We acknowledge Gabriela Bîlbîie (lecturer at the University of Bucharest) for helping us to recruit participants in our four experiments and Labex EFL & Smarts Up (Idex Université Paris-Cité) for co-funding our work.
1Pronoun resolution, i.e. the question of how pronominal expressions retrieve their antecedents in discourse, is probably one of the most studied questions in (psycho)linguistic research (see for recent studies, Holler & Suckow, 2016; Arnold, 2010; Kehler & Rohde, 2019; Schulz et al., 2021; Colonna et al., 2012). Quantitative research has focused on language-specific and language-independent factors to identify what triggers cross-linguistic variation. Among the anaphoric phenomena studied, null and pronominal subject alternation in so-called pro-drop languages has attracted particular attention (Carminati, 2002; Chamorro, 2018; Contemori & Di Domenico, 2021; Torregrossa et al., 2020).
- 1 Even though a number of previous studies use the term “overt subject”, in this paper we make a dist (...)
2Null and pronominal subjects1 in Romance languages have shown a division of labour effect in corpus as well as experimental studies in that null subjects in an upcoming sentence or subclause such as [1a]-[1d] prefer a subject antecedent while pronominal subjects prefer a non-subject antecedent.
[1a] |
Flavia |
ha |
telefonato |
a |
Paola |
quando |
Ø/lei |
era |
in |
ufficio. |
|
Flavia |
aux.3sg |
call.pst |
to |
Paola |
when |
Ø/she |
was |
in |
office |
|
‘Flavia phoned Paola when (she) was in the office.’ |
|
(Italian; Carminati [2002: 378]) |
[1b] |
Juan |
vio |
a |
Pedro |
cuando |
Ø/él |
caminaba |
por |
la |
playa. |
|
Juan |
see.pst.3sg |
dom |
Pedro |
when |
Ø/he |
walk.pst |
by |
def |
beach |
|
‘Juan saw Pedro when (he) was walking on the beach.’ |
|
(Spanish; Fuente & Hemforth [2013: 29]) |
[1c] |
O |
atleta |
consultou |
o |
ortopedista |
quando |
Ø/ele |
|
def |
athlete |
consult.pst |
def |
orthopaedist |
when |
Ø/he |
|
regressou |
da |
viagem |
a |
Itália. |
|
return.pst |
from |
trip |
to |
Italy |
|
‘The athlete consulted the orthopaedist at the hospital when (he) returned from the trip to Italy.’ |
|
(Portuguese; Fernandes et al. [2018: 1990]) |
[1d] |
La |
Marta |
escrivia |
sovint |
a la |
Raquel |
Ø/ella |
vivia |
als |
|
def |
Marta |
write.pst |
frequently |
to |
Raquel. |
Ø/she |
live.pst |
at |
|
Estats |
Units. |
|
States |
United |
|
‘Marta wrote frequently to Raquel. (She) lived in the United States.’ |
|
(Catalan; Mayol [2010a: 128]) |
3One account of the division of labour observed in null/pronominal subject alternation is the Position of Antecedent Hypothesis (PAH – Carminati, 2002, 2005), predicting that: i) null subjects prefer a subject antecedent, whereas ii) pronominal subjects prefer a non-subject antecedent. This hypothesis is supported by evidence from several Romance languages: for Italian, Carminati (2002, 2005) showed a very strong preference of null subjects for subject antecedents (77%) within sentences (with temporal embedded clauses) while pronominal subjects were shown to prefer object antecedents (83%). A similar pattern was found for European Portuguese (Fernandes et al., 2018) with a 79% preference for subject antecedents retrieved by null subjects and 76% for object antecedents in the case of pronominal subjects.
4However, the division of labour effect for null and pronominal subjects can also be accounted for in terms of choice of anaphoric expressions (Givón, 1983). Givón’s Topic Continuity Theory suggests that the usage of referential expressions depends on the degree of topicality that a given entity has in discourse. According to the Accessibility Theory (Ariel, 1990, 1994), referential forms can be ordered on an explicitness scale, ranging from full names to zero expressions. The more accessible an entity is in discourse, the less complex its corresponding anaphoric expression will be. Even though we do not know all the factors that influence the degree of accessibility, syntactic and pragmatic factors seem to play a role. Arnold (2010) proposed two factors responsible for prominence in discourse, information structure and grammatical function: a topical referent will be more accessible than a non-topical one and the same applies to subject antecedents vs. non-subject antecedents.
5Most studies, including the ones presented in this paper, investigate pronoun resolution in SVO (subject-verb-object) sentences. Following the Given-before-New principle (e.g., Gundel, 1988), pre-verbal subjects are often regarded as default topics. Looking only at canonical SVO sentences makes it difficult to disentangle the impact of antecedent function and discourse status accounts for antecedent preferences in pro-drop languages. Thus, as claimed by Runner and Ibarra (2016), the null subject preferences for pre-verbal subject antecedents may be due to topichood in the absence of controlled studies manipulating information structure. Runner and Ibarra provide experimental results on Spanish showing that the topic status of the antecedent plays a role beyond its grammatical function. By comparing referents representing either given or new information, their study suggests that null subjects are more strongly preferred when retrieving given antecedents (i.e., pre-verbal subjects). New referents, on the other hand, seem to be referred to equally with null or pronominal subjects, with no significant differences.
6To be comparable to most of the existing literature, we will also look at canonical subject first sentences in Romanian in this paper. Since we will not be able to distinguish a preference for subjects or topics in our experiments, we will stick to the less theory-driven description of a division of labour effect in the following.
7Interestingly, there is some variation across Romance languages (as well as other languages allowing null subjects) with respect to the strength of the division of labour effect. The preference for a subject antecedent seems to be quite robust for null subjects across languages, despite some existing variation (see, for Spanish and Italian, Filiaci et al., 2013; for Spanish, Fuente & Hemforth, 2013; for European and Brazilian Portuguese, Fernandes et al., 2018; for Greek and Italian, Torregrossa et al., 2020). However, with respect to pronominal subjects, more variation has been found across languages. Spanish and Greek for example are much more flexible than Italian regarding the antecedent of a pronominal subject (Sorace & Filiaci, 2006; Sorace et al., 2009; Filiaci et al., 2013; Chamorro, 2018; Contemori et al., 2019; Contemori & Di Domenico, 2021; Torregrossa et al., 2020).
