Pronominal subject anaphora resolution has been shown to be highly sensitive to syntactic factors. A very strong preference of personal pronouns for subject antecedents has been widely attested in previous studies in non-null subject languages (Kaiser & Trueswell 2008 for Finnish; Arnold 2010; Kaiser 2011 for English; Bader & Portele 2019 for German). In null subject languages, capitalizing on the Position of Antecedent Hypothesis (PAH) proposed by Carminati (2002) for intra-sentential anaphora in Italian, null and overt pronouns have been claimed to display a division of labor in their preferential interpretation choices. This hypothesis predicts that, in referentially ambiguous contexts, null pronouns have a strong preference for antecedents in Spec,IP (i.e., subject antecedents), whereas overt pronouns show a clear bias for antecedents in lower syntactic positions (i.e., object antecedents). As indicated in (1) by the subindex, when ambiguity is at stake, the null pronoun (pro) typically prefers to retrieve the subject antecedent in the preceding clause and the overt pronoun (ell/ella, ‘he/she’), the object antecedent. Although Carminati (2002) proposed the PAH for Italian, these patterns have also been attested for Catalan (Mayol & Clark 2010; Bel & García-Alcaraz 2018) and for other Romance and non-Romance null subject languages (e.g., Papadopoulou et al. 2015 in Greek; Rinke & Flores 2018 in Portuguese; Contemori & Di Domenico 2021 in Spanish; Wolna & Durlik & Wodniecka 2022 in Polish), both at intra- (1–3) and inter-sentential level (4).

Concerning anaphora resolution in Catalan, only the two aforementioned studies have previously experimentally examined null and overt pronoun interpretive biases in globally ambiguous contexts within the framework of the PAH. Bel & García-Alcaraz (2018) analyzed semantically ambiguous intra-sentential contexts using an acceptability judgement task and found Catalan to show well-defined PAH-like biases of both null and overt pronouns in subordinate-main clause order, as illustrated in (2), and only a clear bias of overt pronouns towards object antecedents in main-subordinate sequences, as shown in (3). Indeed, the order of clauses and the structural organization of linguistic information (as we will see in the next section) have been shown to introduce nuances in the validity of the PAH, particularly evident in subordinate-main sequences (see Chamorro 2018; de Rocafiguera & Bel 2022 for a discussion).

The second study mentioned on Catalan anaphora resolution, by Mayol & Clark (2010), assessed inter-sentential contexts using a two-alternative forced-choice task and a self-paced reading task. Although these are not the specific contexts for which the PAH was formulated, null pronouns were also found to prefer subject antecedents and overt pronouns to prefer object antecedents, as in (4) and in line with the PAH, expanding its predictions formulated for intra-sentential contexts.

Nevertheless, Carminati (2002) recognizes that pronoun interpretation may also be sensitive to discourse factors such as the information status of the antecedents, which is examined in the next section.

Assuming Rizzi’s (1997) articulated structure of the expanded left periphery shown in (5) (from Fábregas 2016: 4), Carminati (2002: 184) further hypothesizes that referents introduced in positions higher than Spec,IP, such as topic positions (i.e., Spec,TopP), i.e. the position that typically host topics such as clitic-left dislocations (CLLD) in Romance languages, may be similarly preferred for null pronouns as antecedents in Spec,IP. She speculates that topicalized referents would compete with subject antecedents without overriding the preference of null pronouns for antecedents in Spec,IP. Focalized referents (i.e., in Spec,FocP) such as fronted focus, on the other hand, would not compete with subject referents. She attributes this asymmetry between topicalized and focalized referents to discourse factors. Null pronouns typically retrieve old referents in discourse, associated with topicalized antecedents, whereas focalized referents often introduce new information into the discourse.

More recently, considering the different position of subjects and topics, Torregrossa & Andreou & Bongartz (2020) proposed a reformulation of the PAH, defining the relative prominence of a constituent based on its hierarchical height. Their proposal is grounded on Rizzi (2018)’s claim that null subjects in Italian are mainly sensitive to “aboutness”, a property shared by both subject and topic positions, rather than to “subjecthood”. More specifically, Torregrossa et al. (2020: 9) propose the principles in (6) to account for the interpretation of null pronouns.

