Cross-linguistic influence at the discourse–syntax interface: Insights from anaphora resolution in child second language learners of Italian

Abstract

This paper reports the results of an experimental study on the resolution of intra-sentential anaphora in Italian by two groups of 13–14-year-olds: monolingual native Italian speakers and highly proficient child second language (L2) learners of Italian whose native language is Croatian. In a picture selection task the participants were asked to identify antecedents of null and overt subject pronouns in ambiguous forward and backward anaphora sentences. Our assumption in the paper was that Italian and Croatian do not differ with respect to the antecedent biases of null and overt subject pronouns in the contexts under investigation. As predicted, the L2 learners expressed pragmatically appropriate antecedent preferences in all conditions. They even selected the pragmatically inappropriate subject antecedent for the overt pronoun less often than the native speakers, especially in backward anaphora. The L2 learners’ antecedent preferences closely mirror those established in previous research for their age-matched monolingual Croatian peers. We take this as evidence for our position that cross-linguistic influence is operative at the discourse–syntax influence in child L2 acquisition and that it might be the main cause of non-target-like behaviour of some groups of highly proficient bilinguals concerning certain properties pertaining to this interface.

Keywords

Cross-linguistic influence discourse–syntax interface Interface Hypothesis anaphora resolution forward anaphora backward anaphora null subject pronouns overt subject pronouns child second language acquisition Italian Croatian

Introduction

The expression of pronominal subjects in null subject languages by bilingual speakers has received a lot of attention in recent years. Numerous studies have investigated the interpretation and/or production of null and overt subject pronouns in languages such as Italian, Spanish, Brazilian Portuguese and Greek in different bilingual domains, including adult second language (L2) acquisition (e.g. Belletti, Bennati, & Sorace, 2007; Kraš, 2008a; Lozano, 2009; Montrul, Dias, & Thomé-Williams, 2008; Montrul & Rodríguez Louro, 2006; Pérez-Leroux & Glass, 1999; Rothman, 2009; Sorace & Filiaci, 2006; Tsimpli & Sorace, 2006), bilingual first language (L1) acquisition (e.g. Paradis & Navarro, 2003; Serratrice, 2007; Serratrice, Sorace, & Paoli, 2004; Sorace, Serratrice, Filiaci, & Baldo, 2009), heritage language acquisition (Montrul, 2004) and L1 attrition (Tsimpli, Sorace, Heycock, & Filiaci, 2004). While some studies have reported target-like performance of bilingual speakers with respect to pronominal subjects (e.g. Kraš, 2008a; Rothman, 2009), others have reported non-target-like behaviour, especially with overt forms, even on the part of highly proficient bilinguals (e.g. Serratrice et al., 2004; Sorace & Filiaci, 2006; Tsimpli et al., 2004). In particular, some groups of highly proficient bilinguals have been shown to accept or produce overt subject pronouns referring to infelicitous topical antecedents to a significantly greater extent than monolingual speakers. In other words, they have been shown to overgeneralise overt pronouns to contexts that require the use of null pronouns. In contrast, they interpreted or produced null subject pronouns appropriately.

Based on the observed instances of overuse or overacceptance of overt subject pronouns in null subject languages, which contrast with appropriate use and interpretation of null subject pronouns, and on the assumption that the distribution of null and overt subject pronouns in discourse is regulated at the discourse–syntax interface, Sorace and Filiaci (2006, p. 340) propose that “narrow syntactic properties are completely acquirable in a L2, even though they may exhibit significant developmental delays, whereas interface properties involving syntax and another cognitive domain may not be fully acquirable.” They also predict that the same interface properties may be acquired late in bilingual L1 acquisition and may not be fully retainable in L1 attrition. This proposal is known as the Interface Hypothesis (see Sorace, 2011, for its most recent version).

