You are on page 1of 41

Danish Dependency Treebank

Annotation guide: Verbs


Matthias T. Kromann Department of Computational Linguistics Copenhagen Business School Line Mikkelsen Department of Linguistics University of California, Santa Cruz Dependent Complement roles Adjunct roles subject (subj) external topic (xtop) direct object (dobj) noun vocative (voc) indirect object (iobj) parenthetical apposition (appa) predicative (pred) verbal relative (rel) verbal object (vobj) subject-oriented modifier (mods) verb predicative (pred) object-oriented modifier (modo) quotation object (qobj) modifier (mod) predicative (pred) adjective adjectival object (aobj) modifier (mod) particle (part) locational-directional object (lobj) preposition prepositional object (pobj) modifier (mod) predicative (pred) locational-directional object (lobj) temporal-durational object (tobj) adverbial modifier (mod) adverbial object (avobj) predicative (pred) coordinator (coord) other conjunct (conj) punctuation (pnct)

Verbs Forms
In imperatives, the subject is obligatorily absent (in the semantic interpretation, the subject is retrieved by coreference with the intended listener; thus, a vocative within an imperative always coincides with the subject, but only by coreference). All other complements of the verb are realized normally.

In infinitives, there is no primary subject dependency, but the logical subject is in some cases encoded with a subject filler dependency. See our analysis of raising and control and our analysis of the special pronoun 'at'.

In present participles, the subject is obligatorily absent (in the semantic interpretation, it is retrieved from the noun or verb that the present participle modifies). When the participle modifies a noun attributively, acts as a predicative, or modifies a verb while preceding another of the verbs complements, all of its noun complements must be absent or be realized as morphological complements (as in "den syngende cowboy"). In all other situations, noun complements are realized normally. Most other complements of the participle are realized normally in both cases. Only certain verbs allow the use of the present participle in predicative constructions.

Semantically, the present participle in "Hun fandt ham liggende" can modify either the subject or the object. In this respect, present participles are just like adjectival modifiers (eg, "ngen" in "Lgen undersgte patienten ngen"). In the treebank analyses we have chosen not to include information about which participant adjectives or present participles modify, since it would make the manual annotation too complicated, and there is not always enough information in the corpus to decide this question. Gerunds behave like nouns in the external syntax and should be tagged as such. In the internal syntax, they are still verbs and can take prepositional objects and (slightly more unusually) nominal objects in accordance with the verb's valency frame. Subjects are never realized in gerunds, and even though the subject coincides with the possessor in a possessive, the gerund is still marked as a "possd" complement of the possessor, with no explicit subject filler between the possessor and the verb.

In perfect participles, there is no primary subject dependency, but all other complement roles are realized as normal. Perfect participles can function as modifiers of nouns and verbs, as verbal objects if they are uninflected, and as predicatives if they appear with adjectival "-n" or "-e" inflection (eg, "skret"-->"skren"/"skrne"). When the perfect participle acts as a predicative or as a modifier to a noun or a verb, there is no filler (semantically, the missing subject is retrieved from the modified noun, or from the subject or direct object of the modified verb).

When the participle acts as a verbal object (eg, of the auxiliaries "vre" ("be"), "have" ("have"), and "blive" ("become")), a subject filler is used to pass on the subject argument to the possibly passivized perfect participle.

When acting as modifiers of verbs, perfect participles should be analyzed as subject-oriented (mods) or direct object-oriented (modo) modifiers, if possible.

Perfect participles have an active form and possibly one or more passive forms. Verbs can be categorized with respect to whether the active or passive form of the participle is selected when the participle acts as verbal complement to the verbs "have", "vre", and "blive", or when it acts as modifier. This is illustrated by the following table: have vre blive verb classes and examples modifier (have) (be) (become) accusatives: ane bede bo bre dele f fle fre give gre have hbe hre koge kbe lade lave lege leve lide love lyde lre lse lbe mene mle nyde opn rive rre se sige sl sy sge tabe tage tale tro vide vise vre yde bne active passive passive passive
Hun giver brnene en ballon --> Hun har givet brnene en ballon (subj) --> ?Brnene er givet en ballon (?iobj) --> ?En ballon er givet brnene (?dobj) --> Brnene bliver givet en ballon (iobj) --> En ballon bliver givet brnene (dobj)

unaccusatives: blive d? falde forsvinde glde? komme lykkes ske stige synes
Kursen falder --> *Kursen[subj] har faldet --> Kursen[subj] er faldet --> *Kursen[subj] bliver faldet

active

active

neither: begynde foreg fortstte flge g komme kre medvirke mde n optrde rejse ske skynde slippe slutte sove springe standse starte stege stige st trde trnge undg vende virke
Vi gr en tur i skoven --> Vi[subj] har get en tur i skoven --> Vi[subj] er get en tur i skoven --> Ruten[dobj] er get af os mange gange fr --> Ruten[dobj] bliver get af os i morgen --> *Vi[subj] bliver get en tur i skoven

active

active/ passive passive

passive

vobj or pred? We have chosen to analyze uninflected perfect participles as verbal objects ('vobj') rather than subject-oriented predicatives ('pred'). The intuition behind the predicative analysis is that perfect participles are just like adjectives. The predicative analysis does have some advantages: it makes the analysis of perfect participles coherent with the analysis of copula constructions, explains why perfect participles have '-e' and '-n' inflected forms that closely resemble adjectives, and explains why perfect participles are so easily coordinated with adjectives (eg, "Han er vred og skuffet" ("He is angry and disappointed")). However, the predicative analysis also has some disadvantages: it gives a non-uniform analysis of perfect participles ('vobj' in "Han har spist" ("He has eaten"), but 'pred' in "Den er spist" ("It is eaten")), and does not seem to obey the usual number- and gender-agreement between subjects and predicatives (eg, "Mange *lsladt/lsladte/*god/ gode mnd" ("Many released/good men") vs. "Mange mnd er lsladt/?lsladte/*god/ gode" ("Many men are released/good")). One possible explanation is that perfect participles are genuinely ambiguous between being verbal objects and predicatives. However, we have been unable to find tests that would allow us to distinguish between these two alternative uses in practice. For this reason, we

have chosen to always prefer the 'vobj' analysis over the 'pred' analysis, except when the participle is explicitly marked with the adjectival '-e' or '-n' inflections. These considerations also apply to examples like "f den stjlet" ("have it stolen (from you)"). Such an example might be analysed as an example of a resultative construction. But it is definately not resultative to the same degree as "gre nogen glad" ("make somebody happy"). Besides rewording the predicative-like part ("den er stjlet" ("it is stolen"), "nogen er glad" ("somebody is happy")) reveals that following the considerations above, "stjlet" in "den er stjlet" should be analysed as a "vobj" to "er", not "pred", whereas "glad" in "nogen er glad" should be analysed as "pred".