In this proposal, the hierarchical position of an antecedent is granted special importance in terms of making an antecedent prominent, or more preferred, for subsequent null pronouns. As displayed in (6b), the preferred antecedent of a null pronoun will be the one appearing in a hierarchically higher position, typically occupied by subjects or topics. In contrast to Carminati (2002), the authors predict that null pronouns may not similarly prefer subject antecedents in Spec,IP and antecedents in Spec,TopP. Instead, the preference for the hierarchically higher position may prevail (i.e., Spec,TopP). In this regard, they suggest that “the greater the difference between constituents in terms of hierarchical height, the more evident this bias is [the bias of null pronouns toward the hierarchically higher antecedent]” (Torregrossa et al. 2020: 9). As such, the bias of null pronouns towards subject antecedents is argued to be more pronounced when the object occurs in situ than when it is, for instance, left-dislocated. When the object occurs in situ there is greater distance in terms of hierarchical height between Spec,IP (i.e., the subject) and the object constituent. However, these observations were not empirically tested in their study. Finally, in (6c), they acknowledge the need of adopting a multifactorial approach to account for the possibility for null subjects to corefer with object antecedents, given that pronominal preferences are not absolute. However, they lack sufficient evidence to define how, or to what extent, these multiple factors interact (Torregrossa et al. 2020: 22).

Further research and experimentation are needed to address this gap. In the current study, we will use clitic-left dislocations as a topicalization device and it-cleft constructions as a focalization device to alter the syntactic and pragmatic positions within the phrase structure to serve as potential loci for hosting antecedents for null and overt subject pronouns.

From a broader perspective within discourse pragmatics the notions of topicality and givenness have often been intuitively related to prominence (see Cowles & Walenski & Kluender 2007; Kaiser 2011; von Heusinger & Schumacher 2019). Topical or given information has been considered to be more prominent and thus more likely to be pronominalized than new information (Chafe 1976; Prince 1981; Givón 1983). In agreement with Torregrossa et al. (2020)’s proposal (see above), most experimental studies demonstrate that antecedents in topic position tend to be good candidates for subject pronouns in reference assignment processes (e.g., Kaiser 2011 for English; Colonna & Schimke & Hemforth 2012 for German; de la Fuente 2015 for Spanish). In these studies, the preference for topic antecedents does not seem to override the preference for subjects, only to compete with it, as already hypothesized by Carminati (2002).

Apart from the pragmatic features carried by antecedents, null or overt subject pronouns themselves convey a discourse pragmatic meaning. In null subject languages, null pronouns have traditionally been assumed to be specialized in conveying topic maintenance (thus preferring to corefer with topical antecedents) and overt pronouns in conveying topic shift (thus preferring to corefer with non-topical antecedents) (e.g., Lozano 2009; Sorace et al. 2009; García-Alcaraz & Bel 2019). This has been mainly supported by the analysis of (semi)spontaneous production. However, as demonstrated in several corpora studies, null pronouns can also be felicitously used to express topic shift (e.g., Lubbers-Quesada & Blackwell 2009; Lozano 2016; García-Alcaraz & Bel 2019; Giannakou & Sitaridou 2022) (see also Leonetti 2021). Bel & Perera & Salas (2010) examined anaphoric third person subject pronouns in oral and written narratives in Catalan. Their findings showed that null pronouns mainly expressed topic maintenance and preferred to refer to antecedents in subject position, whereas the scarce number of overt pronouns did not show a clear coreference pattern.

Languages employ topicalization mechanisms that are crucial for our analysis. Clitic-left dislocation (CLLD), in particular, serves as a topicalization device (as opposed to focalization). At this point, it is relevant to cite Mayol (2010b) since it empirically examined the role of word order and topicalization via CLLD in the resolution patterns of both null and overt pronouns. As shown in the examples below (from Mayol 2010b: 129), she contrasted SVO unmarked sentences (7a) and OVS CLLD sentences (7b) in a two-alternative forced-choice task.

