! 1

Clausal Conjunction cannot be the only source of

Closest Conjunct Agreement: A Picture-Matching Study

A prominent and elegant hypothesis about agreement phenomena generalized as closest conjunct

agreement takes this pattern to result from clausal coordination, in which within each conjunct

verb agrees with a non-conjoined subject, while a later clause-reduction operation deletes the

repeated part of the clause within the second conjunct that results in apparent closest conjunct

agreement (Aoun et al. 1995, 1999). Closest conjunct agreement is the dominant agreement

pattern in the South Slavic languages Slovenian and Bosnian/Croatian/Serbian (B/C/S). A

natural question is whether closest conjunct agreement in these varieties may indeed be analyzed

as entirely derived from conjunction reduction. In this paper, we report on two experiments

conducted on Slovenian and B/C/S to test this, using a picture-matching study to more closely

control the interpretation associated with each sentence. The results of our experiments reject

the hypothesis as far as these languages are concerned.!

Keywords: Agreement, Conjunction, Closest Conjunct Agreement, Conjunction Reduction,

Forced Choice Picture Experiment, Matching-Estimation Experiment, South Slavic

1. Introduction

In a range of typologically and genetically diverse languages, including Hindi, Arabic and

different members of the Slavic family, a conjoined subject may trigger verbal agreement with

! 2

only one of its conjuncts, as in (1) (Munn 1993, Aoun et al. 1994, 1999, Babyonyshev 1996,

Marušič et al. 2007, 2015, Bhatt & Walkow 2013, a.o).1

(1) Qaraʔat ʕaliyaa wa ʕumar l-qişşa. Standard Arabic

read.3FSg Alia and Omar the-story

'Alia and Omar read the story.' Aoun et al (1994: 207)!

In some languages, this pattern only obtains with the first conjunct (First Conjunct Agreement,

FCA), and in others only with the last (Last Conjunct Agreement, LCA), while in some both

options are found, additionally conditioned by the surface ordering between the subject and the

verb, and according to some researchers also by whether the language displays head-initial or

head-final syntax (Bhatt & Walkow 2013, Polinsky 2014).

Slovenian and Bosnian/Croatian/Serbian (B/C/S) are languages in which preverbal subjects

trigger both types of single conjunct agreement. However, with postverbal subjects FCA (a

subtype of CCA) is dominant, and LCA is clearly degraded (Willer-Gold et al. 2016, Arsenijević

& Mitić 2016a, Arsenijević & Mitić 2016b). Consider (2a), where the preverbal subject licenses

FCA (neuter plural, NPl), LCA (feminine plural, FPl) and a default or resolved gender value

(masculine plural, MPl). By contrast, the postverbal subject in (2b) shows very strongly degraded

LCA and a very low production rate of default agreement. !

(2) a. Penkala i olovke su pronađena/pronađene/pronađeni.

fountain-pen.NPl and pencil.FPl AuxPl found.NPl/FPl/MPl!

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!1 Single conjunct agreement of adnominal elements is not discussed in this paper.`!

! 3

b. Pronađena/??pronađene/pronađeni su penkala i olovke.

found.NPl/FPl/MPl AuxPl fountain-pen.NPl and pencil.FPl

‘Fountain-pens and pencils have been found.’2

Next to the given patterns available with different gender conjuncts, there is an additional

strategy of resolution when the conjuncts are of the same gender: agreement in a non-masculine

gender shared by the conjuncts (where at least for some speakers resolution differs from default

agreement, see Arsenijević & Mitić 2016a and Willer-Gold et al. 2016 for discussion).

(3) Majka i ćerka su stajale / %stajali ispod kišobrana.

mother.F and daughter.F AuxPl stood.FPl/MPl under umbrella

‘Mother and daughter stood under the umbrella.’!

The availability of a broad spectrum of cross-linguistically attested agreement patterns with

conjoined subjects makes Slovenian and B/C/S fruitful ground for more detailed research of

conjunct agreement, especially for single conjunct agreement.

A number of competing analyses have been offered for single conjunct agreement

phenomena in postverbal order (Munn 1993, Aoun et al. 1994, 1999, Babyonyshev 1996, Sadler

2004, Tantalou & Badecker 2005, Soltan 2006, Marušič et al. 2007, 2015, Bhatt & Walkow

2013). One of the analyses that keeps coming back is the one involving clausal conjunction

followed by a type of ellipsis referred to as conjunction reduction. Aoun et al (1994, 1999) argue

that this configuration is the sole source of the single conjunct agreement in Arabic, but

explicitly refrain from ascribing it a universal cross-linguistic status. Bhatt & Walkow 2013 and

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!2 All the examples in the paper, where not specified otherwise, come from B/C/S. As the research has

targeted both B/C/S and Slovenian, it is important to clarify that near counterparts were used in Slovenian, which

did not differ in any aspect immediately relevant for the purposes of this research.!