Dependents
Noun complements: subjects, direct objects, indirect objects, and predicatives
Noun complements can be phrases headed by pronouns, common nouns, type-shifted adjectives, and gerunds; infinitival and sentential constructions headed by the special pronoun "at"; and embedded wh-questions headed by interrogative pronouns like "om", "hvem", "hvornr", "hvorfor", etc. Verbs may license four different kinds of noun complements: subjects ('subj'), direct objects ('dobj'), indirect objects ('iobj'), and predicatives ('pred'). Verbs must license exactly one subject ('subj') or expletive subject ('expl'), and never license more than one noun complement of each type -- in particular, expletives turn subjects into direct objects. Licensed noun complements are always optional, ie, they are not necessarily realized when the verb is actually used. For instance the following sentence licenses a subject, but does not realize it:

Noun complements should not be confused with the different kinds of noun adjuncts that a verb may have: external topics ('xtop'), vocatives ('voc'), and parenthetical appositions ('appa'). Verbal complements of modal or auxiliary verbs share the subject via a subject filler dependency (see our analysis of raising):

Different noun complement roles can be disambiguated with the following linguistic tests: subj vs dobj: (1) question test, (2) modal verb test, (3) adverbial test, (4) case marking test, (6) expletive test subj vs iobj: (1) question test, (2) modal verb test, (3) adverbial test, (4) case marking test subj vs pred: (1) question test, (2) modal verb test, (3) adverbial test, (4) case marking test dobj vs iobj: (1) question test, (6) expletive test, (8) PP-paraphrase dobj vs pred: (5) paradigm test, (6) expletive test, (7) agreement test

iobj vs pred: (5) paradigm test, (6) expletive test, (7) agreement test, (8) PP-paraphrase The linguistic tests are described in detail below: (1) question test: form the corresponding question, then the noun complements must occur in the order subj/expl < iobj < dobj/pred, and all adverbials must occur after the subject (eg, "At lse nsker hun ikke" --> "nsker hun ikke at lse?").

(2) modal verb test: insert a modal verb, then the subject either appears right after the finite verb, or in fronted position provided there is no noun complement or expletive right after the finite verb (eg, "Bogen giver Marie til Jonas" --> "Bogen vil Marie give til Jonas").

(3) adverbial test: insert an adverbial like "not" after the verb, then the adverbial must come after the subject, but can be inserted before any other noun complement (eg, "Fjolser var de" --> "Fjolser var de ikke").

(4) case marking test: replace the entire noun complement with a personal pronoun with nominative case-marking, then non-extracted subjects without restrictive modifiers and coordinators should have nominative case, whereas all other noun complements should have accusative case (eg, "Peter og ham legede ofte" --> "De/*dem legede ofte").

(5) paradigm test: replace the noun complement with non-NPs; a noun complement is a predicative if and only if it can be replaced with PPs and APs that can appear predicatively (other than type-shifted APs) (eg, "Han bliver en god bager" --> "Han bliver glad").

(6) expletive test: insert an expletive "der" in the sentence and change potential direct objects to indefinite noun phrases, then the result should be ungrammatical if the verb licenses a direct object or predicative, and grammatical otherwise (see expletives); notice that expletives turn subjects into direct objects (eg, "Pengene tilfalder hende" --> "Der tilfalder hende nogle penge" vs. "Pengene tiltrkker hende" --> "*Der tiltrkker hende nogle penge").

(7) agreement test: change number and gender of adjectival predicatives; adjectival subject predicatives should agree with subjects in number and gender; adjectival object predicatives should agree with direct objects in number and gender (eg, "De kalder dem *god/*godt/gode", "Hun gjorde den god/*godt/*gode igen", . (8) paraphrase: predicatives can usually be paraphrased by APs (cf. Hudson); indirect objects can usually be paraphrased as PPs (cf. Jennifer Herriman, 1995. "The Indirect Object in Present-Day English". Acta Universitatis Gothoburgensis. Gteborg, Sweden.); direct objects usually cannot be paraphrased by PPs or APs. De var venner => De var venlige mod hinanden Direct objects ('dobj') also occur in constructions with object-control (see control). Non-subject complements headed by "at" are usually analyzed as direct objects, but can also be predicatives (see our analysis of at).

Note that the reflexive object of socalled "inherently reflexive" verbs as well as "measure objects" are tagged as 'dobj', although these constructions do not allow passivization: Han skyndte sig. De skammer sig.

Den vejer to kilo. Den koster en hel herregrd. In the Dutch treebank (CGN) these two types of objects are tagged with separate dependency labels (SE for obligatorily reflexive complement and ME for measure complements). The motivation for distinguishing these constructions from regular transitive constructions (with direct objects) is that they do not passivize. For us, passivizability is not a necessary criterion for being a direct object. Instead, we propose to tag reflexive complements and measure complements as 'dobj' and encode the lack of passivization (and the restrictions on what can function as 'dobj' with these verbs) in the lexical entry for the verb.

Indirect objects
We use "iobj" for indirect objects of ditransitive verbs. Det har taget mig lang tid. Det blev forldrene for meget. Jeg kbte mig en ny cykel. Han er mig noget langsom.

Note that when this argument is expressed with a preposition ("Han viste den til hende") it is tagged as a prepositional object, and not as an indirect object. An alternative is to tag also the prepositional expression of the beneficiary as "iobj". Our reason for not doing so is the principle that "iobj" and "dobj" are always nominal. Another alternative is to tag the prepositional expression as a modifier ("mod"). The main reason for not doing this is that the "til"-expression behaves more like a complement than a modifier. It is (almost) obligatory, it is ordered before any modifiers (modulo extraposition), and it seems to express the same semantic role as the "iobj", which is definitely a complement.

Verb complements: vobj, pred, and qobj


Verbal objects
The label vobj is used for verbal complements. Bare infinitives occur as verbal complements of modal verbs where they share the subject of their governor via a subject filler. Bare infinitives also occur as complements to verbs of perception, and some object control verbs.

Past participles occur as vobj of auxiliary verbs, including the passive auxiliary "blive".

More complex verbal clusters form a chain of local vobj dependencies:

Predicatives
Verbal predicatives occur with present participles and with inflectet past participles:

Also verbal predicatives occur with object predicatives:

Quotational objects
The label "qobj" is used to indicate a connection bewteen a verb and a quotation. The label demands, that the sentence is clearly a quotation, and it demands that the verb semantically allows a quotation, like for instance "sige" and "sprge". Here are some typical verbs that take a quotational object: sige sprge svare mene tilfje rapportere konstatere fastsl

skrive oplyse forklare fortlle mumle grine smile Here are some examples of analysis:

If the quotation is split in two, the analysis is unchanged: "Spis din mad, s du kan blive en stor dreng," sagde Carla. "Spis din mad," sagde Carla, "s du kan blive en stor dreng".

Adjective complements: pred, aobj , part


Predicatives
Predicative complement The dependency label "pred" is used for predicative complements. Semantically, these can be oriented towards the subject or the object. A predicative complement can be (headed by) a preposition, noun, adverb, adjective, article (indefinite pronoun), or present participle. Subject-oriented predicative complements occur with copula verbs like "vre", "synes", "blive":

Object-oriented predicative complements are found in resultative constructions: De drak ham fuld. Hun lb skoene i laser. De gjorde manden gal. Vi holder byen ren/bilen i gang. and with verbs like "finde" and "kalde": Jeg finder ham stdig. De kaldte ham et fjols. Hun havde penge til_gode. Hun flte sig respekteret.