Based on her results, Mayol (2010b) argues that null pronouns are sensitive to syntactic factors but not to the manipulation of information status or word order. Topicalizing an object antecedent via CLLD did not affect the preference of null pronouns for subject antecedents. On the other hand, overt pronouns showed sensitivity to both syntactic and pragmatic factors, preferring object antecedents only in the unmarked SVO condition (non-subject, non-topic constituents1), and not showing a clear preference in the CLLD OVS condition. It is worth noting that Mayol (2010b)’s results on null pronouns do not support the predictions of Carminati (2002), who hypothesized that null pronouns should be sensitive to both subjecthood and topicality cues. However, de la Fuente (2015), who similarly analyzed the impact of topicalization via hanging topic left dislocations (HTLD) on the preferences of null pronouns in Spanish, found a significant increase in the preference of null pronouns for topicalized object antecedents.

Another linguistic device that can influence accessibility is focalization. In this context, clefting is a focalization mechanism that manipulates information structure. Research on anaphora resolution in marked information structures, such as it-clefted clauses, shows mixed results. Psycholinguistic research has found that linguistic focus increases the accessibility of focused entities (Foraker & McElree 2007; Káldi & Babarczy 2021). However, it is not always straightforward that focused entities are chosen as the preferred antecedent to interpret a subject pronoun (Blything et al. 2021). In fact, focusing it-cleft structures have also been found to make an antecedent less accessible for an anaphoric subject pronoun, leading to an “anti-focus effect” (Colonna et al. 2012; Colonna & Schimke & Hemforth 2015; de la Fuente 2015; Patterson & Esaulova & Felser 2017; Patterson & Felser 2020).

No previous studies have addressed the impact of it-cleft structures on anaphora resolution in Catalan, but de la Fuente (2015) examined the effects of both subject and object focalization via it-cleft structures in Spanish. His results revealed that null pronouns consistently showed a higher preference for the non-clefted antecedent, conveying presupposed, topic-like, information. Similar to Colonna et al. (2012; 2015) for French, de la Fuente (2015) concluded that focusing an antecedent in an intra-sentential (but not in an inter-sentential) context leads to an anti-focus effect. His design, however, did not include overt pronouns, and to our knowledge, there has not been any other research examining the effects of information status on the interpretive preferences of subject pronouns in a null subject language. The present study will expand the findings of Mayol (2010b) in Catalan and de la Fuente (2015) in Spanish by exhaustively investigating the interpretation of both null and overt pronouns in both topicalization and focusing structures (clitic-left dislocation and it-clefts).

A consequence of manipulating the information status of the antecedent using syntactically marked structures is that word order, or the sequential position of the antecedents, is also altered. In this regard, the initial position has been argued to be a privileged position for an antecedent to be picked up by a subsequent pronoun, leading to a first-mention advantage (Gernsbacher & Hargreaves 1988) attributed to general cognitive processes (see also Carreiras & Gernsbacher & Villa 1995). However, subsequent studies assessing this first-mention advantage, such as Fukumura & van Gompel (2015) in an eye-tracking study on English subject pronouns, have demonstrated that the preference of pronominal subjects for subject antecedents is stronger than the preference for first-mentioned antecedents (Cowles et al. 2007; Kaiser & Trueswell 2008; see also Bader & Portele 2019). On the other hand, Järvikivi et al. (2005) showed no first-mention advantage but claimed that both syntactic function and order of mention play a role in the interpretation of personal pronouns in Finnish, and d-pronouns in German were also claimed to be sensitive to order of mention, different from personal pronouns (Bader & Portele 2019).

In the present study, we will interpret a preference for first-mentioned antecedents as a preference for those in the highest structural position, as suggested by Carminati (2002) or Torregrossa et al. (2020). This proposal implies a prevalence of syntactic/structural factors over sequential factors, although they overlap. We will be studying how the information status of the antecedent interacts with its syntactic position, being aware that manipulating information structure to alter the pragmatic status of referring expressions also change their surface position.