! 4

Polinsky 2014 discuss it as one possible underlying structure in other languages too – yielding

different surface agreement patterns depending on whether the language is head-initial or head-

final. Schein (2017) argues that coordination is always at the clausal level and thus CCA, as well

as any other agreement pattern, necessarily involves clausal coordination followed by

conjunction reduction.

According to this type of analysis, a sentence like (4a) would roughly be derived from a

structure which coordinates two clauses, and then elides the material from one of them, as in

(4b). The surviving verb simply agrees with its local clausal subject, yielding what looks at the

surface as single conjunct agreement. A similar derivation would underlie (4c, d).!

(4) a. U supi su se kuvale knedle i rezanca.

in soup AuxPl Refl cooked.FPl dumpling.FPl and noodles.NPl

‘Dumplings and noodles simmered in the soup.’!

b. U supi su se kuvale knedle i u supi su se

in soup AuxPl Refl cooked.FPl dumpling.FPl and in soup AuxPl Refl

kuvala rezanca.

cooked.NPl noodles.NPl

‘Dumplings simmered in the soup and noodles simmered in the soup.’

c. Knedle i rezanca su se kuvala u supi.

dumpling.FPl and noodles.NPl AuxPl Refl cooked.NPl in soup

‘Dumplings and noodles simmered in the soup.’

d. Knedle su se kuvale u supi i rezanca su se

dumpling.FPl AuxPl Refl cooked.FPl in soup and noodles.NPl AuxPl Refl

kuvala u supi.

! 5

cooked.NPl in soup

‘Dumplings simmered in the soup and noodles simmered in the soup.’

Let us refer to this analysis, where phrasal conjunction yields default (DEF) or resolution (RES),

and single conjunct agreement (e.g. CCA) is always an epiphenomenon of conjunction reduction,

as the Type of Conjunction Hypothesis (TCH).!

Marušič et al (2007, 2015), in particular for Slovenian, argue against TCH. They offer

examples manifesting the compatibility of single conjunct agreement with collective predicates

as in (5) as a crucial argument against the conjunction reduction analysis. They assume that if the

underlying structure of sentences with single conjunct agreement involved clausal conjunction

(as in 5b), a semantic violation should emerge when their predicate is of the collective type. (5a)

is therefore not derived from (5b).!

(5) a. Krava in teleta so se pasla skupaj Slovenian

cowF and calvesN are refl grazeN-PL together

‘A cow and her calves were grazing together.’ !

b. *Krava se je pasla skupaj in teleta so se pasla skupaj.

cow refl aux grazed together and calves aux refl grazed together

However, Aoun et al. reject a similar argument formulated by Munn 1999 for Arabic.

Moreover, there are analyses, such as Schein (2017), on which sentences like (5a) are indeed

equivalent to those like (5b), but both types have a biclausal interpretation, i.e. a two-event

interpretation – roughly: A cow took part (together with other participants) in a grazing event,

and calves took part (together with other participants) in a grazing event. Pragmatics may then

impose the identity of the two events, but this is not entailed by the underlying structure. !

! 6

In this paper, we aim to test whether it is possible to maintain that all sentences with the verb

which agrees with the linearly closest conjunct (closest conjunct agreement, CCA) are

underlyingly biclausal with two independent events. We exploit the implication of such an

analysis that underlyingly biclausal sentences will be more prone to accept a reading with two

separate events, one corresponding to the clause with the hypothesized first conjunct, and the

other to that with the hypothesized second clausal conjunct.

We report on two experiments that we designed and conducted to offer answers to these

questions. They test the availability of two-event interpretations for sentences with conjoined

subjects, depending on the pattern of agreement shown. The biclausal analysis of CCA predicts

that sentences with CCA will strongly prefer the two-event interpretation to the one-event

interpretation with a joint, mixed, participation of the referents of both conjuncts, while for

sentences with the default or resolved agreement no precise prediction is made – considering that

a distributed interpretation of the predicate is generally available, unless explicitly suppressed. In

order to achieve this effect of suppressing the 2-event interpretation, we additionally introduced

the variable predicate type (with 2 levels: collective and non-collective). Collective predicates

are supposed to suppress the 2-event interpretation, and if Munn’s (1999) and Marušič et al’s

(2007) argument is correct, these should also disfavor the biclausal underlying parse of sentences

with conjoined subjects.

As an illustration, consider the sentences in (6) and the picture in Figure 1.

(6) a. U bici su se sudarali koplja i sablje

in battle Aux.Pl Refl collided.MPl spears.N and swords.F

b. U bici su se sudarala koplja i sablje

in battle Aux.Pl Refl collided.NPl spears.N and swords.F

! 7

‘Swords and spears collided in the battle.’

c. U bici su se sudarala koplja i sudarale su se

in battle Aux.Pl Refl collided.NPl spears.N and collided.FPl Aux.Pl Refl

sablje

swords.F

‘Swords collided in the battle and spears collided in the battle.’

