
assume that a metaphysical condition, i.e., a result relation of (immediate) possible
future events, is just a special case of a generic condition, which relates possible events
in a relevant time interval that is not restricted to the immediate future.13
A solution along these lines would help to explain why (18a) does not get the same
conditional reading as (18b). Since we no longer have to stipulate a necessity operator
hidden in the first connect, we are free to assume that the overt necessity operator in
(18a) has to be part of the propositional content that enters the restrictor, if at all.14
Unfortunately however, the sketched idea does not help to account for the condi-
tional readings of (17) and (19) either, and I will have to leave the issue as an interest-
ing open problem for future analysis.

3.  Conjunctive disjunction

There are certain standard cases of natural language disjunction that seem to have in-
formed the notion of logical, truth-conditional disjunction. (20) is a case in point.
(20) I don’t know exactly what John did. He either went to the cinema, or he stayed
at home all night.

In (20) the connects of the disjunction are epistemic alternatives to one another. The
speaker commits himself to the truth of neither. This is different for a class of examples
that I would like to discuss in this section, where the speaker commits himself to both
connects. The speaker in (21), for instance, intuitively says two things: that coherence
is required and that lack thereof has undesirable consequences.
(21) Therefore, we must start to be coherent in our intentions or we will be indulging
in fine theoretical discourses which have absolutely no substance outside this
Chamber. (Europarl (en) (EU-EN), 2698860)

The discourse separating or as in (21) gets a speech-act conjunction reading. It is in

this respect that I will speak of conjunctive disjunctions.15

.  If this assimilation of metaphysic conditions to generic conditions could be verified, there
are hardly any cases left uncovered, because epistemic conditions (“If the gardener has not killed
the baroness, then the butler has.”) or speech-act related conditions (“If you’re hungry, there are
biscuits on the shelf.”) cannot be expressed by conditional conjunctions, if only very marginal.
.  We can thus account for the fact that the only available conditional reading is the one men-
tioned earlier: “If you have to drink one more beer, then I’ll leave.”
.  An anonymous reviewer correctly remarks that my terminology may be confusing. So let
me clarify: conditional conjunctions are conjunctions with a conditional reading at content level,
whereas conjunctive disjunctions are conjunctions at speech-act level.
 Pseudo-imperatives and other cases of conjunction/disjunction 

To the best of my knowledge, the fact that disjunction can get such conjunctive
readings has received little, if any, attention in the literature, although this phenom-
enon is far from infrequent, as I will show with a multiplicity of wild examples. I would
like to substantiate this claim in the following by a cartography of conjunctive disjunc-
tions. The main aim, still, is to corroborate the CH. But, in addition to this, I think that
the examples discussed in this section might be of independent interest for semantic
analyses of disjunction. That is why I will go into this in some more detail.

3.1  Hypothetical conjunctive disjunctions

The most frequently occurring examples of conjunctive disjunction are, like (21),
­connections with expressions of necessity in the first connect. Interestingly, necessity
can be expressed not only by an overt modal as in (21), but also rather indirectly as
in (22):
(22) a. The key is to keep it simple or it will not work, it will not remove the
obstacles.
(Europarl (en) (EU-EN), 272055)
b. With regard to control – as I said at the end of the speech – we propose
that the report we present every Spring to Parliament should include an
assessment of the sustainability element of the decisions taken. This is
crucial, or we will end up with conflicting decisions.
(Europarl (en) (EU-EN), 5718224)

All of the examples in (21) and (22) are discourse separating connections: speech-act
conjunctions, in analogy to our previous analysis of DisjPIs in (12). A speech-act con-
junction analysis of conjunctive disjunctions is further corroborated by the examples
in (23) where the first connect is an explicit performative.
(23) a. I would therefore ask you to clarify this point, or I will not be a bearer of
good news for the Italian farmers.
(Europarl (en) (EU-EN), 2637413)
b. Mr Spencer is asking for the floor again. I would ask him not to make
personal allusions or we will never finish.
(Europarl (en) (EU-EN), 18178233)

The same holds of example (24) where the speaker commits himself to some future
action in the first connect and gives a reason for his preference for this course of action
in the second connect.
(24) Mr President, I shall not now echo the compliments paid to Mr Lamy or he will
start to blush.
(Europarl (en) (EU-EN), 28355082)

Similar to this are cases where the speaker makes a promise in the first connect and
then gives a reason why he made it:
 Michael Franke

(25) a. No cookies, no pop-ups and no evil tracking devices, I promise. Or you can
have a bite of my dinner.
(http://absolutely-fuzzy.com/blog/?m=200406, 16.7.2006)
b. I will have these things judged by this weekend! Po will make me do it!!!!! I
promise or you can kick my butt!!!!!!
(http://www.writersco.com/the%20Day%20Before%20The%
20 World%20Ends, 16.7.2006)

As a final example, consider (26) where the first connect expresses a hope and the
second connect delivers a reason for having that hope.
(26) Senator Petten and I met quite frequently, and we would discuss the business that
was to be done that week. We would then say, ‘I hope to God the leaders do not
find out or we will never get it done.’ Therefore, do not tell them what you are
planning for that week, and you will get things accomplished.
(Hansard Senate e (HANSARD_SENATE_E), 1733312)

All the examples, (11) and (21)–(26), have a particular discourse format in common:
they are all conjunctive disjunctions “X or Y” whose first connect X presents a topic
χ, which I will refer to henceforth as the χ-component (of the disjunction “X or Y”),
as necessary, preferred or desirable. The second connect is interpreted in the context
of not-χ. χ is a mere (immediate-)future possibility which is why we could speak of
hypothetical conjunctive disjunctions here in contrast with a factual variety that will be
discussed in the next section.
Crucially, the χ-component of “X or Y” need not be identical to the semantic con-
tent of X. It may not even be contained in or referred to explicitly in X, as example (27)
makes clear.
(27) That is enough points of order, or we will never get on to the items on the
agenda. (Europarl (en) (EU-EN), 4859513)

The χ-component of the conjunctive disjunction in (27) is the content of an indirect

speech act associated with the first connect, namely that one should stop collecting
more points of order. So, if we assume that α is the speech act associated with X in
context, we can represent the speech-act conjunctive reading of “X or Y” as in (28).
(28) α(χ) & inform(“if not χ, then Y”)

(28) gives the general discourse format of a hypothetical conjunctive disjunction. It

now seems plausible that DisjPIs with their discourse format in (12) are a special case
of hypothetical conjunctive disjunctions.
For hypothetical conjunctive disjunctions “X or Y” whose first connect X presents
a not-yet-realized state or action χ as necessary, preferred or desirable, we expect that
the content of the second connect Y has a negative connotation. Indeed, the preference
bias noted for DisjPIs generalizes to all hypothetical conjunctive disjunctions, for a
(hearer‑)desirable second connect seems impossible:
 Pseudo-imperatives and other cases of conjunction/disjunction 

(29) ? It is necessary/preferred/desirable that this happens, or you will get my

­Ferrari.

This raises the question whether we also find the reversed pattern: are there hypotheti-
cal conjunctive disjunctions “X or Y” whose χ-component is presented as undesirable
in X, but whose second connect Y has a positive connotation? Although conceivable
in principle, this pattern does not seem to occur. Constructed examples sound decid-
edly odd too:
(30) ? It is undesirable that this happens, or you will get my Ferrari.

That means that the previously attested bias puzzle in connection with DisjPIs is part
of a larger whole. Hypothetical conjunctive disjunctions all present their χ-component
as desirable and have a negatively connoted second connect which is interpreted as an
epistemic alternative to χ, i.e., have a double preference bias.16

3.2  Factual conjunctive disjunctions

This double preference holds only for hypothetical conjunctive disjunctions. There is
another class of conjunctive disjunctions where this preference bias is not present.
These could be called factual conjunctive disjunctions, because, as we will see shortly,
their χ-components refer to true, established or presupposed states of affairs. There are
furthermore two kinds of factual conjunctive disjunctions: one is evaluative, the other
evidential.
In evaluative factual conjunctive disjunctions some true state of affairs χ is present-
ed as fortunate or unfortunate in the first connect, while the second connect states an
epistemic alternative to χ which is either negative or positive. As χ is presented as a true
state of affairs, the second connect usually contains a counterfactual would-modal. The
examples in (31) present some true state of affairs χ as fortunate and give a negatively
connoted counterfactual epistemic alternative.
(31) a. “Lucky we know the Forest so well, or we might get lost,” said Rabbit half
an hour later, and gave the careless laugh which you give when you know
the Forest so well that you can’t get lost.
(A.A. Milne, The House at Pooh Corner, ‘Tigger is unbounced’)
b. . . . . . . . . the wilderness . . . . . . . . remember the jumping over the drains and
the endless footpath? :lol: …..that was fun :) yup. Luckily we did that, or we
could have been walking farther away up to the Kallang Basin and Nicoll
Highway! :eek:
(http://skyscrapercity.com/archive/index.php/t-106394.html, 16.7.2006)

.  Gómez-Txurruka (2002) also stipulated this double preference bias for readings of or that
she called “conditional alternative”.
 Michael Franke

In contrast, the examples in (32) present some true state of affairs χ as unfortunate and
give a positively connoted counterfactual epistemic alternative.
(32) a. It is truly unfortunate that you are so far from us, or we would be weekly
guests at the gallery.
(http://www.cordair.com/accolades.htm, 12.7.2006)
b. Christine, it’s unfortunate that their version of the Creative Commons
license does not allow derivative versions to be created or we could fix this
problem (and also correct their numerous misspellings and other typos).
(http://www.foresight.org/nanodot/?p=2295, 21.8.2006)

In all examples in (31) and (32) the second connect with its particular negative or
positive connotation seems to give a reason why the state of affairs χ is fortunate or
unfortunate, respectively. A preference bias, as for hypothetical conjunctive disjunc-
tions, does not exist.
Finally, there are evidential factual conjunctive disjunctions such as (33) which do
not involve preferences at all, but which are entirely epistemic in character.
(33) a. It is not a very exciting topic to be discussing but, as we all know, it is very
important or we would not be here.
(Europarl (en) (EU-EN), 11272804)
b. Finally, I would also appeal to our colleagues in the national parliaments,
since they apparently have little interest in the matter either, or they
would be pressing their governments and raising a debate in the national
parliaments on why the Member States are so perverse when it comes to
taking the necessary decisions.
(Europarl (en) (EU-EN), 16403391)

Whereas in the evaluative cases, (31) and (32), the second connect gives a reason for
the evaluative judgment expressed in the first, in the evidential cases (33) the second
connect gives a reason for the epistemic judgment that the χ-component is believed
true, or at least highly probable.
The discourse format of factual conjunctive disjunctions in (34), be they evalua-
tive or evidential, is basically the same as that of the hypothetical variant in (28).
(34) α(χ) & inform(“if had not χ, then would Y ”)

The only difference is that, due to the factuality of χ and evidenced by the frequent would-
modal in the second connect, the conditional relation involved is c­ ounterfactual.

3.3  Remarks on the semantics of disjunction

The above exposition raises the question whether conjunctive disjunctions and the
discourse analysis that I suggested in (28) have any relation to free-choice readings
of disjunction (Ross 1941; von Wright 1968) and recent conjunctive semantics for
 Pseudo-imperatives and other cases of conjunction/disjunction 

disjunction (Zimmermann 2000; Geurts 2005) that have been proposed to account
for free-choice readings.17 Let me briefly comment.
The free-choice reading of disjunction (35a), which is present to a lesser degree
also in (35b), is the conjunctive reading according to which the speaker permits that
the hearer may take an apple and that he may take a pear (but not both).
(35) a. You may take an apple or a pear.
b. You may take an apple or you may take a pear.

In very vague outline, this free-choice effect can indeed be conceived of as a special
case of a conjunctive disjunction. The analysis of (35) according to the scheme in (28)
would then be a reading in which the modal may gets a performative reading (Lewis,
1979) to the effect that a permissive speech act is given:
(36) permit(“take an apple”) & inform(“if you don’t take an apple, you
may take a pear”)

We might then assume that the modal in the asserted conditional gets a performative
reading too, so that the assertion in (36) becomes a conditional permission and the
whole sentence comes out as a speech-act conjunction where the speaker allows the
hearer to take an apple and a pear.18 Suffice it to say that it is at least not entirely im-
plausible that the free-choice conjunctive reading of disjunction could be conceived of
as a discourse separating disjunction.
However, it is widely held that the free-choice effect does not arise in all contexts,
or that it can be cancelled (Kamp 1978) as in (37) in which case we get an epistemic
reading with the modal used descriptively: the speaker does not know what is permit-
ted (by some other source).
(37) a. You may take an apple or a pear. But I don’t know which.
b. You may take an apple or you may take a pear. But I don’t know which.

Now it is interesting to see that for most, if not all, examples of conjunctive disjunc-
tion that I have given above the performative reading with its discourse format (28) is
most salient, but that an epistemic reading is also available. For instance, example (21),
repeated here, could also get an epistemic reading in the vein of (20).

.  Thanks to various commentators and an anonymous reviewer for raising this issue.

.  A more complete investigation has to wait for another occasion. But let me nevertheless
make two short comments: (i) Depending on the interpretation of the conditional permission
we may account for the felt exclusivity that the hearer may not take both apple and pear at the
same time. (ii) Notice however that this sketchy idea depends on taking the long form (35b)
as basic for the free-choice effect and the short form (35a) as elliptical. This is at odds with the
observation that the free-choice reading is not the preferred reading of the long form (35b).
 Michael Franke

(21) Therefore, we must start to be coherent in our intentions or we will be indulging

in fine theoretical discourses which have absolutely no substance outside this
Chamber.
(Europarl (en) (EU-EN), 2698860)

In this case, the speaker of (21) would say that the true state of affairs is one out of two
possible candidates: either coherence is required or the addressees will be indulging in
a particular form of discourse.
It seems, then, that conjunctive disjunctions have performative readings, not only of
certain modals, but also of explicit performatives (23), commitments (24), permissions
(25) etc. All of these may in principle also have an epistemic reading, no matter how im-
plausible the context would be that would give us such a reading. So the relation between
free-choice disjunction, whose leading examples are permissions and obligations, and
conjunctive disjunctions seems to be that the former are a special case of the latter.
Do we need a specialized semantics of disjunction as a conjunction of modalized
propositions in the vein of Zimmermann (2000) or Geurts (2005)? Not necessarily, it
seems.19 For the difference between the performative and the epistemic reading need
not necessarily lie in the semantics of the disjunction. The difference may simply be the
difference between a performative or a descriptive use of some linguistic form.20
In case of a performative use of the disjunction “X or Y” we associate with X some
speech-act α(χ) with the content χ. It is this χ that is then filtered out as the epistemic
content of the speech-act α(χ) and taken up by the disjunction. In contrast to this,
in case of a descriptive reading of the disjunction “X or Y” the whole propositional
content of X is the epistemic unit that the disjunction is sensitive to. It is then perhaps
not too far-fetched to assume that in both performative and descriptive readings of a
disjunction “X or Y” the semantics of the disjunction are the exact same, only that dif-
ferent propositions are considered the epistemic content of the first connect, depend-
ing on its use. So for a performative reading of X in “X orY” we get (38a), while for a
descriptive reading of X in “X or Y ” we may get (38b).
(38) a. α(χ) & inform(“χ or Y ”)
b. inform(“X or Y ”)

This behavior of or is plausible, if we think of it as a connector of epistemic possibili-

ties: disjunction seems sensitive to the epistemic minimal unit (cf. Pasch et al. 2003)
of the first connect.21

.  In fact, it is hard to conceive how such an approach could deal with (27) where the
χ-component is not even mentioned in the first connect.

.  Whether an expression is interpreted as being used performatively or descriptively may very
well be a matter of whether the speaker is taken to be competent or not (cf. Zimmermann, 2000).
.  We are now in a position to explain the relation between or and otherwise: or may only
refer to the epistemic minimal unit of the first connect, whereas otherwise has much wider
 Pseudo-imperatives and other cases of conjunction/disjunction 

3.4  Embedded conjunctive disjunctions

Let us finally note a set of examples that slightly complicate the picture sketched so
far. An analysis of conjunctive disjunctions as discourse separating connections, i.e.,
as speech-act conjunctions, seems very plausible for all examples discussed. If correct,
examples of embedded conjunctive disjunctions will require special scrutiny.
(39) a. They are there, threatening campesinos that they must plant coca or they
will die.
(Europarl (en) (EU-EN), 7760017)
b. Aerie forces you to tell her you love her or the romance is over.
(http://www.gamebanshee.com/baldursgateii/npcs/ aerie.php, 26.6.2006)
c. That person showed me a letter from a banking institution asking for
additional guarantees or the institution would demand full payment of its
loan within 48 hours.
(Hansard House e (HANSARD_HOUSE_E), 8151944)

To account for embedded occurrences of conjunctive disjunctions the notion of dis-

course separation simply has to encompass reported speech.

3.5  Intermediate summary

In conclusion, conjunctive disjunctions are a highly interesting topic with a lot of chal-
lenges for semantics which are beyond the scope of this paper. At present our main
concern is still the PoPA. This section tried to corroborate the part of the CH that
claimed that there are discourse-separating disjunctions, by providing a wide range of
parallel cases of conjunctive disjunctions.

4.  The Generic Challenge

4.1  The problem of generic DisjPIs

So far we have made plausible an account of the PoPA based on the CH by pointing out
that conditional conjunctions and conjunctive disjunctions are phenomena that go well
beyond PIs. We should therefore accept the claim that some ConjPIs are conditional
conjunctions and that some DisjPIs are conjunctive disjunctions. What is left to argue
is that all DisjPIs are conditional disjunctions.

anaphoric possibilities. This is shown nicely by the following examples from Webber et al.
(2003):
i. a. If the light is red, stop, otherwise you might get hurt.
b. If the light is red, stop, or you might get hurt.
ii. a. If the light is red, stop, otherwise you may proceed.
b. ? If the light is red, stop, or you may proceed.
 Michael Franke

I have already indicated that this might not be clear for generic DisjPIs (13), re-
peated here for convenience:
(13) a. Working for a successful start-up is no fun. The atmosphere is so tensed.
Agree to everything he says, or your boss fires you immediately.
b. Speak at least six languages or you are not a cosmopolitan.

The problem with generic DisjPIs is that it is not entirely clear what discourse format
generic DisjPIs have, in particular whether they are discourse separating connec-
tions or not. Intuitively, we would probably like to conceive of them as one commu-
nicative event, but then not as a statement with mere conditional content, but rather
as a single goal-oriented directive that presents something as necessary or required
for a particular purpose. But if generic DisjPIs are to be interpreted as one commu-
nicative event only, albeit some sort of informative directive, this contradicts the CH
in its present formulation which claimed that all DisjPIs are discourse separating
connections.
In order to decide the question whether generic DisjPIs are discourse separating
or not, it does not suffice to rely on intuition alone. Therefore I would like to consider
evidence for and against semantic subordination in conditional conjunctions and con-
junctive disjunctions that was proposed by Culicover and Jackendoff (1997). I will
extend their discussion with special emphasis on generic DisjPIs, but conclude that
the evidence is not decisive.

4.2  Pseudo-coordination vs. asymmetric coordination

Since we are in doubt whether all occurrences of DisjPIs are discourse separating, we
should not speak of conjunctive disjunctions when referring to all of them. So, let us
call the use of or in all DisjPIs and other clear cases of conjunctive disjunction more
generally ‘explanation-or’.
Culicover and Jackendoff (1997) argue that conditional conjunctions are pseudo-
coordinate, i.e., coordinate in syntax, but subordinate in conceptual structure, while
explanation-or is, though asymmetric, coordinate on both levels of analysis. Two ar-
guments are given for why conditional conjunction is a subordination in semantics,
but explanation-or is not. These two arguments rest on observations concerning NPI-
licensing (see section 4.2.1) and the possibility of cataphoric binding by quantifiers
across connects (see section 4.2.2), respectively.22

.  Culicover and Jackendoff assume that if a connection behaves like a conditional with re-
spect to NPI-licensing and cataphoric binding, then this is an argument for it being a case of
subordination semantically. It is a further step to take conditional-likeness as evidence for dis-
course integration, but this is what I will do. If this further step is incorrect, so much the better
for the Connector Hypothesis.
 Pseudo-imperatives and other cases of conjunction/disjunction 

4.2.1  NPI-licensing
The first argument for semantically subordinate and, but coordinate or in PIs is that
only the former allows negative polarity items (NPIs) in the first connect. (40) is Culi-
cover and Jackendoff ’s (53c).
(40) Say anything and/*or I’ll call the police.

As conditionals quite clearly allow NPIs in their if-clauses, Culicover and Jackendoff
argue that this parallel is evidence for subordination in case of conditional conjunction
and for coordination in case of explanation-or.
However, the impossibility of NPIs in the first connect of explanation-or is not
an argument against subordination. This is so, because the parallel in NPI-licensing
behavior between conditionals and PIs is actually more complicated than suggested by
Culicover and Jackendoff ’s argument (cf. Lawler, 1975). NPIs are licensed in ConjPIs
only if the imperative connect gets a negative or, to some minor extent, a generic in-
terpretation. This then is entirely parallel to the NPI-licensing in the if-clauses of the
corresponding conditionals, as evidenced in (41) and (42).
(41) a. If you say anything to anyone about this, I’ll kill you.
b. Say anything to anyone about this and I’ll kill you.
(42) a. ? If you say anything to anyone about this, I will give you my Ferrari.
b. ? Say anything to anyone about this and I will give you my Ferrari.

The crucial point here is that not all conditionals license NPIs in their if-clause. This
was observed by Lakoff (1970) who gave the contrast pair in (43), which shows how
NPI-licensing has a pragmatic dimension and is susceptible to the kind of argument
that is being made with the assertion of a conditional.
(43) a. If you eat any loxo, I’ll {batter you / ??give you whatever you like}.
b. If you eat some loxo, I’ll {?batter you / give you whatever you like}.

But then, if there are conditionals which do not license NPIs in their if-clauses due
to some rhetorical effect of NPIs, then the impossibility of NPIs in the first connect
of explanation-or-connections should not be an argument against subordination, un-
less one were willing to claim that also certain conditionals are not subordinate in
semantics. Rather, NPI-licensing should be dismissed as a criterion for semantic sub-
ordination for these reasons, or should at least not be counted as an argument against
a subordination in the semantics of explanation-or-connections.

4.2.2  Cataphoric binding

Culicover and Jackendoff (1997, pp. 203 – 4) present interesting binding data to support
the idea that conditional conjunction is subordinating in conceptual structure, while expla-
nation-or is not. They observe that conditional conjunction allows for cataphoric binding
by a quantifier phrase with every across the connects, just as conditionals do (44a), but that
apparently explanation-or does not (44b) (=(53b) in Culicover and Jackendoff 1997).
 Michael Franke

(44) a. If you give himi enough bribes, every senatori will vote for the president’s
proposal.
b. Give himi enough bribes and/*or every senatori will vote for the president’s
proposal.

Possibilities of cataphoric binding by quantifiers across connects might prima facie

seem like a genuine test for subordinateness, but there is room for doubt that this is so.
Although the or-variant of sentence (44b) seems odd, especially after having read the
and-variant, this might be due to quite different reasons. Very likely we are inclined
to assume by default that senators need to be bribed to vote for the president’s propos-
al, not against. This scenario is made salient by the and-variant of (44b), but renders
the or-variant of (44b) unintelligible. In contrast to that, there are other examples of
explanation-or with cataphoric binding across connects where little doubt about their
acceptability arises. Drawing on juvenile gender stereotypes, for instance, (45a) was
judged acceptable by 7 out of 9 informants, marked by 2, while no one found it entirely
unacceptable on a scale including ‘acceptable’, ‘odd’, ‘marked’ and ‘unacceptable’. This
was, nota bene, the exact same distribution as that obtained for the conditional sen-
tence in (45b).
(45) a. Tell heri every now and then that you love heri, or every girli will leave you
sooner or later.
b. If you don’t tell heri every now and then that you love heri, every girli will
leave you sooner or later.

From this we should conclude, contra Culicover and Jackendoff, that at least under
certain circumstances cataphoric binding by quantifiers across disjuncts is possible.
As a matter of unfortunate fact, the situation is, as usual, far more complicated.
Cataphoric binding possibilities seem to depend crucially on the kind of modality
that is at stake. Unlike the examples in (45) where a generic relationship is expressed,
examples (46) relate to the concrete future course of events, i.e., they express a meta-
physical conditional relationship. The striking observation is that although 7 out of 9
informants found the conditional (46b) acceptable or odd, 7 out of 9 informants found
the DisjPI (46a) marked or unacceptable.
(46) a. Invite heri, or every girli from our rugby team will feel neglected.
b. If you don’t invite heri, every girli from our rugby team will feel neglected.

Moreover, in the group of 7 informants who found (46b) acceptable or odd, 5 found
it odd. This means that there is an acceptability contrast not only between DisjPIs and
conditionals in case of metaphysical modality, but also between generic and meta-
physical variants of either sentence.
One possible conclusion to be drawn from all this is that we have to be careful with
Culicover and Jackendoff ’s second test for subordination, as judgments seem to depend
on a variety of factors, such as default expectations and the kind of conditional relation-
ship that is expressed. Yet if we were to hold on to the test, which had some prima facie
 Pseudo-imperatives and other cases of conjunction/disjunction 

plausibility on its side, another possible conclusion is that there are some conditional-
like, subordinate instances of DisjPIs, namely those that express a generic idea.

4.3  The Connector Hypothesis revisited

I will leave the matter whether generic DisjPIs refute the CH undecided. But suppose
that generic DisjPIs are not discourse separating connections. If we nevertheless wanted
to account for the PoPA, we could resort to a weaker formulation of the CH which in-
cludes the possibility of discourse integrating DisjPIs. We could then try to defend that
such discourse integrating DisjPIs are not merely unbiased conditional statements, but
rather a conventionalized way of expressing purpose-related necessity. This would still
explain the preference bias and accord with intuition. But unfortunately, this weaker
hypothesis could no longer be made plausible by parallel uses of or alone. In its defense
we would have to find support for the claim that if a disjunction communicates a con-
ditional relationship it communicates purpose-related necessity.
In conclusion, if we accept that there are discourse integrating, conditional-like
uses of disjunction, like generic DisjPIs, the PoPA is replaced by a different problem,
namely the problem of conditional disjunctions: why do generic DisjPIs express pur-
pose-related necessity only and not also unbiased conditional relationships? Seen in
this light, the CH is a possible explanation with independent evidence for the bulk of
cases that fall under the PoPA but also gives rise to an interesting, more refined, follow-
up problem.