8Regarding Spanish (Filiaci et al., 2013; Fuente & Hemforth, 2013, among many others), the subject vs. object asymmetry is well established: null subjects prefer subject antecedents as in Italian or European Portuguese (66.32%), while pronominal subjects display a preference for object antecedents (63.19%). The two diatopic variants of Portuguese (European and Brazilian) show impressive variation with respect to the division of labour (Fernandes et al., 2018). In Portuguese, the authors found a general preference for a subject antecedent for the null subject of the subordinate clause (79% in European Portuguese vs. 74% in Brazilian Portuguese). However, there was a significant difference regarding the interpretation of the pronominal subject. In European Portuguese, the object was clearly the preferred antecedent of the pronominal subject of the subordinate clause whereas in Brazilian Portuguese there was no clear preference between the subject antecedent and the object antecedent. Preferences displayed in Catalan (Mayol, 2010a) show a particular pattern, i.e. the subject antecedent preference of null subjects was weaker (59.1%) than the object antecedent preference of pronominal subjects (64.8%).
9A potential problem with the crosslinguistic comparison of preferences in the existing literature is that different tendencies in the division of labour effect may appear due to differences in the experimental design. Different preferences between Romance languages might be triggered by language specific factors, which is what we are interested in this paper, but they could also be due to variation in the experimental material. Thus, the difference in the division of labour found in Catalan might for example be due to an important difference in experimental material, since Mayol (2010a) looked for a subject antecedent preference between simple sentences (and not within complex sentences as other experiments did). Differences in pronoun interpretation across and within sentences have been proposed from a theoretical perspective by Lasnik (1976), Karttunen (1971), Bach (1970), and confirmed by empirical evidence. As suggested by Miltsakaki (2002), anaphora within a discourse segment (two clauses related by a connective) behave differently from anaphora across two discourse segments (two separate sentences that are not related by a connective; see also Colonna et al. [2015] for similar evidence on German and French). Miltsakaki (2002) argues that grammatical function may have a larger impact when a pronoun refers to an antecedent in a previous sentence. However, within the same sentence semantic factors such as verb bias and discourse relations (triggered by the subordinating connective) might have a larger impact on pronoun resolution. The antecedent preferences shown for Spanish in Fuente and Hemforth (2013) in intrasentential contexts compared to those of Alonso-Ovalle et al. (2002) who investigated intersentential contexts may be explained by differences in their material. Both studies found a preference for a subject antecedent for null subjects while antecedent choices presented by pronominal subjects were less clear in Alonso-Ovalle et al. (2002).
Table 1 – Antecedent preference in Romance languages (in percentages)
Language |
Null subject |
Pronominal subject |
Subject ant. |
Object ant. |
Subject ant. |
Object ant. |
Spanish (Fuente & Hemforth, 2013) |
66 |
37 |
34 |
63 |
European Portuguese (Fernandes et al., 2018) |
79 |
21 |
24 |
76 |
Brazilian Portuguese (Fernandes et al., 2018) |
74 |
26 |
46 |
54 |
Italian (Carminati, 2002) |
77 |
23 |
17 |
83 |
Catalan (Mayol, 2010a) |
59 |
41 |
35 |
65 |
10Also, for Italian (Carminati, 2002), the experimental items were not completely uniform (see e.g., [2a]-[2d]). While for Spanish (Fuente & Hemforth, 2013) or Portuguese (Fernandes et al., 2018) preferences were tested in embedded temporal clauses introduced by when with a subject and direct object antecedents (proper names) in the root clause, the Italian experiments by Carminati (2002) exhibit considerable variation in the material, even though there were not any suggested item-level differences. In the given examples, the noun phrase (NP) introducing the second referent has different syntactic functions: [2a] (direct object), [2b] and [2d] (prepositional complement), [2c] (modifier).
[2a] |
Roberto |
ha |
insultato |
Ugo |
quando |
Ø/lui |
era |
ubriaco. |
|
Roberto |
aux.3sg |
insult.pst |
Ugo |
when |
Ø/he |
was |
drunk |
|
‘Roberto insulted Ugo when he was drunk.’ |
[2b] |
Maria |
fa |
lo |
shopping |
con |
Lucia |
quando |
Ø/lei |
è |
libera. |
|
Maria |
do.3sg |
def |
shopping |
with |
Lucia |
when |
Ø/she |
is |
free |
|
‘Maria goes shopping with Lucia when she is free.’ |
[2c] |
Marinella |
è |
diventata |
amica |
di |
Sandra |
dopo |
che |
Ø/lei |
|
Marinella |
aux.3sg |
become |
friend |
of |
Sandra |
after |
that |
Ø/she |
|
ha |
divorziato. |
|
aux.3sg |
divorced |
|
‘Marinella became the friend of Sandra after she divorced.’ |
[2d] |
Pietro |
non |
parla |
con |
Vito |
da |
quando |
Ø/lui |
|
Pietro |
neg |
talk.3sg.prs |
with |
Vito |
from |
when |
Ø/he |
|
ha |
dato |
le |
dimissioni |
|
aux.3sg |
give.pst |
def |
resignation |
|
‘Pietro hasn’t talked to Vito since he resigned.’ |
11Thus, to get a proper full picture of Romance languages with respect to the division of labour of null and pronominal subjects, one needs parallel experimental studies, i.e. the same grammatical function of antecedents and the same type of anaphora. In this paper, we kept our experimental designs on Romanian as close as possible to the constructions investigated for Spanish by Fuente and Hemforth (2013) as well as for European and Brazilian Portuguese in Fernandes et al. (2018). This is particularly useful because it will enable us to compare our data to a language with a weak to moderate division of labour effect such as Spanish as well as to a language with a more categorical effect such as European Portuguese. We hypothesize that Romanian follows a similar pattern to Spanish (Fuente & Hemforth, 2013).
- 2 We also included the modality here (written vs. spoken corpus).
12Based on the studies on European and Brazilian Portuguese by Fernandes et al. (2018), we hypothesize that the general use of null subjects in a language at least partly predicts the strength of the division of labour effect. Most Romance languages are pro-drop languages, with an overall preference for null subjects; however, the frequency of null subjects can vary considerably. Previous quantitative studies in the literature reveal the frequency of null subjects2 (corpus studies): 62% in Catalan (Casanova, 1999; oral corpus), 78% in European Portuguese (Duarte, 2000; written corpus), 44% in Brazilian Portuguese (Soares, 2017; oral corpus); preferences for null subjects (experimental studies): 77% in Italian (Torregrossa et al., 2020). We also took into consideration the type of modality (written vs. spoken corpus), since it seems to have an impact on the division of labour. In Romanian, pronominal subjects are less frequent than null subjects in the written modality (Istrate et al., 2023). This tendency may be due to the influence of prescriptive grammar in Romanian, recommending the use of pronominal subjects only to emphasize referents (Avram, 2001). However, overt pronouns seem to be more frequent in spoken corpora, which might be explained by a noisy channel approach (Gibson et al., 2013). Given the increased noise in oral communication, a rational speaker may choose to be more redundant to get the message across (i.e., to use more overt pronouns). However, the percentages mentioned may not be directly comparable because spoken and written corpora vary in more ways than just the mode of production. Controlled experimental studies would be useful to test this difference in more detail.
