Agreement processes in English and Spanish: a

completion study*

Sara María Riveiro Outeiral & Juan Carlos Acuña-Fariña

University of Santiago de Compostela

Proposal for a short title: full title is under 65 characters.

Corresponding author:

Juan Carlos Acuña-Fariña

University of Santiago de Compostela

English Department

Facultad de Filología. Avd. Castelao s/n. Campus Universitario Norte

15782. Santiago de Compostela

Spain

carlos.acuna.farina@usc.es

tel. +34 66299999111900

1
 Abstract

The nature of agreement has been the topic of extensive debate in the recent literature of both linguistics

and psycholinguistics. In contrast to either fully syntactic or fully semantic accounts, so-called

‘constraint-satisfaction models’ (Haskell et al. 2010, among others) posit that all grammatical encoding is

subject to a number of influences (syntax, semantics, pragmatics, frequency, etc.) which compete to

dominate every computation, including agreement processes. After briefly considering psycholinguistic

work on attraction (Wagers et al. 2009 and references therein), we try to shed light on this debate by

observing how agreement operates in certain structures which were previously tested by Berg (1998) in a

comparison of German and English. Here, we establish the same type of comparison between Spanish

and English, and conclude that: 1. agreement is resolved after a constant tug-of-war between the syntactic

and the semantic, a process in which semantics is likely to interfere in formal operations when these are

performed in the context of a weak morphology; 2. agreement resolution is effectively subject to various

linguistic influences, including the morphological characteristics of each language, but also the domain in

which agreement is realised; and 3. agreement is responsible for shaping overall linguistic systems in the

sense that, as noted by Berg, it may motivate left–orientation (as in English) or not (as in Spanish) as a

general default strategy for locating subjects.

2
Agreement processes in English and Spanish: a completion study

1 Introduction

Over the past two decades, agreement has become a topic of special interest in both

linguistics and psycholinguistics, since there is evidence that the understanding of

agreement processes may reveal relevant aspects of the nature and the functioning of

language in general. It seems evident at first that agreement is responsible for building

up the structure of the clause in many languages, since often it is precisely the presence

of agreement ties which signals that a clause has been assembled. Furthermore,

agreement is a pervasive phenomenon in the sense that some form of it is present in

over 70% of the world's languages (Mallinson & Blake 1981). However, in spite of its

importance, both linguists and psycholinguists disagree as to whether agreement is

essentially a semantic (Pollard & Sag 1988; Barlow 1999; Vigliocco et al. 1996a;

Thornton & MacDonald 2003; Vigliocco & Hartsuiker 2002; Haskell & MacDonald

2003) or a syntactic phenomenon (Chomsky 1995, 1999, 2001; Bock & Eberhard 1993;

Eberhard 1997; Levelt, Roeloffs & Meyer 1999; Carminati 2005; Franck et al. 2006).

Corbett (2006:3) and Eberhard et al. (2005) in fact view it as arguably the major

interfacial problem between morphology and syntax, a fact that makes it difficult to

understand if viewed only from the core of either component. In the words of Anderson

(1992:103): “agreement is a quite intuitive notion which is nonetheless surprisingly

difficult to delimit with precision”.

Here we focus on subject-verb agreement ties in English and in Spanish.

Essentially we take up a suggestion by Berg (1998), pursued also by Acuña-Fariña

(2009), that agreement is resolved differently depending on whether languages possess

3
either a complex morphological system (with marks for number, gender and case, like

German) or a limited one (like English). It is claimed that morphological attrition results

in semantic interference in agreement processes, as well as a blind left-orientation in

locating agreeing subjects. So we provide a completion test similar to that of Berg

(though more circumscribed in scope) in which we seek to provide evidence for the

following main hypotheses.

Firstly, English is likely to show semantic interference when the subject denotes

a multiple referent of the form ‘a gang of thugs (ARE)’, as, in such cases, the language

system tends to rely on meaning (establishing agreement with the plural referent), rather

than on morphology (establishing formal agreement with the morphologically singular

head noun). Secondly, when there are two potential subjects within the same identifying

clause (one in pre-verbal and another one in post-verbal position; e.g. the key IS/ARE

the problems that decision would cause), English stops using meaning as the first

criterion to establish agreement and instead resorts to a left-orientation strategy of

placing subjects blindly to the left of the verb, thus simply respecting SVO. Thirdly,

contrary to English, Spanish verbs are more likely to show agreement in the singular

when a multiple referent with overt singular morphological marks is presented as the

subject, thus simply using morphology to project the overall structure of the NP,

disregarding conceptual influences. Finally, however, Spanish is hypothesised to show

more sensitivity to notional concord in other cases, such as in identifying clauses with

two possible subjects, due to the fact that it is less dependent than English on strict,

predictable tree geometries. It is well-known that Romance languages in general use

their relatively free word order to place new information conveniently (Lambrecht

1994). Here we intend to cast some light on how agreement in particular interacts with

that general tendency, especially in cases of feature mismatch.

4
 In essence, we seek to demonstrate that agreement is not a fixed, encapsulated

syntactic reflex that is set in the same way for all structures and all languages, but rather

a much more complex operation that is sensitive to specific constructional properties,

and to the general language type on the rich/poor morphology spectrum.

2 A working definition of agreement and some descriptive facts

Even though no definition of agreement has ever been generally accepted (Corbett

2006:4), a consensus view of it might well be that “agreement refers to some systematic

covariance between a semantic or formal property of one element and a formal property

of another” (Steele 1978:610). That is, agreement ties usually involve morphological

features (such as gender, number or person) of a controller (such as a subject noun

phrase) being replicated onto a target (such as a determiner, adjective or verb) in a

certain domain (such as the noun phrase in the case of articles and nouns, the clause, in

subject-verb ties, or the sentence, in cases involving pronouns).