Fixed expressions: st parat/klar There are also constructions that could be considered prepositional predicative complements (at least this is what they are called in PFGA p. 77): De ans ham for utilregnelig. De regnede ham for idiot. Hun betragtede det som et uheld. Han opfattede det som en fornrmelse. These could be analyzed in terms of pred as follows:

Predicative complements must be distinguished from predicative adjuncts/modifiers, which describe some circumstance of the event denoted by the verb: Man br ikke kre bil fuld. (subject-oriented modifier) Hun kbte huset umbleret (object-oriented modifier) An alternative is to analyse the preposition as a pobj to the verb and the adjective as a dependent of the preposition. The problem with this analysis is that the preposition "for" does not normally allow adjectival complements. An alternative is to (more or less) follow the word class of the complement and tag prepositional complements as pobj, (pro)nominal complements as dobj, adverbial complements as part, adjectival complements as aobj, and present participle complements as vobj. This, however, seems to miss the unity of these constructions. The lexical entry for "vre" (and other copula verbs) in STO seems to take a middle position. Pred is used for nominal and adjectival complements, but prepositional complements are listed as pobj (e.g. "Sim var over ham med det samme", "Nogen havde vret efter ham i skolen", and "Hun var fra den egn"), and some adverbs (prepositions without a complement) are tagged as particles (e.g. "Bttens lg var p"). Til den sidste type hrer muligvis ogs udtryk som "vre til", "vre p", og "vre med". Copula constructions: These are basically covered in the sections on 'pred' and 'expl'. One remaining question is whether we should distinguish between predicative and equative/ specificational/identificational copula clauses, when both dependents are nominals: Han er en strk svmmer. (predicative) Den strkeste svmmer p holdet er ham. (specificational) Han er Peter. (equative) Peter er ham. (equative) Det er Peter. (identificational) Analysis of "som X"-constructions is done as follows: (1) Analyze "som X" as a relative, if possible:

(2) If you can insert "s vel" ("as well") in front of "som", then analyze "som X" as a coordination:

(3) If "som X" functions like a predication "er X" (ie, "historien som fortalt" ---> "historien er fortalt" / "story as told" ---> "story is told", then "som X" is analyzed just like "er X" ). -that is, "X" is a predicative "pred", unless "X" is a verb, in which case we choose "vobj" instead:

Moskva er ganske rigtigt en by i Rusland, som hvdet af Marie. If the verb is finite, the construction is analyzed as a relative clause; if the verb is perfect, "som" is analyzed as the head, and the verb as the "pred" complement of "som". The English Dependency Treebank distinguishes equative from predicational copular clauses, using an inversion test (equatives invert, predicational clauses do not) and an embedding test (predicational clauses embed as small clauses without "be", equatives do not). In predicational clauses the (head of the) post-copula phrase is the head of the clause, and the copula and the subject are its dependents. In equatives, the copula is the head, and the post-copular clause is an object. The Penn Treebank treats all copular clauses as predicational, including ones with two nominals. The post-copular phrase is uniformly analyzed as XP-PRD. The Corpus Gesproken Nederlands does not discuss copula clauses explicitly. Some predicational copula clauses are given under the discussion of the complement label 'PREDC'. This leaves two options for the Danish Dependency Treebank: (1) Treat all copula clauses as predicational -- tag all predicate complements as 'pred'. (2) Distinguish predicational copular clauses from equative/specificational/identificational copular clauses -- tag the complement of the former as 'pred' and the complement of the latter as 'dobj'.

Adjectival objects
Adjectival complements bear the dependency label aobj. Han krte trt. De for vild. Hunden gik amok.

An alternative is to treat these as predicative complements (pred). There are two arguments against this analysis. First, these verbs do not accept the full range of predicative complements (PPs, NPs, adverbials, present participles etc). Second, the semantics is not (always) predicative: "De for vild" ~/~> "De er vild(e)". Another alternative is to analyze the adjective as a particle (part). Some further examples: tage noget alvorligt have det godt/drligt/vidunderligt An alternative is here, too, to analyse "alvorligt" and "godt" as pred, but these constructions are not truly predicative: "noget er alvorligt" is not the correct analysis (since you can take something funny seriously) and neither is "det er godt" the correct analysis.

Preposition complements: pobj


We use "pobj" for complements headed by a preposition. Some prepositional complement express a (physical) location or direction, others express more abstract relations, including the "af"-phrase in a passive construction. Hun lagde den p bordet. (In this example, "p" would now be analysed as lobj to the verb (see the paragraph on lobj). Alternatively it should be analysed as a modifier, since there is no close connection between the verb and the preposition: "Hun lagde den p bordet/i spanden/under hovedpuden". Also "p bordet" goes with many other verbs as well). Han ventede p dem. Sagen blev undersgt af politiet.

Further special examples: Det virkede/ld/s ud som_om hun var glad. Hun lader/opfrer sig som_om hun var hjemme. Hun stemte ja til forslaget. ("til" pobj til "stemme") Han sagde nej til tilbuddet. ("til" pobj til "sige") The preposition "med" Usually "med" is analysed as mod to the verb, but fixed expressions are exceptions to this rule. vre med: "vre med": "med" is pred to "vre" "vre med til": "til" is pobj to "med". g med: "g med nogen", "g med i byen": "med" mod to "g" "g med hunden", "g med [bekldning]", "det gr [adjectiv phrase] med noget", "[tiden] gr med noget": "med pobj to "g" "g med til noget" ("accept"): "med" pobj to "g", "til" pobj to "med" komme med: "komme med et sted hen": "med" mod to "komme" "komme med et udspil", "komme med [adjectiv phrase] ud af det med nogen": "med" pobj to "komme" flge med: "flge med (nogen)": "med" mod to "flge"; "med" can be analysed as part, only if "med" cannot take an object in the particular construction "flge noget op med noget andet": "med" pobj to "flge" skulle med "det skulle med i billedet": "med" mod to "skulle", "i" mod to "med" have med: "have noget med": "med" mod to "have" "at have det med at grine": "med" pobj to "have" "have med noget at gre": "at" dobj to "have", "gre" vobj to "at", "med" pobj " to "gre", "noget" nobj to "med" lbe med: "lbe med nogen", "lbe med et sted hen": "med" mod to "lbe" "lbe med sladder": "med" pobj to "lbe" synge med: "med" mod to "synge"

hre med: "med" mod to "hre" se med: "med" mod to "se" spille med: "med" mod to "spille" To distinguish prepositional complements from prepositional modifiers, we use the following diagnostics (adapted from Philp (1999)). generality: prepositional complements only apply to particular verbs, whereas prepositional modifiers tend to apply to all verbs. fixed preposition: In prepositional complements that express abstract relations the preposition is fixed in the sense that it cannot be replaced with a semantically similar preposition (without a non-proportional change in meaning). This is not true for prepositional modifiers. Han ventede p dem/??i dem (pobj) Han ventede p stationen/i grden (mod) linear order: Prepositional complements tend to precede prepositional modifiers: Han venter p dem p stationen. --> ??Han venter p stationen p dem. ("p dem" is pobj, "p stationen" is mod) This really is only a tendency and cannot at all be used as an analysing guideline: Search in our tagged corpus reveals that a search for the construction verb + prepositional modifier + prepositional complement comes out with about 50 examples. A search for the construction verb + prepositional complement + prepositional modifier comes out with about 120 examples. The complement + modifier word order is clearly more frequent, but the word order modifier + complement is on the other hand quite common.

Adverbial complements: avobj


The label "avobj" applies to all adverbials analysed as complements (i. e. not displaying the optionality that characterize modifiers) and not interpretable as a particle, a locative-directional object or a time-indicating object (the last two, when applying to adverbs, being subcategories to the category adverbial object. Here are some examples: Han levede op til sit rygte. Foreningen gr ind for dyrevelfrd. Forhandlingerne stod p i 3 dage. Projektet gr ud p at beskytte sen. Hun ville gerne holde op med at ryge. Lad sovsen dampe lidt ind. De dukker aldrig op til tiden. Vi har god tid til at tnke tingene igennem.

The following principles can help distinguish adverbial objects from particles: Word class: Avobj applies only to adverbs, not to prepositions. Particles on the other hand apply to both adverbs (e.g. "give op") and prepositions (e.g. "skrive under"). Prefixes: Particles are separable prefixes, which together with the verb form a synonymous complex word form (skrive under -> underskrive, give op -> opgive). Adverbial objects cannot do this (at least not with a preservation of meaning). Objects: Particles do not take an object. If it takes an object, it is converted into a pobj. Adverbial objects, on the other hand, often combine with a preposition (ind i, op p). This preposition is analysed as a modifier (mod) or a prepositional object (pobj) according to how close the connection between the adverb and the preposition is in the particular phrase, i.e. whether the phrase is a fixed expression ("kre ind til/i/p/under" = modifier; "leve op til" = fixed expression that does not allow a change of the preposition). Adverbial objects differs from lobj, tobj and modifiers too: Lobj apply only if there is a distict locative-directional meaning involved. Tobj apply only if there is a distinct time-indicating meaning involved. Modifiers are caracterized by being optional and combining with most verbal phrases. Lobj (locative-directional object) and tobj (time-indicating object) should be understood as sub-categories to the category avobj. Again, these are complements, i. e. not optional. Below we give some examples and a list of typical locative-directional and time-indicating adverbs and typical verbs/verbal phrases taking a locative-directional or time-indicating object. Locative-directional: Langsomt kravlede de op ad bjerget. Min datter bor stadig hjemme. De unge rejser ud i verden.