13The frequency of null pronouns across languages may vary for several reasons. One of the most striking examples is Brazilian Portuguese where the frequency of null subjects has been decreasing compared to other Romance languages (in particular compared to European Portuguese), which may be accounted for in terms of ongoing parametric change of the language (Duarte, 1993, 1995, 2000, inter alia). In this respect, Soares et al. (2020) suggested that Brazilian Portuguese is on a path from a null subject language toward a language with obligatory phonological expression of subjects (like French). One of the reasons for the decrease in null subjects in Brazilian Portuguese may be the impoverished inflection system (compared to other variants of Portuguese: Duarte, 1995). Soares et al. (2020) also found effects of animacy and specificity of antecedents for third person pronouns in Brazilian Portuguese corpus studies and controlled experiments. Istrate et al. (2023) showed that null subjects in Romanian are more frequent with discourse persons (first and second person) in corpus studies and that discourse relations play a role as well. Parallel corpus studies and experiments across Romance languages are needed to see whether these kinds of fine-grained parameters affect the distribution of null and pronominal subjects in different ways. Fernandes et al. (2018) tested the role of frequency more directly in an experimental context. They showed that increasing the number of null subjects in experimental materials increased the choice of non-subject antecedents for pronominal subjects so that Brazilian speakers’ preferences resembled those of European Portuguese speakers. Fernandes et al. argued that a Rational Speaker approach based on a Bayesian update model can explain the changes in preferences (see also Schulz et al., 2021, for a similar approach comparing antecedent preferences in French and English): Fernandes et al.’s data suggest that the more accessible a null subject is in the language or even in the local environment to refer back to a subject antecedent, the more a pronominal subject will be interpreted to refer to a non-subject antecedent. Since only subject and object antecedents were investigated in Fernandes et al. as in most studies in the literature, we can of course not exclude that other factors affecting topicality or saliency (Landragin, 2020) may play a role.
14Before presenting our experimental data, we first provide a short characterization of Romanian: typologically Romanian is a consistent-null subject language (Biberauer et al., 2009), i.e. the presence of null subjects is allowed irrespective of person, number or verb tense as in [3a]-[3c]. Pro-drop subjects retrieve their antecedents either morphologically (by verbal inflection) or in discourse (by deixis or anaphora).
[3a] |
Va |
alerga |
cinci |
kilometri. |
|
aux.fut.3sg |
run |
five |
kilometres |
|
‘He/She will run five kilometres.’ |
[3b] |
Ai |
obținut |
premiul |
întâi |
la |
concurs. |
|
aux.2sg |
obtain.pst |
prize.def |
first |
at |
competition |
|
‘You were awarded the first prize at the competition.’ |
[3c] |
Luăm |
micul dejun |
la |
ora |
nouă. |
|
take.prs.1pl |
breakfast |
at |
hour |
nine |
|
‘We have breakfast at nine o’clock.’ |
15As in the other pro-drop Romance languages discussed here (Italian, Spanish, Catalan, Portuguese), null subjects in Romanian most often alternate with overt pronominal subjects. The paradigm of subject pronouns in Romanian expresses number and person. Moreover, in the third person, in both singular and plural, there are distinctive forms depending on gender (see Table 2).
Table 2 – Paradigm of personal pronouns in Romanian
Personal pronouns in Romanian |
Person |
Masculine |
Feminine |
1sg |
eu “I” |
2sg |
tu “you” |
3sg |
el “he” |
ea “she” |
1pl |
noi “we” |
2pl |
voi “you” |
3pl |
ei “they” |
ele “they” |
16Even though subject alternation has not been systematically studied in SVO sentences with corpus or experimental data, there are several previous studies, both theoretical and empirical, about pronouns in Romanian. Cotfas (2012) showed that, in contexts like [4] with a local and a non-local subject as possible antecedents, null subjects prefer the local subject as their antecedent while overt pronominal subjects prefer the non-local subject. The null subject of the subjunctive “celebrate” is likely to retrieve the more recent referent “profesorul”, while the pronominal subject prefers the non-local referent “Dragoș”. Mihăilă et al. (2011), on the other hand, investigating zero pronouns in Romanian in a corpus study, found a less consistent effect of distance. In particular for newswire and literature, authors adapt the distances between null pronouns and their referents depending on writing style (i.e., to avoid possible misinterpretations, or to increase the fluency of narrative sequences).
[4] |
Dragoș |
și |
-a |
luat |
licența |
și |
|
Dragoș |
refl |
-aux.3sg |
passed.pst |
graduation exam.def |
and |
|
profesorul |
său |
vrea |
să |
sărbătorească |
reușita. |
|
teacher |
his |
wants |
sbj |
celebrate.prs |
success.def |
|
‘Dragoș passed his graduation exam and his teacher wants to celebrate the success.’ |
A. |
Profesorul |
vrea |
ca |
Dragoș |
să |
sărbătorească |
reușita. |
|
teacher.def |
wants |
that |
Dragoș |
sbj |
celebrate.3sg |
success.def |
|
‘The teacher wants Dragoș to celebrate the success.’ |
B. |
Profesorul |
vrea |
să |
sărbătorească |
el |
însuși |
reușita |
|
teacher.def |
wants |
sbj |
celebrate.3sg |
he |
himself |
success.def |
|
lui |
Dragoș. |
|
of |
Dragoș |
|
‘The teacher himself wants to celebrate the success of Dragoș.’ |
17Ivan (2020) discussed the frequency of pronouns, suggesting a division of labour between personal pronouns and reflexives. In her experiments, she showed that pronouns are generally used in unambiguous environments while emphatic reflexives are preferred in ambiguous contexts. Thus, using an emphatic reflexive even though the given contexts provide enough information about the potential antecedent is considered infelicitous.
18We will investigate three questions in this paper. The first concerns the general use of null subjects in Romanian, the second considers the role of contrast and the last one looks at the strength of the division of labour effect. Our hypothesis with respect to the general use of null subjects is that, like other Romance pro-drop languages, Romanian displays a general null subject preference for subject antecedents.
19However, even subject antecedents can be referred to by an overt pronominal subject. We assume that a variety of factors can influence the choice of a referential expression (null vs. pronominal subjects in our case). The contrast hypothesis (Dobrovie-Sorin & Giurgea, 2013; see also Mayol, 2010b, for a more general perspective; Amaral & Schwenter, 2005) may be one of the factors that allow for the presence of overt pronominal subjects in Romanian (and other pro-drop languages), i.e. the referent is contrasted with other entities. Example [5] shows the use of a contrastive eu in the second sentence: I in contrast to Ion believe that she will come. Based on this hypothesis about the role of contrast, we will propose several factors such as polarity, alternative referents and focus particles in our first three experiments under the assumption that contrast increases the choice of a pronominal subject.