Figure 1

On the TCH, the sentence with CCA in (6b) is expected to be incompatible with -- or at the

very least to strongly disfavor -- the picture in Figure 1, and the sentence in (6a) is expected to

favor for the reading visually presented in this picture. Moreover, the sentence in (6b) is

predicted to show the same pattern (namely incompatibility with the picture) as one with an

overtly biclausal structure such as (6c).!

! 8

We tested these predictions in the group of South Slavic languages in which the relevant

oppositions are morphologically visible (e.g. gender distinctions maintained in the plural),

namely in Slovenian and B/C/S, in four experiments across 7 sites where CCA has been

previously attested as robust: Nova Gorica (Slovenian; NG), Sarajevo (B/C/S; SA), Zenica

(B/C/S; ZE), Zadar (B/C/S; ZD), Zagreb (B/C/S; ZG), Niš (B/C/S; NI) and Novi Sad (B/C/S;

NS). Given that previous studies have shown that CCA in gender is most robustly attested when

both conjuncts are inanimate and plural, we restricted these experiments to such cases.

2. Experiment 1: Sentence-picture match judgment

2. 1 Aim of the experiment

In this experiment, we wanted to test how conjoined subjects and collective predicates affect the

compatibility of a sentence with a single event interpretation. The experiment was based on two

controlled variables: predicate type (2 levels: collective vs. non-collective semantics of the

predicate) and type of subject (2 levels: conjoined (&P) and non-conjoined (NP)), crossed in a

2x2 design to give four conditions, each matched with a picture representing the respective

single event interpretation. We illustrate the four conditions with sentences from B/C/S, noting

that the sentences in Slovenian were their equivalents.

1. conjoined subject with a collective predicate,

! 9

U bici su se sudarala koplja i sablje

in battle Aux.Pl Refl collided.NPl spears.N and swords.F

‘Swords and spears collided in the battle.’

Figure 2 (repeated from Figure 1)!

2. single NP subject with a collective predicate,

U bici su se sudarala koplja

in battle Aux.Pl Refl collided.NPl spears.N

‘Spears collided in the battle.’

! 10

Figure 3

3. conjoined subject with a non-collective predicate,

U trgovini su se prodavala ogledala i lampe

in shop Aux.Pl Refl sold.NPl mirrors.N and lamps.F

‘Mirrors and lamps were sold in the shop.’

Figure 4

! 11

4. single NP subject with a non-collective prediacate.

U trgovini su se prodavala ogledala

in shop Aux.Pl Refl sold.NPl mirrors.N

‘Mirrors were sold in the shop.’

Figure 5

The TCH predicts that sentences with conjunction will generally show a significantly lower

compatibility with a single event reading than sentences with a single NP subject. Since the

sentential stimuli accompanying conjoined subjects included a CCA pattern of agreement,

considered by the TCH to come from underlying clausal conjunction, this condition is predicted

to cause certain conflict with the collective predicates in the respective conditions if indeed

clausal conjunction is the only source of CCA.!

! 12

2.2 Materials and method

2.2.1 Participants

The experiment was carried out at seven research institutions: University of Nova Gorica (tested

also at University of Ljubljana), Slovenia; University of Zagreb and University of Zadar,

Croatia; University of Sarajevo and University of Zenica, Bosnia and Herzegovina; University of

Novi Sad and University of Niš, Serbia. In total, 210 undergraduate students participated in the

sentence-picture match judgment experiment. 30 first or second year students (sex (F=67%,

M=23%), age (Mean=21.6)) participated at each of the seven research institutions. They were all

native speakers of the local language variety, attended the local secondary school and were not

pursuing a university degree in the study of the local language. Their participation was either

voluntary or they received course credits for their participation.

2.2.2 Materials and Design

The sentence-picture match experiment was designed as a version of a standard sentence-picture

matching task in which participants are asked to mark the value of sentence-picture match on a

scale from 0 to 100(%). A single experimental design and procedure was implemented under

equal experimental conditions across all seven research institutions. The language used in the

experiment was adapted to the research institutions’ local neutral varieties. Experimental

materials were created in parallel in the variety of Zagreb Croatian and Slovenian. The Zagreb

Croatian materials were later adapted to the target B/C/S language variety, i.e. those of Niš, Novi

Sad, Sarajevo, Zadar and Zenica. The adaptations were minimal to ensure uniformity across

research locations. They were mostly due to variation in specific lexical items. .!