Time-indicating: Han blev fdt den 22 august 1917. Bogen udkommer p mandag.

Forestillingen varede 3 timer.

Particles
We use the label part for verbal particles. These are adverbs or prepositions without a complement, which form a close semantic union with the verb. Han skrev kontrakten under. De gav op. De gik sagen igennem. Giften slog hunden ihjel.

The particle is obligatory, in the sense that leaving it out or exchanging it for a regular prepositional phrase results in a different semantics of the verb: Han skrev kontrakten [= He wrote the contract] Han skrev kontrakten under [= He signed the contract; "under" is part] Han skrev kontrakten under bordet [= He wrote the contract under the table; "under" is mod] Some, but not all, particles are separable prefixes, which together with the verb form a synonymous complex word form: skrive under ~ underskrive give op ~ opgive melde sig til ~ tilmelde sig [though "til" might be a 'pobj', since it can occur with a 'nobj' "melde sig til kurset"] The following principles can help distinguish particles from adverbial objects (avobj) and prepositional objects (pobj): Word class: Avobj applys only to adverbs, not to prepositions. Pobj applys only to prepositions. Particles on the other hand apply to both adverbs (e.g. "give op") and prepositions (e.g. "skrive under". Prefixes: Particles are separable prefixes, which together with the verb form a synonymous complex word form (skrive under -> underskrive, give op -> opgive). Adverbial objects cannot do this (at least not with the preservation of meaning). Prepositional objects cannot do this either. Objects: Particles do not take an object. If it takes an object it is converted into a prepositional object. Prepositional objects always carry the possibility of an object ("melde sig til" -> "melde sig til kurset"). Adverbial objects

often combine with a preposition (ind i, op p), which is tagged as either mod or pobj. See above under Avobj. For particles that are prepositions by word class an alternative analysis is to tag them as pobj, with the requirement that they be used without a complement. This complicates the analysis of prepositions, which are normally required to take a complement. For particles that are adverbs, an alternative analysis is to tag them as modifiers. Some difficult cases from the set of sample analyses (see also the section on adverbial + PP in the chapter on preposition): De var med. [ex. 29, "med" is probably 'part' and not 'pred', since "med" is not used with the other copula verbs with this meaning (??synes med, *blive med), but it is used with non-copula verbs with the same meaning (komme med, g med, etc.)]"Vre med" is analysed as pred, since "vre" always takes a pred. See under prepositional objects above. Jeg har svrt ved at flge med. [ex. 52, "med" is probably 'part', even though "med" can occur with an nobj ("flge med dem"). These could be considered two different constructions. In the first "med" is `part', in the other it is `pobj'] "Flge med" is usually analysed as mod. See under prepositional objects above. Lnmodtagerne har mulighed for at mde op p generalforsamlingen. [ex. 65, should be tagged as tagged as 'part', and not `pobj' as in the sample analysis of ex. 65] "Op" in "mde op" would be analysed as avobj. lgge sin stemme om til heltetenor [ex. 61 -- "om" should be tagged as 'part', cf. the existence of "omlgge"] In a limited number of cases adjectives can play the part of a particle. Here are some examples: fritstille - stille frit fastholde - holde fast fastsl - sl fast

Locative-directional objects
Lobj applies to adverbs ("rejse ud i verden"), prepositions ("rejse til Frankring"), pronouns ("kre en tur"), nouns ("g tur") and adjectives ("ligge forrest"). Lobj can be seen as a sub-category to avobj, if headed by an adverb, to pobj if headed by a preposition, to nobj if headed by a noun or pronoun. The same demarcating principles, that apply to avobj, pobj and the rest, apply to lobj. Special principles for the use of lobj are that it indicates a locational-directional meaning. Also it should be possible to replace a locative-directional phrase with other locative-directional phrases. Locative-directional objects: Min datter bor stadig hjemme. De unge rejser ud i verden. Det samme vil ske i Danmark. De krte gennem Europa. Han gr tur med hunden hver dag. Det ligger et helt andet sted. Hun stillede posen allerinderst. Vi er get for langt.

List (unfinished) of verbs taking a lobj: arbejde ("Han arbejder i et kontorvarefirma.") befinde sig ("Manden befandt sig i sin lejlighed.") begynde ("Jeg begyndte p den nye skole i august.") bevge sig ("De bevger sig rundt mellem hinanden.") bo ("Hun bor i Gren.") dreje ("Bilen drejede vk fra vejen.") dukke op ("Den dukkede op hos en af hans kammerater.") fare ("Hunden fr ud p gaden.") finde (uden dobj) ("Brevduerne finder selv hjem.") finde sted ("Arrangementet finder sted i London.") flakke ("Journalister flakker verden rundt.") flytte ("Brnene er flyttet til hovedstaden.") foreg ("Filmen foregik i det gamle Kbenhavn.") fragte ("Han fragter udstyret rundt i byen.") fdes ("Marie er fdt p Tsinge.") f ("Kartl fik strke smerter i benene.") gemme sig ("Katten gemte sig under sengen.") g ("Hun gik ind gennem dren.") havne ("Nogle flygtninge havnede i et fjernt land.") hente ("Brandmndene hentede deres ejendele ud.") hjlpe ("Mange blev hjulpet vk fra slagmarken.") hlde ("Hld blandingen i en skl.") jage ("Fuglene skal jages vk fra frugttret.") komme ("Han kom ud fra skuret med spindelvv overalt.") kigge ("Hun kiggede ned p gaden.") kravle ("Barnet kravlede ud p grsplnen.") kre ("Vi krte til Svendborg.") lande ("Landsholdet lander i Kbenhavn i aften.") ligge ("Huset ligger p en bakketop.") lbe ("Alle lb ud p gaden og rbte og hujede.") n ("Mette nede endelig frem til udkigspunktet.") opvokse ("Pigerne er opvokset p landet.") rejse ("De gamle hippier rejste til Indien.")

rende ("Drengen rendte op p sit vrelse og smkkede dren.") rykke ("De fredsbevarende styrker rykker ind i morgen.") se ("Han s ud p mig gennem ruden.") sejle ("Kongeskibet sejler gennem Lilleblt i dag.") sende ("De gamle blev sendt afsted p ferie.") sidde ("Malkepigen sidder p en lille skammel.") skal ("Nu skal de til Polen i stedet for Bulgarien.") ske ("Denne udviklen vil ogs ske i Danmark.") skynde sig ("Efter festen skyndte de sig hjem.") slippe ("Fangerne slap ud af fngslet om natten.") smutte ("Hesten smuttede gennem bningen i hegnet.") spille ("Michael Laudrup spillede ogs i Italien.") stige ("Clara steg ind i bilen.") stille sig ("Hun stillede sig foran ham med hnderne i siden.") st ("De stod lnge p banegrdspladsen og ventede.") tage ("De unge tog i byen, da festen ebbede ud.") tilbringe (tid) ("llet tilbringer mange mneder i de store tnder.") trisse ("Bedstmor trissede ud i haven og satte sig.") trkke ("Varerne er nu trukket tilbage fra butikkerne.") tumle ("Han tumlede ud af sengen.") tffe ("Den gamle bil tffede hen ad vejen.") vandre ("S begyndte de at vandre op i bjergene.") vende ("Nete vendte hjem fra udlandet i sidste uge.")