The criteria followed to establish formal co-variance may be either semantic or

formal. Thus, in a sentence like Mary plays the piano, there is a formal link between

Mary and the –s suffix on the verb in the sense that the features ‘third person’ and

‘singular’ are copied from Mary onto play-s. This action of copying, where the features

of a controller migrate to a target, has been the position taken by many linguists,

especially those supporting the later versions of generative grammar (Chomsky 1995:

II; den Dikken 2001; but see also Gazdar et al. 1985; Shieber 1986; van Riemsdijk &

Williams 1986). However not everybody agrees with this idea. In an example such as

(1):

5
 (1) The committee have agreed that statistics are needed for effective future

advertising.

the singular noun committee does not match formally with the plural verb have, but it

does so in a semantic sense, since committee is conceptually plural. Such examples may

lead one to think that agreement is driven by the overall conceptual representation of the

message. In fact this position has been advocated by those who note that in ordinary

agreement the formal and the conceptual are usually synchronized, but the litmus test

for what really directs agreement operations is precisely when the two are in opposition.

Yet, the semanticist view of agreement is at odds with cases like (2) below:

(2) More than one person has been invited to his house.

since in these, the referent is conceptually plural, as was the case of committee above,

but syntax is responsible for the actual agreement form. These kinds of incongruencies

are not at all unfamiliar in the world’s languages. The mere existence of arbitrary

gender and morphological redundancy in so many of them is hard to reconcile with a

strictly semantic approach. As Langacker (1991a:307) has observed, “agreement

markings are perhaps the archetypal example of sentence ‘trappings’ employed for

purely grammatical purposes, and are supposedly inconsistent with any claim that

grammar might have a semantic basis”.

A third and for present purposes final argument which refutes a purely notional

nature of agreement comes from the fact that semantics can only influence agreement

under certain structural conditions. Example (3) is perfectly possible in English, and it

6
is even the preferred option in British English. On the contrary, (4) is an ungrammatical

noun phrase in all dialects of English:

(3) The committee have agreed that statistics are needed for effective future

advertising.

(4) *These committee...

What these two examples show is that semantic overrides are allowed at clause level,

but less so inside the noun phrase (Corbett 2006:3; also Dikker 2004:38ff on ‘unit-

generation’, locality and discourse acts in these agreement processes). Specifically, they

show that agreement is also sensitive to domains and that these are also defined

syntactically (as in Corbett’s Agreement Hierarchy, which predicts that notional

influences are more likely the more distance exists between controllers and targets; see

Corbett 1979, 2006:207).

3 Conflicting feature specifications in production studies of agreement

3.1 Attraction or proximity concord

Bock & Miller (1991) were probably the first to test agreement operations in an

experimental manner. The structures they used soon became the basic materials used in

a great deal of later research on agreement (Bock & Cutting 1992; Bock & Eberhard

1993; Bock, Nicol & Cutting 1999; Bock, Eberhard, Cutting, Meyer & Schrieffers

2001; Eberhard et al. 2005). These materials were manipulated in various ways in order

to test both formal variables (such as the role of structural, syntactic depth) and

7
semantic variables (such as the role of collectivity or distributivity) in agreement

processes. In all these studies, the aim was from the very beginning to see whether

semantics interfered with the agreement process or, conversely, whether this was

blindly driven by morphosyntactic information.

In this kind of study, participants are presented with preambles such as (5a) and

(5b), which consist of complex noun phrases, and have to simply first repeat and then

complete them by including a verb or a verb plus a predicate:

(5a) The key to the cabinets… (IS BIG)

(5b) The keys to the cabinet… (ARE BIG)

When completing such preambles, speakers sometimes erroneously make the verb agree

in number with the local noun or ‘interloper’ (cabinet-s), thereby disrupting the

agreement process (as in *the key to the cabinets are in the kitchen). This phenomenon

has been labeled attraction or proximity concord and, according to Eberhard et al.

(2005), it causes as many as 13% of agreement mistakes on the verb in English.

(Jespersen 1924; Quirk et al. 1985; Francis 1986; den Dikken 2001; Huddleston &

Pullum 2002:500ff; see Wagers et al. 2009 for a review of psycholinguistic studies). In

this language, attraction errors occur much more frequently when the local noun is

plural and the head noun singular, so some authors have pointed out that this is a

markedness effect: marked plurals are supposed to need and receive a higher degree of

activation (Croft 1991:54; Eberhard 1997). The basic facts of attraction have now been

confirmed in a number of languages, including English, Spanish, French, Italian, Dutch

and Russian, among others (see Wagers et al. 2009 for a review).

8
3.2 Testing semantic variables in attraction

The first batch of experiments on attraction (Bock & Miller 1991; Bock & Cutting

1992; Bock & Eberhard 1993) established a number of important findings. Importantly,

attraction was seen to be insensitive to semantic manipulations: first, collectives like

army did not attract but morphological plurals like ships did. Secondly, single-token

preambles like the bridge to the islands were compared to multiple-token preambles

like the label on the bottles (one bridge versus many labels), but no differences arose.

This meant that underlying distributivity did not matter either. This led to an

inflectional/copying account of agreement whereby a controller which possesses

inherent features passes them on to a target to establish an agreement relation. This is

done in a formally encapsulated manner that is strongly reminiscent of cyclic phases in

linguistics (Gazdar et al. 1985; Chomsky 1995). Collective nouns such as committee are

especially troublesome when dealing with agreement, since when they are controllers

within a sentence, the processor may base agreement with their targets either on formal

or on notional grounds. Bock, Nicol & Cutting (1999) later examined two different

types of targets: pronouns and verbs. What they found was that subject-verb ties were

not influenced by meaning (collectivity), whereas subject-pronoun agreement was more

prone to show notional matches between head and target (see Corbett 2006:3 on

distance and agreement).