As already described, in previous studies neither syntactic nor pragmatic factors have been found to act as an overarching factor governing anaphora resolution. At the same time, it is not clear that they are similarly relevant. Kaiser & Trueswell (2008) propose a form-specific multiple-constraints approach to account for pronominal anaphora resolution, suggesting that these factors may be prioritized differently or carry different weights in shaping coreference preferences depending on the specific pronoun form involved. More specifically, Kaiser & Trueswell analyzed the interpretive preferences of Finnish personal and demonstrative subject pronouns and found the two forms to have different degrees of sensitivity to syntax and information structure: whereas personal pronouns were mainly sensitive to the syntactic function of the antecedent (preferring subjects), the preference of demonstrative pronouns were guided by a combination of information status and syntactic function (preferring discourse-new and object referents). Similarly, Bader & Portele (2019) identified German personal pronouns to be mainly sensitive to syntactic cues and demonstrative pronouns to be sensitive to a combination of linear position and topicality. In the case of Catalan, Mayol (2010b) also suggested that null pronouns may have a simple preference for previous subjects and overt pronouns be sensitive to both syntactic and pragmatic factors (see also Fedele & Kaiser 2014 for the different weight of syntax, semantics and sentence boundaries in the interpretation of null and overt pronouns in Italian).

Still other studies on non-null subject languages have confronted factors from different linguistic domains, focusing solely on personal pronouns. In a study on English, Kaiser (2011) found subjecthood (and agentivity) to have particularly strong effects on pronoun interpretation, regardless of information structure. These pronouns were sensitive to both topicalization and focusing effects, which boosted the preference for topicalized and focused referents. However, these preferences did not overcome the preference for subjects. Similarly, Blything et al. (2021) found subjecthood/agentivity/first-mention cues to have a stronger role in the offline interpretation of English subject pronouns in focus position (via prosodic marking and it-clefts). On the other hand, Colonna et al. (2012) found that the preference of personal pronouns in German toward subject antecedents was only attested when the subject coincided with the sentence topic. When the object was topicalized, personal pronouns remained unbiased, indicating a similar weight of syntactic function and information structure. Finally, Schumacher & Roberts & Järvikivi (2017) argued that thematic role is a more powerful cue in pronoun resolution in German than grammatical function and linear position (personal pronouns preferring proto-agents and d-pronouns preferring proto-patients). However, these preferences guided by thematic role became weakened when these cues were misaligned with grammatical function and word order cues (i.e., proto-agents not being marked with nominative case, and proto-patients not being postverbal or second-mentioned).

In summary, these studies support a multifactorial and form-specific approach, indicating that there are multiple factors involved in the anaphoric interpretation of a particular pronominal form and that they may vary compared to another form. In the present study, we will consider two pronominal forms (null and overt pronouns) and their potentially different interactions with syntactic and pragmatic factors.

The main interest of the present research is to elucidate how syntactic and pragmatic factors interact on pronominal anaphora resolution in null subject languages by using syntactically and pragmatically marked information structures. As a novelty, this approach thus puts together these different dimensions in a single study and it does so in a language that has received little attention. Specifically, we aim at disentangling the contribution in the interpretation of null and overt pronouns of three factors: 1) the syntactic function of the antecedent (subject vs. object), 2) the information status of the antecedent (topical vs. non-topical), and 3) the hierarchical position of the antecedent (higher vs. lower position in the syntactic structure; also interpreted as first-mentioned vs. second-mentioned in alternative perspectives). These factors overlap in unmarked structures, but, crucially, they do not in marked CLLDs and it-clefts. With this purpose, we will test the Position of Antecedent Hypothesis (PAH; Carminati 2002) in non-canonical sentences where the information status of antecedents is manipulated, thereby altering both their linear and hierarchical positions. More precisely, we will contrast the interpretive preferences of anaphoric null and overt subject pronouns in the following four structures:

As already mentioned, the notions of subject, topic, and higher position in the syntactical hierarchy (or initial position) overlap in unmarked structures (8.1). Clitic-left dislocation (CLLD) and it-cleft structures will allow us to disentangle the role of these syntactic and pragmatic factors in the process of solving pronominal anaphoric dependencies, as detailed below.