5.  Conclusion

This paper addressed PIs as sentential connections and presented their diverse dis-
course formats in detail. The Connector Hypothesis was advanced as the basis of a
possible solution to the Problem of Pragmatic Asymmetry.
In support of the CH, PIs were shown to be embedded in a broader context of
non-standard conjunctions and disjunctions. Conditional conjunctions and conjunc-
tive disjunctions presented themselves as overarching hitherto neglected linguistic
troublemakers with their own respective semantic and pragmatic challenges. This
paper contributed an in-depth classification of conjunctive disjunctions and showed
that the PoPA too is part of a larger whole, as all hypothetical DisjPIs display a double
preference bias.
Doubts about the generality of the CH finally arose from generic DisjPIs. These
examples appeared, contrary to the CH, discourse integrating connections with just
one informative communicative event. I concluded that if generic DisjPIs were indeed
discourse integrating, this would diminish the explanatory scope of the CH, but it
would also leave us with a new, smaller and more concrete explanandum which we
might hope is easier to account for than the one we started out with.
 Michael Franke

References

Asher, Nicholas & Lascarides, Alex. 2003. Logics of Conversation. Cambridge: CUP.
van der Auwera, Johan. 1986. Conditionals and speech acts. In On Conditionals, Elizabeth Closs
Traugott, Alice ter Meulen, Judith Schnitzer Reilly & Charles A. Ferguson (eds), 197–214.
Cambridge: CUP.
Bolinger, Dwight. 1979. Is the imperative an infinitive? In Meaning and Form, Dwight Bolinger
(ed.), 152–182. London: Longman.
Clark, Billy. 1993. Relevance and ‘pseudo-imperatives’. Linguistics and Philosophy 16: 79–121.
Culicover, Peter W. 1970. One more can of beer. Linguistic Inquiry 1: 366–369.
Culicover, Peter W. 1972. OM-sentences. Foundations of Language 8: 199–236.
Culicover, Peter W. & Jackendoff, Ray. 1997. Semantic subordination despite syntactic coordina-
tion. Linguistic Inquiry 28(2): 195–217.
Franke, Michael. 2005a. How and how not to employ discourse relations to account for pseudo-
imperatives. In Proceedings of the 15th Amsterdam Colloquium, Paul Dekker & Michael
Franke (eds), 83–88. Amsterdam: ILLC.
Franke, Michael. 2005b. Pseudo-imperatives. MA thesis, Universiteit van Amsterdam.
Geurts, Bart. 2005. Entertaining alternatives: Disjunctions as modals. Natural Language Semantics
13: 383–410.
Gómez-Txurruka, Isabel. 2002. The semantics of natural language disjunction or. Ms, ILCLI
Donostia-San Sebastián.
Han, Chung-Hye. 1998. The Structure and Interpretation of Imperatives. PhD dissertation,
­University of Pennsylvania.
Kamp, Hans. 1978. Semantics versus pragmatics. In Formal Semantics and Pragmatics for ­Natural
Languages, Franz Guenthner & Siegfried Josef Schmidt (eds), 255–287. Dordrecht: Reidel.
Kratzer, Angelika. 1991. Modality. In Semantik: Ein internationales Handbuch der zeitgenossis-
chen Forschung, Arnim von Stechow & Dieter Wunderlich (eds), 639–650. Berlin: Walter
de Gruyter.
Krifka, Manfred. 1995. Focus and the interpretation of generic sentences. In The Generic Book,
Gregory Norman Carlson & Francis Jeffry Pelletier, (eds), 238–264. Chicago IL: Chicago
University Press.
Lakoff, Robin. 1970. Some reasons why there can’t be any some-any rule. Language 45: 608–615.
Lawler, John M. 1975. Elliptical conditionals and/or hyperbolic imperatives: some remarks on
the inherent inadequacy of derivations. In Papers from the 11th Regional Meeting. Chicago
Linguistic Society, 371–382. Chigaco Linguistic Society.
Lewis, David. 1979. A problem about permission. In Essays in the Honor of Jaakko Hintikka,
Esa Saarinen, Risto Hilpinen, Ilkka Niinluoto & Merrill Provence Hintikka (eds), 163–175.
Dordrecht: Reidel.
Mann, William C. & Thompson, Sandra A. 1987. Rhetorical structure theory: A framework for
the analysis of text. IPRA Paper in Pragmatics, 1: 79–102.
Pasch, Renate, Brauße, Ursula, Breindl, Eva & Waner, Ulrich Herman. 2003. Handbuch der
deutschen Konnektoren. Berlin: de Gruyter.
Roberts, Craige. 1989. Modal subordination and pronominal anaphora in discourse. Linguistics
and Philosophy 12: 683–721.
Ross, Alf. 1941. Imperatives and logic. Theoria 25(7): 53–71.
 Pseudo-imperatives and other cases of conjunction/disjunction 

Russell, Benjamin. 2007. Imperatives in conditional conjunction. Natural Language Semantics

15: 131–166.
Schwager, Magdalena. 2006. Interpreting Imperatives. PhD dissertation, Johann-Wolfgang-Goethe
Universtität zu Frankfurt am Main.
von Wright, Georg Henrik. 1968. An Essay on Deontic Logic and the Theory of Action. Amsterdam:
North-Holland.
Webber, Bonnie, Stone, Matthew, Joshi, Aravind & Knott, Alistair. 2003. Anaphora and discourse
structure. Computational Linguistics 29(4): 545–587.
Zimmermann, Thomas Ede. 2000. Free choice disjunction and epistemic possibility. Natural
Language Semantics 8: 255–290.
 From discourse to “odd coordinations”
On Asymmetric Coordination and Subject
Gaps in German

Ingo Reich*
University of Tübingen

Subject gaps in coordinate structures (i.e., SLF-Coordination, or simply SLFC) and

Asymmetric Coordination (AC) in the sense of Höhle (1983, 1990) exhibit specific
semantic (a ‘fusing’ interpretation) and syntactic properties (e.g., verb fronting in
non-initial conjuncts). This paper ties together these characteristics by considering
them an immediate consequence of the fact that in AC and SLFC properties typically
attributed to ‘coordinate’ structures in discourse show up in ‘grammaticalized’
form. As far as the analysis of SLFC is concerned, it is furthermore argued that the
dropping of the subject (and only the subject) is in a sense parasitic on the special
fusing semantics of the construction. That dropping the subject systematically results
in an overt V1-structure is traced back to the way covert operators are licensed and
identified in specifier positions in German (paralleling Topic Drop).

Keywords: asymmetric coordination, subject gaps, ellipsis, event semantics

1.  Introduction

German (like some other Germanic languages, for example, Dutch) exhibits an
­interesting, and at the same time recalcitrant construction, which we can describe
somewhat sloppily as Subject Gaps in Coordinate Structures (SLFC).1 As it turns
out, SLFC is closely related to another kind of ‘non-standard’ coordination ob-
servable in German (see Höhle 1990): so-called Asymmetric Coordination (AC).

*  I’d like to thank Veronika Ehrich, Cathrine Fabricius-Hansen, Angelika Kratzer, Claudia
Maienborn, Jürgen Pafel, Wiebke Ramm, Marga Reis, Arnim von Stechow, Wolfgang Sternefeld,
Hubert Truckenbrodt, Angelika Wöllstein and three anonymous referees for helpful comments
and related discussion. Thanks to Kirsten Brock for checking my English.
.  The term “SLF-coordination” is due to Höhle (1990) and is shorthand for “Subject Lacking
in F-structure,” where “F-structure” in turn stands for “Fronted structure,” a technical term that
refers to sentences which show overt verb fronting; see Höhle (1986) for details.
 Ingo Reich

While previous approaches to SLFC and AC have focused almost exclusively on

their syntactic side (Reis 1993, and Frank 2002, being notable exceptions), this pa-
per takes as its starting point the constructions’ semantic and pragmatic peculiari-
ties, and seeks to pull these together with their more formal ­characteristics. I will
argue that properties typically attributed to coordination in discourse show up in
AC and SLFC in ‘grammaticalized’ form, and thus determine both the construc-
tions’ ­specific semantic and syntactic properties (among other things, verb fronting
in non-initial ­conjuncts). Before laying out the details of my proposal, however, let
me first introduce these characteristics of SLFC and AC, characteristics that in par-
ticular delineate this construction from what we may call ‘standard’ (symmetric)
coordination.

1.1  Subject gaps in coordinated structures (SLFC)

With this aim in view, consider the particularly interesting example (1), taken from
a German online newspaper. Following Höhle (1983) SLFC can be characterized by
three distinct properties: (i) Within non-initial conjuncts the finite verb is fronted; see
tut (‘pretends’) in (1). (ii) Even though property (i) tells us that SLFC is most likely
some case of sentential coordination (verb fronting requires a left periphery), the non-
initial conjunct lacks a subject.2
(1) da stellt sich jemand vor die Mikrofone und
there stands refl somebody in front of the microphone and
tut so, [als könne er etwas erklären]
pretends part, [as if could he something explain]
‘Somebody stands there in front of the microphone, and pretends to be able
to explain something’ (zeit.de, 26 January 2006)

But it is property (iii) that distinguishes SLFC from phrasal C’-coordinations like the
one in (3), and which establishes SLFC as a separate phenomenon: Whereas in the case
of (3) the subject jemand (‘somebody’) – which fills the ‘shared’ prefield (“Vorfeld”)
position left adjacent to the position of the fronted finite predicate – c-commands
(and, thus, binds) the base positions t1 of the fronted subject in both conjuncts, this is
not the case with SLFC; see (2).
(2) da [[stellt sich jemand vor die Mikrofone] und [tut SL so, …]]
(3) jemand1 [[stellt t1 sich da vor die Mikrofone] und [tut t1 so, …]]

.  In a recent paper, Fortmann (2005) argues that it is not the subject per se, but in fact the
highest argument of the verb (which is not necessarily the subject in German) that is dropped in
SLFC; see, however, Reich (2007a) for critical discussion and a rejection of this position.
 Odd coordinations 

In (2) the subject jemand of the first conjunct is located after [!] the fronted finite
­predicate stellt (‘stands’), i.e., within the middle field (“Mittelfeld”) of the first ­sentence.3
If we now try to analyze (2) along the lines of (3), we need to assume that (2) – like (3) –
is a symmetrically coordinated V1-structure sharing a common prefield filled with da
(‘there’). This time, however, the subject of the first sentence does not ­c-command the
base position SL of the ‘dropped’ subject in the second conjunct, which means that we
cannot construe the subject gap SL in (2) as a trace that is bound by the subject of the
first sentence. In other words, given phrasal C’-coordination, the subject gap in (2), in
contrast to the one in (3), cannot be taken to be the result of ‘simultaneously’ ­moving
the subject jemand out of both conjuncts – a movement process which is known as
“across the board (ATB) movement” (see Williams 1978).
There is one alternative, though, that we have to take into consideration here. If
we suppose that examples like (3) are not the result of C’-coordination followed by
ATB-movement, but that they are the result of CP-coordination followed by deletion
of the subject, then we might still put forward a uniform analysis in terms of ellipsis
(see, e.g., Wilder 1997): Both structures could be taken to display instances of so-called
“forward deletion,” an ellipsis process deleting strings at the left periphery of non-
initial conjuncts under identity. In the case of (2) and (3) this amounts to saying that
(i) the subjects in non-initial conjuncts are to be located in the prefield, and (ii) that
they are deleted in this position under identity with the subject in the first conjunct;
see (4) and (5).
(4) [[da stellt sich jemand vor die Mikrofone] und [jemand tut …]]
(5) [[jemand stellt sich da vor die Mikrofone] und [jemand tut …]]

There is good evidence, however, that the lack of the subject in (1) cannot be due to
“forward deletion” or some similar kind of deletion process. In fact, it is all but clear
whether such a deletion process exists at all (see, e.g., Höhle 1991; Hartmann 2002). If
it does, it certainly does not behave in crucial respects the way clear cases of ellipsis in
coordinate structures do.
To see this, contrast the interpretation of elided indefinites in Gapping – see (6)
below – with the interpretation of elided indefinites in structures like (4) and (5) above.
Whereas in the case of Gapping the interpretation of the indefinite etwas (‘something’)
remains constant irrespective of whether etwas is realized overtly or covertly – in both
cases, etwas in the second conjunct refers (in its preferred reading) to a different object
than etwas in the first conjunct – this is not true of (4) and (5) above: In (4) and (5), it
is claimed that there is one person who both stands in front of the microphone and
who pretends to be able to explain something, i.e., the elided indefinite jemand is

.  This is, of course, one of the reasons why in subject-initial languages like, for example, Eng-
lish SLFC seems not to exist: the only syntactic structures in English that comply with property
(iii) are inverted structures as used in English interrogatives.
 Ingo Reich

systematically interpreted as being ‘coreferent’ with its antecedent. Their overt coun-
terparts in (7) and (8), however, are far more liberal: they may be, and in fact are taken
as attributing the properties of standing in front of the microphone and of pretending
to be able to explain something to two different persons.
(6) Peter schenkte Anna etwas und Anna schenkte Peter etwas
Peter gave Anna s’thing and Anna gave Peter s’thing
‘Peter gave Anna something, and Anna Peter’
(7) da stellt sich jemand vor die Mikrofone und jemand tut so, …
(8) jemand stellt sich da vor die Mikrofone und jemand tut so, …

Besides, “forward deletion” is subject to another important formal restriction not

mentioned yet: It is not only the target of deletion that has to be positioned at the
left periphery of its conjunct, but also its antecedent (see, e.g., van Oirsouw 1993).
This, however, immediately contradicts defining property (iii) of SLFC; and somewhat
loosening that restriction would obviously lead to considerable overgeneration. Now,
since the lack of the subject in non-initial conjuncts in SLFC seems to be due neither
to ellipsis nor to ATB-movement, this property of SLFC remains – for the time being
– quite puzzling.

1.2  Asymmetric Coordination (AC)

As Höhle (1990) observes, SLFC seems not to be restricted in its use to root clause
phenomena like (1) above, but also occurs in dependent contexts as, for example, in
the antecedent of a conditional; see (9).
(9) Wenn du nach Hause kommst und siehst den Gerichtsvollzieher vor
if you back home come and see the bailiff at
der Tür, dann […]
the door, then […]
‘If you come home, and you see the bailiff at the door, then […]’

Since SLFC requires fronting of the finite verb in non-initial conjuncts, we encounter,
however, an unexpected word order asymmetry: While the finite verb is fronted in
the second conjunct (overt V1-position), it is – because of the complementizer wenn
(‘if ’) – in final position (VE) in the first conjunct. This kind of word order asymmetry
is in fact more general and is also found in coordinations that do not lack a subject in
the second conjunct; see (10).
(10) Wenn du nach Hause kommst und der Gerichtsvollzieher steht
if you back home come and the bailiff stands
vor der Tür, dann […]
at the door, then […]
‘If you come home, and the bailiff is standing at the door, then […]’
 Odd coordinations 

Coordinations that show this kind of word order asymmetry have been dubbed
­“Asymmetric Coordination (AC)” – in a narrower sense – in Höhle (1990). As it turns
out, SLFC and AC share many crucial properties (see Reich 2007a, for a detailed dis-
cussion), the two most prominent probably being (i) verb fronting, see (11a), and (ii)
a fusing interpretation, see (11b).

(11) a. Verb fronting

In non-initial conjuncts the finite verb is fronted.
b. Fusing semantics
SLFC and AC are subject to a ‘one-event’ interpretation.

That AC as well as SLFC do in fact require fronting of the finite predicate in non-initial
conjuncts is easily illustrated by the following pair of examples:

(12) a. *Kommst du nach Hause und der GV vor der Tür steht, […]
b. *Da stellt sich jemand vor die Mikrofone und so tut, […]

The fact that AC and SLFC are subject to some kind of one-event interpretation is
somewhat harder to illustrate. Höhle (1983: 22) gives the contrast in (13).

(13) a. Stehen da wieder welche rum und verteilen Flugblätter?

stand there again folks around and hand out flyers?
‘Are, again, folks standing around and handing out flyers?’
b. Stehen da wieder welche rum und verteilen sie FB?
stand there again folks around and hand out they flyers?
‘Are, again, folks standing around and are they handing out flyers?’

Whereas the SLFC in (13a) is conceived of as a single question focusing on the distri-
bution of flyers, the symmetric coordination in (13b) is understood as two basically
independent questions asking about two different things.
It is tempting here to blame the overt subject in (13b), for it seems that VP-
coordination also shows some kind of one-event interpretation; see the pair of ex-
amples in (14), modeled after an example taken from Reis (1993).

(14) a. Jetzt geht er in Buchhandlungen und liest Neuerscheinungen.

now goes he in bookstores and reads new publications
‘Now, he goes in bookstores and reads new publications’
b. Er geht jetzt in Buchhandlungen und liest Neuerscheinungen.
he goes now in bookstores and reads new publications
‘Now, he goes in bookstores and reads new publications’

There is one difference though, and this difference is crucial: In the case of SLFC
the one-event interpretation is obligatory, whereas it is only optional in the case of
VP-coordination. This is – for reasons related to sentence mood – best illustrated with
SLFC in a conditional clause. Consider (15).
 Ingo Reich

(15) a. *wenn er Neuerscheinungen liest und geht in Buchhandlungen

if he new publications reads and goes in bookstores
b. wenn er Neuerscheinungen liest und in Buchhandlungen geht
if he new publications reads and in bookstores goes

In (15), the conjuncts are switched, which has the effect of blocking a one-event
­interpretation (since it is a natural course of events to go in a bookstore and read a
new publication, but not vice versa). This change in interpretation does not affect sym-
metric VP-coordination in (15b), but it does affect asymmetric SLFC in (15a), which
is clearly marginal, if not ungrammatical.
This shows several things: (i) The one-event interpretation associated with SLFC
and AC cannot simply be reduced to whatever mechanism is responsible for one-event
­interpretations in the case of symmetric VP-coordination. (This seems obvious with
AC.) (ii) Since the one-event interpretation in AC and SLFC is obligatory, it is most likely
a semantic (and not a pragmatic) property. (iii) Since the other characteristic property of
AC and SLFC – fronting of the finite verb in non-initial conjuncts – is obligatory, too, it
seems quite obvious that both properties are in fact two sides of the same coin.
If this is correct, we have to answer the question of how to characterize the fusing
semantics of AC and SLFC, and how this semantics is linked to verb fronting (in AC
and SLFC). This is the topic of the following section.

2.  From discourse to “odd coordinations”

But where to start when trying to characterize the fusing semantics of AC and SLFC?
The fact that AC and SLFC frequently occur with conjunctional adverbs like dann
(‘then’), dabei (‘thereby’), etc., which explicitly refer to coherence relations connecting
utterances in discourse, suggests that a somewhat closer look at possible coherence
relations may be promising here.

2.1  Discourse structure and the “Occasion” relation

A more recent classification of coherence relations is found in Kehler (2002).4 Kehler
(2002), building on work done by David Hume and Jerry Hobbs, subdivides coherence
relations into three different classes; see table 1.

Table 1.  Classification of coherence relations in Kehler (2002)

Resemblance Cause-Effect Contiguity
Parallel Result Occasion
Contrast Explanation
… …

.  But see Kehler (2002) and this volume for (a discussion of) alternative classifications.
 Odd coordinations 

Whereas resemblance relations and cause-effect relations are construed as relations be-
tween propositions – structured and unstructured, respectively – contiguity ­relations
are to be conceived of according to Kehler (2002) as relating events. The class of con-
tiguity relations is in fact a singleton class consisting only of the coherence ­relation
occasion, which is informally characterized as follows: “Occasion can be seen as a
mechanism for communicating a complex situation in a multi-utterance discourse by
using states of affairs as points of connections between partial descriptions of that situ-
ation” (Kehler 2002: 22). This seems to be exactly what we are looking for. So let’s have
a closer look at the formal definition. Kehler (2002: 22) gives the following character-
ization of the occasion relation (see also Hobbs 1990):
Occasion (i): Infer a change of state for a system of entities from S1, inferring
the final state for this system from S2.
Occasion (ii): Infer a change of state for a system of entities from S2, inferring
the initial state for this system from S1.

Kehler (2002: 23), however, considers this to be only a first approximation to the
­intuition underlying occasion, since “temporal progression in the absence of a com-
mon scenario connecting the events is insufficient in and of itself.”
In this respect, I completely agree with Kehler (2002). What we do not agree
upon is his classification of coherence relations. The fact that (i) the class of conti-
guity ­relations is a singleton, and (ii) the occasion relation relates events rather than
­propositions ­suggests that occasion is in fact not a coherence relation in the strict
sense, but a discourse phenomenon operating on a different level, interacting with
genuine coherence relations. If this is correct, we end up with a much simpler and
homogeneous classification of coherence relations; see table 2.

Table 2.  Coherence relations as relations between (un)structured propositions

Structured Coherence Unstructured Coherence
Parallel Result
Contrast Explanation
… …

This, of course, does not discharge us of specifying exactly what we mean when talk-
ing about “occasion”.5 To get a better idea of what “occasion” is all about, consider the
following sequence of utterances:
(16) a. At 6:30 a.m., the lights are turned on in house No. 3.
b. Then, at 6:32 a.m., the lights are turned on in house No. 10.
c. Then, at 7:01 a.m., the sun comes up.

.  In the following, the typographical difference between occasion and “occasion” is meant to
reflect the aforementioned different theoretical perspectives on the occasion relation.
 Ingo Reich

There are two ways to interpret the sequence of utterances in (16): (i) simply as a se-
quence of events e1, e2, e3, connected by temporal progression, or (ii) as a complex event
e – say the ‘awakening’ of (for example) Downing Street in London – consisting of
the subevents e1, e2, e3 as perceived of by some person observing the street, and what
happens there. What we are interested in is, of course, scenario (ii). Another way to
describe what’s going on in (16) under scenario (ii) is as follows: The first sentence
(16a) sets the scene by describing part of a possibly complex event e; the second sen-
tence (16b), speaking about another event e2, in a sense ‘extends’ the description of e by
subordinating e2 to e. This process of ‘event subordination’ is then reiterated with (16c),
resulting in a complex event e consisting of (at least) the events e1, e2, and e3.6 We thus
can define “occasion” as some kind of event subordination as follows:
(17) Occasion (event subordination)
Occasion is to be construed as event subordination, i.e., for any two events e1, e2,
Occasion(e1, e2) holds iff it is the case that e2 ≤ e1.

Schematically, the situation in (16) can be depicted as follows:

(18) e

S1 (conj) S2 (conj) S3 … (discourse syntax)

p1 co-rel p2 co-rel p3 … (coherence relations)
e ≥ e2 ≥ e3 … (event subordination)

The basic idea is, thus, clear: Sentence S1 introduces an event variable e. If the discourse
stops here, e denotes the minimal event described by the sentence S1, i.e., the event e1.
If, however, an utterance S2 is added, and S2 is meant to continue the description of a
(complex) event introduced by S1, then the minimal event e2 referred to by S2 is taken
to be a subevent of e – which is guaranteed by the condition e2 ≤ e on e2 and e. This is,
I think, exactly what “occasion” does: While “occasion” ensures that S1 and S2 build a
complex structure in the first place, coherence relations like contrast, result, etc. specify
the way the propositions denoted by S1 and S2 relate to each other.
As an alternative to event subordination, one could think of (recursively) merging two
events e1 and e2, i.e., we could define a merging operation ‘⊕’ which gives us for e­ very

.  The term ‘event subordination’ is reminiscent of the term ‘modal subordination’, which is
introduced in Roberts (1989). In fact, event subordination and modal subordination do share
some important formal characteristics. However, as the following contrast between the modal
subordination example in diesem Haus lebt kein Mann oder es ist gut versteckt (‘There is no man
living in this house or is well hiddden’) and the SLFC construction *in diesem Haus lebt kein
SL ist gut versteckt shows, modal suboradination and event subordination are in principle two
independent phenomena; see Reich (2007a) for further discussion.
 Odd coordinations 

e1 and e2 a complex event e = e1 ⊕ e2.7 The problem with such an approach, ­however, is
that it is all but clear how to define ‘⊕’: What exactly does it mean to merge two events?
And given two arbitrary events e1 and e2, is there always a complex event e which can
be considered to be the result of merging e1 and e2? Event subordination avoids such
problems in a transparent way: First of all, the subevent relation ‘≤’ is (one of) the most
basic relation(s) in any reasonable event mereology; see, e.g., Schein (1993). Second,
the subevent relation ‘≤’ can be defined by spatial-temporal inclusion, without losing
any of the events’ internal structure: This is simply because the event variable e always
directly refers to and is conceptualized as an immediately accessible event e, ­without
having to build e from other events e1 and e2.

2.2  Discourse structure and verb order in German

So much for the specification of the Occasion relation in discourse. But how does this
relate to AC and SLFC? To see this, consider examples (19) and (20). It is common
knowledge that the prototypical verb order of an independently used clause in Ger-
man is V2 or V1, see (19a–b), whereas the prototypical verb order of dependently used
clauses is VE, see (20).
(19) a. Lucky Luke sprang auf sein Pferd
Lucky Luke jumped onto his horse
b. Sprang Lucky Luke auf sein Pferd?
jumped Lucky Luke onto his horse?
(20) Rantanplan wußte, dass L.L. gleich auf sein Pferd springt.
Rantanplan knew, that L.L. part onto his horse jumps
‘Rantanplan knew that L.L. was about to jump onto his horse’

This, of course, is not to say that it is absolutely impossible to use VE-clauses in ­German
independently – as (21) shows this is quite possible:
(21) Dass Lucky Luke ständig auf sein Pferd springt!
that Lucky Luke always onto his horse jumps!

There are important restrictions on the use of such clauses though: First of all, they
never carry the force of an assertion, irrespective of whether they are to be classified as
declarative clauses or not. Second, they are so to speak ‘islands in discourse’, i.e., they do
not directly relate to previous utterances in the usual sense. This suggests that there is an
intimate connection between establishing “occasion” (and “occasion”-related ­coherence
relations) on the one hand and word order on the other: Fronting the finite predicate

.  A merging operation along these lines has been proposed, for example, in Eckhardt
(1998).
 Ingo Reich

seems to be a necessary – though not necessarily sufficient (see, for example, the case of
embedded V2-declaratives in German) – condition on establishing “occasion”.
From this point of view, the fact (i) that both AC and SLFC always show fronting
of the finite predicate in non-initial conjuncts, and (ii) that this behavior systemati-
cally correlates with the presence of some kind of fusing semantics does not look like
a coincidence. The question, then, is in what way these processes – one syntactic, the
other ‘semantic’ – relate to each other. Though I’m not (yet) in a position to give a
general answer to this question, I do have a concrete proposal concerning the special
case of SLFC and AC.

2.3  The functional projection “OccP”

The crucial idea of the analysis to follow is that in AC and SLFC the two conjuncts
are in fact systematically linked by the “occasion” relation, which we know well by
now from complex structures in discourse. Whereas in discourse, however, the linking
of two utterances with the help of “occasion” is optional, and thus most likely prag-
matically mediated, this is not the case with AC and SLFC. As we saw above, the fus-
ing semantics of AC and SLFC – now construed as event subordination triggered by
­“occasion” – is an essential characteristic of these constructions. Therefore, establish-
ing “occasion” in AC and SLFC seems to be a semantic rather than a pragmatic pro-
cess.8 But if it is semantic, there needs to be some syntactic expression or feature that
introduces the semantics of “occasion” into the structure of AC and SLFC. Let’s call
this feature for lack of a better term [occ], which is obviously short for “occasion”. In
the following, I will assume that this very feature ­projects – along with the feature [f]
introduced below – a functional projection OccP (more precisely, OccP/FP). In this
sense, “properties typically attributed to coordinate structures in discourse show up in
AC and SLFC in ‘grammaticalized’ form.”
Syntax. Before laying out the details of “OccP,” however, something needs to be
said about the specifics of verb movement and sentence structure in German. In the
following, I will assume – following recent work in syntax – that verb movement in
German is triggered by a (strong) syntactic feature [f] that needs to be checked by a
finite verb in overt syntax. It is important to see here that this feature is neither directly
connected to force nor to sentence mood, for there are also – as mentioned above –
dependent uses of V1- and V2-sentences in German. I, therefore, take it that [f] is
uninterpretable.9 I furthermore assume that [f] is the head of a functional projection
FP, which distinguishes V1- and V2-sentences from VE-sentences (which are headed
by the complementizer projection CP only). A simple sentence like Jan verpetzt Maria
(“Jan tells on Mary”) is thus analyzed along the lines of (22).

.  There is no need, however, to assume that coherence relations are established semantically.

.  But see Truckenbrodt (2006) for a possible interpretation of V2 in German.

 Odd coordinations 

(22) a. Jan verpetzt Maria. (“Jan tells on Maria”)

b. FP

DPi F’

Jan F vP

verpetzt ti v’

v VP

DP V

Maria tv

The syntactic structure in (22b) furthermore illustrates that I essentially follow Larson
(1988) and others in locating subjects within a functional projection vP, i.e., external
to VP. I cannot enter into a discussion on the VP-internal vs. VP-external hypothesis
here for obvious reasons; in what follows, however, it will become apparent why this
seems to be the most adequate way of treating subjects – at least from the perspective
of AC and SLFC.
After this brief excursion on sentence structure in German, let’s come back to the
analysis of AC and SLFC. As I have already mentioned several times, one of the crucial
characteristics of AC and SLFC is the V1/V2-property of non-initial conjuncts. But
what exactly is it that triggers fronting of the finite verb in these cases? As we argued
above, this property seems to be intimately connected to the process of establishing
“occasion” between two sentences. In other words, it is the feature [occ], denoting
the “occasion” relation, which is to be made responsible for this behavior. In terms
of a feature-based syntax, this is – given our assumptions about verb movement –
­essentially equivalent to saying that the feature [occ] selects for the feature [f]; that is,
we have (23).
(23) Verb fronting in AC and SLFC
[occ] selects for [f].