More recent studies on notional agreement effects have modulated the first

findings. The influence of distributivity is now a generally acknowledged fact (Haskell,

Thornton & MacDonald 2010), even in English, ever since Eberhard (1997) improved

the ‘imageability’ of the materials. Similarly, Hupet et al. (1998) and Thornton &

MacDonald (2003) found semantic effects by manipulating the plausibility of the verb

relative to the two nouns in the complex NP. For instance, in an experiment where

9
either of the two nouns could be a plausible passive subject (the album by the classical

composers . . . BE praised) or only the head noun could be so (the album by the

classical composers . . . BE played), there were more agreement errors when both nouns

were plausible subjects than when only the head noun was plausible (Thornton &

MacDonald 2003). Notice that effects were obtained by manipulating a position (the

verb) where, in formal, percolation terms, all agreement decisions are supposed to have

been left behind and the verb’s agreeing form is supposed to be merely inherited via

control (Wagers et al. 2009).

3.3 English and Spanish in attraction

Where do English and Spanish stand in studies of attraction? Both languages have

shown strong attraction effects, even in recent comprehension studies. However, there

are two conspicuous differences that emerge from the latest research (Lorimor et al.

2008; Acuña-Fariña et al, submitted; Foote & Bock, forthcoming). The first is that

strong, consistent asymmetry (sg + plural attracts but pl + singular does not) has not

been found in either Spanish or other richly-inflected languages like Italian, French and

Dutch (Vigliocco, Butterworth & Semenza 1995; Franck et al. 2002; Franck et al. 2006;

Acuña-Fariña et al, submitted).i The second is that contrary to the original ideas in the

90s, it is now seen that semantic interference is more apparent in English than in

Spanish. Interestingly, in a recent contribution, Foote & Bock (forthcoming) contrasted

two varieties of South-American Spanish, Mexican and Dominican, with English acting

as the baseline for comparison. Whereas Mexican Spanish has preserved the rich

morphology of European Spanish, Dominican Spanish is losing it (Lunn 2002). In their

first experiment Foote & Bock found larger distributivity effects in English and in the

Spanish variety with poorer morphology (Dominican). In their second experiment,

10
when participants produced sentences with null-subjects (and therefore did not repeat

the morphologized preambles), there were large notional effects in the two varieties of

Spanish. This suggests that it is the morphological realization on subject NPs in

particular, and not general language type, that drives agreement computations.

3.4 Testing grammatical choices in production: Berg (1998)

As we have seen, both semantics and syntax were demonstrated to be present in

agreement operations. Berg (1998) set out to examine if and how the structure of a

particular language has a role to play with regard to particular forms of agreement. In

his paper, he suggested that certain agreement choices are influenced by the

morphosyntactic system of each language. In particular, agreement could be based

either on notional or syntactic grounds depending on the strength of the morphological

system of each particular language. Berg maintained that English is more prone to show

notional agreement when the morphological and notional features of the subject noun

phrase mismatch (this being a consequence of the limited morphology of the English

language). As is well-known, with collective nouns such as ‘committee’, English often

establishes a form of agreement based on notional grounds, thus resulting in sentences

such as ‘The committee have…’ However, since English is a deeply left-oriented

language (subjects in English are almost invariably located to the left of the verb),

recourse to notionality is not sought when this puts the left location of the subject at

risk. Thus, in a sentence such as ‘The cause of the accident BE… bad brakes’, English

speakers will show a tendency to place the subject in an initial (or left) position, sticking

to an SVO order. This happens even in spite of the fact that from a semantic point of

view, both the preverbal and the postverbal noun phrases may be plausible subjects. On

the other hand, according to Berg, the agreement systems of other languages work in

11
different directions. Subject-verb agreement involving collective nouns in German is

usually resolved morphosyntactically. Thus, agreement with collective nouns in these

two languages usually results in a singular match with the verb. However, when there

are two possible subjects within the sentence, the parser in German can look to both

sides of the verb, and agreement with the postverbal element is possible (and sometimes

even more frequent). This is a consequence of a higher syntactic flexibility, and an

ultimate consequence of the complexity of the German morphological system.

Berg analysed number conflicts in eight different constructions which, in his

view, present concrete difficulties to the agreement processor in both German and

English. Although both languages are defined by Berg as being of the same “West

Germanic type”, their agreement systems seem to be completely different. This

dissimilarity is due to the fact that whilst German is a richly inflected language, English

is not. Table 1 summarizes the eight categories used by Berg.

[INSERT TABLE 1 ABOUT HERE]

Items from these eight categories figured in two similar completion tests (one in

English and the other in German). The tests were distributed to 46 native speakers of

German and 57 native speakers of American English. There was a series of slots which

the participants had to fill in (on a written form) with a correctly conjugated verb. They

were encouraged to react spontaneously in order to obtain natural and almost

unconscious responses which tried to imitate the normal way of speaking (although the

task was not an oral production task). In order to distract the subjects’ attention from the

real purpose of the task, the critical items were interspersed with filler items, empty

verb slots in which agreement was not an issue (Berg 1998:48).

12
 The results that Berg obtained from this completion experiment pointed primarily

to a tendency for the English speakers to react ‘semantically’ when the controller is a

subject NP denoting notional plurality (e.g: the committee were; a gang of thugs were..).