Instability at the discourse–syntax interface has been accounted for in two ways. According to the so-called representational account, proposed by Tsimpli et al. (2004), instability is caused by cross-linguistic influence whereby the language that instantiates a more economical setting with respect to an interface property influences the language that offers a less economical setting. This influence leads to the underspecification of the interpretable feature relevant to the interface property in the language with a less economical setting. In the case of pronominal subject expression, non-null subject languages (e.g. English), which do not allow subject drop, have a more economical setting than null subject languages. The processing account, put forward by Sorace and Filiaci (2006), views cross-linguistic influence as only a secondary cause of interface instability. This account proposes that the main cause of instability lies in that bilinguals have suboptimal processing abilities (stemming from either cognitive resource limitations or cognitive resource misallocation; see Wilson, Sorace, & Keller, 2009, for the latter proposal). Interface structures are assumed to be particularly costly in terms of processing, as they require integration of different types of information. As a result, bilinguals occasionally resort to certain default strategies in processing interface structures, such as the overuse or overacceptance of overt subject pronouns in null subject languages.

According to the representational account, instability should not arise when two grammatical systems do not differ with respect to the interface property under investigation; according to the processing account, instability should occur even in those cases. The present study tests these predictions by looking at anaphora resolution with null and overt subject pronouns in Italian in highly proficient child L2 learners who are native speakers of Croatian, another null subject language. Based on Kraš (2008b), we assume that Croatian does not differ from Italian with respect to the discourse-pragmatic constraints on the distribution of null and overt subject pronouns. In a previous study, Kraš (2008a) showed that L1 Croatian near-native speakers of Italian who have acquired Italian in adulthood resolve this type of anaphora in a native-like way, taking this as evidence in support of the representational account. However, Sorace (2011, p. 24) argues that the reason these speakers do not overgeneralise overt subject pronouns might lie in the fact that they have acquired Italian as their third language (L3; their L2 being English) and, hence, possibly experience less resource competition in resolving anaphoric dependencies than L2 learners. She explains:

There may be less resource competition for L3 learners and for multilinguals in general. Costa and Santesteban (2004) propose that balanced bilinguals may develop a language selection mechanism that does not depend on inhibition and can be applied to subsequent languages. It is therefore possible that trilinguals, or polyglots in general, show less or no overgeneralization of overt pronouns because they do not need to apply as much inhibitory control as bilinguals.

Even though this proposal still awaits empirical validation, in this study we wish to bypass the potentially problematic inclusion of L3 learners by testing participants who have acquired Italian as their “true” L2 – child L2 learners of Italian.¹ The results of the study should be thus comparable with the results of the other L2 studies that looked into the same linguistic phenomenon. As such, our findings should be directly relevant to the debate on the sources of instability at the discourse–syntax interface in bilingual grammars.

The paper is structured as follows. In the next section we describe pronominal subject expression in Italian and Croatian. This is followed by a review of studies that looked into this phenomenon in L1 and L2 acquisition. The following section describes the methodology and the results of the present study. The paper ends with a discussion of the results and some conclusions.

Pronominal subject expression in Italian and Croatian

Italian and Croatian are pro-drop or null subject languages, which means that pronominal subjects can either be expressed or omitted (i.e. realised as overt or null) in finite clauses in both these languages. Whether or not a pronominal subject is dropped depends on discourse-pragmatic factors: in a typical case, null subject pronouns are used to refer to topical antecedents, while overt subject pronouns are used to refer to non-topical antecedents. Continuity in discourse topic is thus typically expressed by null pronouns, while a shift of topic is signalled by overt pronouns (e.g. Sorace, 2000; cf. Pérez-Leroux & Glass, 1999). Sorace (2000) expresses this by means of the interpretable Topic Shift feature: the use of overt pronouns is felicitous in [+Topic Shift] contexts and the use of null pronouns in [–Topic Shift] contexts. This is illustrated in (1) and (2), equivalent examples for Italian and Croatian respectively.

(1) Sara_i ha imparato a nuotare. pro_i/*j È sorpresa/ Lei_?i/j è sorpresa.

Sara has learned to swim pro is surprised she is surprised

“Sara has learned how to swim. She is surprised.”

(2) Sara_i je naučila plivati. pro_i/*j Iznenađena je/Ona_?i/j je iznenađena.

Sara is learned to-swim pro surprised is she is surprised

“Sara has learned how to swim. She is surprised.”