In the experiments a total of 64 experimental sentential items were presented to the

participants. The factors predicate type [Collective, Non-Collective] and category of subject

! 13

[&P, NPPL] were used to yield a 2x2 design, and 8 mixed-gender &P (FN,NF,F,N) items were

created for each of the two levels of predicate type. All items were surface monoclausal. The

items all had the same structure: [ADV Aux (ReflPro) PredCCA &P/NPPL], i.e. a sentence-initial

adverb followed by inflected auxiliary (and optional reflexive pronoun), a predicate, and a final

subject phrase. The 16 mixed-gender items were further manipulated to create the NP condition

of category of subject experimental items by retaining the first NP of the conjunction (F,N).

These were all monoclausal; to create a total of 32 items. All items in the experiment were paired

with mixed-event pictures of the types in Figure 2 and Figure 4. A full list of the stimuli is

presented in Appendix 1.

Filler items were created using same 2x2 design and the same number of items per

condition. The gender of the nouns used for subjects were (F,N,M,MM) in order to balance the

item-gender ratio in the experiment. This created a total of 32 fillers. 50% of fillers were

designed as sentence-picture mismatches (cf. ungrammatical), yielding 16 items, such as

mismatch in the number of items depicting the subject, mismatch in depiction of the NP in the

subject position and mismatch in depiction of the NP in the adverbial phrase.

2.2.3 Methods

Experiment 1 was conducted in May 2016. The experiment was coded and administered using

the on-line experimental platform IbexFarm (Drummond 2011)3. It was presented on a computer

screen in parallel or individually. The sentence-picture match ellipsis experiment contained an

introduction with task description, 6 practice items followed experimental items and fillers

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!3 We are grateful to Alex Drummond for maintaining this platform, and for his direct help in adapting

IbexFarm to enable this type of experiment.!

! 14

presented in randomized order, each one separated by a blank screen. In the case of practice and

experimental items, participants saw a sentence-picture pair and a slider on the screen, and their

task was to position the slider on the scale so to mark their response – the value of the match.

After reading the sentence and observing the picture depicting the event that the picture

describes, their task was to evaluate to what degree the picture matches the sentence by

positioning the slider on the scale. By positioning the slider on the left side of the scale, they note

a degree of low match (in red) and by positioning it on the right side of the slider note a degree

of high match (green). After marking their response they had to press a continue button

(“NASTAVI”/“NADALJUJ”) in order to proceed to the next experimental item. The duration of

the blank screen between the two items was 750ms. Participants typically completed the

experiment within 15 minutes. IbexFarm randomized a new list of items for each participant.

Acceptability judgment responses were automatically recorded by IbexFarm and exported

afterwards for statistical analysis according to the experimental conditions [Collective-&P,

Collective-NP, Non-Collective-&P and Non-Collective-NP].

2.3 Results and discussion

The analysis was conducted using R package lme4, which provides a generalized linear mixed

model fit by maximum likelihood (glmer (glmerControl(optimizer = "Nelder_Mead"))) with the

response value as dependent variable, and predicate type [Collective, Non Collective] and

category of subject [&P, NPPL] as independent variables. The summary of Experiment 1 results

is shown in Figure 6. The results of monoclausal sentences paired with mixed event pictures

show a statistically significant difference in predicate type [Collective, Non-Collective]

(Estimate 0.9245, Std. Error 0.3277, z value 2.821, Pr(>|z|) 0.00479**); thus, sentences with

! 15

collective predicates were generally judged as less of a match for a single event than those with

non-collective predicates. However, as no statistically significant difference in category of

subject [&P, NPPL] (Estimate 0.1278, Std. Error 0.3230, z value 0.396, Pr(>|z|) 0.69222) was

found, no effect of conjunction itself is attested; thus conditions with a conjoined subject were

judged equally good as those without one. Moreover, no interaction was observed between the

two variables.

Figure 6

This is clearly the opposite of what TCH predicts. The absence of a significant difference

between NP and &P with CCA for both verb types shows that conjunction itself imposes no

detectably different representation with an effect on the sentence-picture match. The effect of

collectivity, on the other hand, may be a matter of processing complexity: verifying that swords

are colliding – or that swords and spears are colliding -- for example, involves inspection of each

sword and spear pair. There is no clear theoretical reason that collective predicates in the NP

ns ns***

77% 79%

85% 86%

40%

50%

60%

70%

80%

90%

100%

Collective-&P Collective-NP Non-Collective-&P Non-Collective-NP

Monoclausal sentences

**ns ns

! 16

condition should show a lower sentence-picture match than non-collective predicates, and we

therefore chalk this difference up to factors outside of the TCH itself. Finally, the absence of an

interaction may be attributed to the lack of a significant effect of one of the variables, namely

conjunction vs. simple NPs.