He interpreted that as a consequence of this language’s poor morphology, which forces

the processor to rely on conceptual structure directly in order to project reference to the

top of the NP node. Conversely, for categories 7 and 8, in which there were two

possible subjects (on both sides of the copula) the subjects preferred as the subject the

NP that is placed in preverbal position. Berg (1998:60) explains the situation in English

along the following terms:

As is well known, the inflectional morphology of English is highly

impoverished. Case and gender inflectional morphology have completely
disappeared and number morphology is restricted to the third person in the
present tenses. Thus, there is hardly any opportunity for (syntactically based)
agreement processes to operate in English. It is a general cognitive principle that
frequency impacts upon the strength of a phenomenon. The less frequently it
occurs, the weaker it is. As a consequence, the limited opportunity that the
language provides for expressing syntactically based agreement relationships
involves a weakening of the syntactic force. In view of the complementary
relationship between the syntactic and the semantic influence, this weakening of
the syntactic force implies a strengthening of the semantic force. Hence, the
agreement pattern in English is for the larger part semantically based.

The German results reflected almost the complete opposite. As is well-known,

German is a language that has a rich morphological component with marks for number,

gender and case. This fact allows word order to be much more flexible than in English.

As a consequence, German speakers provided more diversified results when they dealt

with items in categories 7 and 8, but seemed to cling to morphology blindly when

coping with items of categories 1 to 6. Berg (1998:61) provides the following

explanation for how agreement processes work in German:

13
 […] The situation is quite different in German. Unlike English, it has preserved
much of its inflectional morphology, which codes case, gender and number
distinctions across the major and (some) minor word classes. Case, gender and
number agreement processes are rife in German. In particular, the number
agreement between the subject and the verb applies throughout the tense system.
This high frequency of syntactically based agreement processes accords a great
deal of strength to the syntactic principle, which is therefore in a position to
outweigh the semantic influence. As a result, the agreement pattern in German is
for the largest part syntactically based.

In summary, English syntactic and morphological attrition causes speakers to: 1.

always stick to the stipulated word order when there are two potential subjects (a left-

orientation strategy), and 2. resort to semantics instead of morphology when, once the

subject NP is located, this presents internal layering that launches referential

competition. German, however, represents the opposite situation: variability in locating

subjects either to the left or the right of the verb, and blind obedience to morphology to

project the subject’s head features upwards.

In the same vein, Acuña-Fariña (2009) carried out a similar study for the

comparison of English and Spanish. Spanish and German are similar languages in the

sense that they both have a complex morphological system that allows word order

flexibility. If we establish a comparison between Spanish and English, for instance, we

see that in a Spanish sentence such as ‘l-o-s pequeñ-o-s candelabr-o-s blanc-o-s y l-a-s

cómod-a-s sill-a-s roj-a-s estaba-n allí’ up to 17 different morphological marks

conveying agreement (either for gender or number) are used, whereas in the equivalent

English translation (‘the little white candlestick-s and the comfortable red chair-s were

there’) there are only two. The evident richness of the Spanish (or German)

morphological component allows a higher degree of word order flexibility. Thus in

most cases, adjectives can be located either before or after the noun (both ‘candelabros

blancos’ or ‘blancos candelabros’ are possible), and subjects are easily placed after the

verb in certain types of constructions (‘la causa del accidente fueron los frenos’ or ‘los

14
frenos fueron la causa del accidente’), literally ‘the cause of the accident were the

brakes’ and ‘the brakes were the cause of the accident’.

4 Cross-linguistic differences in agreement: an English-Spanish production task

4.1 Theoretical background of the production task

As we have seen, work on attraction has provided evidence that agreement processes

seem to be intriguingly sensitive to competing formal and semantic constraints. The fact

that English in particular shows a strong asymmetric effect (in the local modifying

position, morphological plurals like ‘dogs’ attract, but collectives like ‘army’ do not)

can be easily interpreted as the consequence of a certain degree of formal routinization,

and may be captured via some sort of copying/percolation account such as the one

adopted in both linguistic and psycholinguistic work (Chomsky 1995:II; den Dikken

2001; Bock 2004; Bock et al. 2004; Bock et al. 2006; Eberhard et al. 2005). The fact

the overall semantics of the phrase (as seen in studies on distributivity) also has

attracting power (in all languages examined, including English) indicates that semantics

is equally strong as syntax in guiding agreement operations under certain conditions.

The fact that asymmetry is less strong in the Romance languages (as well as in Dutch)

indicates that the morphological richness of each language is another interacting

principle at work, and particularly the plural feature may not be marked in those

languages with a strong morphosyntactic component.ii This ties in with Berg’s work on

agreement choices in German and English, which we take as the basis for the present

work. The objective of this section is to investigate number conflicts in subject-verb

agreement in a completion test for both English and Spanish. We believe that

comparing the different morphological strengths of Spanish versus English vis-à-vis is a

15
fruitful topic of research, highlighting such issues as feature markedness, inherent

notionality, the syntax/semantics division of labour, and the order of operations inside

putative encoding cycles. Some of the constructions investigated here have already been

tested in a pilot study presented in Acuña-Fariña (2009). However, since the theoretical

repercussions opened up by Berg are extremely relevant, more solid data are needed to

either confirm his findings or to view them more tentatively.

4.2 Methodology

4.2.1 Materials

We used some categories from Berg’s original test and discarded others. There were

various reasons for the omission of certain categories. Categories 2, 6 and 8a were

discarded either because these expressions do not have a Spanish counterpart or because

their most accurate translations entail a change of structure. More concretely, words like

pancakes (category 2), when used in its plural form, refers to a type of dish which has a

close Spanish translation in tortitas, but will always show plural agreement. Category 6

(many an X) can only be translated into Spanish by an impersonal structure such as: Los

hay que..., and so had to be dismissed because the English and the Spanish structures

were not even similar. Finally, cleft sentences (category 8a) were excluded because they

have no feasible translation into Spanish, or at least not one containing a neuter pronoun

like it. In brief, these three structures are not subject to any number conflict in their

Spanish version so they had to be eliminated. Categories 1, 4 and 5 were discarded

because they were considered to produce the same effect as category 3. All these

constructions pointed to a plural referent expressed by a word (or a group of words) in

the singular, thus having a ‘collectivity’ effect similar to that shown by nouns like