In a typical situation, the null pronoun in the second sentence in (1) and (2) refers to the subject of the first sentence (Sara), the topical antecedent, while the overt pronoun in the second sentence refers to an extra-sentential antecedent, probably mentioned in previous discourse. Therefore, the two versions of the sentence describe different scenarios: in the version with the null pronoun it is Sara who is surprised because she has learned how to swim, while in the version with the overt pronoun someone else is surprised because Sara has learned how to swim.

In intra-sentential contexts in Italian and Croatian, null and overt subject pronouns prefer antecedents in different syntactic positions. In both languages, antecedent preferences of the two types of pronouns are largely consistent with the Position of Antecedent Strategy (PAS), a processing principle proposed for Italian by Carminati (2002). According to the PAS, the null pronoun prefers an antecedent in the subject position (typically acting as the discourse topic), while the overt pronoun prefers an antecedent in a non-subject position. This is illustrated in (3) and (4), for Italian and Croatian respectively.

(3) Antonio_i vede Dario_j mentre pro_i/?j/lui_?i/j/k guida la macchina.

Antonio sees Dario while pro he drives the car

“Antonio sees Dario while he drives the car.”

(4) Antonio_i vidi Darija_j dok pro_i/?j/on_?i/j/k vozi auto.

Antonio sees Dario while pro he drives car

“Antonio sees Dario while he drives the car.”

In complex bi-clausal sentences such as these, the null pronoun in the subordinate clause is more likely to take the subject of the main clause as its antecedent (Antonio) than the object (Dario); in contrast, the overt pronoun is more likely to refer to the object than to the subject. The overt pronoun can also refer to an extra-sentential antecedent.

The PAS can be violated by both null and overt pronouns, i.e. the null pronoun can take a non-subject antecedent and the overt pronoun can take the subject antecedent. Given that this results in inappropriateness and not ungrammaticality, Sorace and Filiaci (2006) argue that the PAS operates at the discourse–syntax interface and not within narrow syntax.

Psycholinguistic evidence for the operation of the PAS in Italian has been provided by Kraš (2008a), Belletti et al. (2007), Serratrice (2007), Sorace and Filiaci (2006) and Tsimpli et al. (2004), while equivalent evidence pertaining to Croatian has been provided by Kraš (2008b) and Kraš and Stipeć (2013). The relevant data in these studies come from monolingual adult native speakers. All the studies used a similar picture selection task. The Italian studies, apart from Kraš (2008a), used the task introduced by Tsimpli et al. (2004). Kraš (2008a) used a modified version of this task.² The modified version was translated into Croatian by Kraš (2008b). In all the studies, apart from Kraš and Stipeć (2013), participants read or listened to³ sentences, and then matched each sentence to one of three pictures. The sentences were of the same type as in (3) and (4), i.e. they consisted of a main clause and subordinate clause, and the subordinate clause contained an ambiguous subject pronoun. The pronoun either followed (forward anaphora) or preceded (backward anaphora) possible antecedents. The pictures showed the antecedent as the matrix subject, the matrix object or an extra-sentential referent. The task in Kraš and Stipeć (2013) differed in that it included sentences only with forward anaphora and offered a choice of two antecedents (the matrix subject and the matrix object) in the pictures, instead of three. Also, the sentences were presented aurally.

Among the Italian studies, the results were fully consistent with the PAS only in Kraš (2008a). In the other studies, the null pronoun was interpreted as coreferential with the subject antecedent in forward anaphora only approximately half of the time. The same was true in Kraš and Stipeć (2013). In the other Croatian study, Kraš (2008b), the results were fully in line with the PAS. Evidence from these studies thus suggests that in ambiguous intra-sentential contexts, the Italian and Croatian overt pronoun prefers the non-subject antecedent in both forward and backward anaphora, while the null pronoun clearly prefers the subject antecedent only in backward anaphora. More research is needed to determine to what extent antecedent assignment with the null pronoun in forward anaphora operates in line with the PAS.