3. Experiment 2: Forced-choice picture task

3.1 Aim of the experiment

The aim of Experiment 2 was similar to that of Experiment 1, but employed a different task, to

potentially provide another chance for the TCH to flourish. Experiment 2 aimed at finding a

preference for two-event readings given CCA. The experiment tested TCH, the hypothesis that

CCA as in (7a) is a result of a biclausal structure: a conjunction at the level of the clause, reduced

by ellipsis to the appearance of a phrasal conjunction – as illustrated in (7b).!

(7) a. U bici su se sudarala koplja i sablje

in battle Aux.Pl Refl collided.NPl spears.N and swords.F

‘Swords and spears collided in the battle.’

b. U bici su se sudarala koplja i sudarale su

in battle Aux.Pl Refl collided.NPl spears.N and collided.FPl Aux.Pl

se sablje

Refl swords.F

‘Swords collided in the battle and spears collided in the battle.’!

We conducted this new task because it could have been that the lack of a difference

between &P and NP in Experiment 1 was due to participants’ overall high tolerance: they

tolerated a relatively high degree of semantic mismatch in these sentence-picture matches, and as

! 17

a result ended up not achieving a statistically significant difference between the conjoined and

non-conjoined subjects. Experiment 2 was designed to force the participants to discriminate

these preferences and readings by a forced-choice between two pictures, which represented the

one- and the two-event interpretation.

Figure 7: Picture pairs: 1 uniform description (joint-event, left) and 2 different description

events (split-events, right)

Participants were asked to pick the better-matching picture for a given sentence, which

either had the monoclausal CCA pattern illustrated in (8a) or the overtly biclausal one in (8b).

(8) a. Znamo da su se po nekom kanalu mimoilazile

know.1Pl that Aux Refl over some canal passed_by.FPl !

lađe i druga plovila.

ships.F and other vessels.N

‘We know that ships and vessels passed each other over some canal’

b. Znamo da su se po nekom kanalu mimoilazile lađe

know.1Pl that Aux Refl over some canal pass_by.FPl ships.F

i da su se po nekom kanalu mimoilazila

! 18

and that Aux Refl over some canal passed_by.NPl

other vessels.N

druga plovila.

‘We know that ships passed each other over some canal and vessels passed each

other over some canal’

This variable was crossed with the variable of the predicate type with the values collective (as

(8a-b)) or non-collective, as illustrated in (9a-b).

(9) a. Izgleda da su u nekoj trgovini na prodaju stavljena

seems that Aux.Pl in some shop on sale put.NPl

ogledala i lampe.

mirrors.N and lamps.F!

‘It seems that some shop put mirrors and lamps on sale’

b. Izgleda da su u nekoj trgovini na prodaju stavljena

seems that Aux.Pl in some shop on sale put.NPl

ogledala i da su u nekoj trgovini na prodaju stavljene lampe.

mirror.N and that Aux.Pl in some shop on sale put.FPl lamps.F

‘It seems that some shop put mirrors on sale and that some shop put lamps on sale’

! 19

Figure 8: two pictures shown for non-collective sentences as in ().!

The combinations above illustrate all four conditions in the experiment:

- monoclausal-collective: sentence (8a) and the choice among two pictures in Fig 7,

- biclausal-collective: sentence (8b) and the choice among two pictures in Fig 7,

- monoclausal-non-collective: sentence (a) and the choice among two pictures in Fig 8,!

- biclausal-non-collective: sentence (b) and the choice among the two pictures in Fig 8.!

The TCH postulates the same underlying biclausal structure for both the surface-

monoclausal sentences with CCA and for the surface biclausal sentences each with simple

subjects (the hypothesized source of CCA clauses). Therefore, it predicts that faced with a

forced-choice between the two-event and one-event reading, participants will always select the

picture that represents the two-event interpretation. A significant difference between the levels

monoclausal and biclausal, in particular one where the former as opposed to the latter has a non-

marginal amount of selected 1-event interpretations – would falsify TCH.

The purpose of the variable predicate type was to strengthen the within-clause

interpretation of the subject in the hypothesized biclausal underlying structure. An objection is

possible to the outlined design, namely that even two conjoined clauses may be referring to the

! 20

same event (Schein 2017), implying that participants may still select the one-event picture while

parsing the sentence into a biclausal configuration. It still may be unlikely, at the level of

pragmatics: even if a biclausal structure may refer to a single event, given that there are more

economical expressions (in the sense of Katzir 2007) which can only have this interpretation,

namely monoclausal sentences (including those with a default or resolved agreement pattern),

there should be a clear advantage for the two-event reading for biclausal sentences. However, to

allow for the possibility that the events in the two clauses are interpreted as co-referential,

particularly in non-collective predicates, we included this predicate type variable.

This variable relies on the tendency of certain predicates to have a collective

interpretation. Such is the case, for instance, with the B/C/S verb sresti se and its English

counterpart to meet. These verbs require a plurality as a participant, which can be realized by a

combination of the subject and an indirect object, or by a plural subject alone. In the former case,

the predicate applies to the referents of the subject and of the indirect object, and in the latter

case between the members of the plurality denoted by the subject.