‘committee’. Furthermore, we preferred to concentrate on fewer categories with a

16
higher number of examples per category, and on a finer study of these categories. The

three categories chosen are presented in table 2:

[INSERT TABLE 2 ABOUT HERE]

Note that in addition to the previously mentioned changes with respect to Berg’s

study, categories 2 and 3 were split into two subcategories. In identifying structures of

the ‘X is Y’ kind the sentences were reversed. This was in order to avoid a possible

tendency for subjects to be the most logical (or concrete), or the most grammatical one

independently of their location. That is, structures such as the cause of the accident BE

bad brakes and the structurally reversed Two broken pistons BE the cause of the sinking

were included in the questionnaire. Additionally, with regard to category 1, the items

were manipulated by adding modifiers which emphasized the topicality of one of the

two elements of the complex NP. Thus, there were sentences such as A very small but

quite significant number of children, in which the first singular noun was modified,

and others like A number of famous actors, in which the emphasis was put on the last

plural element of the NP.

4.2.2 Procedure

The task consists in asking subjects to complete a test formed of 51 short paragraphs. In

each small text there were a series of slots which had to be filled in on a written form

with a correctly conjugated verb.iii To facilitate the task, the infinitive form of the verb

they had to conjugate was provided. The questionnaire was distributed to a total of 85

subjects who were native speakers of English (in order to complete the test in English)

or Spanish (in order to complete the test in Spanish).

17
 The critical items were interspersed within the 51 paragraphs. The proportion of

critical items and fillers (sentences where agreement was not an issue) was 24:96, that

is, four times more fillers than critical items. Only very rarely were two critical items

presented without a minimum separation of 2 or 3 fillers. The Spanish and English

questionnaires were almost identical in order to establish a fair comparison between the

responses in both languages. Once the subjects had their questionnaires, they were

induced to provide unpremeditated responses. In order to promote these responses,

participants were told that the aim of the task was to measure the time that native

speakers took to complete simple tasks involving verb conjugation so as to compare

these times to those of second-language speakers. This way, the subjects’ attention was

distracted from the real purpose of the task.

4.2.3 Subjects

A total of 43 native speakers of English and 42 native speakers of Spanish were

recruited for this experiment. The English speakers were either from the United

Kingdom or the United States. No differences based on the dialect of English employed

by each speaker were found in the use of these expressions in particular. Therefore, no

separate analysis of the American and the British data was considered necessary. As for

the Spanish speakers, they were mainly from Spain. No separate analysis based on the

subjects’ origin was made, since as in the case of English, dialectal variants were not

thought to influence this type of agreement. The age range was from 20 to 62, and the

results were also analysed independently of the subjects’ age. In Berg’s (1998) study no

effects concerning age were found, so this variable was overlooked in the present

research.

18
5 Results of the completion study

Overall, the subjects understood the task correctly, though one English participant had

to be discarded because he did not follow the instructions provided at the top of the

questionnaire. In addition, 0.40% of the English data and 0.10% of the Spanish data

were classified as inappropriate and thus did not contribute to the final score.

The percentages of singular and plural responses for each particular sentence are

summarised in the appendix. Overall, the results show a clear trend towards a left

orientation of the agreement processes in English, whereas much more orientational

flexibility is present in Spanish. However, when identification of the basic SVO

syntactic structure is not at risk in English, this language seems to prefer notional

agreement over morphosyntactic agreement. Again a different picture is revealed in

Spanish, where formal agreement seems to lead the process in most structures. The least

that can be concluded from this very general survey is that the characteristics of the

morphological system of a particular language seem to be largely responsible for its

agreement decisions.

We proceed now to a comparison between the English and the Spanish results

obtained in each category:

5.1 Results obtained for category (1)

[INSERT FIGURE 1 ABOUT HERE]

Figure 1. A (+ADJ) number/bunch/gang of + pl. // Un (+ADJ) número/racimo/ramo/banda de + pl.

5.1.1 Discussion of the English results

The results obtained for category one (figure 1) point to the following facts.

19
 There is a clear tendency towards plural agreement in English (82.28% pl. vs.

17.12% sg.). English subjects base agreement choice on the notional plurality of these

expressions rather than on their grammatical singularity. Notice that, as opposed to what

happens in attraction, these choices are grammatical. It is worth bearing in mind that

the sg-pl make-up of the phrases which typically induces attraction effects in the

psycholinguistic literature accounts for up to 13% of incorrect responses at the most

(Eberhard et al. 2005), so attraction alone cannot explain this solid pattern of

behaviour. Direct appeal to conceptually plural referentiality of expressions such as ‘a

gang of thugs’ is then responsible for the choice of a plural verb. These expressions

denoting multiple referents behave in a similar way to collective nouns.

Sentence (16), the bunch of grapes (TO BE) on the kitchen table, presents

additional constraints influencing agreement choice. It shows a robust preference for

plural and may well represent a tendency towards the grammaticalization of all of the

pre-N2 sequence as a sort of complex determiner or quantifier in the way suggested by,

for instance, Brems (2003) and Langacker (1991b:89) (see below on un número de in

Spanish). The interesting thing is that, as Keizer (2007:151) notes (see also Brems,

2003, forthcoming), these complex NPs suggest a cline with three possible analyses and

varying strengths of activation: a. as right-headed structures; b. as referential, left-

headed ones; and c. as hybrids:

First of all, one and the same construction can be interpreted in more than one
way, depending on the way in which the entity in question is conceptualized by
the discourse participant. Thus a construction like a cup of coffee can be
conceptualized either as a concrete object containing some fluid or as a certain
amount of coffee (the volume of an average coffee cup). But, (…), this
evidence is not always unequivocal; instead it seems as though the two
conceptual domains (of containment and quantification) overlap, or blend, in
the mind of the language user, resulting in a construction which exhibits
features of both. In those cases, the construction in question can be regarded as
conceptually situated on the border of two categories…