Italian and Croatian pronominal subjects in language development

To the best of our knowledge, there are no previous studies on the interpretation of Italian pronominal subjects in child L2 acquisition. Among the studies looking into this phenomenon in adult L2 acquisition, of particular relevance for the present study is a group of studies that used a version of Tsimpli et al.’s (2004) picture selection task. Belletti et al. (2007) and Sorace and Filiaci (2006) tested L1 English near-native speakers of Italian. In both studies the near-natives selected the subject antecedent for the overt pronoun significantly more often than the native speakers. In backward anaphora they even preferred the subject antecedent (in contrast to the native speakers, who preferred the extra-sentential antecedent), while in forward anaphora they preferred the object antecedent, similar to the native speakers. Null pronouns were interpreted in a fully native-like way. A different pattern of results was obtained in a study testing L1 Croatian near-native speakers, Kraš (2008a), where the near-natives interpreted both null and overt pronouns in a native-like way. It should be noted that in this study the participants’ preferences for the overt pronoun in backward anaphora were split between the object and the extra-sentential antecedent.

Given that our study includes child L2 learners, previous studies on the interpretation of Italian and Croatian pronominal subjects in monolingual development are also relevant. Among the Italian studies, Serratrice (2007) showed that children aged 6–9 (mean age: 8;6) interpreted the null pronoun in forward anaphora in a fully adult-like fashion and only marginally differed from adults in the strength of their preference for the subject antecedent in backward anaphora. In contrast, they interpreted the overt pronoun as coreferential with the subject antecedent more often than the adults both in forward and backward anaphora, but especially in backward anaphora. However, in both contexts they chose the object antecedent in the majority of the cases (65% in forward anaphora and 54% in backward anaphora), selecting the extra-sentential antecedent only very seldom.

When it comes to Croatian, the youngest monolinguals tested in both forward and backward anaphora were 13–15 years old (mean age: 13.9) (Kraš, in preparation). For forward anaphora we have evidence from children aged 6–7 (mean age: 6.9) (Kraš & Stipeć, 2013). Evidence from the two studies suggests that when it comes to forward anaphora, Croatian monolinguals are already able to make appropriate antecedent choices at the age of 6–7 years, but that they select a non-subject antecedent for the null pronoun and the subject antecedent for the overt pronoun more frequently than adults, even at the age of 13–15 years. In backward anaphora, the 13–15-year-olds make inappropriate antecedent choices for the null and overt pronoun more often than the adults as well, but predominantly select appropriate antecedents. They differ from the adults in that they prefer the extra-sentential antecedent for the overt pronoun, while the adults are undecided between the extra-sentential and the object antecedent.

Summarising the findings on monolingual development, it appears that knowledge of the distribution of Italian and Croatian subject pronouns in discourse develops gradually and slowly, with knowledge of the distribution of null pronouns emerging earlier than knowledge of the distribution of overt pronouns, at least in Italian.

The study

Aims and predictions

The aim of the present study was to investigate how highly proficient child L2 learners of Italian whose native language is Croatian interpret null and overt subject pronouns in forward and backward intra-sentential anaphora in Italian. In particular, we wanted to determine (a) whether the learners show a tendency to overgeneralise overt pronouns to contexts that require the use of null pronouns to a greater extent than monolingual native speakers of the same age (similar to English-speaking adult L2 learners) and (b) whether the learners interpret null pronouns in a native-like way.

We predicted that the L2 learners in our study would not differ from the native Italian speakers in either their overgeneralisation of overt pronouns to null-pronoun contexts or in their interpretation of null pronouns. These predictions are based on the following two assumptions. The first assumption, based on the findings of Kraš (2008a), is that the English-speaking adult L2 learners in Belletti et al. (2007) and Sorace and Filiaci (2006) interpreted overt pronouns in a non-native-like way as a result of cross-linguistic influence and not because of their suboptimal processing abilities. The second assumption, based on the findings of Kraš (2008b), is that Croatian and Italian null and overt subject pronouns do not differ in their antecedent biases in forward and backward intra-sentential anaphora.

Participants

Two groups of 13–14-year-olds took part in the study: monolingual native Italian speakers (n = 48) and native speakers of Croatian who were learning Italian as the L2 (n = 40). The native speakers were attending the third grade of a lower secondary school (It. scuola media) in Trieste (Italy) and the L2 learners were attending the seventh or eighth grade in four Italian-medium primary schools in Rijeka (Croatia).⁴ The L2 learners had started learning Italian between the ages of three and seven,⁵ typically in an Italian-medium nursery in Rijeka. They were all highly proficient, although not near-native in Italian. A non-standardised C-test⁶ (Kraš, 2008a) was used as a general proficiency measure. The learners with the highest scores were included in the study; the approximate cut-off point was set at 66.5% (two thirds of the test). An independent samples t-test showed that the difference between the C-test scores of the native and the non-native group was significant (t = 8.595, df = 26, p < .001). More information on the participants is given in Table 1.