If the feature [occ] projects a functional projection, and if it furthermore selects

for the feature [f], there are in principle two syntactic configurations where we can
expect [occ] to occur: Either it projects a functional projection of its own, thereby
 Ingo Reich

taking FP as its syntactic complement, see (24a), or [occ] attaches to the head F, and
they project a functional projection OccP/FP, see (24b).
(24) a. [OccP Spec [Occ’ Occ [FP Spec [F’ F [ …
b. [OccP/FP Spec [Occ’/F’ Occ/F [ …

As far as I can see, there is no compelling independent evidence for the more elaborate
structure in (24a). I, therefore, assume that non-initial conjuncts in AC are assigned
the syntactic structure in (24b). If we now reconsider example (10) from section 1.2
above, repeated here as (25) for convenience,
(25) Wenn du nach Hause kommst und der Gerichtsvollzieher steht
if you back home come and the bailiff stands
vor der Tür, dann […]
at the door, then […]
‘If you come home, and the bailiff is standing at the door, then […]’

we end up with a syntactic analysis along the lines of (26).

(26)

CP

CP OccP/FP

wenn…kommst und OccP/FP

DPi Occ’/F’

der GV Occ/F vP

steht ti vor der Tür tv

Two comments seem to be in order here. First, whatever the correct analysis of ‘standard
symmetric coordination’ is (see, e.g., Progovac 1998; Reich 2007a), the use of und (‘and’)
in AC and SLFC seems to be somewhat different. This is apparent from the fact that
‘standard (symmetric) coordination’ allows for ATB-movement, while AC and SLFC do
not (Höhle 1983). I, therefore, follow Büring and Hartmann (1998) in analyzing coor-
dination as adjunction in the case of SLFC (and AC). As we will see in a moment, this
assumption is in fact an immediate consequence of the semantics of [occ].
 Odd coordinations 

The other comment concerns the adjunction site of OccP/FP. In the case of
­ ependent uses of AC and SLFC as in (25/26), OccP/FP seems to (have a strong
d
­tendency to) adjoin sentence-externally, i.e., to adjoin to the whole CP. Even though
this is somewhat unexpected from a semantic point of view, there is good evidence
from the binding of pronouns (quantifiers in the first conjunct do not easily bind
­pronouns in the second conjunct) and scope of negation (the scope of negation is
restricted to the conjunct it is part of) that this is in fact correct; see Reich (2007a,b).
Here, I just want to touch on this matter by illustrating the behavior of negation in this
respect. Consider (27).

(27) #Wenn du nicht die Kronjuwelen klaust und (du) wirst dabei
if you not the crown jewels steal and (you) get thereby
erwischt, dann hast du Glück gehabt
caught, then have you luck had

From a pragmatic point of view, the only reading that makes sense here is the one with
wide scope of negation: If it is not the case [that you steal the crown jewels and you
get caught red-handed], then you are lucky. But as a matter of fact, this reading is not
available in (27). The only reading available is the self-contradictory one with scope of
negation restricted to the first conjunct: If [it is not the case that you steal the crown
jewels] and you get caught red-handed, then you are lucky. This is consistent with the
assumption that the second conjunct attaches to CP. Because of space limitations, I
cannot go into details here, but the reader is referred to Reich (2007a,b) for further
discussion, including possible implications for the semantics of conditionals.
Semantics. Having introduced the main syntactic properties of [occ], let’s have a
closer look at its semantics. We already argued in some detail that [occ] denotes the “oc-
casion” relation, i.e., that it implements event subordination. If we want to spell out this
idea in a formal setup, we need to assume – in accordance with the literature on lexical
semantics and ‘Aktionsarten’ in the tradition of Donald Davidson – that each verb intro-
duces an event variable e, i.e., that sentences are taken to denote sets of (possible) events.
Given these assumptions, we can specify the semantics of [occ] as given in (28).

(28) Semantics of [occ]

||[occ]|| = λQλ¡λPλe∃e2[ ¡(Q(e2))(P(e)) & e2 ≤ e ]

According to (28), the feature [occ] implements two conditions: (i) The second
c­ onjunct Q is conjoined with the first conjunct P by some appropriate coordinating
conjunction ¡; this relates to the condition ¡(Q(e2))(P(e)), which can be rewritten
as “P(e) & Q(e2)”, if ¡ is instantiated by and. (ii) The feature [occ] furthermore states
that the event e2 corresponding to the second conjunct Q is a subevent of the event e
introduced by the first conjunct P. While condition (i) prevents the coordinating con-
junction ¡ from projecting its own functional projection (otherwise we run into a type
mismatch), and thus is to be made responsible for the adjunction structure of AC and
SLFC, condition (ii) implements the by now familiar event subordination “e2 ≤ e”.
 Ingo Reich

If we take (essentially following Kratzer 1991) the complementizer wenn (‘if ’) to

be semantically empty, it is straightforward to see that the proposed semantics derives
the interpretation given in (29) for the AC in (26), where CBH stands for the complex
predicate “come back home,” and SAD is shorthand for the other complex predicate
“stands at the door.”

(29) λe∃e2[ [CBH(e, you) & (SAD(e2, bailiff )] & e2 ≤ e ]

According to (29), the AC in (26) denotes that set of (complex) events e which are
­directly characterized as ‘you are coming back home’ events – because of the condition
CBH(e, you) – and indirectly as being at the same time also ‘the bailiff is standing at the
door’ events – because of the condition SAD(e2, bailiff ) mediated by event subordination
e2 ≤ e. This seems to be a good and reasonable approximation to the fusing semantics
of AC and SLFC.
The ‘Vorfeld’ of non-initial conjuncts in AC. Before switching over to SLFC let
me point out an interesting prediction of this approach that concerns overt realiza-
tions of the prefield of non-initial conjuncts in AC. Following Frey (2004), I assume
that the German prefield can be filled in three different ways: (i) by base generation
(of some specific elements, e.g., the prefield-es), (ii) by formal movement (FM),
and (iii) by real A-bar-movement. Whereas real A-bar-movement is semantically
­motivated by some interpretable feature in the left periphery ([topic], [contrast],
[wh], etc.) and may, in principle, start from any position within the middle field,
FM is not triggered semantically (it is just ‘formal’), and it is restricted to short
movement of the leftmost expression in the middle field (which is quite often, but
not always the subject).
Now, what about the prefield of non-initial conjuncts in AC? As we saw above,
the prefield is, in this case, not simply the specifier position of some FP, but the
specifier position of an OccP/FP; and since the feature [occ] is interpretable, we
­expect movement to [Spec,OccP/FP] to be restricted as follows: While base genera-
tion and FM to [Spec,OccP/FP] should be fine (see, e.g., (30)), semantically moti-
vated movement should be out if (and only if) the semantics of the triggering feature
is incompatible with the semantics of [occ]. This definitely excludes wh-movement
to this position; but it also prevents real A-bar-movement of a (lower ranked) object,
which – according to Frey (2004) – always involves some kind of contrastive inter-
pretation; see (31).

(30) Wenn du nach Hause kommst und dir ist schlecht

if you back home come and you.dat feel sick
(31) *Wenn du nach Hause kommst und mit deiner Frau
if you back home come and with your wife
spricht der Gerichtsvollzieher, dann …
speaks the bailiff, then …
 Odd coordinations 

There is far more to say about this intricate topic; alas, I have to refer the reader once
more to Reich (2007a) for a more detailed discussion.

2.4  Subject gaps in coordinate structures

As I argued in section 1, it seems safe to assume that AC and SLFC are – apart from
some minor differences – essentially one and the same phenomenon. Or to put it an-
other way: SLFC is basically AC with a subject gap.
If this is correct, we have to answer at least two questions: (i) Why is it that in AC
subjects (and only subjects) can be dropped, resulting in an SLFC? (ii) What is the
nature of the subject gap – is it just an instance of some well-known kind of ellipsis or
is it due to something special?
As we saw in the introduction, there seems to be no (straightforward) way to
­derive the gap from ATB-movement or forward deletion. Moreover, the fact that
subject gaps (as the name suggests) are restricted to subjects – a restriction which
is, to say the least, not very widespread in ellipsis phenomena – suggests that the
second option, i.e., that the subject gap is something special, is true. The fact that
subject gaps in coordinate structures are limited to AC further suggests that they
are closely tied to some idiosyncratic property shared by these constructions. The
hypothesis I want to pursue in the following is that this very property is the ‘fusing
semantics’ of AC and SLFC, i.e., it is event subordination, implemented with the
help of [occ], that licenses dropping the subject. The crucial idea of the approach
to follow is thus that subject gaps are possible in AC (and only in AC), since event
subordination enables us to ­semantically reconstruct the subject of the second
conjunct from the subject of the first conjunct. This will follow from the ‘subject
condition’.
Syntax and semantics of vP. Before I am in a position to precisely state the ‘sub-
ject condition’, however, I have to introduce some more assumptions about the syntax
and semantics of sentence structure in German. In the previous section, I made the
assumption that subjects – understood as nominative DPs that constitute the highest
thematic argument – are to be located VP-externally in a functional projection called
vP. Since there are verbs in German that do not allow for subjects in this strict sense
(see, e.g., mir[dat] ist komisch zumute ‘I’m in a funny mood’), this means that vP does
not project independently, but is triggered by a (non-interpretable) feature [v] on the
relevant verbs.
In this section, I propose – borrowing ideas from Kratzer (1996) – that the external-
ity of subjects is not restricted to syntax, but extends to semantics. Let me illustrate this
with the following example. In (32a) the subject der Gerichtsvollzieher is severed from the
verb’s semantic representation, and the VP vor der Tür steht (‘stands at the door’) already
denotes an (almost) saturated semantic object of the type ‘set of events,’ namely the set of
events that are ‘stands at the door’-events. The head [v] itself is ­semantically vacuous, see
(32b), but it carries (by assumption) a feature [subj] which is interpretable, and which
 Ingo Reich

states that the specifier of vP (i.e., x) is a subject (to e), see (32c);10 this statement is then
conjoined with the denotation of its complement, see (32d).11
(32) a.
vP

DP v’

der GV v VP
m [subj]
||[subj]||
vor der Tür steht
λe.stands-at-the-door(e)
b. ||v|| = λQ.Q
c. ||[subj]|| = λQλxλe[subj](e)(x) & Q(e)]
d. ||(32a)|| = λe[subj(e)(m) & stands-at-the-door(e)]

The vP as a whole thus denotes the set of events e such that e is a ‘stands at the door’-
event, and the bailiff is the subject in e (i.e., he realizes the highest thematic role in the
‘stands at the door’-event e). This seems correct.
Two notes: First, as a consequence of severing the external argument from its verb, both
VP and vP are now of the same semantic type, i.e., they both denote sets of events. Second,
weather verbs like regnen (‘rain’) do not select for a thematic argument, but for an expletive
subject es (‘it’). In such cases [v] does not carry the feature [subj], i.e., the vP contributes
nothing to semantic interpretation; [v] just assigns nominative case to the expletive es.
The Subject Condition. Let’s consider now an AC with an overt subject in its sec-
ond conjunct which is coreferential with the subject in the first conjunct; see (33). Its
semantic representation is given in (34).
(33) Wenn du nach Hause kommst und du siehst den
if you back home come and you see the
Gerichtsvollzieher vor der Tür, dann […]
bailiff at the door, then […]

.  The property of being a subject relative to a given event e needs to be independently ­defined,
of course. Roughly speaking, x is a subject to an event e – i.e., ||subj(e)(x)|| = 1 – iff x refers to the
highest thematic role participating in the event e; see Reich (2007a) for details.
.  In a step by step fashion the interpretation of vP proceeds as follows: First, the denotation
of VP (λe.stands-at-the-door(e)) functions as an argument to the interpretation λQλxλe[subj(e)
(x) & Q(e)] of the feature [subj]; this results in λxλe[subj(e)(x) & stands-at-the-door(e)] as the
denotation for the node v’. The denotation of v’ then takes the subject m in [Spec,vP] as its argu-
ment, and we end up with the denotation λe[subj(e)(m) & stands-at-the-door(e)].
 Odd coordinations 

(34) λe∃e2[subj(e)(you) & come-back-home(e) &

subj(e2)(you) & see-at-the-door(e2)(bailiff ) & e2 ≤ e ]
The semantic representation in (34), of course, contains two characterizations of the
subject you, namely ‘subj(e)(you)’ and ‘subj(e2)(you)’. Since we know that we can drop
you in the second conjunct, what we’d like to be able to say is that the second charac-
terization is in some sense redundant – for if it were, then the subject in the second
conjunct could be semantically reconstructed on the basis of the subject in the first
conjunct, and the gap would be licensed.
However, since the event variables e and e2 do not necessarily refer to the same event,
‘subj(e)(you)’ does not entail ‘subj(e2)(you)’ nor vice versa. But wait: Since one of the
crucial characteristics of AC is that AC systematically triggers event subordination ‘e2 ≤
e’, it could very well be that this property of AC is the right piece of information that
bridges the gap and licenses the entailment; that is, the question to ask is the ­following:
Is it possible to conclude from the fact that you is the subject of e and the fact that e2 is
a subevent of e that you is also the subject of e2? Again, the answer is “no”. If the answer
were “yes”, then it would be impossible to have an AC with an overt, non-coreferent
subject in the second conjunct. But, as (25) shows, this is possible.
So it seems that we’re stuck. But in fact we are not. What I just showed is that the con-
ditions ‘subj(e)(you)’ and ‘e2 ≤ e’ do not strictly [!] entail – in the sense of a ­material
conditional – the condition ‘subj(e2)(you)’. It is well known, however, that entailments
in natural language typically are not strict, but non-monotonic (see, e.g., Asher & Las-
carides 2003). To see this, consider (35).
(35) a. If John comes, the party will be a success.
b. If John and Mary come, the party will be a success.
If entailments in natural language adhered to the material conditional, it should be
possible to conclude from the truth of (35a) the truth of (35b), since it follows from
‘p1 → q’ that ‘p1 & p2 → q’ for any proposition p2. But in fact we cannot strictly conclude
from (35a) that (35b) is true as well – it may very well be that John behaves completely
different in the presence of Mary.
So let’s assume that the entailment we are after is in fact a non-monotonic entail-
ment (symbolic: ‘»’), i.e., that the ‘subject condition’ in (36) holds.
(36) Subject condition
For any two events e, e2 and any individual x the following holds:
subj(e)(x) & e2 ≤ e » subj(e2)(x)
What (36) tells us is that – in the absence of any other information – we can conclude from
the fact that x is the subject of e, and that e2 is a subevent of e, that x is also the subject of e2.
If we assume the validity of (36), then a dropped subject can be semantically reconstructed
from its antecedent – and this seems to be (almost) all that is needed for licensing a subject
gap in the second conjunct of an AC. Note, by the way, that (36) does not license subject
gaps in initial conjuncts, for event subordination is an asymmetric relation.
 Ingo Reich

That the assumption of (36) is in fact consistent is easily shown, for there are only
three cases to consider. Either the subject of the second conjunct is realized, or it isn’t.
If it is, then it is either coreferent, or it is not. Suppose we have a non-coreferent overt
subject, then the conditions ‘subj(e2)(y)’ and ‘x ≠ y’ are added to the antecedent of (36);
in this scenario the conclusion is blocked, since the added conditions in the antecedent
contradict the conclusion. Suppose, alternatively, that we have an overt, coreferent sub-
ject, then the condition ‘subj(e2)(x)’ is added, and the entailment is trivialized. The last
scenario is the one with a covert coreferent subject, i.e., the ­scenario stated in (36).
Subject gaps and properties of vP. That there are in fact only three scenarios to
consider follows from the syntax of subject gaps: What I do have to assume – and what
I will, henceforth, assume – is that to drop the subject in an AC means to drop the
feature [subj]; see (37).

(37) Subject gaps (Part I)

In an SLFC the head [v] of the second conjunct lacks [subj].

As we saw above, dropping the feature [subj] is equivalent to saying that no condition
of the form ‘subj(e2)(x)’ is introduced in the relevant conjunct, i.e., from the point of
view of syntax no thematic subject is realized. In this respect, the situation in SLFC
seems to be – at least to some extent – comparable to the one with weather verbs
like regnen (‘rain’). There is one important difference, though: Whereas in the case of
weather verbs there simply is no highest thematic role in the event structure of e (the
‘raining event’) that could be realized in syntax, the event structure corresponding to
the relevant SLFC conjunct does of course include a highest thematic role, which is in
need of syntactic realization; nevertheless, the syntactic realization of this argument
can be suppressed in SLFC, for we are – thanks to the ‘subject condition’ – able to link
the not realized thematic role of the second conjunct to the subject of the first one. In
this sense, a coreferent subject in the second conjunct of an AC is redundant, and its
dropping is ‘parasitic’ on the semantics of AC.
What we have not yet talked about is the nature of the gap itself. As it turns out,
the nature of the gap is already completely determined by the assumptions we made
above. It is clear from what has been said that the subject gap is to be located in
the specifier position [Spec,vP] of vP. This position, however, cannot be completely
empty, since the task of the head [v] is to assign nominative to some element in
[Spec,vP], and there is no good reason to assume that this is somehow suppressed
in SLFC. Since the head [v] is not accompanied by the feature [subj], the element in
[Spec,vP] needs to be some kind of expletive (if it were referential, it would have to
be conjoined with the verbs’ interpretation, but there is no way to conjoin an expres-
sion of type e with an expression of type t without the mediating semantics of the
feature [subj]). The only expression fulfilling all these criteria we know of seems to
be the expletive ­subject es. So, is the subject gap a covert variant of es? Most certainly
not, since es carries in addition to the feature nominative the features 3rd person
singular; the finite verb in the second conjunct of an SLFC, however, is not restricted
to 3rd person singular, but seems to agree in person and number with the subject in
 Odd coordinations 

the first conjunct.12 It follows thus that the subject gap is unspecified with respect
to person and number. To sum up, what we know about the nature of the gap is the
following: It is covert, it is an expletive, and it is unspecified with respect to person
and number, but it is specified with respect to case – it carries the case nominative.
The only ­expression I know of that has exactly these properties is the feature [nom]
itself. So let’s assume the following restriction:
(38) Subject gaps (Part II)
In an SLFC [Spec,vP] of the second conjunct is filled by [nom].

In fact, (38) is exactly what we expect in a feature-driven syntax, where each fea-
ture [x] has to be matched with its counterpart to avoid a crashing derivation: The only
task of the head [v] is to match its feature [nom] with a feature [nom] in [Spec,vP] (i.e.,
it ‘assigns nominative’ to [Spec,vP]); since there are no other processes going on within
vP in the case of SLFC, there is no need (and it is even impossible) to realize a different
feature than [nom] in [Spec,vP].
Subject gaps and the V1-property of non-initial conjuncts. There is one last q ­ uestion
we need to answer: Why is it that dropping the subject in an AC systematically ­results in
a V1-structure (rather than a ‘gapped’ V2-structure)? Or to put it differently: Why does
a subject gap block movement to and base generation in the prefield of a V ­ 2-clause?
First note that the subject gap [nom] is a covert expression which has to be –
­according to, e.g., Lobeck (1995) – (i) licensed, and (ii) identified. There are two configu-
rations in which covert expressions are licensed: either in a complement relation, or in
a spec-head configuration. Since the subject gap is not in a complement, but a specifier
position, licensing needs to take place in a spec-head configuration. The phenomenon of
‘Topic Drop’ in German tells us that licensing covert expressions in such a configuration
is in fact possible in German. Essentially following Fries (1988), the syntax of Topic Drop
can be sketched as follows: Dropping the topic is (i) licensed by the matching feature
[topic] in the head [f], and it is identified by the finite verb in [f]; see (39).

(39) FP

DP1 F’

du F[TOPIC] vP
you

hast t1 ja recht
are t1  right

.  This is, of course, a tough nut to crack in approaches like the one proposed in this paper.
There are, as far as I can see, at least two promising ways to tackle the problem: Either there is
some syntactic process like ‘long distance agreement’ available (see, e.g., Sternefeld 2006b, chap.
V.3) or person and number agreement is semantically mediated (see, e.g., Reich 2007a).
 Ingo Reich

Now consider Wunderlich’s (1988) example (40) of a typical root clause SLFC.

(40) In den Wald ging der Jäger und schoss einen Hasen
in the forest went the hunter and shot a rabbit

What we are primarily interested in is, of course, the syntactic structure of the
­second conjunct. If the licensing of the subject gap in examples like (40) is in fact
essentially structurally parallel to the licensing of Topic Drop, then there needs to be
some ­interpretable feature corresponding to [topic] that is located in the head [v]
of vP; the obvious candidate here is of course the feature [occ], the core ­characteristic
of AC, which denotes event subordination, and thus feeds the licensing ‘subject
­condition’ in (36).
The fact that the subject gap is identified by the fronted finite verb then follows
immediately from the syntax of the feature [occ]: Since [occ] selects for the feature
[f], and [occ] is to be located in the head [v] of vP (the subject gap in [Spec,vP] is
licensed in a spec-head configuration), it follows that [f] is to be located in the head
[v] of vP, too. In other words: Whereas an AC with an overt subject is a projection of
OccP/FP (see the previous section), an SLFC is an even more ‘compressed’ projection
OccP/FP/vP; see the syntactic structure given in (41), and its semantic interpretation
in (42).
(41)
[FP In den Wald [F’ gingi [vP der Jäger [v’ ti,[SUBJ]
VP

VP OccP/FP/vP

ti und OccP/FP/vP

[nom] Occ’/F’/v’

Occ/F/v VP

schossj einen Hasen tj

(42) λe∃e2[subj(e)(the-hunter) &

[went-in-the-forest(e) & shot(e2)(a-rabbit)] & e2 ≤ e ]
 Odd coordinations 

In (41) the subject gap [nom] is (i) licensed by the interpretable feature [occ], and
(ii) identified by the overt fronted finite verb schoss (‘shot’). Since the prefield of the
fronted structure in (41) is identical with the specifier position of OccP/FP/vP, and this
position is already (and necessarily) filled with the case feature [nom] (required by the
head [v] of vP), movement to this position as well as base generation of some other
element in this position is blocked.
On where to attach the second conjunct of an AC. If we compare the analysis of the root
SLFC in (41) with the analysis of the dependent AC in (26), we observe another, somewhat
surprising difference: Whereas in (26) the second conjunct is adjoined sentence-externally,
i.e., to CP, it is adjoined sentence-internally in (41). This corresponds to the empirical ob-
servation that in cases like (26) negation cannot (easily) have scope over the second con-
junct of an AC, but in cases like (43), taken from Höhle (1983), it can.
(43) Hoffentlich sieht er uns nicht und zeigt uns an!
hopefully sees he us not and reports us part!
‘Hopefully he doesn’t see us and report us.’

A closer look at the facts reveals that the empirical generalization seems to be as
­follows: If the finite verb in the initial conjunct is fronted (V1/V2), then there is a
strong tendency to attach the second conjunct sentence-internally, i.e., to VP; if it is
not (VE in the first conjunct), then there is a strong tendency to attach the second
conjunct sentence-externally, i.e., to CP. This is, however, not a strict rule, for there are
obvious – but nevertheless systematic – exceptions to this generalization (see Reich
2007a, for further discussion).

3.  Summary

Starting from semantic and pragmatic considerations, I have proposed an analysis of

Asymmetric Coordination (AC) and SLF-Coordination (SLFC) that ties together their
semantic and syntactic characteristics by considering them an immediate ­consequence
of the fact that in AC and SLFC properties typically attributed to coordinate structures
in discourse show up in ‘grammaticalized’ form. Concretely, I have suggested that in AC
and SLFC both conjuncts are linked by the Occasion relation, which I have ­furthermore
argued to express ‘event subordination’. In a feature-based syntax this has been imple-
mented with the help of a syntactic feature [occ] that selects for the feature [f], which
is, in turn, responsible for the fronting of the finite verb in non-initial ­conjuncts.
Finally, the dropping of the subject in the case of SLFC has been argued to be in a
sense ‘parasitic’ on the special ‘fusing’ semantics of SLFC: Given event subordination,
a coreferent subject in the second conjunct of an AC is redundant in the sense that it
can be reconstructed with the help of a non-monotonic inference – the ‘subject condi-
tion’ – from the semantics of the first conjunct. On the syntactic side, the gap has been
construed as the case feature [nom], which – being a covert expression – has to be
 Ingo Reich

licensed (paralleling Topic Drop in German) by the “occasion” feature [occ] in a spec/
head-configuration and identified by the fronted finite verb. The fact that the prefield –
i.e., Spec of OccP/FP/vP – is filled with the feature [nom] derives the V1-property of
non-initial conjuncts in SLFC.

References

Asher, Nicholas & Lascaride, Alex. 2003. Logics of conversation. Cambridge: CUP.
Büring, Daniel & Hartmann, Katharina. 1998. Asymmetrische Koordination. Linguistische
­Berichte 174: 172–201.
Eckhardt, Regine. 1998. Adverbs, events, and other things. Tübingen: Niemeyer.
Fortmann, Christian. 2005. Die Lücken im Bild von der Subjektlücken-Konstruktion.
­Linguistische Berichte 204: 441–476.
Frank, Annette. 2002. A (discourse) functional analysis of asymmetric coordination. In Proceedings
of the LFG02 Conference, M. Butt & T. King (eds), 174–196.
Frey, Werner. 2004. The grammar-pragmatics interface and the German prefield. Sprache
& Pragmatik 52: 1–39.
Fries, Norbert. 1988. Über das Nulltopik im Deutschen. Sprache & Pragmatik 3: 19–49.
Hartmann, Katharina. 2002. Right Node Raising and Gapping: Interface conditions on prosodic
deletion. Philadelphia, Amsterdam: John Benjamins.
Hobbs, Jerry R. 1990. Literature and Cognition. Stanford: CSLI.
Höhle, Tilman N. 1983. Subjektlücken in Koordinationen. Ms., Tübingen.
Höhle, Tilman N. 1986. Der Begriff  ‘Mittelfeld’. Anmerkungen über die Theorie der ­topologischen
Felder. In Textlinguistik contra Stilistik. Akten des VII. Internationalen Germanisten-Kongresses
Göttingen 1985, Bd. 3, W.E. Weiss, H.E. Wiegand & M. Reis (eds), 329–340. Tübingen:
Niemeyer.
Höhle, Tilman N. 1990. Assumptions about asymmetric coordination in German. In Grammar
in Progress: Glow Essays for Henk van Riemsdijk, J. Mascaró & M. Nespor (eds), 221–235.
Dordrecht: Foris.
Höhle, Tilman N. 1991. On reconstruction and coordination. In Representation and Derivation
in the Theory of Grammar, H. Haider & K. Netter (eds), 139–197. Dordrecht: Kluwer.
Kehler, Andrew. 2002. Coherence, reference, and the theory of grammar. Stanford: CSLI.
Kratzer, Angelika. 1991. Conditionals. In Semantics: An International Handbook of Contemporary
Research, A. von Stechow & D. Wunderlich (eds), 651–656. Berlin: De Gruyter.
Kratzer, Angelika. 1996. Severing the external argument from its verb. In Phrase Structure
and the Lexicon, J. Rooryck & L. Zaring (eds), 109–137. Dordrecht: Kluwer Acadamic
publisher.
Larson, Richard. 1988. On the double object construction. Linguistic Inquiry 19: 335–391.
Lobeck, Anne. 1995. Ellipsis. Functional Heads: Licensing, and Identification. NewYork, Oxford:
OUP.
Oirsouw, Robert van. 1993. Coordination. In Syntax: An International Handbook of Contemporary
Research, J. Jacobs, A. von Stechow, W. Sternefeld & T. Vennemann (eds), 748–763. Berlin,
NewYork: De Gruyter.
Progovac, Ljiljana. 1998. Structure for coordination I/II. Glot International 3: 3–6/9.
 Odd coordinations 

Reich, Ingo. 2007a. Asymmetrische Koordination im Deutschen. Habilitationsschrift,

Universität Tübingen.
Reich, Ingo. 2007b. What asymmetric coordination in German tells us about the syntax and
semantics of conditionals. In Proceedings of Sinn & Bedeutung 11, Barcelona. Universitat
Pompeu Fabra.
Reis, Marga. 1993. Satzfügung und kommunikative Gewichtung: Zur Grammatik und Pragmatik
von Neben- vs. Unterordnung am Beispiel ‘implikativer’ und-Konstruktionen im Deutschen.
In Wortstellung und Informationsstruktur, M. Reis (ed.), 203–249. Tübingen: Niemeyer.
Roberts, Craige. 1989. Modal subordination and pronominal anaphora in discourse. Linguistics
and Philosophy 12: 683–721.
Schein, Barry. 1993. Plurals and events. Cambridge, Mass. The MIT Press.
Sternefeld, Wolfgang. 2006a. Syntax. Eine morphologisch motivierte generative Beschreibung des
Deutschen. Band 1. Tübingen: Stauffenburg.
Sternefeld, Wolfgang. 2006b. Syntax. Eine morphologisch motivierte generative Beschreibung des
Deutschen. Band 2. Tübingen: Stauffenburg.
Truckenbrodt, Hubert. 2006. On the semantic motivation of syntactic verb movement to
C. Theoretical Linguistics 32: 257–306.
Wilder, Chris. 1997. Some properties of ellipsis in coordination. In Studies on Universal Gram-
mar and Typological Variation, A. Alexiadou & T.A. Hall (eds), 59–107. Philadelphia,
Amsterdam: Benjamins.
Williams, Edwin. 1978. Across the board rule application. Linguistic Inquiry 9: 31–43.
Wunderlich, Dieter. 1988. Some problems of coordination in German. In Natural Language
Parsing and Linguistic Theories, U. Reyle & C. Rohrer (eds), 289–316. Dordrecht: Reidel.
part iv

Diachronic perspectives
 Old Indic clauses between subordination and
coordination1

Rosemarie Lühr
University of Jena

In Old Indic main clauses, the verb is usually unaccented. If it has a prefix, the
stress lies on the prefix. However, in subordinate clauses that are introduced by
a complementizer or a relative pronoun, the verb does bear an accent, whereas
any prefix remains unstressed. Apart from that, we find clauses in Old Indic that
have the same prosodic properties as subordinate clauses, but are not introduced
by a complementizer or a relative pronoun. This paper deals with two questions:
Are sentences with an accented verb but no complementizer main sentences or
subordinated sentences and under what conditions do sentences with or without
stress on the first verb appear. We will see that information structure plays an
important role in explaining all these phenomena.