[5] |
Ion |
mi- |
a |
spus |
că |
Maria |
nu |
vine |
mâine. |
|
Ion |
cl.dat.1sg |
has |
told |
that |
Maria |
neg |
comes |
tomorrow |
|
Eu |
cred |
că |
va |
veni. |
|
I |
believe.1sg.prs |
that |
will.3sg |
come |
|
‘Ion told me that Maria would not come tomorrow. I think she will come.’ |
|
(Dobrovie-Sorin & Giurgea [2013: 249]) |
- 3 We wish to thank one of the reviewers for suggesting rerunning the experiment consistently with sim (...)
20In Section 2 of this paper, we test the general preference for null and pronominal subjects, including the factors related to contrast. In our first experiment, we tested the role of polarity in question/answer dialogues assuming that negation enhances the contrastiveness of the subject and thus the choice of pronominal subjects. In a second experiment, we restricted answers to simple direct answers to the question but added contexts with one or two referents under the assumption that having an alternative referent in the context will enhance the proportion of pronominal antecedent choices3. A third experiment looked at the effect of the focus particle only that presupposes a set of alternatives not necessarily explicitly mentioned in the context (Rooth, 1992; Kaiser, 2010; Spalek & Zeldes, 2017) thus making the choice of a pronominal subject more feasible.
21Following Fernandes et al. (2018) and Schulz et al. (2021), we expect that the division of labour effect is influenced by the null subject frequency in the language. The more frequent or preferred a null subject is, the more categorical antecedent preferences of both null and pronominal subjects are. The second goal of the current paper is, accordingly, to add pronoun resolution preferences in Romanian to the general picture, a language where, to our knowledge, the question of subject alternation and its role in antecedent preferences has not yet been studied in controlled experiments. Fernandes et al. (2018) argue that the more frequently and consistently null subjects are used in the language, the more pronominal subjects will prefer non-subject antecedents. Depending on the results of the experiments in the first section in this paper, we can make predictions on the strength of the division of labour effect for the second experiment.
- 4 Given the presented stimuli, we acknowledge that there is little control over what prosody particip (...)
22In our first experiment, we focused on the null subject/pronominal alternation in Romanian. We are interested in two questions in this experiment. As discussed above, the use of null subjects varies across Romance languages. One way to determine the baseline preference for null subjects is detailed corpus analyses. In this paper, we take an experimental approach to tackle this question. The second question concerns the role of contrast, as already mentioned (Dobrovie-Sorin & Giurgea, 2013). We therefore took into account the polarity of the sentence assuming that negation may play a role in a contrastive reading of the subject. Our hypothesis is that the subject is more contrastive in a negative sentence. In Example [6], the negation may trigger the assumption that the mentioned individual represents one of several possible alternatives in the communication situation so that he did not go to the show while others possibly did4.
[6a] |
Poate |
Paul |
să |
ne |
însoțească |
la |
spectacol? |
|
May |
Paul |
that |
cl.1pl.acc |
accompany |
at |
show |
|
‘May Paul come with us to the show?’ |
[6b] |
Nu, |
el |
nu |
ne |
poate |
însoți |
la |
spectacol. |
|
No |
he |
neg |
cl.1pl.acc |
may |
accompany |
at |
show |
|
‘No, he may not come with us to the show.’ |
23In the following experiment, we presented participants with polar questions and affirmative or negative answers. The dependent variable was a forced choice between two sentences containing a null subject or a pronominal subject. The experiment was constructed according to a one-factor only design, i.e. two-level polarity {affirmative, negative}. The time to complete the questionnaire was less than 15 minutes.
24Participants: Forty-nine participants (6 males and 43 females, mean age = 21) completed our questionnaire. They were all native speakers of Romanian, either students at the University of Bucharest, attending classes at the Faculty of Foreign Languages and Literatures, or people recruited on social media. Participation was voluntary with no obligations imposed during the classes. For the statistical analysis of the data, we excluded only one participant who did not reach the 75% threshold of correct answers to the comprehension questions.
25Materials: We constructed 20 experimental items. To control for possible gender effects, we proposed an equal distribution between sentences containing male and female referents. In our experimental study, only 3rd person human referents are included. Each experimental item consists of a closed question and its corresponding response. The first condition contains an affirmative polarity answer, while the second condition contains a negative polarity answer, as illustrated in [7].
[7] |
Question: |
|
Participă |
Ioana |
la |
concursuri? |
|
participate.prs.3sg |
Ioana |
at |
competitions |
|
‘Does Ioana take part in the contests?’ |
|
Condition (1): affirmative polarity |
|
Da, |
Ø/ea |
participă |
la |
concursuri. |
|
Yes |
Ø/ea |
participate.prs.3sg |
at |
competitions |
|
‘Yes, she takes part in the contests.’ |
|
Condition (2): negative polarity |
|
Nu, |
Ø/ea |
nu |
participă |
la |
niciun |
concurs. |
|
No |
Ø/ea |
neg |
participate.prs.3sg |
at |
no |
competition |
|
‘No, she takes part in no contest.’ |
|
Comprehension question |
|
Ioana |
a |
văzut |
-o |
pe |
Alexandra |
când |
s- |
|
Ioana |
aux.3sg |
see.pst |
cl.3sg.f.acc |
dom |
Alexandra |
when |
refl |
|
a |
urcat |
în |
autobuz. |
Pe |
cine |
a |
văzut |
Ioana? |
|
aux.3sg |
get.pst |
in |
bus |
dom |
who |
aux.3sg |
see.pst |
Ioana? |
|
‘Ioana saw Alexandra when she got on the bus. Whom saw Ioana?’ |
26In Figure 1, we show an example of an experimental item that participants see on the Ibex Farm platform while answering the questionnaire.
Figure 1 – Screen sample: binary forced-choice task (Experiment 1)
27Half of the experimental items were followed by a comprehension question (yes/no) to make sure that participants were reading the question-answer pairs for comprehension, see Example [7]. In addition to the items that focus on subject alternation, we also included items from an independent experiment that serve as distractors to prevent participants from developing a response strategy.
28The items were presented according to the Latin square design in a random order, with each participant viewing each item in only one of its two conditions. Participants were unable to revisit and alter their choices while completing the questionnaire.
29Procedure: The following procedure was applied to all three experiments in this section. The binary choice task was created online on the Ibex Farm platform at Université Paris Cité, created by Alex Drummond and maintained by Achille Falaise. Each experiment began with instructions for the task as well as a series of demographic questions (age, gender, first language, i.e. language spoken since early childhood). Participants gave their informed consent to the use of their anonymized data for research purposes.
30We ran Bayesian models using the brms package (Bürkner, 2018; Bürkner & Charpentier, 2020), the R version 4.3.2 (R Core Team, 2023) and the Rstudio interface (version 2023_12_1 – Posit team, 2023) for this as well as the following experiments. All our dependent variables are binomial. We therefore fitted Bayesian binomial regression models to our data using the Bernoulli family. The models were run with 4 MCMC chains and 9,000 iterations per chain. We will report results where the 95% credibility interval (CrI) does not include 0 as strong evidence and effects with a probability of 0.90 to 0.95 as weak evidence.