Table 1.

Participant characteristics.

	Age of testing		Age of first exposure		Proficiency score
	Range	Mean	Range	Mean	Range	Mean	SD
L1	13–14	13.38	N/A	N/A	71.5–98.5	85.85	7.16
L2	13–14	13.53	3–7	4.33	63–81.5	71.61	5.02

Materials and procedure

The participants’ ability to resolve intra-sentential anaphora in Italian was assessed by means of the task used by Kraš (2008a). There were four experimental and four control conditions, containing ambiguous and unambiguous sentences, respectively. The purpose of the control sentences was to make sure that the pictures were drawn in a clear way and to serve as fillers. All the sentences in the task consisted of a main and subordinate clause. In half the sentences the main clause preceded the subordinate clause and in half it followed the subordinate clause. The main clause included two singular noun phrases (NPs) of the same gender (equal numbers of male and female) in the role of the subject and the object.

The structure of the subordinate clause differed in the experimental and control sentences. The experimental sentences contained temporal clauses introduced by the conjunction mentre (“while”). These clauses featured a third person singular null or overt subject pronoun matched in gender with the subject and object of the main clause. The pronoun could thus ambiguously refer to either of the two NPs in the main clause. Sentences with main-subordinate clause order instantiated forward anaphora and those with subordinate-main order instantiated backward anaphora. Examples of the four experimental conditions are given in (5).

(5) a. (Forward anaphora with a null pronoun)

La figlia_i chiama la mamma_j mentre pro_i/j guarda la tivù.

the daughter calls the mother while pro watches the TV

“The daughter calls her mother while she watches TV.”

b. (Forward anaphora with an overt pronoun)

La figlia_i chiama la mamma_j mentre lei_i/j guarda la tivù.

the daughter calls the mother while she watches the TV

“The daughter calls her mother while she watches TV.”

c. (Backward anaphora with a null pronoun)

Mentre pro_i/j guarda la tivù, la figlia_i chiama la mamma_j.

while pro watches the TV the daughter calls the mother

“While she watches TV, the daughter calls her mother.”

d. (Backward anaphora with an overt pronoun)

Mentre lei_i/j guarda la tivù, la figlia_i chiama la mamma_j.

while she watches the TV the daughter calls the mother

“While she watches TV, the daughter calls her mother.”

The four control conditions are illustrated in (6). In two of the control conditions, shown in (6a) and (6c), the subordinate clause contained a non-finite verb in the form of a gerund controlled by the subject of the main clause. In the condition shown in (6a), the subordinate clause followed the main clause and, in the condition shown in (6c), the subordinate clause preceded the main clause. In the remaining two control conditions the subordinate clause contained a finite verb. The subordinate clause again either followed or preceded the main clause; in the former case, illustrated in (6b), the subordinate clause was a relative clause that modified the object of the main clause and, in the latter case, shown in (6d), it was a temporal clause that contained the third NP, in the role of the subject, matched in number and gender with the subject and object of the main clause.

(6) a. (Postposed gerundive clause)

La figlia chiama la mamma guardando la tivù.

the daughter calls the mother watching the TV

“The daughter calls her mother while watching TV.”

b. (Relative clause)

La figlia chiama la mamma che guarda la tivù.

the daughter calls the mother who watches the TV

“The daughter calls her mother who watches TV.”

c. (Preposed gerundive clause)

Guardando la tivù la figlia chiama la mamma.

watching the TV the daughter calls the mother

“While watching TV the daughter calls her mother.”

d. (Temporal clause)

Mentre la nonna guarda la tivù, la figlia chiama la mamma.

while the grandmother watches the TV the daughter calls the mother

“While the grandmother watches TV, the daughter calls her mother.”