20
 Sentence (26), That bunch of beautiful big red roses (TO BE) the best thing I

could have bought her, is also interesting in that it provided the majority of the (anti-

trend) singular responses. In fact, sentence (26) is the only sentence within this category

which shows a higher number of singular than of plural responses in English (27:15). It

is noteworthy that this sentence was manipulated by adding dependents to the second

noun of the complex NP, a fact that was intended to increase its topicality. This also

separated the head noun and the verb more than in standard cases of attraction. Yet the

processor preferred agreement with the head, against the general tendency of the

construction, and that of the language at large. This suggests a strong level of

entrenchment of a bunch of FLOWERS construction as a left-headed structure and is

precisely in line with Brems’ idea (2003: 294ff) that the original (unbleached) meaning

of bunch, which used to be applied to only a restricted set of nouns such as flowers,

grapes and so on, retains its strong lexicality with such modifiers (therefore not

promoting its own grammaticalization as a quantifier). Since in general no effects of the

topicalization provided by increased adjectivization were obtained, no extra reference

will be made to this.

5.1.2 Discussion of the Spanish results

The Spanish results showed much more variability, though generally subjects seemed to

prefer the singular, causing the verb to agree with the grammatical features of their

nominal expressions (rather than with the notional ones). However, a high number of

plural responses were also provided by the participants in this experiment. Closer

inspection shows that these plural answers were mainly found in one construction only,

namely un número de… (‘a number of’). The reason for this disparate behaviour might

be that number, unlike gang or bunch, does not evoke a referent easily unless it is

21
associated with another noun that provides more imaginable referentiality (Eberhard

1999). This is at least partial evidence for a syntactic analysis of such desemanticized

strings as complex determiners, along the lines of recent work by Keizer (2007:VI),

cited above, where such ‘pseudo-partitive’ constructions are seen as ‘right-headed’. By

contrast, the duality of a gang of thugs or a bunch of flowers is evident in that one can

easily profile the entire set – as opposed to the containing units in the set – in them,

making them potentially ‘left-headed’ (while still also potentially right-headed, hence

their duality).

Figures (2) and (3) show the difference in the percentages obtained for

singular and plural in un número de + pl. and un racimo/ramo/banda de + pl:

[INSERT FIGURES 2 AND 3 ABOUT HERE]

Figures 2 & 3. Different results obtained for un numero de + pl. and un racimo/banda de + pl.

5.2 Results obtained for category (2a)

[INSERT FIGURE 4 ABOUT HERE]

Figure 4. The cause of the accident (TO BE) bad brakes // La causa del accidente (SER) los frenos.

5.2.1 Discussion of the English results

The results for category 2a (Figure 4) show an inclination of the English speakers to

choose the NP which is located to the left of the verb (in this case the one in the

singular) as the subject of the sentence. This is a consequence of the fact that English

speakers tend to find the subject in a preverbal position, due to the strong SVO bias of

this language. As maintained by Berg (1998:41), this left-orientation is most probably

caused by its morphological attrition, since English cannot use morphological marks in

order to match distant elements within a clause. Thus, subjects are almost always

22
located to the left of the verb, since word order is the only way to assure that there will

be a relational link between subject and predicate.

5.2.2 Discussion of the Spanish results

A comparison with Spanish shows an almost perfectly opposite pattern, as this language

prefers agreement with the most ‘informative’ NP, that is, with the NP that is

semantically and/or pragmatically richer, irrespective of its position within the structure

of the sentence. The general tendency of the language no doubt contributes to this trend

due to the fact that, as is well-known, in Spanish information structure is usually

accommodated via word order arrangements, which results in the frequent postposition

of focal subjects (Lambrecht 1994).

5.3 Results obtained for category (2b)

[INSERT FIGURE 5 ABOUT HERE]

Figure 5. Two broken pistons (TO BE) the cause of the sinking // Dos pistones rotos (SER) la causa del
hundimiento.

5.3.1 Discussion of the English results

As illustrated in figure 5, here both languages show a strong tendency towards plural

agreement. In the case of English, this can be predicted in the face of syntactic

constraints which cause the verb to agree with the element at its left, which in this case

is plural.

5.3.2 Discussion of the Spanish results

As for Spanish, the syntactic flexibility of this language allows agreement with either of

the two NPs. Here, the Spanish participants seem to prefer (as in 2a) the NP which is

23
more contentful (dos pistones rotos, dos mil euros…) instead of the cohesive, low-in-

information-value NP after the copula (la causa, la consecuencia, el motivo…).

5.4 Results obtained for category (3a)

[INSERT FIGURE 6 ABOUT HERE]

Figure 6. What I actually adore (TO BE) horror films // Lo que a mí realmente me gusta (SER) las
películas de terror.

5.4.1 Discussion of the English results

Overall, the pattern does show that even English can overcome its solid SVO bias under

certain specific circumstances. It is interesting that Berg (1998:57) also attested this

reversal of the left-orientation reflex of English in his study. He suggested that all and

what:

…are relatively neutral in terms of number. The pronoun all for example can
modify a singular noun (all the time) or a plural noun (all the students). This
relative neutrality does not interfere with the semantic principle, which may
consequently deploy its full vigor. (Berg 1998:57)

However, a distinction must be made in English between pseudoclefts beginning

with what and those beginning with all. Figures 7 and 8 show that the results for these

two types of pseudoclefts are not similar:

[INSERT FIGURES 7 & 8 ABOUT HERE]

Figures 7 & 8. Different results obtained for pseudoclefts introduced by what… and all that…

In fact, pseudoclefts beginning with what in English promote verb agreement with the

constituent to the left of the verb (the what- clause), while pseudoclefts beginning with

all show a stronger inclination toward plural verb agreement. So the majority of the

24
English anti-trend results for this category pertain to the specific case of the latter

construction. A possible explanation for this is that the word all has such strong

inherent plurality in its conceptualization that the competition between semantics and

syntax is clearly won by the former. Berg (1998) did not make any distinction between

these two types of expressions and thus did not obtain clear results for pseudoclefts.