31The Bayesian binomial regression model included polarity as the fixed factor as well as random effects for participants and items and random slopes for polarity. Polarity was mean centre coded with “affirmative” approaching -0.5 and “negative” 0.5. We found strong evidence for a preference of null subjects across conditions (coef.est = 1.27, CrI = [0.35, 2.21], P(β > 0) = 0.996). There was also weak evidence for an effect of polarity (coef.est = -0.394, CrI = [-0.872, 0.110], P(β < 0) = 0.944, see Figure 2). While null subjects remained the most frequent choice, participants chose the pronominal subject more often in the context of negation.
Figure 2 – Results Experiment 1
32Our experimental study showed a clear preference of Romanian speakers towards null subjects with 70% null subjects in the case of affirmative polarity and 65% in the case of negative polarity. A further (weak) effect is that of negative polarity interacting with subject type, slightly increasing the choice of a pronominal subject. This effect is compatible with the assumed effect of contrast. The pronominal subject seems to be interpreted as a negated alternative to the question serving as a context, confirming the initial hypothesis that pronominal subjects become more acceptable as a choice in a contrastive context.
33It should however be mentioned that the answers proposed in our items are quite heterogeneous. On the one hand, some of the experimental items present a propositional negation, followed by the antecedent verb in the negative form. In other words, there the negative particle nu appears twice, once as a prophrase adverb (discourse particle) and another time as a verbal affix (propositional negation): Poate Paul să ne însoțească la spectacol? (“Can Paul come with us to the show?”) and Nu, (el) nu ne poate însoți la spectacol (“No, he cannot accompany us to the show”). On the other hand, some items with negative polarity include the particle nu, followed by an affirmative sentence, which has a different lexical content from the previous question: S-a specializat Paul în lingvistică? (“Did Paul specialize in linguistics?”) and Nu, el s-a specializat în biologie (“No, he specialized in biology”). This means that some of our answers provide a direct answer to the “question under discussion (QUD)” (Roberts, 2012) while others go beyond a simple answer and even provide a direct contrast. In the first experiment, we did not find any significant differences between the two types of answer. The following two experiments aimed at disentangling the effects of polarity and contrast on referential choices.
- 5 Thanks to our reviewers who pointed out this aspect.
34In a second experiment we aimed to test the role of contrast regarding null and pronominal subject preferences more directly. Our assumption in the first experiment was that negation enhances a contrastive reading. A more direct way to evoke contrast is to add other referents in the context. Given the heterogeneity of the items in the first experiment5, we built pairs of questions and corresponding answers that directly answered the question without adding more information (both in positive and negative polarity). We examined the following research questions:
- Does negative polarity increase the preference for pronominal subjects, by inducing a contrastive reading? (questions/answers being minimally different, by the presence of “yes” or “no”)
- Do alternative referents in the context play a contrastive role, thus increasing the preference for pronominal subjects?
35Our hypothesis is that negation may lead to a slight preference for pronominal subjects (as in our first experiment), possibly because of an implicit contrast in the context. Moreover, by including a second referent beside the main one, we expect a clear preference for pronominal subjects in items where we use more than one referent.
36Participants: The participants group consisted of 68 Romanian native speakers (51 females and 17 males, mean age = 27). The recruitment procedure was identical to the first experiment (see 2.1.1).
37Materials: The experiment was constructed with a 2 by 2 design, i.e. polarity with two levels (affirmative, negative) and context alternatives (one referent, two referents). The dependent variable was “choice of subject” with “null subject” and “pronominal subject” as possible choices. This gave rise to four conditions:
Context:
[8a] |
Ioana |
este |
elevă |
de excepție. (one referent) |
|
Ioana |
is |
student |
exceptional |
|
‘Ioana is an outstanding student.’ |
[8b] |
Alexandra |
și |
Ioana |
sunt |
eleve |
de excepție. (two referents) |
|
Alexandra |
and |
Ioana |
are |
student.pl |
exceptional |
|
‘Alexandra and Ioana are outstanding students.’ |
Question:
[9] |
Participă |
Ioana |
la |
concursul |
de |
matematică? |
|
participate.prs.3sg |
Ioana |
at |
competition.def |
of |
maths |
|
‘Is Ioana taking part in the maths contest?’ |
Choice for negative answer:
[10a] |
Nu, |
nu |
participă |
la |
concursul |
de |
matematică. |
|
No |
not |
participate.prs.3sg |
at |
competition.def |
of |
maths |
|
‘No, she is not taking part in the maths contest.’ |
[10b] |
Nu, |
ea |
nu |
participă |
la |
concursul |
de |
matematică. |
|
No |
she |
not |
participate.prs.3sg |
at |
competition.def |
of |
maths |
|
‘No, she is not taking part in the maths contest.’ |
Choice for affirmative answer:
[11a] |
Da, |
participă |
la |
concursul |
de |
matematică. |
|
Yes |
participate.prs.3sg |
at |
competition.def |
of |
maths |
|
‘Yes, she is taking part in the maths contest.’ |
[11b] |
Da, |
ea |
participă |
la |
concursul |
de |
matematică. |
|
Yes |
she |
participate.prs.3sg |
at |
competition.def |
of |
maths |
|
‘Yes, she is taking part in the maths contest.’ |
38Each affirmative or negative answer pair ([10], [11]) was preceded by a context sentence such as [8a] or [8b] and a question regarding the mentioned referent or, respectively, one of the two mentioned referents (Example [9]). We presented 20 such items, controlling the gender of the antecedents (half masculine and half feminine) and the tense of the verb (half in the present and half in the compound past of the indicative mood). Since SVO order is the most frequent in declarative unmarked sentences (Pană Dindelegan & Maiden, 2013) and we presented written stimuli to participants (thus no intonation cue), we decided to construct questions with post-verbal subjects.
39We also added 20 items coming from a parallel experiment testing the role of focus particles for overt and null pronoun choice in Romanian. Additionally, we included 7 fillers to avoid participants developing a response strategy while completing the task.
Figure 3 – Screenshot: binary forced-choice task (Experiment 2)
40Procedure: Same as in Experiment 1.
41Due to a coding error, two of the items had to be excluded from the analyses. The Bayesian binomial regression model included polarity and context as fixed factors as well as random effects for participants and items and random slopes for polarity and context. We found strong evidence for a preference of null subjects across conditions (coef.est = 1.286, CrI = [0.651, 1.933], P(β > 0) = 1; see Figure 4). We did, however, not find evidence for an effect of polarity (coef.est = 0.122, CrI = [-0.440, 0.709], P(β > 0) = 0.66). On the other hand, the prediction that the presence of an alternative referent in the context plays a role in the choice of the referential expression was confirmed. There was strong evidence for an effect of context in that choices of pronominal subjects in the two-antecedent condition were more frequent than in the one-antecedent condition (coef.est = 0.644, CrI = [-1.243, -0.080], P(β < 0) = 0.99).