Each sentence in the task was accompanied by three pictures, showing the matrix subject, the matrix object or the extra-sentential referent as the agent of the action described in the subordinate clause. The pictures were numbered one, two or three, depending on the linear order in which they appeared on the screen. The relative position of the three types of pictures was varied systematically. A set of pictures corresponding to the sentences in (5) and (6) is shown in Figure 1. Picture number one depicts the subject, picture number two the object and picture number three the extra-sentential referent performing the action described in the subordinate clause.

Figure 1.

Example of a picture set.

There were 48 sets of pictures in the task, with 48 corresponding sentences. These 48 sentences were distributed over eight presentation lists in a Latin Square design. Each list thus contained six items per condition. There were also four practice items in the experiment, the same in all lists, taking the form of the sentences in the control conditions.

The task was implemented with SuperLab Pro 2.0 and run on a laptop with a 14.4" screen. The presentation of the sentences preceded the presentation of the pictures. The sentences were presented word by word under speeded conditions⁷ in order to prevent the subjects from spending too much time thinking about the sentences. The length of time for presenting each word was adjusted to the length of the word: it was 301 ms per word plus 17 ms per letter (e.g. 352 ms for a three-letter word). The three pictures appeared on the screen after the last word of the sentence had disappeared. The participants were asked to select one picture to match the sentence. Responses were entered on a keyboard. The time for response was not limited. The sentences were presented in a different random order for each participant. The experiment lasted 10–15 minutes for each participant.

The study also included a questionnaire, used to elicit relevant background information on the participants. Questionnaires were filled in together with C-tests before the administration of the experiment. This part of the study was done in a group.

Results

In what follows, we present the results of the study, starting with the experimental conditions. We examine the distribution of the three responses (subject, object, extra-sentential referent, i.e. “other”) in each condition by participant group. The values we present are the mean percentages of responses, calculated by averaging the individual percentages of responses obtained from the counts of the number of times each possible response was chosen by each participant in each condition.

The distribution of responses in the two forward anaphora conditions is shown in Figure 2. In the null pronoun condition both participant groups interpreted the pronoun as coreferential with the subject in most cases; they considered the object as a less likely, but still plausible antecedent for the pronoun. The extra-sentential referent was selected very infrequently. In the overt pronoun condition the two participant groups preferred the object as the antecedent for the pronoun. The second choice of antecedent was the subject and the third the extra-sentential referent. The subject antecedent was chosen by the L2 learners slightly less frequently than by the native speakers(i.e. in 20% versus 24.10% of the cases).

Figure 2.

Responses in the forward anaphora conditions.

Figure 3 shows the distribution of responses in the two backward anaphora conditions. In the null pronoun condition both participant groups preferred the subject antecedent for the pronoun. The object and, especially, the extra-sentential antecedent, were chosen far less frequently. The overt pronoun condition was the only one in which the two participant groups expressed different antecedent preferences. More precisely, the native speakers interpreted the pronoun as coreferential with either the subject or the object antecedent, whereas the L2 learners were undecided between the extra-sentential and the object antecedent for the pronoun, showing a slight preference for the former. Importantly, the L2 learners selected the subject antecedent less often than the native speakers (i.e. in 22.08% versus 37.5% of the cases).

Figure 3.

Responses in the backward anaphora conditions.

To assess the significance of the trends shown in Figures 2 and 3, we performed the polytomous logistic regression analysis. We fitted the statistical model to the data using the polytomous package (Arppe, 2013) in the statistical software R (R Core Team, 2012). The outcome variable was the choice of response (subject, object, extra-sentential referent) and the predictor variables were pronoun type (null versus overt), anaphora type (forward versus backward) and participant group (L1 versus L2), together with three interactions we were interested in (pronoun type × anaphora type, pronoun type × participant group, anaphora type × participant group). Given that none of the three response choices could be treated as a meaningful baseline, the chosen polytomous regression method was one-versus-rest, meaning that each outcome was contrasted with the other two (grouped together) in a separate binary regression model (see Arppe, 2008, p. 120; Arppe, 2013, p. 22).⁸ The obtained parameters are shown in Table 2.

Table 2.

Summary of results of the statistical analysis.