5.4.2 Discussion of the Spanish results

In category 3a (figure 6), Spanish again shows a tendency for the more contentful NP to

be the controller of agreement regardless of its syntactic position within the sentence.

As in the previous category, Spanish speakers seem to disregard quasi-empty,

anaphorically-cohesive nominals as agreement controllers and opt for rich, focal

referents to connect with the predicational core. For instance, they normally prefer to

establish verb agreement with a noun phrase such as horror films rather than with a

sentence beginning with either what or all (e.g., what I actually adore) which is more

semantically inert and much less easily imaginable and topicalizable. This trend is

extremely strong.

5.5 Results obtained for category (3b) -pseudoclefts-

[INSERT FIGURE 9 ABOUT HERE]

Figure 9. Ballet performances (TO BE) what I actually love // Las representaciones de ballet (SER) lo
que en realidad me gusta.

5.5.1 Discussion of the English results

In category 3b (figure 9), both languages show a preference for plural agreement. The

coalition between English habitual left-orientation and the NP1’s inherent topicality

explains this robust effect.

25
5.5.2 Discussion of the Spanish results

As predicted, the lack of left-orientation leaves content/topicality as the main instigator

of Spanish responses and, consequently, a more flexible effect emerged in this

language.

6 General discussion

The most comprehensive conclusion we can draw from the present findings is that

agreement is not an on-off switch which, when pressed, works in the same

predetermined ways for all the languages of the world. We obtained results similar to

those in Berg (1998) and to the pilot study of Acuña-Fariña (2009), all indicating that

English behaves differently from German and Spanish, and that the latter two languages

pattern in broadly similar ways. The cross-linguistic comparisons that the three

languages afford are limited both quantitatively and qualitatively since they constitute a

very poor sample, and besides this, the three do not represent extremely different

linguistic genealogies. Despite this, the fact that they behave in the expected way (given

Berg’s ideas about the repercussions of morphology in setting grammatical and

processing biases) is surely relevant.

It is not worthwhile to simply emphasize the poor morphology of English

compared to the rich morphology of Spanish and German, as this is well-known.

Similarly, it is commonly understood that as English gradually lost its once rich

morphology, it veered towards and finally settled for a rather rigid word order

(Sonderegger 1998), whereas Spanish and German did not (due to retention of their

ebullient morphology). The theoretical significance of what Berg (1998) suggested,

which the present work tries to bolster, is that far from displaying a frozen picture of

26
either a formal or a semantic nature, agreement uses more semantic or more syntactic

regulation depending on the morphological status quo of each language. This

architectural opportunism is, however, perfectly synchronized with the way languages

align themselves in terms of the interaction between word order and information

structure, affording a series of neat, formal predictions. Firstly, all things being equal

(such as the domain of agreement), then the stronger the morphological component of

each language, the less likely it is that semantics will penetrate into feature percolation

operations inside the noun phrase. This was evident from the results for category 1 in

Spanish (the ‘a NOUN of NOUN’ construction). Conversely, the less morphology is

employed, the stronger the need to have resort to semantics in order to project

referential phrases, and subject phrases in particular – as shown by the English results

for the same category. Furthermore, with less morphology, there will be more need for

subjects to be located in stable locations, resulting in word order fixity in the sense

suggested by Hawkins (1994:372; 2004:160; see also Siewierska 1998). As the results

of category 2b show, this tendency is noticeably not subject to ceiling effects, since the

results are even more robust when left-orientation and content are aligned. The results

also suggest that formal and semantic regulation of agreement operations cannot be set

a priori as a fixed Chomskyan cycle of encapsulated computation, as in current phase-

based forms of generative grammar (Chomsky 2001:46; 2005:13). Instead they depend

on the particulars of each language, as these have varying degrees of morphological

strength. Finally, and this was evident in both our results and in Berg’s, intracategorial

variability was observable for every category in the three languages studied (e.g. ‘a

bunch of’ construction in English and the ‘un número de’ construction in Spanish). This

means that in production, semantics can readily interfere with cyclic domains of

agreement, even in richly-inflected languages like Spanish or German in cases where

27
semantic construals are inescapably useful. In English, complex NPs may show a

continuum ranging from right-headedness (with first nouns reanalyzed as parts of

complex determiners) to left-headedness (and ‘standard’ nominal syntax), with

intermediate cases as suggested by Brems (2003) and Keizer (2007). The data from the

identifying structures are particularly interesting in that they show the fluid nature of the

choices: 1. In English the NP subject is expected to occur left of the VP through its

position in an underlying SVO template, and meaning is not needed, thus not consulted;

2. However, after another cycle of computation, meaning has an additive effect when it

aligns with left-oriented choices, as well as a competing effect when it does not, as is

the case with the pseudo-cleft construction involving the conceptually plural all; 3. the

Spanish NP BE NP sequences also show how a language that punctiliously observes

morphological information when constructing phrasal constituents disregards that

information during the functional assembling of the sentence structure. Indeed, one of

the study’s more conspicuous findings was that a central component of sentence

structure – the identification of the subject phrase of which something is predicated – is

resolved differently within the two languages examined.