42Comparing these results to those shown in Experiment 1, we could not confirm the polarity effect found in the previous study. In negative sentences, pronominal subjects seem to be as frequent as in affirmative sentences. However, adding a second referent (“Ioana” vs. “Alexandra and Ioana”) increased the number of choices of pronominal subjects both in positive and negative polarity sentences. Thus, we suggest that a contrastive reading might be triggered by including more referents in the discourse, confirming that pronominal subjects can be used to mark contrast (Pešková, 2014).
Figure 4 – Results Experiment 2
43The third experiment was conducted with the aim of refining the role of contrast in the division of labour between null and pronominal subjects, using the focus particle “only” that presupposes a set of alternatives. No explicit alternatives were mentioned but we assumed that the presupposition of a set of alternatives may increase the choice of pronominal subjects over null subjects.
44Participants: All details concerning the recruitment process of participants in the third experiments are found in Section 2.2.1. Items for this experiment were used as fillers in Experiment 2.
45Materials: In this experiment, the only factor varied was “Focus”, with each item constructed as an SVO question with either no focus particle or the particle doar or numai (“only”). As in the previous experiments, participants had to choose between answers with null or pronominal subjects. We constructed questions with pre-verbal subjects to ensure the homogeneity of items (the ones without and with focus particle), being preceded by the same context sentence across conditions (Example [12]). Different from Experiment 2 (in order to properly serve as distractors), we replaced the simple “yes/no” answers with more indirect answers that did not directly answer the QUD.
Context:
[12] |
Prietenele |
mele |
cumpără |
mereu |
cărți |
de pe |
|
Friend.f.pl.def |
poss.1sg |
buy.prs.3pl |
always |
books |
on |
|
acest |
site. |
|
this |
site |
|
‘My friends always buy books on this website.’ |
Question:
[13a] |
Ramona |
beneficiază |
de |
o |
reducere? |
|
Ramona |
benefit.prs.3sg |
from |
indf |
discount |
|
‘Does Ramona receive a discount?’ (no focus particle) |
[13b] |
Numai |
Ramona |
beneficiază |
de |
o |
reducere? |
|
Only |
Ramona |
benefit.prs.3sg |
from |
indf |
discount |
|
‘Only Ramona receives a discount?’ (with focus particle) |
Answer:
[14] |
Ø/ea |
plătește |
mai puțin |
decât |
de obicei. |
|
Ø/she |
pay.prs.3sg |
less |
than |
usual |
|
‘She pays less than usual.’ |
46Similar to Experiment 2, we controlled the gender of referents (half masculine and half feminine) and the tense of the verb (half present and half compound past of the indicative mood).
Figure 5 – Screen sample: binary forced-choice task (Experiment 3)
47Procedure: The procedure was the same as in Experiment 2.
48The Bayesian binomial regression model included focus as fixed factor as well as random effects for participants and items and random slopes for focus. As a general tendency, we found a stronger preference for pronominal subjects in both conditions, with (67.9%) and without focus particle (57.8%), compared to null subjects (coef.est = -0.736, CrI = [-1.256, -0.226], P(β < 0) = 0.997). Moreover, in the condition containing the focus particle, pronominal subjects were more frequent than in the condition without the focus particle doar or numai (coef.est = -0.626, CrI = [-0.974, -0.283], P(β < 0) = 0.999; see Figure 6).
Figure 6 – Results Experiment 3
49The first two experiments in this section provide evidence for an overall preference for null subjects in Romanian. We found this tendency in “yes/no” items where the question-answer pairs used the same lexical material (Experiment 2) or closely related lexical material (Experiment 1). The first experiment, containing no explicit context before the target question, revealed a slight increase in pronominal subject choices in negative sentences. However, in Experiment 2 we found no robust effect of polarity when a preceding context was added and we used a direct answer to the QUD using the same words, the only difference being the alternation between “yes” and “no”. The effect of contrast in Experiment 2 seemed to be induced by the presence of alternatives in discourse rather than polarity. For both affirmative and negative answers, participants chose more pronominal subjects when they had an alternative referent in the context (i.e. two proper names) even though the question only concerned one of the referents.
50Furthermore, a similar preference for pronominal subjects was observed with presupposed alternatives containing focus particles numai and doar (“only”). When the given referent in the context was preceded by the focus particle, it was retrieved more often by pronominal subjects than in SVO questions without a focus particle. This is compatible with our hypothesis that the presupposition of a set of alternatives increases contrastiveness. However, somewhat unexpectedly, we also observed a general preference for pronominal subjects across the two conditions in Experiment 3. We suggest that this tendency might be due to the construction of items, as the proposed answers to the given questions went beyond a simple response. An answer such as “She paid less than usual” is not a direct answer to the QUD “Did only Ramona receive a discount?”. It may be the case that there is a preference for a null subject in the case of a simple direct answer but a preference for a pronominal subject for an answer that goes beyond the QUD. More experimental studies are necessary to test this hypothesis directly. Interestingly, a very similar “division of labour” was found by Miller and Hemforth (2014) for English verb phrase ellipsis (“do”) and verb phrase anaphora (“do it”, Miller & Pullum, 2022), where direct answers to the QUD preferred the shorter form “do” whereas answers that go beyond the QUD preferred “do it”.
51To sum up the experiments of this section, we find a general preference for null subjects in two out of three experiments that is, however, modulated by the explicit or presupposed presence of alternatives as well as possibly by the relation of the answer to the QUD.
52Based on the results in Section 2 of this paper, we suggest as the baseline for null subject preferences in Romanian the (3rd person) null subject affirmative answers (70%) in our first experiment since this is the simplest case of all our experiments. This allows us to place the status of the pro-drop in Romanian in a broader framework, if we compare it to experimental results obtained in other Romance languages. Compared to the frequency of (3rd person) pro-drop in Italian (77%, see Torregrossa et al., 2020) and Catalan experiments (62%, see Casanova, 1999), the status of null subject in Romanian occupies an intermediate position in relation to the two other languages. In contexts like those used in our experiment, Contemori and Di Domenico (2021) found that Italian speakers produced null subjects in 87% of the cases while this was only the case in 66% of the cases for Mexican Spanish speakers. Romanian speakers appear to be more aligned with Spanish speakers with respect to their choices.
53Following Fernandes et al. (2018) as well as Contemori and Di Domenico (2021), we assume that the overall frequency of null subjects is closely related to gradual variation in division of labour effects across Romance languages: the higher the percentage of the null subject baseline, the stronger the preference to associate a pronominal subject with a non-subject antecedent.