Keywords: verb accentuation, Old Indic, information structure, subordination,

coordination

1.  The Problem

This paper is about a certain verb accentuation in Old Indic that at first glance seems to
indicate subordination rather than coordination. A possible solution is influenced by
the information-structural and prosodic description of a certain German coordinated
clause type. If the following analysis is correct, the description of the German phe-
nomenon is relevant for other languages as well and can be applied not only to modern
Indo-European languages but also to historic ones.
In Old Indic main clauses, the verb as a rule does not bear the accent. If it has a
prefix, the prefix bears the accent. However, in subordinate clauses that are introduced
by a complementizer or a relative pronoun, the verb does bear accent, whereas any
prefix remains unstressed.
(1) main clause subordinate clause
verb vérb
préfix verb prefix vérb

.  I thank Augustin Speyer and Jonathan Gress-Wright for the translation into English.
 Rosemarie Lühr

(2) RV V,31,6
a. prá te puØrvāп» i káraп» āni
pfx of thee-gen.sg early-acc.pl.n deeds-acc.pl.n
Vocam prá nuØtanā
I will proclaim-1.sg.inj.aor.act pfx newest-superl.acc.pl.n
maghavan
rich in giving-voc.sg
b. yaØ cakártha /
which-rel.acc.pl.n hast done-2.sg.ind.pfv.act
c. śáktīvo yád vibhárā
powerful-voc.sg that-sub dividest-2.sg.sbjv.pres.act2
ródasī ubhé
worlds-acc.pl.n both-acc.pl.n
‘I will proclaim your former deeds, the most recent ones that you, full of
gifts, have done, that you, powerful one, divided both halves of the world.’

In Old Indic there are also clauses which are stressed like subordinate clauses, but
which do not have an overt complementizer or relative pronoun to introduce them.
(3) clause verb
préfix verb
(4) RV V,45,3
a. ví3 párvato jíh¿Fta
apart-pfx mountain-nom.sg.m opens-3.sg.inj.pres.med
saØdhata4 dyáur
reaches goal-3.sg.inj.pres.med sky-nom.sg.m/f
b. āvívāsanto dasayanta
trying to win-prt.pres.act.nom.pl.m they wear down themselves-
3.pl.inj.pres.med
bhuØma
earth-acc.sg.n
‘The mountain opens up, the sky reaches the goal. To win the earth, they
(the An. giras) wear down themselves (with spells).’

Clauses such as (4) are viewed controversially in Indo-European philology: Dunkel

(1985: 51ff.) thinks they are paratactic and interprets the accent on the verb as a sign
of emphasis. Hettrich (1988: 166f.), on the other hand, assumes that “Verbalbeto-
nung […] vom vedischen Sprachteilhaber primär als Signal der Unterordnung emp-
funden [wurde], sobald der implizit gegebene semantische Bezug des betreffenden

.  The subjunctive refers to past events here (Hoffmann 1967: 244).

.  The prefix is stressed because of emphasis in initial position.

.  On the stress of saØdhata see below.

 Between subordination and coordination 

Satzes zu seinem Nachbarsatz dies zulieߔ5. But these main clauses that are stressed
like subordinate clauses are “illokutiv selbständig”6. Therefore he dubs such clauses
“Ergänzungssätze”7 (Hettrich 1988: 157). Oldenberg (1906) thought on similar lines:
Inasmuch as it is about “Nebeneinander- oder Gegenüberstellung”8 he regards these
clauses – which are stressed like subordinate clauses – as subordinate.9 In that he, for
part of the clauses, follows Delbrück (1888: 42), who wrote in his Altindische Syntax:
“The first clause is the basis for the second one”.
Some examples can in fact be interpreted in this way. The relation postea ergo
propterea, which is depicted in the order of the clauses (Abraham 1976: 14; Lühr 1989),
could underly (5):
(5) RV VI 4,7
tvaØm. hí mandrátamam
thee-acc.sg namely most pleasing-superl.acc.sg.m
arkaśokaír vavr® máhe
songs of flames-instr.pl.m we have chosen-1.pl.ind.pf.med
máhi nah· śrós·y agne /
big-acc.sg.n of us listen-2.sg.imp.aor Agni-voc.sg.m
‘For you, the most pleasing one, we have chosen with songs of flames, [because
we wish], hear our big [prayer], oh Agni!’

In that case it would be a final relation: The consequence would be in accordance with
the wishes of the referent and the proposition of the hí-clause is something that is – in
the speaker’s opinion – the condition for the consequence in the background sentence
(“As we worshipped Agni with songs of flames, our wish that he hears our prayer is
legitimate”) and that is therefore realized by the referent (Pasch 1987: 138ff.).
But if one interprets these two hypotheses in an information-structural way and
first scrutinizes the subordination hypothesis, then the difference in verb stress could
be seen in connection with the in-focus-form and out-of-focus-form found in many
languages (Drubig & Schaffar 2001: 1096ff.; cf. also Gussenhoven 1992). Alternatively
one could see the structural strategy of different verb stress at text level as a signal
of ‘embossing’, in that the syntactic difference between main and subordinate clause
corresponds to the communicative difference between foreground and background

.  ‘The speaker of Vedic saw verb stress primarily as a sign of subordination, if the implicitly
given semantic reference of this sentence to its neighboring sentence allowed for that’ (trans. RL).
.  ‘independent with respect to their illocution’ (trans. RL).
.  ‘supplement clauses’ (trans. RL).

.  ‘juxtaposition or opposition’ (trans. RL).

.  Klein (1992) has discussed Oldenberg’s material and uses intonation as explanation for the
different verb stresses, as I do in this paper. But he does not refer to information structure.
 Rosemarie Lühr

information (Hartmann 1984; Klein & von Stutterheim 1992: 70ff.). As with all other
subordinate clauses, the speaker would “durch die Wahl der Nebensatzform zu erken-
nen [geben], dass es sich nicht um assertierte Information handelt, sondern um eine
Voraussetzung zur Einschätzung der eigentlichen Aussage, die ihrerseits in Form eines
Hauptsatzes präsentiert wird” (Hetland & Molnár 2001: 630).10 Hartmann (1984) goes
one step further: By using a subordinate clause a speaker can signal that the content of
the subordinate clause will not be part of the subsequent discourse.
In the other case, i.e., the hypothesis of verb emphasis, contexts must be given
in which the verb is actually emphasized. In German we find such cases. The Verum
Focus, for example: Here the feature [FOC] is assigned to the finite verb.-
(6) A: Jura will never agree to rework his article.
B: Jura WILL rework the article.

In such cases the emphasis lies on the fact that is associated with the verb. The verb
must be previously mentioned (Szucsich 2002: 221). As Anita Steube (2001: 236)
showed with sentences like in (7), Verum-focus is one of several correction sentences
with contrastive focus:
(7) a. Peter hat es [geSAGT]CF corrects lexical meaning
b. Peter [HAT]CF es gesagt corrects time
c. Peter [HAT]CF es gesagt - Verum-focus corrects sentence force
d. Peter [SAGte]CF es corrects lexical meaning
e. Peter [sagTE]CF es corrects time etc.

I had better say right now that neither of the two hypotheses, neither the subordina-
tion hypothesis nor the hypothesis of verb emphasis, can be applied to the following
material. This can be seen if a main clause and a clause of the problematic type, i.e.,
with stress typical for a subordinate clause, are compared. Note however that also in
real main clauses the verb can be stressed under certain circumstances. It is stressed,
for example, if it is the first word of the clause (cf. above to sádhata in (4)).

2.  Analysis

2.1  Obligatory Stress

True, among the Old Indic sentences that are stressed like a subordinate clause there
are no sentences such as (7) where the verb is in contrastive focus (cf. also Hettrich
1988: 165f.). Contrast, however, is a conditioning factor for verb stress in any case.

.  ‘[the speaker would] signal by choosing the subordinate mode that it is not asserted in-
formation but a presupposition of the actual statement that is presented in the main clause’
(trans. RL).
 Between subordination and coordination 

This is demonstrated by certain bipartite structures. They consist of a preposed clause

with accent on the verb and a main clause that contains a verb, regularly unstressed.
Regularly there is verb stress in the first conjunct of bipartite structures introduced by
vā – vā ‘either …. or …’; cf. (8):
(8) VII 104,9
yé pākasam̆ ˙ sám viháranta
· ·
which-nom.pl.m simple speech-acc.pl.n dishevel-3.pl.ind.pres.med
évair yé vā bhadrám
·
as usual-instr.pl.m which-nom.pl.m or good-acc.sg.n
dūsáyanti svadhaØbhih· /
·
make bad-3.pl.ind.pres.act by own impetus-instr.pl.f
áhaye vā taØn pradádātu
snake-dat.sg.m either the-acc.pl.m shall surrender-3.sg.imp.pres
sóma       aØ vā dadhātu
Soma-nom.sg.mpfx or shall bring-3.sg.imp.pres
nírr® ter upásthe
destruction-gen.sg.f
‘Those who conscientiously twist simple speech or wantonly make bad the
good, those Soma either shall surrender to the snake or bring to the womb of
destruction.’

In addition, there are the examples with ca – ca ‘X as well as Y’; cf. (9):
(9) RV I 120,9
rāyé ca no mimītám ·
wealth-dat.sg.m and us-dat.pl help-2.du.imp.pres.act
vaØjavatyai/ is· é ca no
bringing reward-dat.sg food-dat.sg.f and us-dat.pl
mimītam · dhenumátyai
help-2.du.imp.pres.act rich of milk-dat.sg
‘Entitle us to wealth, bringing reward, and entitle us to food and cows, full of
milk!’ (Oldenberg 1906: 716) (but cf. 2.3. to the identical verbs).

Whereas the first vā or ca occurs here after an element of the first conjunct, simple vā
‘or’ and ca ‘and’ appear after a word of the second conjunct; cf. (10) and (11).
(10) RV I 6,9
átah· parijmann aØ gahi
thence wandering-voc.sg.m here-pfx come-2.sg.imp.pres.act
divó vā rocanaØd ádhi
sky-gen.sg.m or light-abl.sg.n from
‘Come here from there, oh wandering one, or from the sky’s empire of light!’
(11) RV I 10,5
ukthám índrāya ˙ syam
śám̆ ·
song-nom.sg.n Indra-dat.sg.m must be recited-nom.sg.n
 Rosemarie Lühr

várdhanam purunis· s· ídhe /

nourishment-nom.sg.n giving many gifts-dat.sg
śakró yáthā sutés· u n· o
powerful-nom.sg.m so that Soma-juices-loc.pl.m of us-gen.pl
rārán· at sakhyés· u ca
may have fun-3.sg.subj.pres.act societies-loc.pl.n and
‘A song be recited for Indra for his nourishment; him, to whom we are obliged,
so that the powerful one be joyful in our company, with our soma-juice.’

Thus, in bipartite structures with double coordinator, the position of vā and ca in the
first conjunct signals to the hearer that a second conjunct is to come.
A third kind of sentences with fixed verb stress on the first clause are those con-
taining anyá- – anyá- ‘the one – the other’. Here the first verb is always stressed, when
both this clause and the following clause form a Pāda, i.e., a verse (12) (Oldenberg
1906: 724):
(12) RV I 123,7
ápānyád11 éty abhy
away-pfx-one-nom.sg.n goes-3.sg.ind.pres.act to it-pfx
ànyád eti12 vís· urūpe
other-nom.sg.n comes-3.sg.ind.pres.act in different shapes-nom.dual.n
áhan¿F sám · carete /
day and night-nom.dual.n together-pfx go-3.dual.ind.pres.med
‘The one (half of the day) goes, the other comes: both dissimilar halves of the
day meet.’

The hearer knows that it is about day and night both from the context (‘the burning
flames have risen. The shining Us· as, the dawns, bring to light the treasures, worth to be
desired, that have been hidden in darkness’) and from his world knowledge.
Types (8) through (9) and (12), i.e., the types with fixed verb stress, have in com-
mon that the hearer expects a second conjunct: ‘either X … or Y’, ‘both X … and Y’,
‘the one … the other’. The hypothesis presented here is that the verb stress in the first
conjunct is caused by the expectation of a second conjunct.
Searching for comparable structures in German, we have to refer to Ewald Lang’s
(2004: 58) analysis of bipartite coordinated clausal structures for (13):
(13) Was machen denn deine Eltern?
l*h l*h h %
[[[Mein VAter ]T [ist ernsthaft KRANK ]FIP]
→

.  Cf. (19) as to stressing the prefix.

.  Cf. Tichy’s (2000: 43) assumption that verb stress in main clauses is caused by an ‘antithetic
accent’.
 Between subordination and coordination 

l*h h*l l %
[0 [meine MUtter ]T [geht ARbeiten ]FIP]U]

→
(Lang & Umbach 2002: 155)
‘What are your parents doing? – My dad is seriously ill, my mom goes to work.’
(Transl. RL)

The two essential properties of this construction are:

a. Within the utterance domain [...U] the subjects mein Vater, meine Mutter are
evoked by the previously mentioned deine Eltern and thus belong to the background.
Prosodically they are contrastive topics, marked by “ ” and “L*H H”.

→
b. Both conjuncts form prosodically symmetrical Intonational Phrases (IPs) with a
difference at the end of both conjuncts: The first conjunct has , the second . Also

→
“der steigende Akzent bildet ikonisch13 Offenheit bzw. Unabgeschlossenheit ab und
evoziert dadurch eine Alternativmenge mit einer nicht leer laufenden offenen Proposi-
tion …”14 (Molnár & Rosengren 1996: 82; Jacobs 1997: 124).

We now have to check whether we find these properties also in the Old Indic
example (12)! Property (a) is present in any case: ‘The one day-half goes, the other
comes’ contains two contrastive topics whose referents are known from the context
and because of world-knowledge. About property (b) my approach is: The accent on
the clause-final éti ‘he goes’ in the first conjunct is comparable to the rising contour
L*H H in the German example.15 In fact, the Old Indic accent marked by the Udātta
must be a rising contour, as the name (‘high’, ‘exalted’) suggests. The Udātta in the
contrastive topics anyád – anyád ‘the one – the other’, on the other hand, was probably
accompanied “von einer Stimmverstärkung”16 (Wackernagel 1896: 284; Klein 1992:
86).
A contrast can be expressed also by antonyms, as exemplified in (14). The struc-
ture is:
(14) first conjunct … vérb / … préfix verb second conjunct
→

.  The term ‘indexalic’ is to be preferred here (O.Panagl, p.c.).

.  ‘the rising accent denotes openness iconically or that it is not yet finished and by that evokes
an alternative set with an open proposition that is going to be filled’ (trans. RL).
.  Klein (1992: 38ff.) also traces back the verb accent in such bipartite structures to “incom-
pleteness”. He refers to it as “intonational accent”, or as “anti-final contour” (1992: 89). In addition
he assumes – justifiably – “two fundamentally different phonetic bases for verbal accentuation in
the Rigveda: salience/emphasis and heightened intonation” (1992: 91).
.  ‘by a rise in volume and/or pitch’ (trans. RL).
 Rosemarie Lühr

(15) RV I,152,3
r® tám píparty ánr·tam
·
truth-acc.sg.n fosters-3.sg.ind.pres.act untruth-acc.sg.n
ní tār¿F
down-PFX presses-3.sg.ind.pres.act
‘Truth he fosters, untruth he suppresses.’

Since (15) as a whole is the answer to the focus question ‘what does the god do?’, the
antonyms ‘truth’ and ‘untruth’ are contrastive foci. As in other languages too, ‘contrast’
is no intrinsic property of either focus or topic constituents (Molnar 2002; Vallduvi &
Vilkuna 1998).
Thus the consequence of this interpretation of the Old Indic verbal stress in the first
conjunct is: The Udātta on the verb serves to structure the text.17 By choosing a rising
verb accent on the verb of the first intonational phrase it is hinted that this is an initial
conjunct of a coordinated bipartite structure and that a final member is to follow.18
(16) first conjunct … vérb /… verb second conjunct
→ →

… prefix vérb / préfix verb

(17) first conjunct … vérb / … préfix verb second conjunct
→

Let us look at two other comparisons of Vedic and German.

Firstly, Delbrück (1871) utters the following to the state of affairs in Proto-Indoeu-
ropean (but compare his contradicting statements quoted in 2.1):
Haupt- und Nebengedanken und folglich Haupt- und Nebensätze existirten
schon in der Periode des einfachen Satzes vor der Entstehung des Relativums und
der Conjunctionen, nur dass sie kein sprachliches Zeichen hatten, ausser dem
freilich sehr mächtigen und mannigfaltigster Nüançen fähigen Satzbetonung19.
 (Delbrück 1871: 98)

.  Klein (1992: 88), too, interprets verb stress in Old Indic subordinate clauses as a signal that
“the accented clause is incomplete”. If, however, the emphasis on the subordinate clause verb is
a signal that the hearer should expect another clause, in this case a main clause (as ‘two-verb’
coordinated structures suggest, Lühr 2007), this verb emphasis would have to have started in
preposed subordinate clauses and from there would have to have been transferred to postposed
subordinate clauses (H.Hettrich, Th.Krisch, p.c.; Klein 1992: 91ff.).
.  Cf. German, when guiding person into a parking space: Weiter!, Weiter!, Weiter!, Stopp! ‘On!
On! On! Stop!’ (Peters 2005: 111).
.  ‘main thoughts and subordinate thoughts, consequently main clauses and subordinate
clauses, were in existence already in the period of the simple sentence, before complementizers
and relative pronouns came into being. They only had no linguistic sign assigned to them, save
for the powerful and finely grained sentence stress’ (transl. RL).
 Between subordination and coordination 

Furthermore, Delbrück assumed that the special stress of a verb in a subordinate

German clause is comparable to the Vedic accentuation and that it has something to
do with importance (Kümmel, p.c.):
Der Nebensatz [enthält] sehr oft gerade das …, was im Zusammenhange der Rede
das Allerwichtigste ist, und im Sanskrit, wo z.B. die Relativsätze zum allergrössten
Theile ‘nothwendige’ sind, ist das besonders häufig der Fall. Dadurch allein schon
ist eine stärkere Betonung des Nebensatzverbums gerechtfertigt. Nun kommt
noch die Gewohnheit des Sanskrit hinzu, die Nebensätze voranzustellen, wodurch
unläugbar in dem Hörenden eine Spannung auf den Hauptsatz hervorgerufen
wird.20 (Delbück 1871: 96ff)

At least the last assumption – that the hearer expects a following phrase, if a verb is
stressed in Vedic – is compatible with our solution, while the importance criterion of
the subordinate clause does not convince.
Secondly Tichy (2000: 43) assumes that clausal structures in Vedic and Proto-
Indo-European had a similar intonation as a certain Modern German sentence type.
She compares:
(18) Hätt’ ich’s gewusst (/), hätt’ ich’s gesagt (\).
‘If I had known it, I would have said it.’ (Transl. RL)

Under certain circumstances also the main clause verb could have been stressed, “so
bei der Gegenüberstellung mit einem inhaltlich kontrastierenden Verbum (‘antithe-
tischer Akzent’)”21:
(19) Der eine kommt (/), der andre geht (\)
‘One comes, the other leaves.’ (Transl. RL)

and “ferner in Verbindung mit affirmativen (= bekräftigenden, wie ved. íd) oder kau-
salen (= begründenden, wie ved. hí) Partikeln;22 vgl. nhd. Ich hab’s ja gewusst (/).23

.  ‘The subordinate clause often contains exactly that which is most relevant in the context of
the text, and this goes especially frequently for Sanskrit, in which e.g., nearly all relative clauses
are ‘necessary’. This alone justifies a stronger emphasis on the subordinate clause verb. In addi-
tion there is the tendency of Sanskrit to prepose the subordinate clauses. This, beyond any doubt,
evokes suspense for the main clause in the hearer’s mind.’ (Transl. RL).
.  ‘if put in opposition to another verb with contrasting content (antithetic accent)’ (transl. RL).

.  ‘furthermore in connection with affirmative (as Vedic íd) or causal (as Vedic hí) particles’
(transl. RL).
.  ‘Well, I knew it’ (transl. RL).
 Rosemarie Lühr

Diese Verhältnisse [seien] aus dem Vedischen erschlossen, wo die angegebenen Regeln
gelten (allerdings gibt es dort keine konjunktionslosen Nebensätze mehr)”.24
However, we do not gain much from a comparison with Modern German clauses
that contain a complex predicate (Hätt’ ich’s gewusst (/), hätt’ ich’s gesagt), as Proto-
Indo-European had certainly no such predicates. The assumption that main clause
verb stress started from cases with contrastive focus on this part of speech (Der eine
kommt (/), der andere geht (\)) is not very likely either, for there are no examples in
­Vedic where the verb is in contrastive focus as mentioned above. Expectation of an-
other proposition seems to be a much more plausible starting point.
After having dealt with the conditions under which sentences having an accented
verb but no complementizer appear we can give an answer to the question if those
sentences are main clauses or subordinated ones: The rising contour in the first into-
national phrase does not signal subordination but coordination.

2.2  Optional stress

In the examples (8) through (9), (12) and (15) the rising contour on the verb (together
with lexical means such as the bipartite vā … vā, ca … ca, anyá- … anyá-, or ant-
onyms) causes an expectation for a second conjunct. We now have to scrutinize cases
in which the bipartiteness in both conjuncts containing a verb is expressed purely by
accent and not by lexical means. The relevant examples are of the following kind: The
verb of the first conjunct is stressed and in final position, as in (12). The verb of the
second conjunct is stressed as well. It is in clause-initial position and consequently
shows the already mentioned main clause stress.

(20) first conjunct … vérb / vérb … second conjunct

→

Oldenberg (1906: 728; cf. Klein 1992: 33ff.) describes such structures as follows: In front
of the first verb and behind the second verb usually there are elements that correspond
to each other, such as two subjects or two objects. One also finds heterogenous mate-
rial on both sides, e.g., first place nominative, second place instrumental. The middle of
the Pāda, “wo die beiden [betonten] Verben zusammenstoßen”25 he terms culmination
point (“Gipfelpunkt”). But I assume also here that the Udātta in conjunct-final position
on the verb of the first conjunct is a signal for bipartiteness, cf. (21):

(21) RV I 135,8
yám aŚvatthám upa­tís» t» hanta
which-acc.sg.m AŚvatta-tree-acc.sg.m reach-3.pl.ind.pres.med

.  ‘This state of affairs has been concluded from the Vedic situation where the afore men-
tioned rules apply; there are, however, no more subordinate clauses without covert complemen-
tizer’ (transl. RL).
.  ‘where both (stressed) verbs clash’ (transl. RL).
 Between subordination and coordination 

jāyávo 'smé té
winners-nom.pl.m at us-loc.pl those-nom.pl.m
santu jāyávah» / sākám»
shall be-3.pl.imp.pres.act winners-nom.pl.m simultaneously
gaØvah» súvate
cows-nom.pl.m/f give birth-3.pl.ind.pres.med
pácyate yávo
ripens-3.sg.ind.pres.med grain-nom.sg.n
‘Those who stepped underneath the AŚvattha‑tree [tree of life] as winners, those
shall be winners with us. At this time cows give birth and the grain ripens.’

The hearer knows that the soma sacrifice causes wealth and happiness for the human
beings. Thus ‘cows’ and ‘grain’ are again contrastive topics, and both conjuncts are the
answer to the focus question ‘what wealth is given to the human beings?’
Example (22) also refers to the situation of a sacrifice:

(22) RV I 135,7
yátra graØvā vádati tátra
where millstone-nom.sg.m sounds-3.sg.ind.pres.act thither
gachatam» gr® hám índraś 26 ca
come-2.dual.imp.pres.act house-acc.sg.m Indra-nom.sg.m and
gachatam / ví sūnr®' tā
come-2.dual.imp.pres.act pfx glory-nom.sg.f
dádr® śe rØF yate
will be seen-3.sg.ind.pres.med flows-3.sg.ind.pres.med
ghr® tám
lard-nom.sg.n
‘Where the stone sounds, thither come, you and Indra, come into the house!
The glory is to be seen, lard is flowing.’ (Kümmel 2000: 233)

The focus question here could be: ‘what happens at the soma sacrifice, if rightly ex-
ecuted?’
In ví sūnr®' tā dádr® śe rØF yate ghr® tám not only is the verb dádr® śe in the first conjunct
stressed, but also the initial prefix ví. This is in accordance with Old Indic language us-
age: Prefixes standing at the top of the clause are optionally stressed. As for optionality,
cf. (23):

(23) RV I 40,8
upa ks» atrám pr® ñcītá
pfx government-acc.sg.n might multiply-3.sg.opt.pres.med

.  A peculiar use of ca in the RV is to add one noun (nearly always in the nominative) to
another, which has to be supplied.
 Rosemarie Lühr

hánti raØjabhir
hits-3.sg.ind.pres.act kings-instr.pl.m
‘He may extend his government, he hits (the enemy) with the kings.’

Thus (23) has the following intonational structure:

(24) First conjunct prefix … vérb / vérb … second conjunct

→
Now we are in a position to analyse example (4), which also has the accent on the
clause-initial prefix. The example is repeated below.

(4) RV V 45,3
a. asmaØ ukthaØya párvatasya
this-dat.sg.n saying-dat.sg.n mountain-gen.sg.m
gárbho mahØF nām· janús· e pūrvyaØya /
body-nom.sg.m exalted-gen.pl.f birth-dat.sg.n first-dat.sg.n
b.i ví párvato jíhīta
apart-pfx mountain-nom.sg.m opens-3.sg.inj.pres.med
b.ii saØdhata dyáur
reaches goal-3.sg.inj.pres.med sky-nom.sg.m/f
‘The body of the mountain opens up to this spell for the birth of the exalted
(dawns). The mountain opens up, the sky reaches the goal.’27

(4b.i) here refers to the preceding context (4a). Note that the missing verb in (4a)
must be supplemented from (4b.i): ‘The body of the mountain opens up to this spell
for the birth of the exalted (dawns). The mountain opens up’. Thus (4b) is an answer
to the focus question: ‘What happens if the body of the mountain [does X] to this
spell for the birth of the dawns?’ Answer: ‘The mountain opens up, the sky reaches
the goal’.
Whereas the examples up to now have been coordinations of two conjuncts, there
are also examples with four conjuncts (25): The preceding context is: ‘The lion’s thun-
ders are rolling from afar, when Parjanya makes rain-bringing clouds.’