The same architectural opportunism is revealed in studies of attraction, as

evidenced by our brief review in Section 2. Primarily after the first series of studies by

Bock & collaborators and Vigliocco & collaborators (e.g. Bock & Cutting 1992; Bock

& Eberhard 1993; Vigliocco et al. 1995, 1996a, 1996b, etc.), it was believed that

conceptual distributivity was to be more expected in the Romance languages because

their rich morphology provided frequent opportunities to consult meaning

representations directly linked to conspicuous form cues. Nowadays, the opposite

theory is gaining ground (Acuña-Fariña et al, submitted). A recent attraction study by

Lorimor, Bock, Zalkind, Sheyman, & Beard (2008) in Russian, a richly-inflected

28
language like Spanish or German, showed that this language was less, rather than more,

affected by semantic interference than English. Additionally, after conducting a meta-

analysis of cross-linguistic attraction experiments showing differential sensitivity to

semantic interference, Bock & Carreiras (submitted) concluded that, if anything, a

richer morphology meant meaning was less likely to find a place in agreement

computations. Haskell et al. (2010) show that data from English corpora reveal a

tendency to grammaticalize semantic interference. They even suggest that a process

similar to structural priming contributes to the observed asymmetry of attraction

mistakes (sg-pl combinations attract, but pl-sg ones do not) via a speaker’s past

experiences with related agreement constructions (which contain singular subjects with

plural predicates but not plural subjects with singular predicates). Haskell et al.

explicitly advocate probabilistic learning of these biases. Acuña-Fariña (2009) reviewed

the recent psycholinguistic literature on the processing of agreement (not attraction),

notably using electrophysiological measures (like Event Related Potentials (ERP)), and

concluded that the blind routinization of agreement operations which characterizes

agreement in the Romance languages is evident even at the far right end of Corbett’s

Agreement Hierarchy, with pronouns separated by long distances, where semantic

interference is expected to be strongest. For instance, in Italian, epicenes like la vittima

or il personaggio (which can refer to male and female referents but have either feminine

or masculine morphology) enter agreement relationships by preferentially ‘bonding

with’ (Garrod & Terras 2000) pronouns of the same morphological gender. This

happens even in contexts which have been carefully biased semantically or

pragmatically to produce a different pattern of co-indexation (Cacciari et al. 1997).

Interestingly, nouns like erede (‘heir’), which do not mark gender (and are thus like

ordinary English nouns), are sensitive to semantic manipulations and preferentially

29
bond with pronouns that match contextual biases. Finally, the same opportunism – that

is, the same tendency to use the very resources that preferentially abound in each

particular language – is also evident in a recent completion study by Franck et al.

(2008) in which they manipulated the morphology of the article + noun sequences in

Spanish, French, and Italian. This showed that speakers tend to trust the morphological

cue that is most reliable: the noun ending in Italian, both the article and the noun in

Spanish (as the o/a ending in Spanish nouns is only transparent in 70% of cases), and

the article in French (as nominal endings have eroded in this language). All this

suggests that a system of grammar and processing makes use of multiple, probabilistic,

soft constraints rather than fixed serial access to information, automatic feature

percolation and strict phrasal cycles. This does not mean that such directionality and

cyclicity is unlikely. On the contrary, it may be useful in the rich morphology languages

with strong alliterative form redundancy in comprehension, as in Spanish, where form,

not meaning, drives the initial processing cycles (see Acuña-Fariña 2009 and the

psycholinguistic evidence cited therein). The interesting aspect of cross-linguistic

psycholinguistic research, the kind of research done by Berg (1998) and that pursued

here, is the global picture which shows: 1. opportunistic use of strong existing

constraints in each type of language; 2. gradability in both representation and

performance; and 3. a framework of options for advantageous use which is mainly

reduced to a constant conflict between form routinization and semantic interference,

which is resolved both at the level of each particular language and at the level of each

particular construction.

Since agreement operations, word order biases, and redundancy largely form the

core of grammar systems, it makes sense to expect that the way each language resolves

such biases will have ramifying consequences for the remainder of those systems. In his

30
closing remarks, Berg (1998:64ff) himself mentions a number of fairly specific

ramifications of the different behaviour of agreement in English and German: animacy

in the Saxon Genitive, the presentative construction with there, the grammar of NP

determination and definiteness, and even the use of punctuation. In all cases, he notes

that the two languages align in exactly the same way as they do in the case of agreement

proper: with more use of semantics in the English constructions and less in German.

Finally, Acuña-Fariña (2009:415) extends the ramifications to the markedly different

constraints on the use of misrelated participles in Spanish and English (where a

dependent participial clause recovers its subject geometrically by looking at the subject

of the matrix clause almost without exception) and suggests that even gap-filling and

long-distance dependencies may be resolved differently in the two languages. Exploring

these expected ramifications in detail may prove worthwhile in the future.

Notes:

*This research was funded by the Spanish Ministry of Science and Innovation (grant number PSI2009-
11748) and the Regional Government of Galicia (grant number INCITE09204014PR). These grants are
hereby gratefully acknowledged.
 

i
Vigliocco et al. (1996a) did find a relatively robust effect in Spanish.

ii
The grammar of the interaction of all these principles is still being written. For instance, Haskell et al.
(2010) point out that singular subject NPs that trigger plural agreement on the verb are a characteristic
feature of English attested in corpora (whereas the opposite pattern – plurals establishing singular
agreement with verbs – is never found). This introduces a frequency dimension in current research that
may make the interaction of all these forces even more complex.

iii
Much research on agreement processing has used acoustic/oral methodologies (Bock & Miller 1991;
Vigliocco, Butterworth & Semenza 1995; Vigliocco, Butterworth & Garrett 1996; Franck et al. 2008,
etc.). Although here the subjects had to respond to the stimuli presented on a written form, they were
encouraged to imitate the manner in which they normally speak. Furthermore, they were induced to react
quickly, thus not giving them time to think about the structure of the sentences with which they were
presented. This is the same methodology used by Berg (1998).

31