(10) a. Dečak se sreo sa devojčicom.

boy Refl met with girl

‘The boy met with the girl.’

b. Dečaci su se sreli.

boys Aux Refl met

‘The boys met (with each other).’

c. Dečak i devojčica su se sreli.

boy and girl Aux Refl met

‘The boy and the girl met (with each other).’

! 21

d. *Dečak se sreo.

boy Refl met

‘*The boy met.’!

When two clauses with identical predicates are conjoined, it is much harder to establish

co-reference between the two events when the predicates are collective.

(11) a. Dečaci su se sreli i devojčice su se srele.

boys Aux Refl met and girls Aux Refl met.

‘The boys met (with each other) and the girls met (with each other).’

‘#The boys participated as experiencers and themes in a meeting event, and the girls

participated as experiencers and themes in the same meeting event.’!

b. Dečaci su pevali i devojčice su pevale.

boys Aux sung and girls Aux sung.

‘The boys sang and the girls sang.’

‘The boys participated as agents in a singing event, and the girls participated as agents

in the same singing event.’

The collective predicate type should even more strongly block the one-event, joint

interpretation (represented by a the picture) for any biclausal structure. Therefore, in particular

the conditions represented by (8b), and, according to the TCH under the biclausal analysis of

(8a) as well, cannot be linked with a reading in which ships pass by not only other ships, but also

other vessels, i.e. the 1-event picture in Figure 7.

Similarly, the role of the variable surface clause size (overtly biclausal conditions vs.

those monoclausal on the surface) was to test that the claim that the biclausal structure favors the

2-event interpretation, as well as that the 2-event pictures manage to properly represent this type

! 22

of interpretation. In other words, the overtly biclausal conditions were supposed to provide a

baseline for the pattern of unquestionably biclausal conjunction. If the assumption that biclausal

conjunction favors 2-event interpretations was not confirmed by the biclausal conditions, then

the validity of these stimuli would be questionable.

3.2 Materials and Methods

3.3.1.1 Participants

In total, 90 undergraduate students participated in the forced-choice picture experiment. 30 first

or second year students (sex (F=72%, M=28%), age (Mean=20) participated at three research

institutions (University of Nova Gorica (tested at University of Ljubljana), Slovenia; University

of Sarajevo, Bosnia and Herzegovina; University of Novi Sad, Serbia). They were all native

speakers of the local language variety, attended the local secondary school and were not pursuing

a university degree in the study of the local language. Their participation was either voluntary or

they received course credits for their participation.

3.3.1.2 Materials and Design

The forced-choice experiment was designed as a version of a forced-choice task in which a

participant is presented with one sentence alongside two different pictures and their task is to

choose between the two in order to determine which one is a better match for the presented

sentence. A single experimental design and procedure was implemented under equal

experimental conditions across all research institutions. The language used in the experiment was

adapted to the research institutions’ local neutral varieties. Experimental material was first

created in Slovenian, and was later adapted to the target language variety, i.e. those of Sarajevo

! 23

and Novi Sad. The adaptations were minimal to ensure uniformity across research locations, and

were due to variation in the gender of specific lexical items. The same pictures and noun phrases

used in conjunctions were used as in Experiment 1.!

In the experiment a total of 64 experimental sentential items were presented to the

participants.

Experiment 2 used predicate type [Collective, Non-Collective] and clause size

[Monoclausal, Biclausal] as factors yielding a 2x2 design, with event picture [1-event, 2-events]

as the dependent variable. 8 mixed-gender combinations of nominal expressions between the

feminine (F) and the neuter (N) were matched with 8 selected predicates for each of the two

levels of predicate type. Each of the resulting 16 matches were then manipulated into a

monoclausal and a biclausal realization along the clause size variable, as illustrated in (8) and (9)

to create a total of 32 items (8 per each of four conditions). All monoclausal items had the same

structure: [ADV Aux (ReflPro) PredCCA NPNPL&NPFPL] – an initial adverb followed by an

inflected auxiliary clitic, a potential reflexive clitic, the verb and a final conjoined subject phrase.

All the biclausal items had the same structure: [Adv Aux (ReflPro) PredCCA NPNPL & Adv Aux

(ReflPro) PredCCA NPFPL] – an initial adverb followed by an inflected auxiliary clitic, a potential

reflexive clitic, a predicate and a subject phrase the conjunction, and then the same sequence

repeated in the second clause. Special attention was paid to the naturaleness of the resulting

examples, across all four conditions. A list of all stimuli is included in Appendix 2.

All items in the experiment were paired both with a one-event and a two-event picture.