(25) RV V 83,4
prá vaØtā vaØnti  /
forward-pfx winds-nom.pl.m blow-3.pl.ind.pres.act
patáyanti vidyútah· úd
fall-2.pl.ind.pres.act flashes-nom.pl.f upward-pfx
ós·adhīr jíhate  / pínvat
plants-nom.pl.f rise-3.pl.ind.pres.med streams-3.sg.ind.pres.med

.  Hettrich (1988: 160) here assumes a modal-adverbial relation and illustrates this by
the following translation: “ebenso wie der Berg sich auftut, [gelangt] zum Ziel der Himmel”
(Hoffmann 1967: 174: ‘Der Berg tut sich auf und gelangt zum Himmel’).
 Between subordination and coordination 

svàh· /
sky-nom.sg.m/f
‘The winds start blowing, flashes fall, the plants rise, the sun overflows.’

Everybody knows what is going to happen if Parjanya excites a thunderstorm. The

poet formulates the answer in two parallel Pādas with two conjuncts each and a rising
tone at the end of the first conjunct. The structure is:

(26) first conjunct préfix … vérb / vérb … second conjunct

→ →
third conjunct prefix … vérb / vérb … fourth conjunct

If the structure is tripartite, only the first and the second conjunct have the intonation-
al structure with an Udātta in the first conjunct, as in the answer to the focus question
‘What does the Old Indic god Savitar effect?’:
(27) RV I 35,9
ápaØmīvām baØdhate
away-pfx-pressure-acc.sg.f dispel-3.sg.ind.pres.med
véti suØryam abhí
drive-3.sg.ind.pres.act sun-.acc.sg.m towards-pfx
kr® s·n· éna rájasā dyaØm
black-instr.sg.n mist-instr.sg.n sky-acc.sg.m/f
r® n· oti
reaches-3.sg.ind.pres.act
‘He dispels sickness, drives the sun; he reaches the sky with his black mist.’
(28) a. first conjunct prefix … vérb / vérb … second conjunct
→

b. third conjunct préfix … verb

Structure (28) with a stressed verb before a stressed verb in the middle of the Pāda is
not compulsory, as (29) shows: In the first conjunct the verb carati is unstressed before
the stressed verb of the second conjunct titvis· é. The passage is about Indra’s battle in
the Vala-myth:

(29) RV I 52,6
pár- īm· ghr® n· aØ carati  /
around-pfx him heat-nom.sg.f goes-3.sg.ind.pres.act
titvis.é Śávo
is incited-3.sg.ind.pf.med power-nom.sg.n
‘The glowing heat goes around him, his power is incited.’
(Kümmel 2000: 224)
(30) first conjunct préfix … verb / vérb … second conjunct

By using the unstressed verb in the first conjunct the poet obviously signals that he
regards both conjuncts as confined informational units. The subjects of the two con-
juncts differ without being in contrast to one other. Thus, if no contrast is expressed
 Rosemarie Lühr

lexically, in the first conjunct both structures are possible, the one with verb stress and
the one without. This is true if both conjuncts meet in the middle of the Pāda.
(31) a. first conjunct prefix … vérb / vérb … second conjunct

→
b. first conjunct prefix … verb / vérb … second conjunct

But type (31a) is more common (Oldenberg 1906: 726ff.).

Furthermore, verb stress in the first conjunct is optional if both conjuncts are ex-
tended over several Pādas. Whereas in (32) the verb in the first conjunct is stressed, in
(33) it is not:
(32) RV I 164,20
dvaØ supan» aØ sayújā
two-nom.dual.m birds-nom.dual.m connected-nom.dual.m
sákhāyā samān» nám. vr® ks·ám
companions-nom.dual.m together-adv tree-acc.sg.m
pári s·asvajāte /
around-pfx keep hugged-3.dual.ind.pf.med
táyor anyáh.
of them-gen.dual.m the one-nom.sg.m
píppalam. svādv átty
berry-acc.sg.n sweet-acc.sg.n eats-3.sg.ind.pres.act
ánaŚnann anyó abhí
not eating-prt.pres.act.nom.sg.m the other-nom.sg.m towards-pfx
cākaŚĪti
looks on-3.sg.ind.pres.act
‘Two birds, close companions, are hugging the same tree. One of them eats the
sweet berry, the other looks on, not eating.’ (Kümmel 2000: 591)
(33) RV I 93,6
aØnyám» divó mātaríŚvā
PFX-the one-acc.sg.m sky-gen.sg.m/f Mātarisvan-nom.sg.m
jabhāra ámathnād
has brought-3.sg.ind.pf.act tore-3.sg.ind.ipv.act
anyám pári Śyenó ádreh·
the other-acc.sg.m hither-pfx eagle-nom.sg.m rock-gen.sg.m
‘MātariŚvan brought one from the sky, the eagle tore the other away from the
rock.’ (Kümmel 2000: 338ff.)

The lack of lexical marking of contrast and the distribution of conjuncts on two Pādas
thus cause optional verb stress in the first conjunct. It is unknown whether this is a
metrical or a linguistic phenomenon, or both.28

.  We know from other areas of Vedic syntax that the poets artificially equated pada and
sentence, so we get pada-initial topicalization (Hale 1987: 8ff.).
 Between subordination and coordination 

2.3  Obligatory non-stress

Turning to cases in which the verb in the first conjunct is compulsorily unstressed, we
see that there are only two patterns: Firstly, the verb in the first conjunct is unstressed
if it is identical to the verb of the second conjunct.
(34) RV I 103,5
sá gaØ avindat
he-nom.sg.m cows-acc.pl.m/f found-3.sg.ind.ipv.act
só avindad áŚvān
he-nom.sg.m-part found-3.sg.ind.ipv.act horses-acc.pl.m
sá ós» adhĪh· só apáh·
he-nom.sg.m plants-acc.pl.f he-nom.sg.m-part water-acc.pl.f
sá vánāni
he-nom.sg.m trees-acc.pl.n
‘He (Indra) won the cows, he won the horses, the plants, the rivers and seas, the
trees.’
f. also (35):
C
(35) RV X 94,1
praíté vadantu
forward-pfx-these-nom.pl.m shall speak-3.pl.imp.pres.act
prá vayám. vadāma
forward-pfx we-nom.pl want to speak-1.pl.subj.pres.act
‘They shall start to speak, we want to start to speak.’
(36) RV I 134,5
tvám · víśvasmād bhúvanāt pāsi
thou-nom.sg all-abl.sg.n world-abl.sg.n protect-2.sg.ind.pres.act
dhárman· ā -suryàt pāsi
kind-instr.sg.n Asura-power-abl.sg.m protect-2.sg.ind.pres.act
dhárman· ā
kind-instr.sg.n
‘You protect from all the world, as it befits your kind, you protect from the
Asura-power, as it befits your kind.’
With regard to deaccenting, the examples above are comparable to German sentences
showing broad focus and deletion of an identical element or gapping (Hartmann 2000:
171), as for instance in (37):
(37) a. weil [IP HánsF [VP [ eine ÈRDbeere]f isstf]f]f und
because John          a STRAWberry eats and
[PéterF [VP eine KÌRsche]f isstf]f]f]f.
Peter a CHERry
‘because John eats a strawberry, and Peter a cherry’
Those sentences, in which the second verb is deleted, are mostly to be found in spoken
New High German. They may be reputed to be a marked option in comparison with
 Rosemarie Lühr

(37)(b) with deletion of the first verb. This is usual in written language. In the litera-
ture also the terms “Einsparung” (Duden 2005: 912) or “coordinate ellipsis” (Schwabe
2000: 248ff.) are used:
(37) b. weil [IP HánsF [VP [ eine ÈRDbeere]F isstF]F]F
because John a STRAWberry
und [ PéterF [VP eine KÌRsche]f isstF]F]F]F.
and Peter a CHERry eats

Nevertheless, sentences like (37a) do occur. Cf. also examples (38)–(39) with distinct
morphological features (Lang & Umbach 2002: 161):
(38) ER trank BIER und WIR tranken WEIN.
he drank beer and we drank wine
(39) MAX wurde KRANK und WIR wurden GESUND.
Max became ill and we became healthy
‘Max went ill and we recovered.’

Here, the two identical verbs are focus-marked by focus-projection rule (40ii):

(40) Focus Projection

i. F-marking of the head of a phrase licenses the F-marking of the phrase.
ii. F-marking of an internal argument of a head licenses the F-marking of the
head (Selkirk 1995: 555; Hartmann 2000: 126).

(40ii) means, that the whole utterance can only be focused (maximal or wide focus) if
the constituent carrying the nuclear accent (focus exponent) is in its base position and
in the sister position of the verbal head, as for instance in (41):
(41) [CP Maria hat behauptet, [CP dass [vp die Tante [V’ die Nichten
Maria has claimed    that     the aunt the nieces
[vo begrüßt hat]]]]
welcomed has
‘Maria claimed that the aunt welcomed the nieces.’
Maria hat behauptet, dass [die Tante [die NICHten begrüßt hat]
‘Maria claimed that the aunt welcomed the NIEces.’

Moreover, (41) shows that the main stress of a phrase is assigned to its most deeply em-
bedded constituent. If there is no other focus structural information given, the focus
projects (Stolterfoht & Bader 2004: 261).
The main factor in our examples, however, is maximal contrast of the determiner
phrases in (42):
(42) Maximal Contrast Principle
In a Gapping construction the number of contrasting remnant-correspondent
pairs is maximized.
 Between subordination and coordination 

This principle ensures that there is no remnant without an adequate correspondent.

A verb will always remain unaccented, as long as it is not a focus (Hartmann 2000:
165, 170).
Thus, the Maximal Contrast Principle concerns phonological deaccenting in con-
trastive structures. It applies if two constituents are identical. In that case, gapping
either of the first or the second constituent takes place. The result is increasing stress
on the remaining contrasting pairs, here the determiner phrases.
In Old Indic such contrastive accents are obviously created intonationally by the
complete lack of accent on identical verbs. While both verbs are preserved here, in
German deaccenting results in deletion of one verb on the level of Phonological Form
in consequence of source and target identity (Hartmann 2000). On the other hand, we
expect a ‘fortition of voice’ on contrasting corresponding pairs, following the Maximal
Contrast Principle as in the German examples.
An exception from our deaccenting rule are cases involving ca … ca ‘both … and’
and identical verbs, as in example (9), repeated here:
(9) RV I 120,9
rāyé ca no mimītám ·
wealth-dat.sg.m and us-dat.pl help-2.du.imp.pres.act
vaØjavatyai / is· é ca no
bringing reward-dat.sg food-dat.sg.f and us-dat.pl
mimītam · dhenumátyai
help-2.du.imp.pres.act rich of milk-dat.sg
‘Entitle us to wealth, bringing reward, and entitle us to food and cows, full
of milk!’

Cf. also (43):

(43) RV I 123,12
áśvāvat¿Fr gómat¿Fr
bringing horses-nom.pl.m bringing cows-nom.pl.m
viśvávārā yátamānā
wished by all-nom.pl.m contesting-part.pres.med.nom.pl.f
raśmibhih· suØryasya / párā ca yánti
rays-instr.pl.m Sūrya-gen.sg.m away-pfx and go-3.pl.ind.pres.act
púnar aØ ca yanti bhadraØ naØma
again-adv hither and go-3.pl.ind.pres.act lucky names-acc.pl.n
váhamānā uśaØsah·
bringing-prt.pres.med.nom.pl.f dawns-nom.pl.f
‘Bringing horses and cows, wished by all, contesting with Sūrya’s rays the dawns
go and come, with names bringing luck.’

Remember that the Vedic Udātta on the verb of the first conjunct is not considered as
a means of marking prominence but as a signal to expect another conjunct. But since
ca already signals a following conjunct, stress on the identical verb of the first conjunct
 Rosemarie Lühr

seems to be redundant. Thus, there might exist a rule which overrides deaccenting of
the first of two identical verbs in the surroundings of ca … ca.29 In Optimality Theory
such a rule would be comparable to the constraint MAX(LEX),30 which is not domi-
nated by other constraints.31
On the contrary, expected deaccenting appears in the first conjunct with ná … ná
‘neither… nor’, nearly without exception.32 This is our second pattern for obligatory
non-stress; cf. the sentences (44) and (45), which contain antonyms:
(44) RV X 129,1
naaØsad ās¿Fn nó
neither-nonexistence-nom.sg.n was-3.sg.ind.ipf.act nor-part
sád ās¿Ft tadaØn¿Fm
· naØs¿Fd
being-nom.sg.n was-3.sg.ind.ipf.act then-adv not-was-3.sg.ind.ipf.act
rájo nó vyòmā paró yát /
air space-nom.sg.n not-part sky-nom.sg.n above-adv
‘There was neither non-existence nor existence; neither was air space nor the
sky above it.’
(45) RV I 113,3
ná methete ná
not are in mutual fight-3.dual.ind.pres.med not
tasthatuh·
stand still-3.dual.ind.pf.act
‘They do not start a fight, they do not stand still’ (Oldenberg 1906: 717)

In both examples sentence negation occurs, in (44) in combination with constituent

negation. It has been unclear until now why the first verb is unstressed. We propose an
explanation following Trautwein’s (2005: 216f.) analysis of the semantics of sentence
negation. Here, his assumption that negative sentences “tend to denote almost noth-
ing” is decisive. Looking at a single negated sentence like
(46) Eli did non run/laugh/wait.

.  The assumption that accentuation of the verb following the first ca is due to fact that there
is no determiner phrase to put the stress on holds for example (43) but not for example (9), cf.
rāyé at the beginning of the sentence.
.  Cf. Wunderlich (2003): Max (lexF): Every lexically assigned (positively-valued) feature in
the input has a correspondent in the output.
.  Admittedly, as we are dealing with a dead language claims about obligatoriness are uncer-
tain. We cannot extend the corpus by tests (as we do when dealing with living languages, com-
bining intuition and corpus search). But every language has rules that are fixed in the lexicon.
.  Counterexamples are to be found in RV I 62,12 und III 53, 14 with “Doppelgegensatz”
(Oldenberg 1906: 717 footnote 2).
 Between subordination and coordination 

we find that almost everything could be the case in the situations denoted by (46). Al-
though Eli did not run, it is possible nevertheless that she was moving somehow (driv-
ing, cycling etc.). Only the presupposed configuration which would also precede the
situation denoted by the corresponding positive sentence remains “as the possible truth-
maker of the negative propositional description”. But many verbs, including several state
and activity verbs presuppose no specific scenario at all. Anyway, the most important
semantic feature is that sentence negation neutralizes the complete aspectual informa-
tion. Therefore, no continuation of any event or state is expected if you do not have
double coordinators like English neither … nor or German weder … noch containing
different lexemes but only a general expression meaning ‘not’ and no specific word order
information is available. Exactly this is the case in Old Indic (for word order cf. the Old
Indic examples with double vā … vā or ca … ca). The verb of the first conjunct may be
completely deaccented since nothing lets one expect a continuation of the first conjunct
following the regularly accented ná ‘not’ at the top of the sentence. However, besides the
negator the negated determiner phrases are stressed. Thus, negated sentences are in-
stances of contrastive structures, too. They are similar to the above mentioned structures,
which contain identical unstressed verbs but contrastive stress on determiner phrases.

3.  Summary

All Old Indic examples that have been put forward are cases of coordination and not of
subordination, although the verb in the first conjunct tends to be stressed as in subordi-
nate clauses. Verb stress in the first conjunct, however, functions as a signal that a second
conjunct is to follow. It is a rising tone, as is the usual realisation of the Old Indic Udātta.
We find an Udātta of that kind obligatorily in bipartite phrases of the types ‘either X or Y’,
‘both X and Y’, ‘the one… the other…’ and with antonyms. The Udātta is optional if such
lexical means are missing and two verbs clash in the middle of the verse, for instance.
Generally, the Udātta is not present if the two verbs of the conjuncts are identical, as here
the contrastive accents on the determiner phrases seem to be the main factor, similar to
German gapping constructions. But sentences with ca … ca ‘both X and X’ and identical
verbs show redundancy, since the first verb is stressed. Not only the semantics of ca but
also the verbal accent signal expectation of a second conjunct. To explain this phenom-
enon there was no other way left but a lexical rule. On the other hand, in sentences with
ná – ná ‘neither – nor’ both verbs are unstressed. Here, the determiner phrases are maxi-
mally contrasted and bear contrastive accents as in the gapping constructions. Probably,
the specific semantics of sentence negation is of importance.
This paper is an attempt to make statements about information structure and ac-
centuation even for very old languages.33

.  Cf. Klein 1992: 95.

 Rosemarie Lühr

References

Abraham, Werner. 1976. Die Rolle von Trugschlüssen in der Diachronie von Satzkonnektoren.
In Opuscula slavica et linguistica. Festschrift für A. Issatschenko [Schriftenreihe Sprachwis-
senschaft. Universität für Bildungswissenschaften], Heinz Dieter Pohl & Nikolai Salnikow
(eds), 11–72. Klagenfurt: Heyn.
Delbrück, Berthold. 1871. Der Gebrauch des Conjunctivs und Optativs im Sanskrit und Griechi-
schen [Syntaktische Forschungen 1]. Halle: Waisenhaus.
Delbrück, Berthold. 1888. Altindische Syntax [Syntaktische Forschungen 5]. Halle: Waisenhaus.
Drubig, Hans Bernhard & Schaffar, Wolfram. 2001. Focus constructions. In Language Typol-
ogy and Language Universals [Handbücher zur Sprach- und Kommunikationswissenschaft
(HSK) 20(2)], Martin Haspelmath & Ekkehard König (eds), 1079–1104. Berlin: de Gruyter.
Duden. 2005. Duden. Die Grammatik [Der Duden in zwölf Bänden 4]. Mannheim: Dudenverlag.
Dunkel, George. 1985. IE hortatory *ey, *eyte: Ved. éta … stávāma, Hitt. ehu=wa it, Hom. εi δ’
α
" >ε. Münchener Studien zur Sprachwissenschaft 46: 47–79. ˘
Gussenhoven, Carlos. 1992. Sentence accents and argument structure. In Thematic Structure. Its
Role in Grammar, Iggy Roca (ed.), 79–106. Berlin: Foris.
Hale, Mark. 1987. Syntax in the Comparative Syntax of the Oldest Indo-Iranian Languages. PhD
dissertation, MIT.
Hartmann, Dietrich. 1984. Reliefgebung: Informationsvordergrund und Informationshinter-
grund in Texten als Problem von Textlinguistik und Stilistik. Zur Verwendung sprachlicher
Mittel zum Ausdruck von Haupt- und Nebeninformation. Wirkendes Wort 4: 305–323.
Hartmann, Katharina. 2000. Right Node Raising and Gapping. Interface Conditions on Prosodic
Deletion. Amsterdam: John Benjamins.
Haspelmath, Martin & König, Ekkehard (eds). 2001. Language Typology and Language Univer-
sals [Handbücher zur Sprach- und Kommunikationswissenschaft (HSK) 20.2]. Berlin: de
Gruyter.
Hetland, Jorunn & Molnár, Valéria. 2001. Informationsstruktur und Reliefgebung. In Language
Typology and Language Universals [Handbücher zur Sprach- und Kommunikationswis-
senschaft (HSK) 20.2], Martin Haspelmath & Ekkehard König (eds), 617–633. Berlin: de
Gruyter.
Hettrich, Heinrich. 1988. Untersuchungen zur Hypotaxe im Vedischen [Untersuchungen zur in-
dogermanischen Sprach- und Kulturwissenschaft 4]. Berlin: de Gruyter.
Hoffmann, Karl. 1967. Der Injunktiv im Veda. Eine synchrone Funktionsuntersuchung. Heidel-
berg: Winter.
Jacobs, Joachim. 1997. I-Topikalisierung. Linguistische Berichte 168: 91–133.
Klein, Jared. 1992. On verbal accentuation in the Rigveda [American Oriental Society: Essay nr.
11]. New Haven CT: American Oriental Society.
Klein, Wolfgang & von Stutterheim, Christiane. 1992. Textstruktur und referentielle Bewegung.
Zeitschrift für Literaturwissenschaft und Linguistik 86: 67–92.
Kümmel, Martin Joachim. 2000. Das Perfekt im Indoiranischen. Eine Untersuchung der Form
und Funktion einer ererbten Kategorie des Verbums und ihrer Weiterentwicklung in den alt-
indogermanischen Sprachen. Wiesbaden: Reichert.
Lang, Ewald. 2004. Schnittstellen bei der Konnektoren-Beschreibung. In Brücken schlagen.
Grundlagen der Konnektorensemantik [Linguistik – Impulse & Tendenzen 5], Hardarik
Blühdorn, Eva Breindl & Ulrich Hermann Waßner (eds), 45–92. Berlin: Walter de Gruyter.
 Between subordination and coordination 

Lang, Ewald & Umbach, Carla. 2002. Kontrast in der Grammatik: Spezifische Realisierungen
und übergreifender Konnex. In Sprachtheoretische Grundlagen der Kognitionswissenschaft:
Sprachliches und nichtsprachliches Wissen [Linguistische Arbeitsberichte 79], Anita Steube
(ed.), 145–186. Leipzig: University of Leipzig.
Lühr, Rosemarie. 1989. Ist die Basiskategorie TEMPORAL als Ausgangspunkt für konjunk-
tionale Bedeutungen ein Universale? Zur Polyfunktionalität von Konjunktionen in den
indogermanischen Sprachen. Historische Sprachforschung 102: 153–173.
Lühr, Rosemarie. 2007. Verbakzent und Informationsstruktur. In Struktur und Semantik der
Verbalphrase, Rosemarie Lühr & Susanne Zeilfelder (eds).
Molnár, Valéria. 2002. Contrast from a contrastive perspective. In Information Structure in a cross-
linguistic perspective, Cathrine Fabricius-Hansen et al. (eds), 147–162. Amsterdam: Rodopi.
Molnár, Valeria & Rosengren, Inger. 1996. Zu Jacobs’ Explikation der I-Topikalisierung. Sprache
& Pragmatik: Arbeitsberichte 41: 49–88.
Oldenberg, Hermann. 1906. Vedische Untersuchungen. 16. Die Verbalenklisis im Rigveda.
Zeitschrift der Morgenländischen Gesellschaft 60: 707–760.
Pasch, Renate. 1987. Illokutionsstrukturtypen und Typen der Verknüpfung von Inhalten illoku-
tiver Handlungen. In Satz, Text, sprachliche Handlung [Studia Grammatica 25], Wolfgang
Motsch (ed.), 119–161. Berlin: Akademie-Verlag.
Peters, Jörg. 2005. Intonation. In Duden. Die Grammatik, 7 Aufl. [Duden Band 4], 95–128. Mann­­
heim: Dudenverlag.
Schwabe, Kerstin. 2000. Coordinative ellipsis and information structure. In Ellipsis in Conjunc-
tion [Linguistische Arbeiten 418], Kerstin Schwabe & Ning Zhang (eds), 247–269. Tübin-
gen: Niemeyer.
Selkirk, Elisabeth. 1995. Sentence prosody: Intonation, stress, and phrasing. In The Handbook of
Phonological Theory, John A. Goldsmith (ed.), 550–570. Cambridge: Blackwell.
Steube, Anita. 2001. Correction by contrastive focus. Theoretical Linguistics 27: 215–249.
Stolterfoht, Britta & Bader, Markus. 2004. Focus structure and the processing of word order
variation. In Information Structure. Theoretical and Empirical Aspects [Language, Context
and Cognition 1], Anita Steube (ed.), 259–275. Berlin: de Gruyter.
Szucsich, Luka. 2002. Informationsstruktur in komplexen russischen Sätzen. In Sprachtheoretische
Grundlagen der Kognitionswissenschaft: Sprachliches und nichtsprachliches Wissen [Linguis-
tische Arbeitsberichte 79], Anita Steube (ed.), 213–233. Leipzig: University of Leipzig.
Tichy, Eva. 2000. Indogermanistisches Grundwissen für Studierende sprachwissenschaftlicher
Disziplinen. Bremen: Hempen.
Trautwein, Martin. 2005. The Time Window of Language. The Interaction between Linguistic and
Non-Linguistic Knowledge in the Temporal Interpretation of German and English Texts [Lan-
guage, Context and Cognition 2]. Berlin: de Gruyter.
Vallduvi, Eric & Vilkuna, Marina. 1998. On rheme and contrast. In The Limits of Syntax [Syntax
and Semantics 29], Peter W. Culicover & Louise McNally (eds), 79–108. San Diego CA:
Academic Press.
Wackernagel, Jacob. 1896. Altindische Grammatik, I: Lautlehre. Göttingen: Vandenhoeck &
­Ruprecht.
Wunderlich, Dieter. 2003. Optimal case patterns: German and Icelandic compared. In New Per-
spectives on Case Theory, Ellen Brandner & Heike Zinsmeister (eds), 331–367. Stanford
CA: CSLI.
 Rhetorical relations and verb placement in the
early Germanic languages
A cross-linguistic study

Svetlana Petrova and Michael Solf

Humboldt-Universität zu Berlin

The paper investigates pragmatic principles determining clause structure in

the early Germanic languages. Previous observations on discourse-related
properties of V1 vs. V2 in Old High German are reconsidered on a larger scale
and compared with evidence from other early Germanic languages. It is claimed
that the position of the inflected verb is a device of marking coordination vs.
subordination as the two basic types of discourse relations according to the model
of SDRT (Asher & Lascarides 2003).

Keywords: Early Germanic, verb placement, information structure, discourse

relations

1.  Introduction

Throughout the mass of literature exploring the principles of word order in the early Ger-
manic languages, numerous remarks suggest that the placement of the inflected verb is
sensitive to a complex set of factors pertaining to information packaging and discourse
organization (Ries 1880; Behaghel 1932; Fourquet 1974; Lenerz 1984). ­Following this,
Hinterhölzl & Petrova (2005) take a first attempt at describing word order variation in
early Germanic in a dynamic model of discourse relations as outlined in the Segmented
Discourse Relation Theory (SDRT) by Asher & Lascarides (2003). On the basis of data
from Old High German (henceforth OHG), it is claimed that the position of the finite verb
is a device of differentiating coordination vs. ­subordination as the two major types of rhe-
torical relations outlined in the framework of SDRT. This function mainly manifests itself
in the opposition of verb-initial (V1) vs. verb-second (V2) as the two most common root
patterns in OHG. V2 structures with ­referential material placed before the verb typically
occur in contexts giving supportive, e.g., ­explanatory or descriptive information and there-
fore constitute discourse parts that are subordinated in text structure. By contrast, V1 and
functionally equivalent ­patterns in which only a frame-setting element (a particle or an
adverbial) precedes the verb are generally found in sentences carrying the main story-line
of the narration and are therefore viewed to attribute to the coordinating type of linking.
 Svetlana Petrova and Michael Solf

Following these observations it may be concluded that syntactic patterns other than the
widely assumed basic verb-final order started to emerge in early Germanic for ­reasons of
discourse organization and rhetorical explicitness. In earlier studies, it has been already
observed that verb fronting in early Germanic is related to phenomena like illocution
and sentence mood. Kiparsky (1996) and Van Kemenade (1997) ­postulate residual V2 in
the case of syntactic operators like interrogative phrases, ­negation elements or sentence
adverbials (e.g., þa/þonne in Old English) situated in the left periphery of the clause and
regularly attracting the verb to the second ­position in the sentence. ­Additionally, Eythórs-
son (1996: 111) assumes a phonologically empty operator yielding verb-initial structures
in imperatives, while Lenerz (1984) – building upon Fourquet’s (1974) idea that verb
fronting highlights the contents of the whole sentence – accounts for special stylistic
usages of V1 in declaratives. In an overview on verb placement in Old English, Pintzuk
(1996: 380) assigns to all these cases the overall label of some common discourse-related
functions. Following this, we aim at discussing further type of data, predominantly de-
claratives, to provide more empirical support for the claim that verb placement serves
special discourse needs in the early Germanic languages. In this respect, we are indebted
to previous observations by Hopper (1979) on Old English and Leiss (2000: 84–109)
on Old Norse who relate verb placement to grounding in discourse and show how this
interacts with the aspectual reading of the verbs involved. Both Hopper (1979) and Leiss
(2000) state that V2-clauses provide background information in Old English and Old
Norse respectively. Hopper relates foregrounding to peripheral verb placement, which
means both Vend and V1. He also observes a strong tendency of perfective verbs to ap-
pear in foregrounding (i.e., V1) units. Leiss draws the attention at the complementary
distribution of the “historic present” and V1 in different types of sentences serving the
aim of foregrounding the message. Her main claim is that V1 is a device of perfectiviza-
tion in Old Norse. However, this view cannot be applied to all cases of V1 in older Ger-
manic. As will be shown later, there are examples in which verbs in initial position retain
their durative semantics, cf. (1a), (17b) and (21a) below. Therefore, we aim at deriving a
common functional definition of V1 with both perfective and statal/durative predicates
in early Germanic which in our view is achievable if we look at word order from the
perspective of text structure and discourse analysis.
The paper is organized as follows: starting from an analysis of the distribution of
V1 and V2 in OHG, we compare our findings with the situation in the remaining early
Germanic languages attested from nearly the same period of time and then propose a
formal model of discourse-semantics which is suitable to account for the distinctions
observed.