Firstly, in CLLD structures (8.2), object antecedents are topicalized and appear first-mentioned (OVS word order), in the hierarchically highest position of the syntactic structure (Spec,TopP). Subject antecedents convey focal (or non-topical) information2 and appear second-mentioned, in a postverbal position, and in a hierarchically lower position than objects (i.e., either in Spec,IP or in situ in a VP-internal position)3.

Secondly, in subject it-cleft structures (8.3), subject antecedents are focused and appear first-mentioned (SVO word order), in the hierarchically highest position of the syntactic structure (Spec,FocP). Object antecedents convey presupposed (topical) information4 and appear second-mentioned, in a hierarchically lower position than subjects.

Finally, in object it-cleft structures (8.4), object antecedents are focused and appear first-mentioned (OVS word order), in the hierarchically highest position in the syntactic structure (Spec,FocP). Subject antecedents convey presupposed (topical) information5 and appear second-mentioned, in a postverbal position, and in a hierarchically lower position than objects.

As reviewed in the introduction, recent proposals, like those by Torregrossa et al. (2020), suggest a need to reevaluate the framework grounded on the influential PAH reconsidering the relative prominence of antecedents in terms of hierarchical height. While previous research has highlighted the significance of information status and word order in anaphora resolution, there are discrepancies regarding the importance of syntactic function versus information structure and sequential word order. We identify gaps in the literature concerning the interaction between these levels, particularly in cases there these cues are misaligned. The current study aims to explore how these factors interact in non-canonical sentence structures. By analyzing experimental data, manipulating the information status of the antecedents, and analyzing their linear and hierarchical positions, we aim to elucidate the mechanisms governing pronominal anaphora resolution in Catalan. By doing so, we seek to contribute to a deeper understanding of language interpretation mechanisms in null subject languages and inform future research in this area.

The research questions and hypotheses that will be tested in the present study are the following:

In the light of the studies that have previously addressed these contexts (Mayol & Clark 2010; Bel & García-Alcaraz 2018), we expect to attest the biases predicted by the Position of the Antecedent Hypothesis (PAH; Carminati 2002): null pronouns should preferably corefer with subject antecedents (topical, in first position) and overt pronouns with object antecedents (focal, in second position). Given that consistent evidence has been found in these previous studies on Catalan, we expect to replicate previous findings. However, these previously attested robust biases could be affected by two different factors. First, although previous research has shown that pronominal biases are more pronounced in subordinate-main sentences, we opted for main-subordinate sentences to ensure consistency across the four analyzed structures. Opting for the subordinate-main order would result in ungrammatical structures in the case of it-cleft sentences, as they would not fit within that sequence (see section 3.2 below). Second, we consistently used the conjunction quan (‘when’) to introduce the second subordinate clause, minimizing variability caused by the connector (de Rocafiguera & Bel 2022; Martín-Villena 2023).

As already stated, we do not expect the interpretive preferences of null and overt subject pronouns to be guided by an only factor and different pronominal forms could be differently sensitive to each factor (i.e., syntactic function, information status or hierarchical/linear position; see Kaiser & Trueswell 2008; Bader & Portele 2019; among others). However, there could be an outranking factor having a higher strength in defining the interpretive patterns of null and overt subject pronouns. In this regard, three hypotheses should be put forward:

The participants in the present study are 53 Catalan-dominant speakers (mean age: 22.26, SD: 3.81, range: 18–35) that were born and raised in Catalonia, and that were studying a university-level degree or had already graduated from university. Since there are no monolingual speakers in Catalonia with Catalan as their first and only language, we aimed to be rigorous by selecting participants with the highest degree of exposure and mastery of Catalan (referred to as Catalan-dominant speakers). Catalan speakers are also exposed to Spanish in the society and end up acquiring both languages. To identify the language dominance profile of the participants, they completed a background questionnaire based on the Bilingual Language Profile (BLP; Birdsong & Gertken & Amengual 2012)6 before participating in the experiment.