54The aim of our fourth experiment was to observe how Romanian speakers interpret a null or pronominal subject taking into consideration the syntactic function of the antecedent (subject or object). We examined the following research questions:
- Is there a complementarity in how null and pronominal subjects retrieve their antecedents depending on the antecedent function (as suggested by Carminati, 2002)?
- Does Romanian display a strong division of labour, like Italian (Carminati, 2002), even though the frequency of null subjects seems to be slightly smaller?
55Our hypothesis is that a division of labour effect should be found in Romanian as in other Romance languages with null subjects. Given the slightly lower frequency of null subjects in Romanian, the division of labour effect may be less categorical than for example in Italian.
56Participants: The participants group consisted of 67 native Romanian speakers (35 females and 27 males, mean age = 24). Other additional details concerning the recruitment of participants are given in Section 2.1.1.
57Materials: The experiment was constructed with a one-factor design, i.e. subject type with two levels {null, pronominal}. In addition to our experimental items, we also included items from an independent experiment (testing the semantic notion of avertivity – counterfactual implicature – in progressive verbal periphrases), to avoid a potential response strategy of participants. In addition, half of our experimental items included comprehension questions, to check participants’ attention during the task. The average time to complete the questionnaire was less than 15 minutes.
58To test the division of labour effect, we constructed 20 experimental items containing a complex sentence and a question about the antecedent of the null or pronominal subject. The complex sentence consisted of a root sentence containing a transitive verb with human subjects and objects (proper nouns of the same gender) and a temporal subordinate clause containing either a null subject or a pronominal subject. Participants were asked to choose between the two options of the person who performed the action referred to in the subordinate. In all 20 items, the temporal subordinates were introduced by când (“when”). Half of the items contained male antecedents and half contained female antecedents. We controlled the 20 experimental items for subject or object bias of the verb and for verb mood and tense (10 items contained verbs in the compound past tense while 10 items contained verbs in the imperfect tense). This experimental design gives rise to two conditions as illustrated in [15a]-[15b], followed each time by a partial question in cine (“who”) about the subject referent of the subordinate and a choice between two answers corresponding to the possible antecedents (see Figure 7).
Condition 1: null subject
[15a] |
Albert |
l |
-a |
vizitat |
pe |
Sebastian |
când |
|
Albert |
cl.3sg.m.acc |
aux.3sg |
visit.pst |
dom |
Sebastian |
when |
|
era |
student. |
|
was |
student |
|
‘Albert visited Sebastian when he was a student.’ |
Condition 2: pronominal subject
[15b] |
Albert |
l |
-a |
vizitat |
pe |
Sebastian |
când |
|
Albert |
cl.3sg.m.acc |
aux.3sg |
visit.pst |
dom |
Sebastian |
when |
|
el |
era |
student. |
|
he |
was |
student |
|
‘Albert visited Sebastian when he was a student.’ |
59Figure 7 shows a screen sample of an experimental item that participants saw on the Ibex Farm platform when answering the questionnaire.
Figure 7 – Screen sample: binary forced-choice task (Experiment 4)
60Half of the experimental items were followed by a full comprehension question (see Example [16]) to check that participants were reading the question-answer pairs for comprehension.
[16] |
Dragoș |
l |
-a |
ajutat |
pe |
Răzvan |
când |
|
Dragoș |
cl.3sg.m.acc |
-aux.3sg |
help.pst |
dom |
Răzvan |
when |
|
a |
locuit |
în |
Cluj. |
|
aux.3sg |
live.pst |
in |
Cluj |
|
‘Dragoș helped Răzvan when he lived in Cluj.’ |
|
Question: |
|
Ce |
a |
făcut |
Dragoș? |
|
What |
aux.3sg |
do.pst |
Dragoș? |
|
‘What did Dragoș do?’ |
|
Answer (a): |
|
Dragoș |
l |
-a |
ajutat |
pe |
Răzvan. |
|
Dragoș |
cl.3sg.m.acc |
-aux.3sg |
help.pst |
dom |
Răzvan. |
|
‘Dragoș helped Răzvan.’ |
|
Answer (b): |
|
Dragoș |
l |
-a |
abandonat |
pe |
Răzvan |
|
Dragoș |
cl.3sg.m.acc |
-aux.3sg |
quit.pst |
dom |
Răzvan |
|
‘Dragoș left Răzvan.’ |
61Procedure: All binary forced-choice experiments described in this paper followed the same procedure. The only difference in this experiment is that participants had to choose between the two antecedents instead of null or pronominal subjects.
62Similar to other Romance pro-drop languages, we found that the null subject of the subordinate is more often associated with the subject of the root sentence (64.2%), while the pronominal subject is more often interpreted as referring to the object of the root (58.2%), as illustrated in Figure 8.
Figure 8 – Results Experiment 4
63The Bayesian binomial regression model included subject type as a fixed factor as well as random effects for participants and items and random slopes for subject type. We found strong evidence for a difference in preferences in antecedent choice for null and pronominal subjects (coef.est = -0.616, CrI = [-0.945, -0.230], P(β < 0) = 1).
64Crucially, the results show that Romanian preferences for null and pronominal subjects are in general similar to other Romance pro-drop languages. We found a preference for the subject antecedent of null subjects and a preference for a non-subject antecedent of pronominal subjects. However, we should note that overt pronominal subject preferences seem to be less strong than what we observe for null subjects in the intrasentential constructions we used. We will discuss similarities and differences to other Romance languages in more detail in the following section.
65To compare our results from Romanian to other Romance languages, we repeat Table 1 as Table 3, including the Romanian data. The results we found for Romanian are very similar to the Spanish preferences but also to those found for Catalan preferences. At the same time, they are much less categorical than the preferences found for Italian and European Portuguese, while the division of labour effect seems stronger than in Brazilian Portuguese. Given that the constructions we used for our experiments were very similar to those used in the cited Spanish and Portuguese experiments, comparisons are particularly relevant in these cases. Similarities and differences in comparison to Catalan and Italian might be due to the use of intersentential anaphora resolution in the case of Catalan or more variation in the constructions for Italian.
66From a very general perspective, we deduce that the division of labour effect is close to categorical in Italian and European Portuguese, since preferences were found to be stronger in these Romance languages. In most of the Romance languages (Romanian, Spanish, Catalan, Brazilian Portuguese), we can observe a strong interaction between subject type and antecedent function although less strong than in Italian and European Portuguese (see Table 3). In Brazilian Portuguese (Fernandes et al., 2018), object antecedents seem to be referred to by a null or a pronominal subject, with no clear preference.