2.  The initial hypothesis: verb placement in the Old High German Tatian

In order to investigate the role of information structure in the syntax of OHG,

­Hinterhölzl et al. (2005) pursue an approach that especially concentrates on the
 Rhetorical relations and verb placement in early Germanic 

r­ elationship between the given/new status of discourse referents and their placement
with respect to the inflected verb in the sentence. For several methodological reasons
outlined in Hinterhölzl et al. (2005: 4–6), the empirical basis of the analysis is ­restricted
to examples from the Tatian translation (9th century) in which the vernacular text
departs from the word order of the underlying Latin original. The analysis provides
significant points in favour of the interdependence between verb placement and infor-
mation structure in OHG which can be best demonstrated on sentences representing
the thetic vs. categorical type of judgements (Sasse 1995). By definition, categorical
sentences have a bipartite structure divided into a predication base, or topic of the
sentence and a comment on this topic. By contrast, thetic utterances are “monominal
predications” (Sasse 1995: 4) in which no particular constituent is taken as the predica-
tion base of the utterance; rather, the entire sentence, including all participants, is as-
serted as a unitary whole. Therefore, thetic utterances are viewed as all-focus sentences
in which no topic-comment division applies (see also Drubig 1992).
Typical representatives of the thetic kind of judgements are presentational sen-
tences introducing a new discourse referent to the context (Lambrecht 1994, pp. 127–
131, 137–146 and 177–181). In OHG, such sentences basically display two syntactic
patterns. In the first one, the finite verb1 occupies the position at the beginning of
the entire sentence yielding V1, see (1a). This use of V1 has been already reflected in
previous studies, e.g., Lenerz (1984: 151–153) or Ramers (2005: 81) who claim that V1
in OHG is typical for utterances containing new information only. Apart from V1, a
second pattern occurs in which a particle or an adverbial is allowed before the verb,
see tho ‘then’, thar ‘there’ in (1b):

(1) a. uuas thar ouh sum uuitua (T 201, 2)

was there too [a] certain widow
‘There was a widow there, too’
vidua autem quædam erat
b. tho uuas man In hierusalem (T 37, 23)
then was [a] man in Jerusalem
‘There was a man in Jerusalem’
homo erat In hierusalem

In both patterns, the newly added referent is placed after the finite verb, see sum uuitua
‘a widow’ in (1a) and man ‘a man’ in (1b). As for the structure of the left periphery of
these sentences, there is evidence to assume that pure V1 is the preferred pattern in
these contexts since particles or adverbials added against the original are also found in

.  The finite verb in both OHG and Latin, as well as in all instances from other early Ger-
manic languages considered later in this paper, is underlined for clarity. A slash stands for end
of verse or text line according to the graphical representation of the instances in the manuscripts
­respectively.
 Svetlana Petrova and Michael Solf

postverbal position thus preserving V1, see thar in (1a). From this we may conclude
that V1 is strongly associated with the functional type of presentational, or thetic/all-
focus sentences in the early period of German.
By contrast, categorical sentences are systematically realized as V2 structures
against various word order patterns occurring in the Latin original. See the second
conjunct in (2) predicating on the discourse entity guot hirti ‘a good shepherd’ intro-
duced in the previous sentence:

(2) (ih bin guot hirti. ‘I am a good shepherd’) [T 225, 16])

guot hirti/ tuot sina sela furi siniu scaph (T 225, 16–17)
good shepherd does his soul for his sheep
‘The good shepherd gives his soul for his sheep.’
bonus pastor/ animam suam dat pro ouibus suis

A general property of V2-instances like those in (2) distinguishing them from ­structures
like (1b) consists in the type of phrase that occupies the pre-verbal domain. In V2-
structures of the categorical kind, the inflected verb singles out a discourse ­referent that
is meant to provide the starting point, or the aboutness topic of the ­utterance (Reinhart
1981). This does not hold for structures like (1b) where the sentence-initial elements
provide the temporal or local orientation of the situation, thus sharing ­properties of
frame-setting elements.
This analysis strongly suggests that the position of the finite verb in OHG is sensi-
tive to the information structure as reflected by the status of the discourse referents
in the sentence. This allows for an initial generalization concerning the placement of
the finite verb in early German. In all cases considered above, the verb occupies one
and the same position, namely the one at the beginning of the new-information focus
­domain, by additionally setting apart the aboutness topic from the rest of the ­utterance.
This generalization is represented in (3):

(3) thetic a. FOC[Vfin … DRnew …]

b. Frame FOC[Vfin … DRnew …]
categorical TOP[DRgiv] FOC[Vfin ...]

Given this conclusion, however, we are in need of an explanation for V1 in sentences

containing discourse-given material as in (4):

(4) bigonda ther phariseus innan imo/ ahtonti queden (T 126, 4–5)
began this Pharisee inside him thinking [to] speak
‘This Pharisee began to speak thinking by himself ’
Phariseus autem coepit intra se/ reputans dicere

Here, the definite expression ther phariseus ‘this pharisee’ refers to a discourse-given
entity which is a suitable topic candidate. Nevertheless, it occupies a position shown
as typical for new referents in all-focus sentences. A further examination of the Tatian
as well as a look at other OHG texts reveals that V1 with discourse-given material is
 Rhetorical relations and verb placement in early Germanic 

a wide-spread phenomenon at this stage of German. Consequently, it is unjustified to

restrict V1 to sentences with new information only. Rather, we have to look for the
conditions leading to V1 and the subsequent postverbal realization of discourse-given
subjects in root declaratives in OHG.
On closer inspection, cases of V1 in OHG show a clear positional and functional
distribution. First, it is well-known that they typically occur in text-initial sentences
or at the beginning of new episodes (Schrodt 2004: 199). A similar fact is reported
for some colloquial registers of modern German as well as for the beginning of some
orally transmitted genres like jokes etc. (Lenerz 1984: 153; Önnerfors 1997: 53). For
the Bavarian dialect, Simon (1998: 145) describes a similar function of V1 sentences
used to open a new dialogue or to continue a dialogue after a break.2 In Tatian, which
is a bilingually attested translation of a gospel harmony, episode onsets, for instance
the shift to another place of reference in the source text of the New Testament, are sig-
nalled by concordance notes in the left-hand margin of the Latin column or between
the Latin and the OHG text. Additionally, it is known that Carolingian manuscripts of
both Latin and vernacular texts use to mark the beginnings of new text units by means
of punctuation and graphical representation in order to distinguish coherent parts in
written discourse (Bästlein 1991: 59 and 1991: 214–242). As for the manuscript of the
Tatian, Simmler (1998: 306–307) remarks that the strategy of dividing episodes and
sub-episodes through initial capital letters predominantly applies for the Latin section
of the text and only rarely occurs in the OHG part. At the same time, we observe that
the graphical distinction of new episodes in the Latin original correlates with the regu-
lar pre-posing of the finite verb in the OHG translation, see (5a–b):
(5) a. uuard thô gitân In then tagon (T 35, 7)
[it] became then done in those days
‘It happened in those days’
Factum est autem In diebus illis
b. Intfiengun sie tho the heilantes lichamon (T 321, 29)
took they then theGen SaviourGen bodyAcc
‘Then they took the body of Jesus’
Acceperunt autem corpus ihesu

This suggests that the syntactic means of verb fronting systematically applies for mark-
ing episode boundaries in OHG as a functional equivalent of the graphical highlight-
ing of the episode onsets in the Latin original. The strong preference for V1 at the

.  But note that Simon points at further restrictions on V1 in Bavarian. Unlike the situation
in the standard language, V1 in Bavarian appears to be most common with modals, less com-
mon with auxiliaries and highly restricted with main verbs (Simon 1998: 148). Moreover, V1
is preferred in sentences highlighting the reference to an addressee. As such, they represent
no declarative statement but fulfill different illocutionary functions in the domain of request
(Simon 1998: 149).
 Svetlana Petrova and Michael Solf

beginnings of new episodes not only accounts for the post-verbal position of full DPs
as in (4) but also for the positioning of pronominal subjects inserted against the Latin
original, see sie ‘they’ in (5b).
Furthermore, it has been observed that V1 regularly occurs with certain verb
­lexemes in OHG (for an overview see Schrodt 2004: 199) which according to our data
classification constitute some stable classes of predicates. The most common group of
predicates favouring V1 – apart from those in presentational sentences – are motion
verbs (6a), verbs of saying (6b) as well as punctual, especially inchoative verbs (6c)
signalling the initiation of a new state of affairs (very often this is a new physical or
cognitive state of the referent):
(6) a. quam thara gotes engil (T 35, 32)
came there GodGen angel
‘There appeared God’s angel’
& ecce angelus domini
b. antlingota thô sîn muoter Inti quad (T 30, 24)
responded then his mother and said
‘Then his mother responded and said’
& respondens mater eius & dixit
c. uuard tho giheilit ther kneht in thero ziti (T 84, 7)
became then healed the boy in these times
‘Then the boy was healed at this very moment’
& sanatus est puer in illa hora

How can the initial position of the finite verb in these examples containing discourse-
given material be re-unified with the fact that the same structure occurs in presen-
tational ­sentences with brand-new referents as well? One suggestion that might be
uttered in this respect is that the predicates favouring V1 are intransitive unaccusatives
in which the ­surface subject is an underlying object actually3. However, our data pro-
vides examples of V1 with transitive verbs as well, see (7a–b); consider that the inser-
tion of the subject pronoun as well as of the adverbial tho in OHG does not affect the
initial position of the inflected verb:
(7) a. Quad her tho zi then giladoten/ ratissa (T 180, 9)
said he then to the guests parableAcc
‘Then he told a parable to the guests’
Dicebat autem & ad Inuitatos/ parabolam
b. furstuont siu thó in ira lihhamen/ (T 95, 14)
understood she then in her body
(thaz siu heil uuas fon theru suhti ‘that she was healed from her suffering’
 [T 95. 15])

.  See Sasse (1995: 6) for a brief discussion on this matter.

 Rhetorical relations and verb placement in early Germanic 

‘Then she realized with her body’

& sensit corpore/ quod sanata ess& a plaga

We suggest that a plausible explanation of the postverbal position of both given and
new subjects in V1 sentences discussed so far can be gained if next to the ­informational
status of referents, discourse relations among sentences are considered as well. From
the point of view of discourse organisation, examples (4)–(7) do not act as categorical
sentences providing a comment on a given referent but rather as event-reporting sen-
tences answering implicit questions like “What happened then?/How does the story
go on?” etc. This makes clear that the discourse referents contained in the instances
under scrutiny are not mentioned as the starting point or the aboutness topic of the
utterances but as being involved in the new state of affairs reported here. From this
perspective, sentences (4)–(7) have to be viewed as all-focus sentences just like the
presentational ones given in (1). Since the post-verbal position is associated with new-
information focus (see (3) above), fronting of the finite verb is used as a special strat-
egy to highlight the entire proposition and to disable a topic-comment separation,
which otherwise would have applied, especially as far as discourse-given material is
contained in the sentence.
Another point in favour of this account on V1 comes from the lexical meaning
of the predicates involved. Motion verbs, inchoatives predicates as well as verbs of
­saying affect the main characteristics providing the deictic orientation of the narra-
tive situation according to ‘place’, ‘time’ and ‘participant/perspective’. In discourse-
analytic approaches like Brinton (1996) these characteristics are considered relevant
for distinguishing episode boundaries in running discourse. Changes with respect
to one of these characteristics mark the beginning of a new episode in the structure
of the text. This observation fits to the fact that the shift of place and/or participant
is a typical instance in which episode onset is marked by an initial capital letter in
Carolingian manuscripts (see Bästlein 1991: 168 and 1991: 192). Following these
considerations, we can conclude that predicates like those favouring V1 in OHG
operate as inherent indicators of episode boundaries. Presentational sentences ap-
pear to be a subset of this group of sentences establishing a new situation via change
of personnel.

3.  Comparison to other early Germanic languages

The hypothesis about the role of verb placement as a discourse-structuring device

in OHG would gain strong support if it turned to hold in other early Germanic lan-
guages as well. In the following analysis, the contexts and factors favoring V1 vs. V2
in OHG shall be reconsidered on the basis of data from other early Germanic lan-
guages. For this reason, texts from Old English (henceforth OE), Old Saxon (hence-
forth OS) and Old Norse (henceforth ON) shall be considered. Remarks on age,
 Svetlana Petrova and Michael Solf

size and genre of the evaluated texts will be given at the beginning of each section,
respectively.4

3.1  Old English

For OE, we chose the text of the Beowulf poem, the oldest epic narrative of all early
Germanic literature.5 The text, comprising some 3.000 alliterative lines, is composed
in the 8th century in the Anglian dialect but written down in the later half of the 10th
century in Late West Saxon – the standard OE dialect at that time – though a number
of original Anglian forms remain (see Lehnert 1960, vol. I, p. 43).
Starting with a review of typical V1-occurrences in Beowulf, it is important to
note that these are by no way rare or uncommon for this text of the early Germanic
period. First of all, Beowulf tends to use V1 regularly in text-initial position as well as
at the ­beginning of a new text section. New chapters (called ‘fits’) are easily detected
in Beowulf since they are marked by Roman numbers in the manuscript. At the begin-
ning of such chapters, V1 goes with all types of main-verb predicates. So in (8a) the
sentence at the beginning of a fit XII describes a statal (durative) condition, whereas in
(8b) the predicate at the beginning of fit XIX clearly describes a punctual event:
(8) a. Nolde eorla hleo […] / cwealm-cuman
not wanted noblemenGen protector murderous visitorAkk
cwicne forlætan (Beo 791f.)
alive let go
‘The protector of the warriors did not wish to let the monster go alive’
b. Sigon þa| to slæpe (Beo 1252)
sankPl then to sleep
‘They sank then into sleep’

Next to episode onsets, V1 in Beowulf appear to favour the same groups of verbs
which were also described for OHG above. Among these, motion verbs constitute the

.  Gothic, which apart from the runic inscriptions provides us with the earliest written records
of the Germanic group, leaves only little ground for any reliable conclusions about authentic
word order. The basic text of the corpus, the translation of the New Testament from Greek made
by Wulfila in the 4th century and attested in fragments of copies from the 5th and 6th century
(see Braune & Heidermanns 2004, § E5, p. 6), shows an overwhelming identity with the word
order of the parallel Greek text. If this text is indeed the source for the Gothic translation, future
work could address especially sentences deviating from the syntactic structure of the original,
an approach similar to the one pursued for the Tatian above. According to Fourquet (1938:
234–281), such sentences in Gothic really exist. A full sample of these would provide a basis for
a subsequent analysis of verb placement in Gothic, too.

.  On the possibility to acquire suitable material for syntactic analysis from Bewoulf, see
­Pintzuk (1996: 386).
 Rhetorical relations and verb placement in early Germanic 

­ verwhelming part. In such instances, both context-given and context-new referents

o
are involved, see (9a–b). Note that in (9b) Wealhtheow, the wife of the Danish king
Hrothgar, is mentioned for the first time in the narrative:

(9) a. Com þa| to lande lid-manna helm (Beo 1623)

came then to land sailorsGen protector
‘Then the protector of the sailors approached the shore’
b. eode wealh-þeow forð cwen hroð-gares (Beo 613)
went Wealhtheow forth wife HrothgarGen
‘Then came Wealhtheow, Hrothgar’s wife’

V1 is also common with verbs of saying appearing both at the beginning of a fit (10a)
as well as within one (10b):

(10) a. Heht ða þæt heaðo-weorc to hagan biodan (Beo 2893)

ordered then that battle-toilAcc to castleDat announce
‘He ordered to tell the people in the castle about this difficult fight’
b. Spræc/ ða ides scyldinga (Beo 1168)
spoke then [the] queen [of the] DanesGenPl
‘Then the queen of the Danes spoke’

With verbs of saying, the V2 pattern as in (11) is also very common:

(11) Bio-wulf maðelode bearn ecg-ðioes (Beo 1999)

Beowulf spoke child EcgtheowGen
‘Beowulf spoke, the son of Ecgtheow’.

However, a typical property of this type of expression is that a discourse-given ­referent,

mainly a changing interlocutor in a dialogue, is placed before the verb and an appo-
sition follows it. Apparently, the verb lexeme in this type of structure never changes
which indicates a kind of idiomatic expression standing beyond the analysis of word
order variation in this functional domain.
Furthermore, V1 also correlates with different main verbs sharing the property
of perfective, punctual semantics which – used within a fit – denote the beginning of
a new state of affairs. This pattern is especially frequent when a new or extraordinary
important event is announced, e.g., a turning point in the course of the narration. See
(12) which relates that Beowulf suddenly detects the weapon with which he is going to
win the battle against Grendel’s mother:
(12) GE-seah ða on|searwum sige-eadig bil (Beo 1558)
saw then in battle triumphantAcc bladeAcc
‘In the middle of the battle he saw a triumphant blade’
Also in striking parallelism to the picture drawn for OHG above, V1 sentences may
­contain particles or adverbials like OE þa ‘then’ placed after the verb. V1 as an indica-
tor of a change within the narrative setting is thus preserved. Additionally, instances of
 Svetlana Petrova and Michael Solf

sentence-initial þa followed immediately by the verb co-occur as functional equivalents

to V1-sentences in all contexts described so far,6 see (13):
(13) a. Đa com of more […] gre-/ ndel gongan (Beo 710)
Then came from moorDat […] Grendel goPastPart
‘Then from the moorland […] Grendel came’
b. ða| wæs swigra secg sunu/ eclafes (Beo 980)
then was more silent man son EcglafGen
‘Then this man, the son of Ecglaf, became more silent’

To sum up, the context features and predicate groups distinguished as triggers of V1 in
OHG appears to be a wide-spread pattern in the Beowulf as well. In the previous ­literature,
the placement of a finite verb before all arguments has already been ­associated with the
purpose of highlighting “a new or surprising subject” (Stockwell 1984: 576). Due to the
property of V1 sentences to carry further the discourse, we may now extend the function
of this pattern to that of focussing not only the subject but the entire proposition.
Turning to sentences of the categorical kind and comparing the results from OE
with the preferred V2 in OHG, we encounter a basic difference between the two lan-
guages. V2 with a left-peripheral topic constituent dominating in this pragmatic do-
main in OHG is indeed found in part of the evidence from OE. This is the case in (14)
which clearly allows for an interpretation as an identificational sentence answering the
preceding question ‘Who are you?’:

(14) we| synt gum-cynnes/ geata leode (Beo 260)

we are kinGen GeatsGen people
‘We are by kin of the clan of Geats’

The same kind of topic marking also occurs in parenthetic constructions providing
additional information on an entity just mentioned:

(15) wulfgar maþelode þæt wæs wendla leod (Beo 348)

Wulfgar spoke this was WendelsGen chieftain
‘Wulfgar spoke – this was the Wendles’ chieftain’

Due to the pragmatic status of the referents and on the basis of discourse interpreta-
tion, it can be concluded that the finite verb in structures of the type in (14) and (15)
is set to distinguish the aboutness topic from the new information supplied by the rest
of the sentence.
Nevertheless, patterns other than V2 also appear in categorical sentences of OE.
Consider the following small discourse: after his return to his home land, Beowulf

.  For a detailed investigation on the discourse functions of OE þa see Enkvist & Wårwick
(1987) who describe parallel functions of sentence-initial þa followed by the verb as those
claimed here for V1 in OE.
 Rhetorical relations and verb placement in early Germanic 

relates his adventures with the Danes and is eager to present the gifts that he has ­obtained
from them as an award for his successful fight against Grendel. So he asks to bring these
gifts and as they lay in front of his counterparts, he utters the sentence quoted in (16):

(16) me| ðis hilde-sceorp/ hroðgâr sealde (Beo 2155)

me these ornaments of war Hrothgar gave
‘These ornaments of war were given to me by Hrothgar’

The context of this text passage bears strong indications for the interpretation of the
sentence as a categorical one. It is also clear that the definite expression ðis hildesceorp
‘these ornaments of war’ best qualifies to be the aboutness-topic of the utterance be-
cause due to the preceding context it is the expected starting point of the next sentence
providing more information on these weapons. However, this constituent is not sepa-
rated from the rest of the utterance by means of verb placement as demonstrated for
the parallel cases in (2) from OHG and (14)–(15) from OE. Rather, the aboutness topic
of the utterance shares the same syntactic domain as referents belonging to different
information-structural categories, for example the familiarity topic me ‘me/to me’ as
well as the focus of the sentence, the donor of the weapons Hrothgar.
The examples discussed in this section provide important points concerning the in-
teraction between verb placement and discourse structure in OE. On the one hand, the
cases of V1 confirm the findings for OHG. On the other hand – in contrast to the situation
in OHG – V2 turns out not to serve as a topic-marking device in categorical statements.

3.2  Old Saxon

After having pointed at a crucial difference between the syntactic realization of categorical
sentences in OHG and OE, we turn to the investigation of evidence from the OS period.
The data is based on the most representative text of the OS corpus – the Heliand – a
9th/10th-century poetic gospel harmony comprising 5.983 alliterative lines (see Rauch
1992: 1).
It has been pointed out in the previous literature that sentences in which the
­finite verb precedes all arguments in main clauses are extremely frequent in OS syn-
tax. Rauch (1992) estimates sentences initiated by a particle followed immediately by
the finite verb to be the most common pattern in OS and therefore accounts them to
be “[t]he unmarked word order of the OS independent declarative sentence” (Rauch
1992: 24), followed in number by pure V1 sentences. As early as in the revealing
work of Ries (1880), the kind of logical relations between sentences in context have
explicitly been accounted for as factors triggering this kind of fronting of the finite
verb in OS.
Looking at the correlation between the finite verb form and the pragmatic features
of discourse referents, we encounter a slightly different situation in OS in compari-
son to OHG. On the one hand, in presentational contexts, the type of clause-initial
­particles followed by the finite verb seems to be more frequent than pure V1, see (17a)
 Svetlana Petrova and Michael Solf

vs. (17b). Thus, these two patterns in presentational sentences establish a quantitative
relation opposite to that in OHG:
(17) a. Than uuas thar ên gigamalod mann (Hel 72)
then was there an old-aged man
‘Then it was an old-aged man there’
b. Lag thar ên felis bio( an (Hel 4075)
lay there a stone upon
‘A stone lay there upon [the entry of the tomb]’

On the other hand, categorical sentences – for example those directly following
­presentational ones – exhibit the structure established for OHG already, i.e., they use
to fill a single position before the finite verb form with the topic of the utterance, most
usually in form of an anaphoric pronoun referring backwards to the entity just intro-
duced to context, see (18):

(18) a. that uuas fruod gomo (Hel 73)

that was wise man
‘that was a wise man’
b. That uuas sô sâlig man (Hel 76)
that was so blessed man
‘This was such a blessed man’

This situation leads to some conclusions about the interaction between information
structure and syntax in OS. First, it points at a higher stage of generalization of the
V2-rule in OS as this structure appears in different contextual types of main sentences
despite of the type of constituent or the pragmatic status of the referents involved.
Second, it shows that new information in all cases follows the finite verb and therefore
confirms the view of a right-peripheral focus domain in early Germanic as stated so
far. This is also demonstrated by other instances bearing a brand-new referent, for
example in the object position of a transitive verb, see (19):

(19) Thar fundun sea ênna gôdan man (Hel 463)

there foundPl they aAcc goodAcc manAcc
‘There they found a good man’

However, OS displays some more peculiarities. There are cases showing patterns other
then V1 or V2, see (20):

(20) Ic is engil bium (Hel 119)

I his angel am
‘I am his angel’

The sentence provides more information on a discourse-given referent, thus it ­classifies

as a categorical one, with the pronoun OS ic as the aboutness topic of the utterance.
Nevertheless, the sentence shows a mixed topic-focus-domain in front of the finite
 Rhetorical relations and verb placement in early Germanic 

verb, a peculiarity which reminds us of comparable instances in OE like (16) repeated

here for convenience:

(16) me|ðis hilde-sceorp/ hroðgâr sealde (Beo 2155)

‘These ornaments of war were given to me by Hrothgar’

Like OE, OS does not show a tendency to distinguish the aboutness-topic from the
rest of the utterance as was observed for OHG, thus confirming OS to linger on a
­continuum between the other West Germanic dialects.
On the other hand, although V1 turns out to be less frequent in presentational
­contexts, it is interesting to look for further utilizations of this pattern aside from these
classical cases of all-focus sentences. As a matter of fact, such instances really occur in OS
exactly in the conditions under which they systematically appear in OHG and OE dis-
cussed above. Also quite similar to the situation in these languages, V2 with a preceding
frame adverbial acts as an optional variant in the pragmatic domains of V1-sentences.
First, we shall turn to sentences at the beginning of a new text section. There is no
problem to isolate such instances in Heliand as the text is divided in chapters termed
‘fits’ (‘per vitteas’) in the Latin preface of the poem. V1 occurs with all sorts of predi-
cates describing both states and actions, see (21a) vs. (21b):
(21) a. Stôd imu thô fora themu uuîhe  (Hel 3758)
stood PronRefl then in front of the temple
Crist uualdandeo
almighty Christ
‘The almighty Christ stood in front of the temple’
b. Uurðun thô thea liudi umbi thea êra Cristes, / umbi
becamePl then the people about the doctrines ChristGen about
thiu uuord an geuuinne (Hel 3926f.)
the words in quarrel
‘Then these people began quarreling about the message of these words’

Second, V1 regularly occurs in sentences containing the predicate groups distin-

guished as triggers of verb-initial placement in OHG above. Like in OE, the most com-
mon ­examples are provided by sentences containing motion verbs. In general, these
­occupy the initial position in the sentence despite of the pragmatic status of the referent
involved or the position of the utterance in global text structure. As (22a) vs. (22b–c)
show, both new and given referents are found in post-verbal position, and only (22b)
is one at the beginning of a new fit, the rest of the examples signal a change of the situ-
ation within one and the same episode:
(22) a. Thô quam thar ôc ên uuîf gangan (Hel 503)
then came there also a woman goPastPart
‘Then a woman came there, too’
b. Giuuitun im thô eft an Galilealand Ioseph endi
wentPl PronRefl then back to Galilee Joseph and
 Svetlana Petrova and Michael Solf

Maria (Hel 780)

Maria
‘Then Joseph and Mary went back to the land of Galilee’
c. Fôrun thea bodon o(ar all (Hel 350)
wentPl the messengers over all
‘The messengers went all over the country’