Time-indicating objects
Tobj applies to adverbs, prepositions, pronouns and nouns. Tobj can, as lobj, be seen as a sub-category to avobj, if headed by an adverb, to pobj if headed by a preposition, to nobj if headed by a noun or pronoun. The same demarcating principles, that apply to avobj, pobj and the rest, apply to tobj. Special principles for the use of tobj are that it the phrase expresses a time-indicating meaning and could be replaced by other time-indicating phrases. Time-indicating objects: Koncerten vil finde sted i_dag. Bogen udkommer p mandag. Han blev fdt den 22 august 1917. Forestillingen varede 3 timer.

List (unfinished) of verbs taking a tobj: arbejde ("Han arbejder kun et par timer om dagen.")

begynde ("Afstemningen begynder mandag klokken 12.") dukke op ("Den ukendte indspilning dukkede op i 1988.") finde sted ("Koncerten vil finde sted i_dag.") foreg ("Filmen foregr i trediverne.") ske ("Det sker hver eneste dag.") slutte ("Konferencen slutter klokken 5.") tage ("Turen derop tager nsten 10 timer.") udkomme ("Bogen udkommer om to dage.") vare ("Kurset varer 2 uger.")

Modifiers
Most verbal modifiers are annotated as mod (the three exceptions to this are relative clauses, vocatives, and external topics). The modifers tagged as mod include: locative and temporal expressions: mandag, i haven, tidligt, altid, aldrig, mens vi venter, efter de gik, imens, ofte, da vi kom hjem, allerede, nu, inden det bliver mrkt, her, lnge, hidtil, nr, p femte r manner adverbials: med tryk p, langsomt, aggressivt, godt(?), hvordan (discourse) particles: dog, nok, jo, vel, nu(?), alligevel, nemlig, imidlertid(?), faktisk(?) expressions of modality: mske, muligvis, sikkert, faktisk negation: ikke focus particles: kun(? - jvf. ex 6), ogs, igen(?), purpose clauses: for at trkke penge til, hvorfor conditional clauses: hvis det trkker, gr du det en gang til absolutives: med/uden hende p holdet others: i mit tilflde, til gengld, sagt p en anden mde, ligeledes, klart, -mssigt, foruden ham, bl.a., for femte gang, derfor, idet at vi er forberedte p det, selv, In some cases it is not clear from the syntax which element the modifier modifies (e.g. when a PP can modify either the direct object or the verb). In these cases, we follow the principle of low attachment: attach the modifier as low as possible (e.g. to the direct object rather than to its governor). This should not be interpreted as a claim about what speakers/language users do, but simply as a way of choosing an analysis in cases where the syntax allows more than one.

Series of modifiers
When several modifiers follow each other and they are more tied to each other than to the verb, the first modifier should be made the head of the entire modifying phrase:

Special modifiers
Modo and Mods Modifiers attached to the verb can sometimes be directed very specifically towards the subject of the verb or the object of the verb. In that case the more specific labels "modo" (object modifier) and "mods" (subject modifier) can be used. A test for whether a modifier is directed towards the subject or object is that modo's and mods's can function also as a modifier in the nominal phrase, thus: Lad dejen hvile tildkket i en time -> Lad den tildkkede dej hvile i en time Brnene sider vandkmmede og trtte ved bordet -> De vandkmmede og trtte brn sidder ved bordet Here are some examples of analysis: Examples of modo:

Examples of mods:

Modp Another sub-category to the class of modifiers are parenthetical modifiers (modp). Some examples follow here: Nu er det jo ikke ministeren, der finder p dette (sknt han for_tiden er temmelig opfindsom). Sledes har jeg personligt, op til flere gange, set Dronningen g tur p Strget. Her ser vi - overraskende - at kabinen er krympet en anelse.

Constructions
Passives
Passivization converts the subject of an active verb into an "af"-complement (tagged as pobj), and converts either a direct object, an indirect object, or (more limited) the nominal object of a prepositional object into a new subject (subj). Passivization can be expressed morphologically (verb stem + -"s") or syntactically ("blive" + past participle). Sagen undersges af politiet. [dobj -> subj] Sagen bliver undersgt af politiet. [dobj -> subj] Peter blev givet en stor gave. [iobj -> subj] Sagen bliver set p af politiet. [nobj of pobj -> subj]

Almost all verbs in Danish have a passive infinitive, present, past and past participle form. The passivization converts the subject into an "af"-complement of the verb, and converts either a direct object, and indirect object, or the nominal object of a prepositional object into a new subject. Passive verb forms occur as complements (vobj) to modals and "be": Han skal undersges. Han er blevet undersgt. Alternatively, one could analyze "af"+nominal as a modifier to the verb.

Expletive subjects
Expletive subjects (labeled "expl") mostly occur with the word "der", but locative PPs and the place adverbial "her" is sometimes used as expletive subject as well. Any verb that does not have a direct object can undergo expletive shift -- that is, the subject of the verb is converted into a direct object (labeled "dobj") with the restriction that the new direct object must be indefinite, and a locative or temporal adverbial (usually "der") becomes the expletive subject (labeled "expl"). Der mangler en gaffel. Her er tale om et klokkeklart mord. Der er tilfaldet den ldste datter en stor pengesum. I en drm vil benbares Guds forml med dig, Johannes.

Verbal complements of the finite verb share the expletive subject via a [expl] filler.

It is important to distinguish expletive subjects from fronted adverbials. In clauses with a complex verbal form, this is done by the position of the logical subject: if the logical subject immediately follows the finite verb (below left), the initial constituent is a modifier (mod) and the logical subject is subj. If the logical subject follows the right-most verbal form (below right), the adverbial is expl and the logical subject is dobj.

If there is only one verb, try inserting an auxiliary "vil". Alternatively, use the definiteness of the post-verbal nominal as a guide: if the post-copular nominal is definite (below left), it is the subject of a non-expletive construction (and the initial "der" is a modifier). If the post-copular nominal is indefinite (below right), it is a direct object of an expletive construction (and "der" is an expletive subject).

An alternative analysis is to tag the expletive as a regular subject (subj), but this analysis obscures the fact that the presence of the expletive subject affects the valency of the verb: it is only with expletive subjects that these verbs allow direct objects. Another alternative is to analyze the transformed subject as a subject rather than a direct object. However, this fails to capture the fact that the post-verbal nominal in most respects behaves as a direct object and not as a subject: it is accusative (not nominative), it appears exactly in the position of direct objects, and it does not allow the presence of another direct object.

Vocatives
The adjunct label voc is used for nominals that refer to the intended listener, but serve no complement function with respect to the verb. Luk s den dr, Marie! Marie, vil du/I godt lukke den dr?

Marie, jeg lukker dren nu. Vocatives are characterized by being either: pronouns (2nd person): du, I definite adjectives (pos./superlative): (du/I/min) elskede, sde, gamle, sdeste indefinite noun: dreng, gamle mand, far, smed, hest, brumlebi The vocative must be placed either at the beginning or the end of the sentence. A vocative can also be placed in the right field of a verbal phrase before any nominal or prepositional objects, and before any adverbials, but possibly before verbal objects and "at"-infinitives. Examples: Marie, luk s den dr! Luk s den dr, Marie! Marie, vil du/I godt lukke den dr? ?Vil du/I, Marie, godt lukke den dr? Vil du/I godt lukke den dr, Marie? Marie, jeg lukker dren nu. Jeg lukker dren nu, Marie. Vil du [Marie] vre s venlig [Marie] at lukke den dr nu [Marie]? Far/sde/elskede/min sde, vil du komme med kaffen? Vil du komme med kaffen, far/sde/elskede/min sde? Mnd/*drengene/fremmede Broder/gamle mand/kammerat, hent lige avisen. *Jeg/*mig/du der/*ham-hende-den-der/*vi/I drenge/jer drenge/dem luk s den dr. Luk nu den dr, du/I der/I. Luk nu, drenge, den dr! Kom nu herhen, hest!