67Potential factors accounting for these crosslinguistic differences, in particular for Italian and Spanish, may concern discourse-pragmatic factors. Contemori and Di Domenico (2021) showed that Italian speakers use more null subjects with implicit causality verbs to refer either to a subject or an object antecedent. The authors found a less strong tendency for Spanish. Moreover, Filiaci et al. (2013) suggested that differences observed in the division of labour between null and pronominal subjects in Italian and Spanish might be due to differences in the architecture of pronominal systems. While Italian displays two types of overt subject pronouns (strong and weak), Spanish only has strong pronominal forms, which may account for a less categorical division of labour in Spanish.
Table 3 – Antecedent preference in Romance languages (in percentages)
Language |
Null subject |
Pronominal subject |
Subject ant. |
Object ant. |
Subject ant. |
Object ant. |
Spanish (Fuente & Hemforth, 2013) |
66 |
37 |
34 |
63 |
European Portuguese (Fernandes et al., 2018) |
79 |
21 |
24 |
76 |
Brazilian Portuguese (Fernandes et al., 2018) |
74 |
26 |
46 |
54 |
Romanian |
64 |
36 |
42 |
58 |
Italian (Carminati, 2002) |
77 |
23 |
17 |
83 |
Catalan (Mayol, 2010a) |
59 |
41 |
35 |
65 |
68Overall, the expected pattern for the division of labour effect was found for Romanian. Null subjects were shown to retrieve subject antecedents while pronominal subjects seemed to prefer a non-subject antecedent (i.e. direct objects in our experiment). These results are generally compatible with the PAH (Carminati, 2002) but also with a preference of the null subject to choose the most salient antecedent and the pronominal subject to choose a less prominent one (Runner & Ibarra, 2016).
69However, the experimental and corpus studies mentioned in this paper are not entirely comparable due to the absence of fully parallel materials. Since there were no previous experimental studies on Romanian subject alternation and their antecedent preferences, our experiments add evidence to the general picture of null and overt pronouns in comprehension. Our main goal in future studies will be to investigate crosslinguistic variation between Romance languages (such as Romanian, Spanish and Italian) more directly by using fully parallel experimental materials and providing data from parallel corpora. This should enable us to carry out a more accurate comparison in the division of labour between null and pronominal subjects. This is particularly necessary in order to shed more light on the role of the main factors proposed in the literature (discourse factors such as discourse prominence or syntactic factors such as antecedent function).
70From a Romance perspective, we have shown that Romanian preferences are very similar to those found for European Spanish (Fuente & Hemforth, 2013), but also Mexican Spanish (Contemori & Di Domenico, 2021: 62% subject choices for null subjects, 58% object choices for pronominal subjects). In Italian (Carminati, 2002, 2005) and European Portuguese (Fernandes et al., 2018), both null and pronominal subject preferences are close to categorical and, thus, stronger than those found for Spanish and Romanian.
71The results from the first three experiments provide empirical evidence in support of the hypothesis that pronominal subjects convey contrast whereas null subjects are much more frequent in general in pro-drop languages, such as Romanian. Our first experiment confirmed the null subject preference in Romanian, null subjects being strongly preferred (70.2%) compared to pronominal ones (29.8%). Even though null subjects were chosen in both positive and negative contexts, we observed a slightly higher preference for pronominal subjects in negative polarity. Nevertheless, we were unable to confirm the polarity effect in our second experiment, suggesting that contrast is mostly induced by the presence of alternative referents in the context, which triggers a higher preference for pronominal subjects in answers to a direct “yes” or “no” question. In addition, we have shown a robust effect of contrast using focus particles (“only”), observing similar tendencies regarding pronominal subjects. Thus, these results are also compatible with the hypothesis of contrast (Dobrovie-Sorin & Giurgea, 2013; Pešková, 2014).
72Coming back to the comparison with other Romance languages, the preference for null subjects seems to be less strong than in Italian (77%) and European Portuguese (78%) corpus studies. Also, compared to the production study in Contemori and Di Domenico (2021), the results of our Experiment 1 are closer to those found for Mexican Spanish speakers in one referent context (66%) than for Italian speakers (87%). Some reasons for these crosslinguistic differences may be differences in inflectional paradigms but possibly also differences in sensitivity to semantic or pragmatic factors across languages. We are planning parallel crosslinguistic studies to shed light on these questions.
73Putting these results together, they are compatible with our hypothesis that the frequency of null subjects in a language correlates with the strength of the division of labour effect. The frequency of alternative constructions has been shown to play a role in the interpretation of ambiguous constructions including null subjects in non-finite constructions in English and French (Schulz et al., 2021) but also Saxon genitives for relative clause attachment (Gilboy et al., 1995) or scope preferences (Hemforth & Konieczny, 2019). Although further studies will certainly be necessary, we suggest that the division of labour effect for null and pronominal subjects may at least partly be explained by more general principles in human language processing.
- 6 We acknowledge that we cannot know exactly to what extent prosody could have induced an effect of c (...)
74For further investigation, we wonder if there might be a difference in written vs. spoken language depending on how stimuli are presented to participants. The effect of contrast may be stronger in spoken language since prosodic focus markers can be taken into account when creating the experimental design and materials6.
75Additionally, for Experiments 1, 2 and 3, we took only 3rd person referents into consideration. There might be some variation in the language in null and pronominal subject alternation depending on person. For instance, corpus studies (Soares et al., 2020, Brazilian Portuguese) show that null subjects are preferred to retrieve 3rd person [-animate] and [-specific] referents. However, based on corpus investigations in Romanian, Istrate et al. (2023) have argued that null subjects are more frequent for discourse persons while pronominal subjects often refer to 3rd person referents.
76It also seems necessary to shed more light on pronoun production in order to observe to what extent PAH (Carminati, 2002) predicts antecedent preference. As we forced our participants to choose between only two conditions (null or pronominal subject), we cannot be sure whether Romanian native speakers would prefer a pronominal subject or a proper noun to mark contrast (see Contemori & Di Domenico, 2021). Future experiments such as sentence completion tasks will be necessary to answer these questions as well.
77In this paper, we contributed to the general question of the use of null and pronominal subjects and their antecedent preferences in Romance languages by adding data from Romanian, a language for which quantitative data are very scarce. We found a general preference for null subjects that is, however, less strong when contrast plays a role. We also found a division of labour effect in that null subjects prefer subject antecedents and pronominal subjects prefer non-subject antecedents. This division of labour effect seems to be less categorical than in Italian or European Portuguese and closer to Spanish and Catalan. We suggest that the gradience of the division of labour effect across languages that has been found in this paper but also by other authors (Greek vs. Italian: Torregrossa et al., 2020; Italian vs. Mexican Spanish: Contemori & Di Domenico, 2021; European vs. Brazilian Spanish: Fernandes et al., 2018) might be explained by the frequency of anaphoric expressions in the language. Fully parallel studies across languages will be necessary to confirm this hypothesis.
Data and materials are accessible at: https://osf.io/ra5qc/?view_only=5ac605b5c22b4d85b52c0326d95c5cb2.