Half of the stimuli had the 2-event picture on the left and the 1-event picture on the right, and the

other half had the opposite arrangement, and the different arrangements were randomly ordered. !

A total of 32 filler items sentences were used in the experiment – involving the same

! 24

fillers used in experiment 1. New matching and mismatching pictures were created, in order to

be paired with those from Experiment 1, as required by the forced-choice design of Experiment

2.

3.3.1.3 Methods

The forced-choice experiment was coded and administered using the on-line experimental

platform IbexFarm. The experiment was presented on a computer screen in parallel or

individually. It contained an introduction with task description, 6 practice items followed by

experimental items and fillers presented in randomized order. In the case of practice and

experimental items, participants saw a single sentence paired with two pictures. The task, which

was first practiced on the training items, was to read the sentence and observe the pictures and to

choose the picture that was a better match for the sentence by clicking on it. After clicking on the

better matching picture, they had to press a continue button (in the language of the participants)

in order to proceed to the next experimental item. IbexFarm produced a new random ordering of

items for each participant. Picture choice responses were automatically recorded by IbexFarm,

and exported and transcoded (into 1-event or 2-events) afterwards for statistical analysis

according to the number of choices per condition.

3.3 Results and discussion

The results of Experiment 2 show that monoclausal sentences with CCA yield exactly the inverse

patterns of the biclausal ones. For the biclausal conditions, the participants chose the 2-event

picture in over 90% of cases, and for the surface-monoclausal conditions in less than 10% of

cases. The variable predicate type does not appear to have any effect.

! 25

Figure 9: Picture choice per condition

A Two Factor Repeated Measure ANOVA test confirms these observations. The effect of

the factor clause size is highly significant (p < 0.00001; F = 1880.599), while the effect of the

factor predicate type does not even approach significance (p = 0.83238; F = 0.04563). No

interaction between the two factors is attested either (p = 0.88777; F = 0.02028).

The null result for the variable predicate type may be a matter of insufficient sensitivity

of the experiment, but it may very likely be a consequence of the strong effect of the other

variable. Namely, the collective level of the variable predicate type was intended to strengthen

the within-clause interpretation of the predicate, i.e. to have a closer correspondence between the

reading chosen and the underlying structure (1-event for the monoclausal underlying structure, 2

events for the biclausal one). Our experiment shows that the participants selected the 1-event

interpretation for the monoclausal conditions and the 2-events interpretation for the biclausal

condition, with a very strong contrast. This contrast might have simply been so strong that the

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

Collective-Mono Collective-Bicl Non-Collective-Mono Non-Collective-Bicl

Mono- vs Bi-clausal Sentences

2E

1E

! 26

room for strengthening by the variable predicate type was too small to afford significance to this

variable’s effect (see Arsenijević 2017 for additional discussion).!

The results for the surface-biclausal conditions confirm our assumption that the biclausal

structure, i.e. true clausal conjunction, has a strong preference for a 2-event reading represented

by the 2-event pictures. The fact that in a small number of cases the 1-event reading is still

chosen is potentially explained by the potential co-reference between the two events in the two

clauses, in which case they match the 1-event picture. Moreover, the fact that for the

monoclausal condition there were a number of cases where the 2-event interpretation is chosen

can be explained by the availability of the biclausal parse with a reduced clausal conjunction for

this structure – but clearly not as the only structure which derives CCA, or even a dominant one.!

The results reject TCH, since as discussed in section 2, this hypothesis predicts that the

surface-biclausal and the surface-monoclausal levels of the variable clause-size will have

identical results, both with a strong preference for the 2-event pictures. Exactly the pattern that

rejects the hypothesis was attested: the two levels factored the opposite extremes, with the

biclausal level inducing preference for 2-event interpretations, and the monoclausal level for the

1-event readings.

Our results pose a strong challenge for the theories of conjunction in which it always

targets the clausal level, and may receive the phrasal appearance only through conjunction

reduction (Schein 2017). The fact that the 1-event pictures are chosen in a number of cases

agrees with Schein’s view that for collective predicates co-reference is established between the

events in the conjoined clauses. However, the fact that this type of interpretation is only

marginally available for the overtly biclausal sentences, yet it is nearly the only available one for

the monoclausal ones, can hardly be explained under a strict clausal conjunction analysis.