External topics
External topics are sentence-initial or sentence-final phrases that are duplicated by a pronoun within the sentence. These are tagged with the special adjunct edge xtop. Marie, hun er da lidt underlig idag. Er hun ikke lidt underlig idag, hende Marie? Marie, hende har jeg ikke set meget til p det seneste.

If the main sentence is V2 ("declarative"), then there is a strong preference for the resumptive pronoun to immediately succeed the external topic by topicalization. If

the main sentence is V1 ("interrogative"), then normal word order is preserved, and the resumptive pronoun does not have to immediately succeed the external topic. The construction "nr X, s Y" is analyzed as an external topic construction as well:

External topics are sentence-initial phrases that are duplicated by a pronoun within the sentence. If the main sentence is V2 ("declarative"), then there is a strong preference for the resumptive pronoun to immediately succeed the external topic by topicalization. If the main sentence is V1 ("interrogative"), then normal word order is preserved, and the resumptive pronoun does not have to immediately succeed the external topic. Marie, hun er da lidt underlig idag. Marie, er hun ikke lidt underlig idag. Marie, hende har jeg ikke set meget til p det seneste. ??Marie, jeg har ikke set meget til hende p det seneste. Og brnene, har du hrt noget til dem for nylig? ??Og brnene, du har vel ikke hrt meget til dem for nylig? Grn, det er nu ikke nogen pn farve. Grn, synes du virkelig det er en pn farve? Skrive opgave i dag, det kan jeg ikke forestille mig at han vil. Skrive opgave, tror du han vil det? Nr han er sulten, s spiser han altid for meget.

Raising and control


In raising and subject control structures, the infinitive shares the subject of the raising/control verb via a subject filler:

In object control structures, the subject filler for the infinitive is the direct object of the control verb:

Note that some control verbs take a bare infinitival complement (without "at"):

For a good analysis of raising and control, see Hudson's Internet Encyclopedia. Control verbs may produce both object fillers as well as subject fillers. Here is an example of a direct object filler:

The motivation for including the subject fillers for verbal complements is to indicate that they necessarily have the same subject (semantically speaking) as their governor. Is the analysis of "jeg anser ham for at vre plidelig" compatible with the analysis suggested for "jeg anser ham for plidelig"? Below is a list of the control verbs for which we have consequently added a subject filler dependency: afvise agte begynde beslutte elske forekomme foretrkke form forsge hbe love magte n pleje prve synes (se ud til) sge tilstrbe vente vlge nske This is an incomplete list. We have not yet had the resources to work on a more complete list. There are also some problems concerning parallel constructions with "V + P + at", e.g. "Han vil prve p at komme til tiden" ("He will try to be on time"). These constructions should have a subject filler too, but again we have not had the resources to do the necessary analysing work in order to have a complete picture of these constructions. We have therefore not added a subject filler in such constructions.

Relatives
Relative clauses that modify a clause are analyzed as adjuncts of the finite verb. For these we use the special adjunct edge rel, which is also used for relative clauses that modify nominals. Jeg har lavet kage, som du bad mig om. De har restaureret kirken, hvilket har pyntet en hel del.

Relatives are constructions where a clause (called the relative clause) attaches itself as a "rel"-adjunct to its governor (ie, the relativized phrase). The relativized phrase must satisfy a secondary role within the relativized clause, either by having a secondary filler dependency to some governor within the relative clause, or by being the antecedent of some relative or interrogative pronoun within the first phrase in the relative clause. The verb in the relative clause must either have V2 word order, or V3 verb order where the first

Relatives can be quite confusing, when there is more than one and it is combined with it-clefts and and expletives:

The relative pronoun is analyzed as a dependent of some head inside the relative clause. In the case of subject and object relatives, the governor is a verb. In other cases the relative pronoun is a dependent of a preposition: kvinden til hvem han sendte brevet ham p hvem vi ser ham som vi ser p det tidspunkt p hvilket han mente vi var skre

Note that whether the preposition is pied-piped with the relative pronoun to the front of the clause or not makes no difference for the dependency structure (without pied-piping we simply have a discontinuous PP). It does make a difference for the choice of relative pronoun: "hv"-pronoun for pied-piping and "som" when the preposition is not pied-piped. This correlation is captured in the lexicon. The relative pronoun is anaphorically dependent on the element that the relative clause modifies, which we indicate with a secondary dependency edge labelled ref. The relative pronoun may itself have dependents, as in the case of the genitive relative pronoun "hvis":

Some terminology: relativized word = word that has been extracted from below the relative verb; relative verb = verb that heads the relative clause; relativizer = optional relative pronoun or prepositional phrase with embedded relative pronoun. In relative clauses without a relative pronoun, the function of the modified element with respect to the internal syntax of the relative clause is indicated via a filler dependency:

Relative clauses that modify a clause are analyzed as adjuncts of the finite verb. For these we use the special adjunct edge rel, which is also used for relative clauses that modify nominals. Jeg har lavet kage, som du bad mig om. De har restaureret kirken, hvilket har pyntet en hel del.

The relative pronoun does not always pick up the reference from the last phrase of the relativized sentence. Especially in constructions with "det + vre + pred", the relativized phrase is "det", if "det" does not refer to something previously mentioned.

A reconstruction reveals the correct analysis: *Den dygtigste, der vinder prisen, er det" Det, der vinder prisen, er den dygtigste Free relatives / embedded wh-questions. Free relatives are relatives where the relativized phrase is a wh-phrase. Some examples are shown below: De spurgte hvem der stjal kagerne. Han spiste hvad de havde. Gr hvad du vil. (PFGA, p. 156) Jeg brd mig ikke om hvad jeg s. (PFGA, p. 156) Han vidste hvornr jeg ankom. Han spurgte hvad de manglede. Hvem de valgte, er ikke afgrende. Debatten handlede om hvorvidt stavepladen var fup. Hvor der ikke er mdepligt, vil der vre opgaver. These examples are analyzed just like normal relative clauses:

Wh-relatives differ from normal relatives by not allowing the relative pronoun "som" for objects and subjects, but otherwise they have the same properties, including the obligatory use of the relative pronoun "der" in subject relatives and the optionality of the relative clause, as in: Hvem der derimod kom ind i stuen, det var Marie. Denn wem die Magie einer solchen Stunde nie bewut geworden, wird ebenso wenig verstehen. "Nogen delagde en rude -- jeg ved ikke hvem" / "Somebody broke a window -- I don't know who"). Hvor dygtig han end er, s fr han ikke stillingen. Hvor gode hans kvalifikationer end er, (s) fr han ikke stillingen. In the "hvor + AN"-constructions it is difficult whether to consider "hvor" a modifier of the adjective or to consider "hvor" as the head of the relative clause. We have decided to follow the "hvor"-as-modifier-analysis:

An alternative would be to follow the intuition that "hvor" must be the head of the relative clause. Such an analysis would look like this:

The problem with this analysis is, that it conflicts with the understanding of "hvor" as a modifier to "svrt" and of "svrt" being the head of the predicative frase. We would not like to say that "hvor" is the [pred] of "vre", and the consequence is that the analysis points out to heads of the frase "hvor svrt". In German (PJ, "Tysk Grammatik"): Es gab Wirtshuser, die zu betreten den Studenten verboten war. (Bll) Der junge Beisem, dem die Regeln der Bruchrechnung beizubringen ich mich verpflichtet hat. (Bll) Ich wei nur ein Gesicht, dessen veredelte Wirklichkeit durch mein Einbildungskraft korrigieren zu wollen sndhaft wre. (Mann) Die den Sinn des Martyrium leugneten, [die] gerieten am leichtesten in Verfolgung und Folter. Du zahlst jetzt [das], was du getrunken hast. Wem Gott Kinder gibt, dem gibt er auch sorgen. Han spurgte hvem der kom" and "Han spurgte hvem. A class of more problematic examples are listed below: Vi opdagede til hvem han sendte brevene. Han redegjorde for p hvilket tidspunkt og under hvilke omstndigheder han kbte maleriet. Han redegjorde for p hvilket tidspunkt han kbte maleriet ?(og) under hvilke omstndigheder. Sprgsmlet om hvor stor (en) andel af forskningsmidlerne forskerne frit kan rde over. Han gav noget til nogen. Jeg ved ikke hvem/hvad/hvor/hvorfor/ hvordan/hvornr/med hvem/til hvem/af hvilken rsag/p hvilket grundlag.

Jeg kan ikke forklare hvor/hvorfor/hvordan/hvornr/med hvem/til hvem. Han gav kuverten til nogen. Jeg s ikke til hvem. Vi havde uforholdsmssigt mange rverier , uden at jeg kan forklare hvorfor. Det er derfor at han siger det. Hvorfor det er han siger det. hvorfor det er at han siger den slags. These are sometimes called head-less relative clause (with reference to the fact that they do not modify an external head), but we prefer the term 'independent relative clause', since, on our analysis, they do have a head, namely the relative pronoun. What is special about this construction is that the relative clause functions as a complement in its own right, and not as a modifier of a complement. This is reflected in our analysis by having the relative pronoun function as a complement to the external governor: Independent relatives have the same form as embedded interrogatives, but the two can be distinguished by their semantics, as reflected by the subcategorization properties of their governor: Han spiste {hvad de havde} [independent relative] Han spurgte {hvad de havde/*den} [embedded interrogative] Embedded interrogatives. Interrogatives can occur as dependents to an external governor. Embedded wh-interrogatives are analyzed like relative clauses. VPs and APs as relativized phrases. Relative clause typically modify nouns, but (non-restrictive) relative clauses can also modify a verb or adjective: Jeg har lavet kage, som du bad mig om. Huset er grnt, hvilket er en dejlig farve.

Reduced relatives. vin kbt specielt til lejligheden (cf. vin som er kbt specielt til lejligheden) huer strikket i ren uld (cf. huer som er strikket i ren uld) en maskine konstrueret efter srlige principper (cf. en maskine som er konstrueret efter srlige principper) vin kbt til lejligheden German data (judgements by Sabine Kirchmeier-Andersen): Wer/*wen die Schler verachtet, is ein schlechter Lehrer. Wen/*wer die Schler verachten, ist ein schlechter Lehrer. Ich verachte wer/*wen die Schler hat. Ich verachte wen/*wer die Schler hassen. Ich frage wer/*wen die Schler verachtet. Ich frage wen/*wer die Schler verachten.

An alternative is to extend the analysis of independent relatives to embedded interrogatives. Under this analysis the "wh"-word is the head of the construction which takes the finite verb as a rel dependent. Arguments in favor of the relative clause analysis: The fact that the two constructions display the same word order follows naturally. The tagger need not distinguish the two constructions in the treebank. It provides a natural explanation for the occurrence of "der" in embedded subject interrogatives, an otherwise puzzling fact (Diderichsen 1957: 183): "Han spurgte hvem der kom." Arguments against the relative clause analysis: It yields a non-uniform analysis of embedded and matrix "wh"-interrogatives. In matrix interrogatives the finite verb is the head, and the "wh"-word is its dependent In embedded interrogatives, the "wh"-word is the head and the finite verb is its dependent. The interpretive difference between independent relatives and embedded interrogatives is not reflected in the syntactic analysis. Arguments in favor of the interrogative analysis: It provides a uniform analysis of embedded and matrix interrogatives. It follows the tradition. It accomodates the semantics of this construction Arguments against the interrogative analysis: It provides no obvious account of the presence of "der" in embedded subject interrogatives. Several analyses are possible, but none of them are without problems. "der" is an expletive subject, "hvem" is a dobj (compare: der kom en mand). advantages: uses existing analysis (of expletive constructions). problems: the use of "der" in embedded subject interrogatives is less restricted than ordinary expletive constructions (han spurgte hvem der havde taget af kagen -- *der havde taget en mand af kagen) "der" is a modifier to the finite embedded verb (I believe this is the analysis proposed in diderichsen (1957:183) where it is stated that in this use "der" is 'clearly an adverbial and not a subject pronoun'. advantages: no comparison with expletive constructions. problems: ad hoc, such modifiers are not allowed elsewhere "der" is a modifier to "hvem". advantages: no comparison with expletive constructions. problems: ad hoc, such modifiers are not allowed elsewhere. "der" is a modifier to the embedding verb. advantages: no comparison with expletive constructions. problems: ad hoc, such modifiers are not allowed elsewhere What the other treebanks do. Corpus Gespokenes Nederlands (CGN Syntactische Annotatie pp. 47-48))) distinguishes independent relatives (`hoofdloze relatiefzinnen') from embedded interrogatives (`afhankelijke

vraagzinnen'), and give some tests for how to distinguish them in the corpus (e.g. whether one can insert a personal pronoun in front of the relative pronoun - if yes, it is a relative clause, if no, it is an embedded interrogative.) Penn Treebank (1995 Bracketing Guidelines pp. 169-170)) also distinguishes independent relatives ('head-less relatives' in their terminology) from embedded interrogatives ('indirect questions' in their terminology). They give semantic criteria for distinguishing the two. In the DG analysis, we always assume that the relativizing phrase lands on the relativized phrase with a "relz" landing edge. This is not shown in the treebank analyses, but corresponds to the following graphs:

Hudson analyzes relative pronouns as heads of the relative clause, which allows him to avoid the use of anaphoric references (he does not deal with relative prepositional phrases, eg, "til hvem") and give a more uniform account of relative clauses with and without relative pronouns. This analyzes may have great advantages during parsing as well, because the relative pronoun can be attached right away. Another (perhaps even more attractive) analysis is to: (1) in subject relatives, the relative pronoun lands on the relative verb and has the relativized word as its antecedent, giving V2 order in the verb; (2) in relatives with any other relative phrase, the relative phrase lands on the relativized word and has the relativized word as its antecedent, giving V2 order in the relative verb; (3) in all other relatives (ie, those without a relative phrase), the relative verb creates a filler node that has the relativized word as its antecedent, also resulting in V2 order. Post-nominal modifiers headed by a perfect participle could be analyzed as reduced subject relative clauses:

Alternatively, this construction can be analyzed as a regular mod adjunct, headed by the perfect participle. The internal structure of the modifier is the same under this analysis. The only difference is that `mod' replaces `rel'.

As discussed in Gerhard Helbig, "Studien zur deutschen Syntax", the traditional distinction between relatives, embedded wh-questions, and clauses headed by a complementizer is somewhat ad-hoc, since it defines the three classes in terms of mutually inconsistent morphological, syntactic, semantic and pragmatic criteria, and many examples are extremely hard to classify in a consistent way.