The same may be seen to apply to verbs of saying. V1 occurs both at the beginning of
a fit as well as within one, see (23a–b). Sentence-initial particles or adverbials followed
by the verb provide a functional equivalent of this pattern, see (23c):
(23) a. Hiet man that alla thea elilendiun man iro
ordered PronIndef that all the foreign people their
ôðil sôhtin (Hel 345)
home country serche3PlSubjII
‘It was ordered that all these foreign people went to their native country’
b. Hêt imu helpen thô / uualdandeo Krist (Hel 4101)
ordered him helpInf then almighty Christ
‘The almighty Christ ordered to help him’
c. Thô sprak thar ên gifrôdot man (Hel 208)
then spoke there a wise man
‘Then a wise man spoke there’

Finally, the same structural variation holds for predicates pointing at the beginning of
a new state of affairs, e.g., a physical or cognitive state of affairs of a given referent, see
(24a–c):

(24) a. uuarð ald gumo / sprâca bilôsit (Hel 172f.)

became old man speechGen bereft
‘The old man became bereft of speech’
b. Uuarð Mariun thô môd an sorgun (Hel 803)
became MariaDat then heart in worries
‘Then Mary became anxious’
c. Thô uuarð hugi Iosepes, / is môd giuuorrid (Hel 295)
then became mind JosephGen his heart worried
‘Then, Joseph became worried’

Apart from these instances in which – quite similar to the other investigated old
­Germanic languages – OS makes use of verb syntax to signal progress in narration,
OS itself offers a number of special cases in this direction. One of these involves the
use of the V1 in explanatory parts of the narration denoting durative conditions on
a discourse-given referent. In the cases analysed so far, we witnessed regular V2 in
OHG and parts of the OE examples, but no V1. However, the OS data give reason to
believe that V1 in explanatory parts of the narration does not occur randomly but
­applies exactly in sentences announcing an outstandingly important event or property,
 Rhetorical relations and verb placement in early Germanic 

i.e., a state of affairs which is crucial for the further development of the narrative (see
Ries 1880: 19 for a similar interpretation of such instances). Applying our previous
account on cases like these, we may conclude that fronting of the finite verb to the
beginning of the sentences is a syntactic means used to highlight the importance of the
entire proposition in relation to the surrounding units of discourse. V1 and V2 form
an ­interesting opposition, as will be shown by the following examples taken from the
story about the nativity of John the Baptist (Luke 1):

(25) a. Than uuas thar ên gigamalod mann, /

then was there an old-aged man
b. that uuas fruod gomo […]
that was wise man
c. That uuas sô sâlig man / […]
that was so blessed man
d. was im thoh an sorgun hugi, / that sie
was him however in worries mind that they
er( iuuard êgan ni môstun
heir own Neg could (Hel 72–86)
‘Then there was an old-aged man, this was a wise man […] This was such a
blessed man […] But they had great sorrow, for they had no child’

The categorical sentences in (25b–c) provide additional information about the

discourse referent introduced in (25a), the information-structural domains of Topic
and Focus are clearly distinguished by means of verb placement. In (25d), however,
a characterization of the referent is given which is more important than the already
provided information about his age and his wisdom. Zachariah’s and Elizabeth’s lack
of a child is crucial for the further development of the story; it is a condition which is
going to change and to motivate a chain of subsequent events constituting the further
course of the narrative. So in this case, a special utilization of verb position is shown
for OS. Whenever an important feature of a referent, or a crucial event of the story is
narrated, the language switches to V1 in order to highlight the whole proposition and
to set it up against other, not so important parts of the discourse.
As a further peculiarity of OS, V1 appears not only at episode edges but within
episodes of temporally successive events called ‘period’ according to the ancient gram-
matical tradition (see Kusmenko 1996: 147). Consider the following example:
(26) a. Thô uuarð thar an thene gastseli / megincraft mikil
then became there in the guesthall crowd big
manno gesamnod […]
menGen gathered
b. Quâmun managa / Iudeon an thene gastseli;
came many JewsGen in this guesthall
c. uuarð im thar gladmôd hugi, / blîði
became they there gladhearted mind happy
 Svetlana Petrova and Michael Solf

an iro breostun: […]

in their hearts
d. Drôg man uuîn an flet / skîri mid scâlun […]
carry PronIndef wine in room pure with bowls
e. Uuas thes an lustun landes hirdi, / huat
was thisGen in joy landGen shepherd what
he themu uuerode mêst te uunniun gifremidi. /
he theDat crowdDat most for joy did
f. Hêt he thô gangen forð gêla thiornun (Hel 2733–2745)
ordered he then goInf forth gay maidenAcc
‘There was a mighty crowd of men gathered together in the guest hall [...]
Many people came into that guest hall; there they became glad-hearted [...].
Wine was brought to the room [...]. The herdsman of the land bethought
him with joy [...]. He hight to go forth the gay maiden’

The example gives a chain of chronologically ordered events all exposing no dependency
relation among each other but being equally situated on the layer of main action. Again,
V1 signals that each sentence in a narrative chain like the one in (26) reports a new event
and thus forms a situation on itself in which no topic-comment distinction applies.

3.3  Old Norse

ON, although exhibiting a rich amount of authentic text material, starts its written
records only very late, in the 12th century (Ranke & Hofmann 1988: 13–18), i.e., at a
time at which in other Germanic dialects we already speak of Middle High German or
Middle English respectively. Moreover, most of the prose is attested to us in copies of
a later time – something we should keep in mind comparing the ON evidence against
the West Germanic. Although the runic inscriptions show us a mixed picture of word
order, see (27) as an example of V-last in a main sentence from Proto-Norse, ON has a
very strong tendency towards V2 in all kinds of sentences, see (28).
(27) ek hlewagastiR holtijaR horna tawido (Golden horn of Gallehus)
I Hlewagast HoltGen hornAcc made
‘I, Hlewagast, from/son of Holt, made the horn’
(28) gud mínn þu ert híalpari mínn (VienPs 17,3)
God my you are helper my
‘My lord, you are my helper’
Deus meus adiutor meus

Whereas a sentence like (27) was possible centuries before the ON literal tradition started,
ON abandoned this pattern in favour of a more rigid syntax, as (28) demonstrates: The
Vienna Psalter shows how even an interlinear version tends to build V2-sentences, here
by inserting the subject pronoun and the finite copula verb. In other words, ON shows
less evidence that the position of the verb differs according to the pragmatic properties of
 Rhetorical relations and verb placement in early Germanic 

sentence constituents. New discourse referents for instance are introduced in sentences
showing regularly V2, see (29), which stands as typical for the Edda prose:

(29) Garðarr hét maðr (Ldn 3)

Garðarr was called man
‘Garðarr was the name of a man’

Findings like these do not differ from instances of anaphoric reference:

(30) Garðarr siglði umhverfis landit (Ldn 3)

Garðarr sailed around land-the
‘Garðarr sailed around the land’

Unlike in the West Germanic dialects, we may suppose, the V2-order seems to be
much more fixed. As for the functional domains of V1 outlined for OHG, ON uses this
verb pattern too, but shows no regular connection between the content of a verb and
its position. Thus we find examples like (31–33) with a verb of motion, and (34–36)
with a verb of saying:

(31) oc fór hann útan þat sumar (Íslb X, 10)

and went he out that summer
‘and he went out that summer’
(32) þá fór hann á braut (Íslb VII, 3)
then went he away
‘then he went away’
(33) En et næsta sumar epter fóro þeir austan (Íslb VII, 6)
but one next summer after wentPl they eastAcc
‘and the next summer they went eastwards’

Even though there are instances of V1 (31) and equivalent V2-order after a frame
­adverbial like in OHG, OE and OS (32), the occurrence of V2 sentences with the
same group of verbs (33) clearly speaks against a functional positioning of the fi-
nite verb ­according to a West Germanic pattern. The same is true for, e. g., verbs of
saying:

(34) oc mællte því mange í gegn (Íslb X, 9)

and spoke thatDat many against
‘and there were many speaking against that’
(35) En þá hóf hann t lo sína upp [...] oc sagþe (Íslb VII, 13)
but then raised he voiceAcc his up and said
‘and then he raised his voice [...] and said’
(36) En hann sagþe cuningom Ólafe (Íslb 7)
but he said kingDat OlafDat
‘and he said to king Olaf ’
 Svetlana Petrova and Michael Solf

Here, V1 (34), sentence-initial frames followed by the verb (35) and V2 with referential
constituents in preverbal position (36) may serve the same discourse function. As can
be seen from the connective elements, the sentences in (31) and (34) – and this is quite
often the case – are noninitial conjuncts. We often find the typical group marker oc (see
Kotcheva 2000: 154) in particular to connect shorter or longer passages of V1-sentences
to chains of conjuncts of the kind already encountered in the OS Heliand, see (37).
(37) Þá var fj rðrinn fullr af veiðiskap, ok gáðu þeir
then was fjord-the full of catch and carePl they
eigi fyrir veiðum at fá heyjanna, ok dó alt kvikfé
not for acres to make hay-the and died all cattle
þeira um vetrinn. (Ldn 5)
their in winter-the
‘There was the fjord full of fish, and because of the fishing they didn‘t care to
make hay, and all their cattle died in the winter’

This phenomenon, known as ‘Narrative Inversion’ (see Sigurðson 1994) often ­applies
within a passage to link closely related and chronologically successive events to consis-
tent units of running discourse called ‘period’ after the ancient Latin syntactic tradition,
see Heusler (1977, § 508, p. 173), Kusmenko (1996) and Donhauser et al. (2006). It is
a reminiscence of the common Germanic discourse-sensitive properties of verb place-
ment in ON although ON allows less freedom of word order than the remaining early
Germanic languages were shown to do.

4.  Theoretical implications

The foregoing analysis revealed that the placement of inflected verb in the early
­Germanic languages depends on properties of discourse organization and text struc-
ture. The aim in this part is to account for a theoretical model reflecting these proper-
ties of early Germanic syntax.
There is a great variety of approaches developed to capture matters of discourse organi-
zation and text coherence. Two recent and well accepted models, the Rhetorical Structure
Theory RST (Mann & Thompson 1988) and the Segmented Discourse ­Relation Theory
SDRT (Asher & Lascarides 2003) share the basic assumption that ­discourse ­coherence is
achieved only if each utterance makes an illocutionary contribution to ­another utterance
in the context. This is achieved when discourse units establish different kinds of rhetorical
relations among each other thus creating a dynamic, hierarchical structure in discourse.
According to the models presented above, the rhetorical relations linking together the
contents of single discourse units can be basically of the following two kinds:

a. two units can display no dependency relation among each other but share the
same level of discourse hierarchy thus creating a multi-nuclear relation in the
terms of RST or a relation of coordination in the terms of SDRT
 Rhetorical relations and verb placement in early Germanic 

b. two units can build a dependency relation creating a hierarchical structure in

­discourse, i.e., a nucleus-satellite relation due to RST or a relation of subordina-
tion due to SDRT.

In order to show how verb placement participated in achieving discourse hierarchy in

texts of the early Germanic tradition, we chose the model of SDRT and provided some
additional characteristics. Although the inventory of individual discourse ­relations is still
under discussion, there is overwhelming agreement on the basic features distinguishing
coordination vs. subordination as the two basic types of linking. Both are associated
with prototypical rhetorical relations displaying some complementary features (Asher
& Vieu 2005). Coordination, which is prototypical for units situated on an equal level
of discourse hierarchy, is typically represented in the relation of narration. Narration is
established when e.g., two discourse units (α, β) display a temporal relation of succes-
sion and β continues the narrative sequence in discourse. By contrast, subordination is
typically represented in elaboration, i.e., when a unit β provides more detail on another
unit α situated on a higher level of discourse hierarchy. In this case, the two events (α, β)
temporally overlap. Further, the rhetorical relation of continuation applies when two or
more subsequent units β and γ are equally situated on a lower level of dependency with
respect to a higher unit α so that both β and γ represent elaboration on α.
Looking at the distinctive features of coordination vs. subordination in SDRT, we
discover a number of parallels between them and the discourse properties of the word
order patterns discussed in the foregoing data analysis. Turning to the distribution of
the V2-pattern in OHG first, we recall the fact that it appears in sentences providing
explanatory or supportive information like descriptions, characterizations, motiva-
tion etc. on a referent or action previously mentioned in the discourse. The events de-
scribed in such sentences temporally overlap with the state of affairs of the governing
situation. Thus, V2 in OHG indicates relations of subordination, i.e., elaboration and
continuation in discourse. By contrast, V1 in OHG never occurs in elaborative pas-
sages. In its first canonical domain – that of presentational or text-opening sentences,
V1 establishes the basis for subsequent elaboration, whereas in its second domain –
episode onsets within the text – it signals that a previous sequence of elaboration or
continuation is suspended and discourse returns to the level of main action. The same
function is also observed in patterns in which the verb is preceded only by a frame
adverbial. Our analysis on the relation between this pattern and pure V1 in OHG
gives reason to believe that the former pattern is an innovation in the system of OHG
which has major effects on the consolidation of the V2-phenomenon in root clauses
of modern German while V1 was marginalized in certain domain of colloquial style.
Comparing this picture with the one in the remaining early Germanic languages,
we have to observe that elaborative parts display patterns in which the preverbal do-
main is not restricted to the aboutness topic of the utterance as was shown for OHG.
However, in each of the languages examined, V1 was detected in different function-
al domains providing progress in narration. Like in OHG, the pattern displaying a
 Svetlana Petrova and Michael Solf

sentence-initial adverbial or particle also occurs in this function as an equivalent of

V1. However, V1 sentences may also contain frame adverbials or a particle in post-
verbal position. This indicates that V1 is a genuine Germanic pattern and a common
formal correlate of coordination in discourse in all early Germanic languages.
In this respect, the results from ON are especially suggestive. Although ON ­displays
a much later stage of language development towards a generalized V2, we nevertheless
encounter domains in which V1 serves to carry further the discourse by providing
chronologically sequenced units failing to establish a hierarchical order among them.
In these cases known as ‘Narrative Inversion’ V1 still displays typical properties of
­coordination in discourse by establishing the level of main action in text structure.

5.  Conclusions

The present study investigates some discourse-related properties of verb placement in

the early Germanic languages. It reconsiders a number of already known ideas about the
syntax of some individual Germanic languages by providing a larger empirical basis and
a more fine-grained classification of the discourse-related properties of ­particular word
order patterns as well as the conditions for their use. Above all, the pragmatic value of
the word order patterns was seen from the comparative perspective in order to show that
we have to deal with no isolated phenomena but with properties of a common Germanic
heritage.
The most important conclusion from the empirical part of the investigation is that
apart from OHG, other early Germanic languages also provide evidence for the claim
that verb placement plays a role in text-structuring and discourse organization. In West
Germanic languages other than OHG, instances of V1 as a main device of discourse
segmentation show up more or less regularly in the functional domains outlined for
OHG, though with a slightly different distribution and frequency than in OHG. A cru-
cial domain of difference in the early Germanic situation was discovered in the field
of supportive text parts providing more information on a preceding referent or situ-
ation. Here, V2 used in OHG but broadly generalized in OS seems to co-occur with
verb-final structures in OE. These differences in the early Germanic situation could be
made responsible for the development of different word order patterns in the modern
systems of these languages.
In order to explain the role of verb placement in early Germanic, we invoke the
distinction between coordinating and subordinating discourse relations as developed
in Asher & Lascarides (2003) and claim that at a certain stage in the history of the
Germanic languages, the position of the verb was a means for distinguishing the type
of rhetorical relation the sentence implies with respect to the previous context. We
outlined the main characteristics of coordination and subordination in the framework
of SDRT and pointed to the parallels which special word order types display with
­regard to these types of discourse linking. In the overall comparison between the
 Rhetorical relations and verb placement in early Germanic 

languages investigated, we discovered that the V1 pattern is a common formal corre-

late of coordination in discourse while relations of the subordinating kind are realized
by different syntactic means in the languages of the early Germanic tradition.

Acknowledgement

The present paper is an extended and revised version of a talk held at the 28th an-
nual meeting of the German Linguistic Society (DGfS) on February 24th, 2006 in
Bielefeld/Germany, within the program of Workshop 08 “‘Subordination’ vs. ‘coor-
dination’ in sentence and text from a cross-linguistic perspective”. We thank all par-
ticipants of the workshop for questions and discussions as well as the anonymous
reviewers for comments on an earlier version of this paper.

Primary texts

[Beo] Beowulf. Reproduced in facsimile from the unique manuscript British Museum Ms.
Cotton Vitellius A. XV. With a Transliteration and notes by Julius Zupitza. Second
edition containing a new reproduction of the manuscript with an introductory note
by Norman Davis. Published for the Early Text Society. London: Oxford University
Press. 1959.
[Hel] Heliand und Genesis. Hg. v. Otto Behaghel. 10. überarb. Aufl. v. Burkhard Taeger.
Tübingen: Niemeyer. 1996.
[Íslb] Íslendingabók. Hg. Wolfgang Golther. 1923. Ares Isländerbuch. 2., neu bearb. Aufl.
Halle: Niemeyer.
[Ldn] Landnámabók Ìslands. Ug. Finnur Jónsson. København: Thiele. 1925.
[T] Die lateinisch-althochdeutsche Tatianbilingue Stiftsbibliothek St. Gallen Cod. 56.
Hg. von Achim Masser, Göttingen: Vandenhoek & Ruprecht. 1994. (Studien zum
Althochdeutschen, Bd. 25).
[VienPs] Vienna Psalter Der Wiener Psalter. Hg. Heiko Uecker. Cod. Vind. 2713. Kopenha-
gen: Reitzel. 1980.

References

Asher, Nicholas & Lascarides, Alex. 2003. Logics of Conversation. Cambridge: CUP.
Asher, Nicholas & Vieu, Laure. 2005. Subordinating and coordinating discourse relations. Lingua
115(4): 591–610.
Bästlein, Ulf Christian. 1991. Gliederungsinitialen in frühmittelalterlichen Epenhandschriften.
Studie zur Problematik ihres Auftretens, ihrer Entwicklung und Funktion in lateinischen und
volkssprachlichen Texten der Karolinger- und Ottonenzeit [Europäische Hochschulschriften.
Reihe I. Deutsche Sprache und Literatur, 1167]. Frankfurt: Peter Lang.
Behaghel, Otto. 1932. Deutsche Syntax. Eine geschichtliche Darstellung. Bd. IV. Wortstellung.
­Periodenbau. Heidelberg: Carl Winter.
 Svetlana Petrova and Michael Solf

Braune, Wilhelm & Heidermanns, Frank. 2004. Gotische Grammatik. Mit Lesestücken und
Wörterverzeichnis. 20 edn, [Sammlung kurzer Grammatiken germanischer Dialekte. A.
Hauptreihe Nr. 1]. Tübingen: Max Niemeyer.
Brinton, Laurel J. 1996. Pragmatic Markers in English. Grammaticalization and Discourse
­Function. Berlin: Mouton de Gruyter.
Donhauser, Karin, Solf, Michael & Zeige, Lars Erik. 2006. Informationsstruktur und Diskursre-
lationen im Vergleich Althochdeutsch – Altisländisch. In Grenzgänger. Festschrift zum 65.
Geburtstag von Jurij Kusmenko [Berliner Beiträge zur Skandinavistik 9], Antje Hornscheidt
et al. (eds), 73–90. Berlin: Nordeuropa-Institut.
Drubig, Bernhard. 1992. Zur Frage der grammatischen Repräsentation thetischer und katego-
rischer Sätze. In Informationsstruktur und Grammatik [Linguistische Berichte. Sonderheft
4], Joachim Jacobs (ed.), 142–195. Opladen: Westdeutscher Verlag.
Enkvist, Nils Erik & Wårvik, Brita. 1987. Old English þa, Temporal Chains, and Narrative ­Structure.
In Papers from the 7th International Conference on Historical Linguistics, Anna Giacalone-Ramat,
Onofrio Carruba and Giuliano Bernini (eds), 221–237. Amsterdam: John Benjamins.
Eythórsson, Thórhallur. 1996. Functional categories, cliticization, and verb movement in
the early Germanic languages. In Studies in Comparative Germanic Syntax, Höskuldur
­Thráinsson, Samuel David Epstein & Steve Peter (eds), 109–139. Dordrecht: Kluwer.
Fourquet, Jean. 1938. L’ordre des éléments de la phrase en germanique ancien. Étude de syntaxe de
position. Paris: Les Belles Lettres.
Fourquet, Jean. 1974. Genetische Betrachtungen über den deutschen Satzbau. In Studien zur
deutschen Literatur und Sprache des Mittelalters. Festschrift für Hugo Moser zum 65. Geburt-
stag, Werner Besch, Günther Jungbluth, Gerhard Meissburger & Eberhard Nellmann (eds),
314–323. Berlin: Erich Schmidt.
Heusler, Andreas. 1977. Altisländisches Elementarbuch. 7 edn. Heidelberg: Carl Winter.
Hinterhölzl, Roland & Petrova, Svetlana. 2005. Rhetorical relations and verb placement in the
early Germanic languages: Evidence from the Old High German Tatian translation (9th
century). In Salience in Discourse. Multidisciplinary Approaches to Discourse, Manfred
Stede et al. (eds), 71–79. Münster: Stichting / Nodus.
Hinterhölzl, Roland, Petrova, Svetlana & Solf, Michael. 2005. Diskurspragmatische Faktoren
für Topikalität und Verbstellung in der ahd. Tatian-Übersetzung (9. Jh.). Interdisciplinary
Studies on Information Structure 3: 143–182.
Hopper, Paul. 1979. Aspect and foregrounding in discourse. In Syntax and Semantics, Talmy
Givón (ed.), 213–241. New York NY: Academic Press.
Kemenade, Ans van. 1997. V2 and embedded topicalization in Old and Middle English. In
­Parameters of morpho-syntactic change, Ans van Kemenade & Nigel Vincent (eds), 326–352.
Cambridge: CUP.
Kiparsky, Paul. 1996. The shift to head-initial VP in Germanic. In Studies in Comparative
­Germanic Syntax, Höskuldur Thráinsson, Samuel David Epstein & Steve Peter (eds),
­140–179. Dordrecht: Kluwer.
Kotcheva, Kristina. 2000. Om konjunktionerna og och en i isländskan. In Kors och tvärs
i nordistiken. Föredrag från det 4:e studentkollokviet i nordisk språkvetenskap [Kleine
Schriften des Nordeuropa-Instituts 20], Jurij Kusmenko & Sven Lange (eds), 24–35.
Berlin: Nordeuropa-Institut der Humboldt-Universität.
Kusmenko, Jurij. 1996. Die Periode – Syntaktische Haupteinheit der altisländischen Prosa. In
Berkovs bók. To Honour of Prof. V. P. Berkov, Jurij Kusmenko & Lilja Popowa (eds), 145–160.
Moscow: Impeto.
 Rhetorical relations and verb placement in early Germanic 

Lambrecht, Knud. 1994. Information Structure and Sentence Form. Topic, Focus and the Mental
Representations of Discourse Referents. Cambridge: CUP.
Lehnert, Martin. 1960. Poetry and Prose of the Anglo-Saxons, Vol. I: Texts. 2nd, revised edn, Halle
(Saale): Max Niemeyer.
Leiss, Elisabeth. 2000. Artikel und Aspekt. Die grammatischen Muster von Definitheit [Studia
Linguistica Germanica 55]. Berlin: Walter de Gruyter.
Lenerz, Jürgen. 1984. Syntaktischer Wandel und Grammatiktheorie. Eine Untersuchung an
Beispielen aus der Sprachgeschichte des Deutschen [Linguistische Arbeiten 141]. Tübingen:
Max Niemeyer.
Önnerfors, Olaf. 1997. Verb-erst-Deklarativsätze. Grammatik und Pragmatik. Stockholm:
Almquist & Wiskell International.
Mann, William C. & Thompson, Sandra A. 1988. Rhetorical structure theory: Toward a function-
al theory of text organization. Text. An Interdisciplinary Journal for the Study of Discourse 8:
243–281.
Pintzuk, Susan. 1996. Clitization in Old English. In Approaching Second. Second Position Clitics
and Related Phenomena, Aaron L. Halpern & Arnold M. Zwicky (eds), 375–409. California
CA: CSLI.
Ramers, Karl Heinz. 2005. Verbstellung im Althochdeutschen. Zeitschrift für germanistische Lin-
guistik 33: 78–91.
Ranke, Friedrich & Hofmann, Dietrich. 1988. Altnordisches Elementarbuch. 5 edn. Berlin: de
Gruyter.
Rauch, Irmengard. 1992. The Old Saxon Language. New York NY: Peter Lang.
Reinhart, Tanya. 1981. Pragmatics and linguistics: An analysis of sentence topics. Philosophica
27: 53–94.
Ries, John. 1880. Die Stellung von Subject und Prädicatsverbum im Hêliand. Straßburg: Trübner.
Sasse, Hans-Jürgen. 1995. “Theticity” and VS order: A case study. Sprachtypologie und Universa-
lienforschung 48(1–2): 3–31.
Schrodt, Richard. 2004. Althochdeutsche Grammatik II. Syntax [Sammlung kurzer Gramma-
tiken germanischer Dialekte. A. Hauptreihe, Nr. 5/2]. Tübingen: Max Niemeyer.
Sigurðson, Halldór Ármann. 1994. Um frásagnarumröðun og grundvallarorðaröð í forníslensku.
Reykjavík: Málvísindastofnun Háskóla ĺslands.
Simmler, Franz. 1998. Makrostrukturen in der lateinisch-althochdeutschen Tatianbilingue. In
Deutsche Grammatik. Thema in Variationen. Festschrift für Hans-Werner Eroms zum 60.
Geburtstag, Karin Donhauser & Ludwig M. Eichinger (eds), 299–335. Heidelberg: Carl
Winter.
Simon, Horst J. 1998. “KinnanS Eahna fei heid gfrein.” – Über einen Typ von Verb-Erst-
Aussagesätzen im Bairischen. In Deutsche Grammatik. Thema in Variationen. Festschrift
für Hans-Werner Eroms zum 60. Geburtstag, Karin Donhauser & Ludwig M. Eichinger
(eds), 137–153. Heidelberg: Carl Winter.
Stockwell, Robert P. 1984. On the history of the verb-second rule in English. In Historical ­Syntax,
Jacek Fisiak (ed.), 575–592. Berlin: Mouton de Gruyter.
 Index of subjects
A see relative clause, see
Across the Board Movement
(ATB)  7, 283–4, 292, 295
adjunct  13, 22, 60, 65–7, 75,
77, 81, 115–7, 126–9, 144,
210–1, 223, 226
adjunction  14, 22, 157, 223,
292–3
adposition  59, 60, 64–67
afterthought (AT)  23, 117–22
anaphoric shift  164–5, 169,
177, 179
and  3, 6–10, 15–9, 37, 45, 53,
62–5, 69–77, 101, 108, 110,
117–9, 129–30, 144, 188, 275
ATB see Across the Board
Movement
attachment point  126, 139, 149,
151–2
B 136, 152–7, 169, 173, 178,
Background  17, 24, 46, 123,
248–53
Background  210–3
backgrounding  19, 22, 108, 122,
123, 127–9
binding  4
cataphoric  274–6
of pronouns  293
quantifier  193
variable  197, 204, 207
C 129–131, 293
Centering Theory  155
clause  115–129, 163, 167–81
adverbial  104–10, 153, 202
complement  138, 172
coordinate  12, 63, 114
dependent vs.
independent  148,
187–214
main  4, 9, 12, 35, 42, 72–5,
162–7, 170–4, 182
superordinate  2, 109
subordinate clause,
see verb second
clause
clause combining (linkage,
linking)  1–7, 11–24,
89–111, 138, 187–9,
207–8, 214
asyndetic vs. syndetic  see relation
6, 15, 60
canonical (prototypical)
vs. non-canonical  145–146
187–209
hypotactic  15
paratactic  5, 7, 14–5, 21–2
clause linkage shift
(in translation)  96–97,
102–13, 152
see sentence splitting
coherence  16–17, 22, 33, 39–40,
225, 249, 266, 272
Common Integrator  10,
69–70, 120
communicative weight  2, 20,
140, 157
conjoining
coordinate vs.
subordinate  6–7
see conjunction,
see coordination
conjunct  10, 62, 115–22,
conjunction  8, 12, 59, 63–4, 67,
71, 118, 120, 26–4
conditional  24, 263–6
coordinating  60, 62, 65,
93, 293
logical  5–7, 14
subordinating  63–5, 68,
241, 245–8, 251–2, 258
verb phrase (VP) see
coordination
connection  58
asymmetric(al) vs.
symmetric(al)  61
discourse  73–76
hierarchical vs.
non-hierarchical 
58–69, 70, 81
semantic  68–73
syntactic  61–8
connective  59–60, 62–8, 71–72,
78–81, 117–9, 127–30,
Consequentiality  121, 140–2,
151, 156
constraint ranking  208
continuum  7, 341
clause-linking  14, 90,
108–12
semanto-syntactic  208
Continuum Approach  110, 112
conversational analysis  78, 80
coordinate
construction  6–14, 120
ellipsis  322
relationship  188–9
structure  290, 295, 301
Coordinate Structure
Constraint (CSC)  7, 9
coordination  24, 59, 60–2,
68–74, 77–82, 115–22,
129–31, 281–6
and-  6–7, 17–9, 24, 130
asymmetric (AC) 
9, 24–5, 274, 281–301
asyndetic (implicit) vs.
syndetic (explicit) 
93, 101
canonical (standard,
prototypical) vs.
non-canonical  7–10
clause (clausal)  7–11, 15,
21, 24, 101–2, 112
multiple  6
og- (Norwegian)  117–9, 144
parenthetical  7
  Index of subjects
phrasal  6
pseudo-coordination  see Occasion
7, 274
symmetric  282, 285, 292
unbalanced  7, 62
verb phrase (VP)  6–9,
101–11, 164, 285–6
see conjunction, see SLFC
coordinator  62–3, 65, 70–3
corpus  96–102,143,156, 246,
251, 253
parallel corpora  3, 21, 98, 136,
143 156
cross-linguistic  89–92, 112–3,
131, 161–2, 172–5, 182
CSC see Coordinate Structure
Constraint
D downgrading  116, 119, 129, 131,
decision hierarchy  161, 176,
179–81
deictic frame  165
Dependence  187–214
dependency  1–11, 95–7,
100–112, 138, 141, 145, 157
dependency cline  96
disjunction
conjunctive  24, 255,
266–70
free-choice reading of  271
or-  24
discourse constituent  78, 252
discourse organization  22, 89,
90, 135–6, 140, 152–3, 157
discourse relation  61, 68, 73–4,
77–82, 120, 123, 135, 217–8,
227– 30, 236, 248, 249–51
asymmetric  77, 140, 329
continuative vs. discon-
tiuative see relative
clause
coordinating vs. subor-
dinating  16, 23, 79,
139–50, 158, 228, 236,
241, 248–9, 252, 348
non-veridical  260
multi-level  21
multinuclear  17, 18, 34,
39, 45, 47, 54
nucleus-satellite  76, 78
symmetric  140, 156
see Background,
see Consequentiality,
see Elaboration,
discourse representation  16
multi-level  21, 48–55
Discourse Representation
Theory (DRT)  16
discourse segmentation  24, 51,
156, 255, 259–62, 348
discourse structure  59–60,
73–82, 115, 120–30, 136,
227, 230, 241–2, 247–52
discourse topic  16, 217–18,
225–7, 230–6
discourse unit  60, 74, 80
disjunction  24, 255, 257, 262,
263, 266–74, 277
free-choice reading
of  270–1
161, 164–5, 169–74, 183
Dutch  161, 162, 164, 170,
174, 182
E hypotaxis  1–2, 7–15, 20, 77, 95,
Elaboration  16–7, 36–37, 48–9,
54, 79, 125, 139, 149, 151,
228  I
embedding  11, 15, 54, 65– 8,
74–9, 109–12
English  130–1, 161–82
epistemic minimal unit  272
event  116–7, 120–30, 163–7,
170–83
landmark vs. trajectory  71
one-event interpretation  information packaging  80,
285–6
event integration  173
event selection  178–82
event subordination  24,
288–90, 293–7, 300–1
see Occasion
explicitness  106, 109, 246, 330
external argument  63, 64, 67,
296
F juxtaposition  5, 15 60, 95–6,
focality  80
focus  120–6, 130, 162, 174, 182
of attention  49, 54
forward deletion  283–4, 295
free dass-clause  23, 187, 199– 202,
207, 210–11, 214
French  3, 21–3, 63, 89–113, 146,
152, 161–77, 182, 217, 219,
226, 232–35, 241, 243, 252
fusing semantics  281–95, 301
G
gapping  8, 283, 321–5
genre  49, 52, 98, 154, 336
German  115–9, 126–31, 161–82
government  4, 63–8
gradient (category, phenom-
enon)  21, 62, 89–90, 108
H
Hauptstruktur  116
see main structure
Head-driven Phrase Structure
Grammar  13, 23, 139,
187–9, 208–11, 214
hierarchy  75, 81, 121, 126, 161,
178–9
HPSG see Head-driven Phrase
Structure Grammar
111, 140, 157, 187, 198
incrementality  95–6, 112,
152, 157
information density  154–5
information extraction
(in translation)
to the left  119, 122
to the right  124, 125, 128
89–92, 152, 155
information structure  11, 52,
124, 130, 138, 141, 162–73,
178–83, 307, 325
discourse  81, 89–91
integration  165–6, 173, 182
internal argument  63–64, 67
intonation  72, 80
J
309
see clause combining
L
landmark  70–1, 74–6
Latin  63, 331–4, 341, 346