! 27

4. Conclusion

One hypothesis about CCA takes this pattern to result from reduced clausal conjunction,

displaying agreement by the verb with a non-conjoined subject in the clause whose content

survives ellipsis (Aoun et al. 1995, 1999). Given that is the dominant agreement pattern in South

Slavic languages in which gender marking is visible in plural forms (Willer-Gold et al 2016), it

was important to see whether indeed it can be reduced to ellipsis. In this paper, we reported on

two experiments conducted on Slovenian and B/C/S with the aim to test whether the Type of

Conjunction Hypothesis applies universally to CCA patterns. Since CCA applies most notably to

conjoined inanimate plurals, we employed picture-matching paradigms, to more closely control

the interpretation associated with each sentence. Experiment 1, a sentence picture-match

judgment, showed a high rate of acceptability for monoclausal sentences paired with mixed event

pictures (i.e. 1-event pictures, unlike those expected given a biclausal source). The experiment

found an effect of collectivity of the predicates, but no effect of conjunction. Experiment 2, a

forced-choice picture task, showed a strong preference for the 2-event reading with biclausal

sentences and a 1-event reading with monoclausal sentences. These results allow us to reject the

TCH hypothesis as far as these languages are concerned, and are consistent with a model in

which CCA, the surface appearance of linear order effects in the grammar of agreement, is

derived by reference to the closest conjunct either syntactically (Bošković 2009, Murphy &

Puškar to appear) or in a division of labor between syntactic and post-syntactic mechanisms

(Bhatt & Walkow 2013, Marušič et al 2015, Willer-Gold et al 2016). While these latter debates

may continue to bear theoretical and empirical fruit, one can safely put to the side the hypothesis

that all cases of CCA in South Slavic are reducible to ellipsis of clausal conjunction.!

! 28

References

Aoun, J, E. Benmamoun, & D. Sportiche. 1994. Agreement, word order, and conjunction in

some varieties of Arabic. Linguistic Inquiry 25:195-220.

Aoun, J., E. Benmamoun, and D. Sportiche. 1999. Further remarks on first conjunct agreement.

Linguistic Inquiry 30:669-681.

Arsenijević, B. 2017. Ellipsis, conjunct agreement and the interpretation of collective predicates.

Ms, University of Graz.

Arsenijević, B. and I. Mitić. 2016a. On the (in)dependence of gender with respect to number in

agreement with coordinated subjects: An experimental study. Journal of Slavic Linguistics

24.1:41–69.

Arsenijević, B and I. Mitić. 2016b. In Novi Sad Workshop on Psycholinguistics, ed. S. Halupka-

Rešetar and S. Martinez-Ferreiro, 41-77. Novi Sad: University of Novi Sad.

Babyonyshev, M. 1996. Structural connections in syntax and processing: studies in Russian and

Japanese. PhD dissertation, MIT.

Bhatt, R. & M. Walkow. 2013. Locating Agreement in Grammar: An argument from Agreement

in Conjunctions. Natural Language and Linguistic Theory 31:951-1013.

Bošković, Ž. 2009. Unifying first and last conjunct agreement. Natural Language and

Linguistic Theory 27:455–496.

Drummond, A. 2011. IbexFarm, Version 0.2.7. Available at spellout.net/ibexfarm/.

Katzir, R. 2007. Structurally-defined alternatives. Linguistics & Philosophy 30.6: 669–690.

Marušič, F., Nevins, A. & B. Badecker. 2015. The grammars of conjunction agreement in

Slovenian. Syntax 18.1:39–77.

! 29

Marušič, F, A. Nevins, and A. Saksida. 2007. Last conjunct agreement in Slovenian. In Formal

Approaches to Slavic Linguistics 16, ed. J. Toman, 210-227. Michigan Slavic Publications.

Munn, A. 1999. First conjunct agreement: Against a clausal analysis. Linguistic Inquiry 30:643-

668.

Munn, A. 1993. Topics in the Syntax and Semantics of Coordinate Structures. PhD thesis,

University of Maryland.

Murphy, A. & Z. Puskar. To appear. Closest conjunct agreement is an illusion. Natural

Language and Linguistic Theory.

Polinsky, M. 2014. What agreement theory can learn from closest conjunct agreement. Talk

given at the Colloquium Grammatiktheorie, University of Leipzig. !

Schein, B. 2017. 'And' : Conjunction Reduction Redux. Cambridge, Massachusetts: MIT Press.!

Soltan, U. 2005. On postcyclic Merge and Agree in syntactic derivations: First Conjunct

Agreement in Standard Arabic Revisited. In Perspectives on Arabic Linguistics 19, ed. E.

Benmamoun, 191-213. Amsterdam: J. Benjamins.

Tantalou, N. & W. Badecker. 2005. Experimental studies of agreement in Modern Greek and

their consequences for grammatical theory. Poster presented at the 18th CUNY Human

Sentence Processing Conference.

Willer-Gold, J., B. Arsenijević, M. Batinić, N. Čordalija, M. Kresić, N. Leko, F. Marušič, T.

Milićev, N. Milićević, I. Mitić, A. Nevins, A. Peti-Stantić, B. Stanković, T. Šuligoj, and J.

Tušek. 2016. Conjunct Agreement and Gender in South Slavic: From Theory to Experiments

to Theory. Journal of Slavic Linguistics 24.1:187-224.