Prepositional phrase as relativized phrase


Usually the relativized phrase is a nominal phrase of some kind or a sentence, that is, the relativized phrase is a verbal phrase. In a few cases the relativized phrase can be a prepositional phrase - a phrase headed by a preposition. This preposition then is the governor of the relative clause. In the corpus we have only come across a few examples, 9 examples of the following 6 expressions: i den udstrkning i det omfang i hvilket omfang i hvilken grad i hvilken kategori for hvem Examples of analysis follow below:

"Ligesom" and "som"


"Ligesom" and "som" (meaning "ligesom") often marks a relative construction, but not always. To determine whether the construction is in fact a relative clause, we have used the following criteria: If the sentence is not complete on its own, the construction is analysed as a relative clause with a filler dependency. If the sentence is complete on its own, the construction is analysed as a non-relative construction, that attaches itself to its governor as a verbal object (vobj). Here are some examples of the different analyses:

Here "som" could be replaced by "ligesom". If we do not analyse this as a relative clause, the analysis would look like this:

Here "bede om" obviously lacks an object, and this indicates, that the relative-analysis is correct. Here is another example:

This is an analysis parallel to the non-relative analysis above, but with this example there is no part missing; the sentence is complete. A relative analysis would therefore be wrong here.

Subordinate conjunctions
Some subordinate conjunctions (CS) (dengang, fr, da, siden, hvis, nr, fordi, end) sometimes have some marks of a relative clause: The sentence they take seems to be missing that something, that the CS represents. This indicates that these constructions should be analysed as relatives. On the other hand, another guideline we have followed in determining whether a construction should be analysed as a relative or not, is whether the CS most likely would be able to take a "som" - making it a relative - or an "at" making it a vobj. And all of the subordinate conjunctions mentioned above most likely would take an "at", which indicates that an analysis that sees them as taking a relative would be wrong. These to tests for whether it is a relative or not, point in two directions. We have decided to go with the vobj-analysis, since the fact that they cannot take the relative pronoun "som" indicates that they are not truly relatives. This on the other hand leaves us with an unsolved problem, namely that these sentences often miss an otherwise obligatory part (direct object, nominal object or even subject).

The adverbial "s lnge"


Regarding the adverbial "s lnge", we believe that the correct analysis is that "s lnge" takes a sentence as a relative. This is because this phrase actually tends to be able to take the relative pronoun "som" (e.g. "Jeg sover lige s lnge, som jeg har lyst til"). Here is an axample of the analysis:

Interrogatives
Polar questions, including alternative questions, exhibit subject-verb inversion. The dependency structure, however, is identical to that of the corresponding declarative:

In wh-interrogatives the "wh"-word usually occupies the sentence-initial position, irrespective of its dependency role: mak Hvem fandt cyklen? Hvad fandt de? Hvordan/hvornr fandt de cyklen? Hvor stor en andel giver han os? Han fandt hvad? Again, the linear order has no effect on the dependency structure, which is identical to that of the corresponding declarative clauses. This holds true also in cases of 'long' extraction, where the governor for the wh-word belongs to an embedded clause (last example).

Note that "hvad" is also used to question verbal complements, but is still tagged as "dobj" in this case: Hvad ville/kunne/m de? (De ville/kunne/m synge) Embedded polar questions are headed by "om", which we analyze as a preposition that takes a verbal complement

Tag questions are adjuncts (mod) to the finite verb of the tagged clause. Note that the verb in the tag bears a filler dependency to an element of the matrix clause.

As an alternative one can analyze "om" as a conjunction, which is linked to the external governor (the finite verb of the matrix clause) by a special complement edge (or possibly as a dobj). Two facts speak in favor of the preposition analysis: "sprge" can take other pobj's (eg, "Han spurgte efter/til kokken.") "om" can also take nominal complements after "sprge" (eg, "Han spurgte om tilladelse").

If "om" in embedded polar questions is analyzed as a preposition the two constructions would receive a unified analysis. A comment on the analysis of "hjem" as a modifier to "gik": alternatively "hjem" could be analyzed as a complement, either part or a locative-directional complement, if we decide to include such a dependency. This has no bearing on the analysis of interrogatives.

VP ellipsis
VP ellipsis is typically realized as VP anaphora, with the proform "det". The proform is tagged as a dobj to the verb:

An alternative is to analyze "det" as a (special type of) vobj. The dobj analysis seems better in so far as it is compatible with "det" being simply a personal pronoun. On the other hand, the vobj analysis reflects the fact that the antecedent for the proform is verbal. VPE constructions without proforms are analyzed in terms of [vobj] fillers also needs to be compatible with the existence of VPE without proforms: Marie kan synge rent, men Susan kan ikke. Vi besgte de byer (som) I gjorde. The last example involves antecedent contained deletion, and that here the proform is impossible: *Vi besgte de byer (som) I gjorde det.

Clefts
Clefts are analysed as copula clauses, where the relative clause is an extraposed dependent of the subject (below left). This analysis is supported by the fact that exactly this dependency structure found without extraposition (below right). These correspond to wh-clefts in English, also known as pseudo-clefts.

Det var pnt af ham at komme. Det var pnt af ham at han kom. Det at komme var pnt af ham.

Det at han kom, var pnt af ham. At komme var pnt af ham. At han kom, var pnt af ham. Det siges/forlyder/foresls at han kommer. Det sagdes/forld at han kom. Det var/blev klart at han ville komme. Det forskrkkede/overraskede os at hun skreg.

Progressive
Progressive aspect is expressed syntactically, either by coordination (analyzed as standard coordination) or by "vre" plus prepositional construction (analyzed as a predicative):

Alternatively, the preposition can be analyzed as a pobj to "vre". The fact that we analyze other prepositional complements to copula verbs as pred, speaks in favor of the pred analysis. On the other hand, this is a construction where "vre" would be said to be be used as an auxiliary verb, under theories that make such a distinction, whereas in predicative constructions "vre" is used a main verb. How do we capture that this is a specialized construction, including: the prepositions cannot be exchanged for others? the nobj for "i" must be one of the nouns "gang" or "frd"? "ved" and "med" must take an "at" + infinitive complement? Also, should the inifinitive share the subject of "vre"? (Probably not)

Other phenomena
Hypothetical statements and conditionals: Vandt jeg bare i Lotto i morgen, kunne jeg kbe et nyt hus. Vinder jeg i Lotto i morgen, kan jeg kbe et nyt hus. Havde han ikke vret s uforsigtig, var det aldrig get s galt. Other: S lille hun er! Parenthetical sentences: De kom halvanden time for sent, (hvad der|hvilket) ikke overraskede os. Den var god, hva'? Har du hrt hvem der kommer i aften? Som X skriver: "Jeg er optimist". Bogen er ikke, som man kunne tro, negativ og hadefuld. et sagt udsagn

A main clause may be modified by another following main clause. Stolen er da smuk, er den ikke? Stolen er smuk, ja. Stolen er smuk, ja den er. Stolen er da smuk. Er den ikke? Hypothetical statements and conditionals: Vandt jeg bare i Lotto i morgen, kunne jeg kbe et nyt hus. Vinder jeg i Lotto i morgen, kan jeg kbe et nyt hus. Havde han ikke vret s uforsigtig, var det aldrig get s galt. De gav hunden gift, hvad han ikke vidste. Han s glad ud, rolig ud, ... Quotation: Jeg er et dejligt menneske, synes Marie. Hvorfor gr de s den slags imod mig? (At) vi bliver inviteret til Firenze, fortalte Marie os i gr. Ergatives: Han knkker grenen --> Grenen knkkede

References
Mller, Stefan (2002) Complex Predicates: Verbal Complexes, Resultative Constructions, and Particle Verbs in German, CSLI Publications, Studies in Constraint-Based Lexicalism.
http://www.id.cbs.dk/~mtk/dtag/ddt/verbs.html last updated by Matthias T. Kromann at 2004-09-23 11:57

You might also like