linear sequence  63–7, 78
Linguistic Discourse
Model  78
M 152, 157
macroplanning  174–181
main clause phenomenon  25
main clause word order  25
main structure  116, 139, 140, 163
movement island  223
multi-level annotation
(MLA)  50–56
N 255–8, 260
narrative  117, 120–1, 124–6,
130, 161–82
Nebenstruktur  116
see side structure
non-integrated constituent  14,
219, 224, 230, 235
see orphan
non-monotonic
entailment  297
NPI (Negative Polarity Item)-
licensing  274–5
nuclearity  17, 21, 33–56
nucleus  36–9, 44, 47, 123
O quaestio  116, 164
Occasion  24, 66, 286–9, 301–2
OE see Old English
OHG see Old High German
Old English  330, 335–6
old (older, early) Germanic
(languages)  3, 25, 329–49
Old High German  25, 329–48
Old Indic  3, 11, 25, 307, 308,
310, 313, 314, 317, 319, 323, 325
Old Norse  330, 335
Old Saxon  321, 335, 339–48
orphan  19, 23, 187–8, 211–3,
218, 223–4, 227, 230, 236
orthographic sentence  4, 7
OS see Old Saxon
otherwise  272–3
P rhetorical  3, 16, 20, 25, 34,
parataxis  1, 7, 14–5, 93, 157,
187–8, 198
see coordination, see
juxtaposition
passive  163–4, 166, 169–71,
174, 178
Penn Discourse Tree Bank  50
PI see pseudo-imperative
PIDO (Principles of Incremental
Discourse Organization)  142, 156
Potsdam Commentary Corpus  (‘Satzrelativsatz’)
40, 51
procedural (operational)
meaning  10, 120
prominence  2, 12, 20, 23, 39,
44–5, 148, 155, 157, 173, 323
prosody  11, 25, 80, 199, 219
pseudo-imperative (PI)  24,
punctuation  3–5, 42, 72, 93,
188, 224, 333
colon  5, 91
comma  4–6, 15, 63, 72,
117–8, 124
full stop (period)  4–7,
15, 19, 63, 72, 93, 118,
122–4, 155, 157, 224
semicolon  5, 93, 155
see juxtaposition,
see orthographic
sentence
Q 89, 140, 248, 346–47
approach  18–9, 139, 150
model  17, 22, 116
Question under Discussion
(QuD)  10, 17
R RST Treebank  39, 42, 51
RC see relative clause
referential linking  67–8
referential structure  148, 150
relation
adjunct  60
asymmetric(al)  2, 297
causal  19, 45, 165–9, 173–4
coherence  34–5, 38–9, 45,
49, 54, 60, 73
conjunctive  40, 52–4
enablement  121
80, 329, 346–8
semantic  6, 54, 77, 245
symmetric(al)  6, 16–17, 77
temporal  6, 16–17, 77
see connection,
see discourse relation
Index of subjects 
relative clause (RC)  14, 60, 68,
106, 113, 117, 126, 128
appositive  14–5, 22, 135,
clause-related
 
143–8
continuative vs. discon-
tinuative  22, 141–2,
148, 152, 155–6
non-restrictive  22, 126–9,
135–8, 143, 148, 150–6
restrictive  12, 14, 198, 206
sentential  13
verb second  187, 197,
202, 214
wh-  23, 187, 195, 197–207,
210–4
‘weiterführender
Relativsatz’  138–9, 156
see weshalb, see wobei
Relevance Theory  8, 19, 117,
120
RFC see Right Frontier
Constraint
Rhetorical Structure Theory
(RST)  17–21, 33–56, 76–8,
right dislocation (RD)  217–22,
235
Right Frontier Constraint  16,
24, 38, 79, 126, 140, 250
RST see Rhetorical Structure
Theory
S
salience  2, 5, 10, 20–1, 33, 35,
40–7, 51–6, 154–5, 313
satellite  36–8, 44, 76–7, 123
scope of negation  193, 198, 201,
207, 210, 213, 293
Segmented Discourse
Representation Theory
(SDRT)  2, 15–9, 22–5, 38,
50, 54, 79, 89, 139–40, 150,
156, 248–9, 329, 346–8
sentence  4–5
categorical vs. thetic 
331–332
all-focus  196–200,
331–332
complex  187–88
  Index of subjects
compound  4, 68
identificational  338
presentational  331–332, 335
sentence boundary  4, 68, 81,
136, 205, 222–30
sentence length  94
sentence splitting  115–8, 124,
129
servitude  96, 108
side structure  17–18, 116
SIP see Strategy of
Informational Parsimony
SLFC (Subject-Lacking-
in-Fronted-structure
Coordination)  24–5,
281–302
state predicate  117, 120–3,
129–30
Strategy of Incremental
Parsimony (SIP)  152,
154, 157
Strong Nuclearity Hypothesis  syntactic structure  62, 67, 74,
38
stylistic preferences  89, 92
subject gap  8, 24, 281–3, 295,
298–301
subject selection  176–82
subjunction (subordinator)  12,
13, 63–73, 137
subordinate clause  2, 9–15, 25,
42, 63–4, 73, 75, 108, 109,
140–1, 161, 167, 169, 220–2,
241–5, 250, 253
canonical vs.
non-canonical  12–23,
187–209
detached  14, 23, 195, 236
embedded  15
integrated vs.
non-integrated  14,
197–209
see free dass-clause, see
relative clause, see
subordination
subordination  59–62, 68–9, 73,
76–81, 115–6, 161–83
clause (clausal)  157, 241,
251
modal  172
semantic  76
syntactic  60–1, 67, 81, 115,
130, 140–1, 144, 149,
152, 155–7
see event, see subordinate
clause
subordinator see subjunction
syntactic complexity  91, 94, 115
77– 81, 98, 128
T verb-second (V2) clause  191–5,
temporal sequence  70, 121
temporal shift  148, 164,
166–182
text comprehensibility  153, 157
text type  47–9, 95, 150, 152
thematic development  43,
48–9, 52
thematic role  75–7, 296, 298
thematic structure  54, 56, 90
topic  116, 161–8, 174, 177, 180–1
aboutness  332–347
topic-comment division
(separation)  331, 335
topic-focus-domain  340
topic shift  49, 54,155, 162, 234
topic time  148,  163, 165, 181
trajector  70–6
translational patterns  117
translational upgrading  141,
148, 150, 152, 156
U
und  61, 130, 292
und zwar  204
upgrading (translational)  22,
129, 135–157
utterance  4, 80, 155, 161, 169,
178, 339
see sentence
V
verb accentuation  307
verb placement (position)  25,
210, 329–1, 335–6, 339, 343,
347–8
202–5
dependent  23, 187, 202–6,
209–10, 214
relative  197
weil-  23, 187, 191, 214
W
weshalb (German)  68, 137, 144
wobei (German)  137, 144
word order asymmetry  284–5
word order constraint  176–83
 Index of names
A C
Aarts, Bas  2, 12, 19, 114
Abraham, Werner  11, 223,
236, 309
Altmann, Hans  191, 217–22
Asher, Nicholas  2, 16–24, 38,
50, 59, 77–81, 89, 126,
139–40, 149–50, 156,
225–9, 247–9, 260, 297,
345–8
Auer, Peter  203, 205,
218, 234
Austin, Jennifer R.  13
van der Auwera, Johan  187,
189, 207, 255, 257, 261
B 264–5, 274–6
Bader, Markus  322
Ballard, Michel  104–6
Banfield, Ann  211
Bästlein, Ulf Christian  de Swart, Henriette  44
333, 335
Bateman, John  37, 39, 45,
50, 52
Bátori, István  5
Beaman, Karen  112
Behaghel, Otto  329
Behrens, Bergljot  24, 116, 135
Bierwisch, Manfred  14, 60, 75
Blakemore, Diane  5, 7, 8, 10,
19, 70, 117, 120, 122
Blühdorn, Hardarik  16, 21,
25, 44, 60, 68–70, 193,
214, 230
Bolinger, Dwight  256, 258
Bosch, Peter  225, 234
Brandt, Margareta  20, 138,
140, 196, 198
Brants, Sabine  51
Breindl, Eva  60, 62, 70
Brinton, Laurel J.  335
Büring, Daniel  4, 11, 80–81,
225, 292
Butulussi, Eleni  205
Carlson, Gregory Norman  39,
42, 47, 51
Carston, Robyn  5, 7–10, 19, 70,
117, 120, 122, 125
Chafe, Wallace L.  80, 95
Chuquet, Hélène  92, 93, 106
Consten, Manfred  222
Cosme, Christelle  15, 20–3,
97–9, 152
Cristea, Dan  38
Cristofaro, Sonia  11–2,
20, 59–62, 76, 90,
108–12, 183
Crysman, Berthold  5, 9
Culicover, Peter W.  222,
D Gussenhoven, Carlos  309
Danlos, Laurence  38, 243
Dik, Simon  62
Dipper, Stephanie  51, 52
Doherty, Monika  3, 90, 91, 92,
152, 157
Donhauser, Karin  346
Drubig, Bernhard  309, 331
Dunkel, George  91, 308
E 292, 321–3
Eckhardt, Regine  289
Eisenberg, Peter  62, 64
Elhadad, Michael  45
F Hetland, Jorunn  2, 20, 310
Fabb, Nigel  211
Fabricius-Hansen, Cathrine  Heusler, Andreas  346
13, 18, 20, 59, 91–2, 95, 98,
115–6, 119, 123–4, 130–1,
138, 141, 144, 152, 157, 187,
189, 191, 207, 214
Fourquet, Jean  329, 336
Frank, Annette  282
Franke, Michael  24, 25, 257–8
Frey, Werner  164, 178, 209,
223, 225, 294
Fries, Norbert  37, 299
G
Geurts, Bart  271, 272
Gibson, Edward  40, 49–50
Gilquin, Gaëtanelle  98
Ginzburg, Jonathan  189, 207,
212
Givón, Talmy  109–11
Golebiowski, Zofia  44
Gómez-Txurruka, Isabel  269
Granger, Sylviane  3, 89, 97
Greenbaum, Sidney  28
Grosz, Barbara  39, 45, 52,
54, 155
Grote, Brigitte  35, 38
H
Haegeman, Liliane  211, 218,
223, 227
Haiman, John  4, 187, 189, 191
Halliday, Michael A.K.  50,
60, 93
Hannay, Mike  93–5
Hartmann, Dietrich  2, 20, 310
Hartmann, Katharina  283,
Haspelmath, Martin  5–7, 10,
59, 62–3
Hasselgård, Hilde  3, 90, 97–98
Heintze, Silvan  53
Hettrich, Heinrich  308–10
Hinterhölzl, Roland  235, 328,
330–1
Hoarau, Lucie  105
Hobbs, Jerry  39, 286, 287
Hoffmann, Karl  308, 318
Hofmann, Ludger  223–4, 230,
234
  Index of names
Höhle, Tilman N.  281–285,
292, 301
Holler, Anke  13–4, 23, 24–5,
59, 138, 139–43, 146, 148,
154, 156, 189, 195, 210, 212,
214, 230
Hopper, Paul  60, 64, 73, 183
Huddleston, Rodney  4–5, 8, 13
J Lohnstein, H.  68, 75, 206
Jackendoff, Ray  264–5, 274–6
Jacobs, Joachim  81, 313
Jasinskaja, Ekaterina  10
Johannessen, Janne Bondi  314
7–9, 12, 59, 62
Johansson, Stig  3, 97–8, 111
K Mackenzie, J. Lachlan  93–4
Kamp, Hans  16, 241, 271
Karhiaho, Izabela  5
Kathol, Andreas  189–90, 207
Kavalova, Yordanka  14, 216
Kehler, Andrew  9, 24, 39, 57,
225, 286–287
Keller, Rudi  191, 193, 194
Kiparsky, Paul  330
Klein, Jared  309, 313–4, 316,
325
Klein, Wolfgang  17, 19,
80–1, 116, 139, 150, 162–3,
177–8
Knott, Alistair  49, 54, 60
König, Ekkehard  189, 207
Kotcheva, Kristina  346
Kratzer, Angelika  265, 278,
281, 294, 295
Krifka, Manfred  11, 265
Kusmenko, Jurij  343, 346
L O
Lakoff, Robin  275
Lambrecht, Knud  178, 222,
232–5, 331
Lang, Ewald  5–7, 10–3, 69–70,
120, 158, 313, 322
Langacker, Ronald W.  11, 74–6
Larson, Richard  291
Lascarides, Alex  2, 16, 38, 50,
139–40, 149–50, 227–9,
237, 248–9, 260, 297, 329,
346, 348
Lawler, John M.  275
Lefèvre, Michel  13, 58
Lehmann, Christian  1, 2–6,
9–15, 90, 110, 152, 157, 187,
189, 191, 207
Lehnert, Martin  336
Leiss, Elisabeth  330
Lenerz, Jürgen  329–31, 333
Levison, Stephen C.  7–9
Lewis, David  271
Lobeck, Anne  299
Longacre, Robert  11, 59, 189
Lötscher, Andreas  48
Lühr, Rosemarie  11, 25, 309,
Lysvåg, Per  111
M
Maienborn, Claudia  217, 289
Mann, William C.  2, 17, 33–50,
59, 73, 76–7, 81, 89, 140,
248, 259, 346
Marcu, Daniel  38, 55
Martin, James R.  40, 52, 93
Matthiessen, Christian M.  93, 115–6, 123, 129–31, 136,
2,  5, 11, 15, 18, 20,  34–8,
44, 50, 60, 73–4, 93,
109–12, 140, 250
McKeown, Kathy  45
Meinunger, André  190, 203–6
Miltsakaki, Eleni  155
Molnár, Valéria  2, 20, 310, 313
Moore, Johanna  39, 50
Moser, Megan  39
N Reyle, Uwe  16, 241, 253
Netter, Klaus  189, 207
Noonan, Michael  13
Oberlander, Jon  225
Oirsouw, Robert van  284
Oksefjell, Signe  3
Oldenberg, Hermann  309–12,
316, 320, 324
Önnerfors, Olaf  333
Oppenrieder, Wolfgang  209
P 210, 212
Paillard, Michel  92–3, 106
Partee, Barbara H.  7
Pasch, Renate  4, 12–3, 21, 52,
60–69, 188, 191, 272, 309
Peters, Jörg  80, 314
Petrova, Svetlana  16, 25,  232,
329
Peyer, Ann  2, 13, 20, 138, 187,
189, 207
Pintzuk, Susan  330
Polanyi, Livia  50, 59, 60,
78–80, 250
Pollack, Martha  50
Pollard, Carl  188,  208, 210
Prasad, Rashmi  50, 52
Prévot, Laurent  24, 248
Progovac, Ljiljana  292
Pullum, Geoffrey K.  4–5,
8–9, 13
Pustejovsky, James  49
Q
Quirk, Randolph  14, 60, 62,
67–8, 85, 108, 109, 111
R
Ramers, Karl Heinz  331
Ramm, Wiebke  14, 16, 18–24,
144, 214
Ranke, Friedrich  344
Rauch, Irmengard  339
Redder, Angelika  5, 10
Reich, Ingo  5, 7–10, 24, 25,
281–2, 285, 288, 292–3,
295–6, 299, 301
Reinhart, Tanya  178, 332
Reis, Marga  2, 13, 187, 190–1,
199–207, 282, 285
Ries, John  329, 339, 343
Roberts, Craige  172, 263, 288
Rondhuis, Jan-Klaas  39, 45, 50
Rooth, Mats  7
Rosengren, Inger  313
Ross, John R.  8, 222, 270
Russell, Benjamin  256, 258
S
Sæbø, Kjell Johan  13
Safir, Ken  211
Sag, Ivan A.  188–9, 207–8,
Sanders, Ted  39
Santorini, Beatrice  189–90
Sasse, Hans-Jürgen  331, 334
Schaffar, Wolfram  309

Schecker, Michael  20, 60
Schein, Barry  289
Schilder, Frank  44
Schrodt, Richard  333–4
Schwabe, Kerstin  5, 322
Schwager, Magdalena  257–8,
264–5
Schwarz-Friesel, Monika  15, 17–8, 20, 33–47, 59–60,
234
Selkirk, Elisabeth  219, 322
Shaer, Ben  209, 223–4, 227
Shopen, Timothy  59
Sidner, Candace  40, 45, 52, 54
Simmler, Franz  333
Simon, Horst J.  333
Smith, Carlota  47
Smith, Michael  220
Solf, Michael  16, 24, 25
Solfjeld, Kåre  8, 15, 19–23, 91,
93, 115–6, 128, 135, 141
Sperber, Dan  9, 152
Stede, Manfred  19, 21, 35, 40,
51, 53, 76–8, 157, 225
Steinbach, Markus  203, 205
Sternefeld, Wolfgang  299
Steube, Anita  310
Stockwell, Robert P.  338
Stolterfoht, Britta  322
Stutterheim, Christiane von  Velde, John te  7–11
17, 19, 22, 80–1, 116–7, 124,
162–3, 180, 310
Sweetser, Eve  193
Szucsich, Luka  310
T Vikner, Sten  190, 203
Taboada, Maite  33, 73, 76–7
Tenbrink, Thora  44
Thompson, Sandra A.  2, 4, 11,
64, 73–77, 81, 109, 110–2,
140, 187, 189–91, 203, 250,
259, 346
Tichy, Eva  315
Tomlin, Russell S.  20
Truckenbrodt, Hubert  190,
203, 206, 290
Txurruka, Isabel G.  18
U Williams, Edwin  283
Uhmann, Susanne  191–5, 212,
218–9, 224, 229
Umbach, Carla  313, 322
Uszkoreit, Hans  189, 207
V
Vallduvi, Eric  314
Van Valin, Robert  59,
187, 189
Vanderauwera, Ria  94–95
Verstraete, Jean-Christophe  Zimmermann, Thomas Ede
12
Index of names 
Vieu, Laure  18, 23–4, 59,
77, 79–81, 89, 126, 139,
228, 248, 347
Vilkuna, Marina  314
Vinay, Jean-Paul  96
W
Wackernagel, Jacob  313
Webber, Bonnie  39, 50,
67, 273
Wechsler, Stephen 
190, 206
Wegener, Heide  60, 191,
193–4
Werlich, Egon  47
Wilder, Chris  283
Wilson, Deirdre  9
Wolf, Florian  40, 49–50
Wright, Georg Henrik von 
270
Wunderlich, Dieter  324
Z
Zeevat, Henk  9, 225
Zhang, Ning  5
Zifonun, Gisela  5, 7, 13, 60,
223–4, 230, 234
 
271–2
Studies in Language Companion Series
A complete list of titles in this series can be found on the publishers’ website, www.benjamins.com

103 Josephson, Folke and Ingmar Söhrman (eds.): Form and Functions. Apsect, tense, mood, diathesis,
and valency. Expected August 2008
102 Goddard, Cliff (ed.): Cross-Linguistic Semantics. 2008. xvi, 356 pp.
101 Stolz, Thomas, Sonja Kettler, Cornelia Stroh and Aina Urdze: Split Possession. An areal-
linguistic study of the alienability correlation and related phenomena in the languages of Europe. x, 546 pp.
Expected May 2008
100 Ameka, Felix K. and M.E. Kropp Dakubu (eds.): Aspect and Modality in Kwa Languages. 2008.
ix, 335 pp.
99 Høeg Müller, Henrik and Alex Klinge (eds.): Essays on Nominal Determination. From morphology
to discourse management. xviii, 366 pp. + index. Expected May 2008
98 Fabricius-Hansen, Cathrine and Wiebke Ramm (eds.): 'Subordination' versus 'Coordination' in
Sentence and Text. A cross-linguistic perspective. 2008. vi, 357 pp.
97 Dollinger, Stefan: New-Dialect Formation in Canada. Evidence from the English modal auxiliaries.
2008. xxii, 355 pp.
96 Romeo, Nicoletta: Aspect in Burmese. Meaning and function. 2008. xv, 289 pp.
95 O’Connor, Loretta: Motion, Transfer and Transformation. The grammar of change in Lowland Chontal.
2007. xiv, 251 pp.
94 Miestamo, Matti, Kaius Sinnemäki and Fred Karlsson (eds.): Language Complexity. Typology,
contact, change. 2008. xiv, 356 pp.
93 Schalley, Andrea C. and Drew Khlentzos (eds.): Mental States. Volume 2: Language and cognitive
structure. 2007. x, 362 pp.
92 Schalley, Andrea C. and Drew Khlentzos (eds.): Mental States. Volume 1: Evolution, function,
nature. 2007. xii, 304 pp.
91 Filipović, Luna: Talking about Motion. A crosslinguistic investigation of lexicalization patterns. 2007.
x, 182 pp.
90 Muysken, Pieter (ed.): From Linguistic Areas to Areal Linguistics. 2008. vii, 293 pp.
89 Stark, Elisabeth, Elisabeth Leiss and Werner Abraham (eds.): Nominal Determination. Typology,
context constraints, and historical emergence. 2007. viii, 370 pp.
88 Ramat, Paolo and Elisa Roma (eds.): Europe and the Mediterranean as Linguistic Areas. Convergencies
from a historical and typological perspective. 2007. xxvi, 364 pp.
87 Verhoeven, Elisabeth: Experiential Constructions in Yucatec Maya. A typologically based analysis of a
functional domain in a Mayan language. 2007. xiv, 380 pp.
86 Schwarz-Friesel, Monika, Manfred Consten and Mareile Knees (eds.): Anaphors in Text.
Cognitive, formal and applied approaches to anaphoric reference. 2007. xvi, 282 pp.
85 Butler, Christopher S., Raquel Hidalgo Downing and Julia Lavid (eds.): Functional
Perspectives on Grammar and Discourse. In honour of Angela Downing. 2007. xxx, 481 pp.
84 Wanner, Leo (ed.): Selected Lexical and Grammatical Issues in the Meaning–Text Theory. In honour of
Igor Mel'čuk. 2007. xviii, 380 pp.
83 Hannay, Mike and Gerard J. Steen (eds.): Structural-Functional Studies in English Grammar. In
honour of Lachlan Mackenzie. 2007. vi, 393 pp.
82 Ziegeler, Debra: Interfaces with English Aspect. Diachronic and empirical studies. 2006. xvi, 325 pp.
81 Peeters, Bert (ed.): Semantic Primes and Universal Grammar. Empirical evidence from the Romance
languages. 2006. xvi, 374 pp.
80 Birner, Betty J. and Gregory Ward (eds.): Drawing the Boundaries of Meaning. Neo-Gricean studies in
pragmatics and semantics in honor of Laurence R. Horn. 2006. xii, 350 pp.
79 Laffut, An: Three-Participant Constructions in English. A functional-cognitive approach to caused
relations. 2006. ix, 268 pp.
78 Yamamoto, Mutsumi: Agency and Impersonality. Their Linguistic and Cultural Manifestations. 2006.
x, 152 pp.
77 Kulikov, Leonid, Andrej Malchukov and Peter de Swart (eds.): Case, Valency and Transitivity.
2006. xx, 503 pp.