All participants declared that they began acquiring Catalan at the same time (50.94%) or earlier (49.06%) than Spanish, most of them were raised in monolingual Catalan-speaking families (73.58%) and they all declared that they use Catalan to talk to their parents and/or siblings. Regarding language use, they reported using Catalan most of their time during the week (mean: 88.34%; SD: 8.21) and showed a much more limited use of Spanish (mean: 13.30%; SD: 6.59) and other languages (mean: 1.83%; SD: 3.47). Concerning self-rated language proficiency, most of them rated it as being ‘very good’ in all skills of both languages, although they perceived their oral production abilities to be slightly higher in Catalan than in Spanish (mean out of 6 in Catalan: 5.91; SD: 0.35; mean out of 6 in Spanish: 4.92; SD: 0.87). None of them rated their abilities in Catalan to be lower than in Spanish. Regarding the global dominance scores provided by the BLP questionnaire, which allows for placing participants within a numerical continuum of language dominance, the participants in the present study received a mean score of –82.59 (SD: 16.01; range: –121.24 to –56.76). The values of the global dominance score in the BLP range from –218 to 218 and are obtained from the subtraction of the score of one language from the score of the other one. In the present study, negative values are indicative of Catalan dominance, and positive values indicative of Spanish dominance. A score close to zero would suggest balanced bilingualism.

One participant was excluded because s/he consistently and systematically considered all the pronouns across conditions as ambiguous (and did not make an interpretive choice, as indicated in the instructions of the task). No items were excluded.8 The Qualtrics Survey interface measured the time that participants took to give a response for each item in the task, which was used to clean the data. All items that had been answered faster than 3 seconds were excluded, since we considered that this amount of time was not enough to carefully read the sentences and provide an answer accordingly. Overall, the removed observations represented a 3.64% of the initial dataset.

At this point, reference is needed to the responses obtained in the combined two-alternative forced choice plus slider task. As shown in Figure 2, participants gave answers that were mainly at the edges of the continuous scale: they expressed an interpretive choice, but the task did not seem to capture a gradation in the strength or certainty of these choices. Given the distribution of the data, the answers were finally reduced to a binary choice (i.e., as subject or object choices), as would have been in a traditional two-alternative forced choice task.9

To analyze the results, a mixed-effects logistic regression model was run in R (v. 4.0.5; R Core Team 2021) using the glmer function of the lme4 package (v. 1.1.27; Bates et al. 2015) and the emmeans package to obtain pairwise contrasts (v. 1.7.0; Lenth 2021). The dependent variable was the choice to interpret an ambiguous pronoun as referring back to either a subject or an object. The mixed-effects logistic regression model included Pronoun (null, overt) and Information structure (unmarked structures, topicalized objects via CLLD, focused subjects via it-clefts, and focused objects via it-clefts) as fixed effects, as well as their interaction. As random effects, varying intercepts for participants and items were added to the model and a by-participant varying slope for the effect of Pronoun. Additional random slopes were tested but they led to estimation problems within the models. The final model had the maximal random structure that converged and was the model offering the best fit (we performed a maximum likelihood ratio comparison between the models that converged). The model’s total explanatory power was moderate (conditional R² = 0.23), and the part related to the fixed effects alone (marginal R²) was 0.08. No multi-collinearity issues were detected (highest VIF = 1.001). This model had a C-index of concordance of 0.76. The dataset and the code used to run the analyses in R are available in the Open Science Framework repository. The coefficient table of the mixed-effects logistic regression model and the table of pairwise contrasts can be found in Appendix C. Supplementary analysis of the responses in their original scale using an ordered beta regression can be found in Appendix D.