Predictor variable	Odds of response selection
Predictor variable	Subject	Object	Extra-sentential referent
Intercept	1.266*	0.651*	0.0382*
Pronoun type (overt)	0.2682*	4.093*	0.8024
Anaphora type (backward)	1.998*	0.4203*	2.241*
Group (L2)	1.073	0.942	1.557
Pronoun type × Anaphora type	0.8433	0.5182*	5.081*
Pronoun type × Group	0.6297*	1.114	1.449
Anaphora type × Group	0.7919	0.9894	0.8535

Specifically, the model parameters are expressed in terms of estimated odds in favour of or against the choice of a particular response given the predictor variables. The Intercept line shows the odds contributed by aggregate default values of the predictor variables – in our case the L1 group’s responses in the forward anaphora with a null pronoun condition. Values <1 indicate that the chance of a specific outcome is decreased, while those >1 suggest that the chance of a specific outcome is increased. Significant values are marked with an asterisk.

As expected, the default case (i.e. forward anaphora with a null pronoun) increases the odds of subject selection (1.3:1) and decreases the odds for object (0.7:1), and especially extra-sentential referent, selection (0.04:1). The pronoun type has a significant impact on response choice, with an overt pronoun increasing the odds of object selection (4:1) and decreasing the odds of subject selection (0.3:1). Anaphora type is also a significant predictor of response choice: in backward anaphora the odds of subject (2:1) and extra-sentential referent (2.2:1) being chosen increase and the odds of object being chosen (0.4:1) decrease. In backward anaphora with an overt pronoun, the odds of extra-sentential referent selection (5:1) are increased and of object selection decreased (0.5:1). The participant group is not a significant predictor of response choice, nor is the interaction between anaphora type and participant group. The opposite is true for the interaction between pronoun type and participant group: when pronoun type = overt and group = L2, odds are decreased for subject selection (0.6:1). This is consistent with what can be seen in Figure 2 and especially Figure 3, and suggests that the subject is less likely to be chosen as antecedent for the overt pronoun by the L2 learners than by the native speakers. In summary, the odds provided by the model confirm the significance of the trends shown in Figures 2 and 3. Most importantly, they confirm that the two participant groups differ with respect to their selection of the subject antecedent for the overt pronoun, with the L2 learners choosing this response less often than the native speakers. We discuss these findings in the next section.

As for the overall performance and fit of the model, they are estimated via two measures, R_L²and Accuracy.⁹ In our case, the R_L² value of 0.1366 is not particularly high,¹⁰ but Accuracy has the value of 59.74% overall, which suggests that the predictions that our model makes are above chance. The values of Accuracy for individual responses are 70.55% for the subject, 55.03% for the object and 37.22% for the extra-sentential referent, indicating that the model is the most accurate in predicting the subject responses and the least accurate in predicting the extra-sentential referent responses.

We close this section by showing the distribution of responses in the four control conditions in Table 3. It can be seen that in all control conditions the two participant groups made correct responses in the majority of cases, confirming that the pictures were clear enough to be interpreted in the intended way. More precisely, in the gerundive clause conditions participants tended to choose the subject response, in the relative clause condition participants tended to choose the object response, and in the temporal clause condition, in which there was a third NP in the subordinate clause, participants tended to choose the response indicating this third referent most frequently. The fact that the temporal clause condition is associated with the lowest rate of correct responses for both participant groups might be due to the fact that sentences in this condition were the longest and, as such, probably the hardest of all the sentences in the experiment to retain in working memory.

Table 3.

Mean percentages of responses in the four control conditions.

Condition	Group	Referent choice
Condition	Group	Subject	Object	Other
Postposed gerundive clause	L1	82.99%	12.50%	4.51%
	L2	79.58%	16.25%	4.17%
Relative clause	L1	17.02%	79.51%	3.47%
	L2	20.00%	73.75%	6.25%
Preposed gerundive clause	L1	83.06%	13.13%	3.82%
	L2	77.50%	17.92%	4.58%
Temporal clause	L1	16.74%	14.03%	69.24%
	L2	9.17%	25.42%	65.42%

Discussion

The aim of this study was to test the abilities of 13–14-year-old highly proficient Croatian-speaking child L2 learners in the domain of forward and backward intra-sentential anaphora resolution in Italian. We predicted that the learners would not show a tendency to overgeneralise overt subject pronouns to contexts where null subject pronouns are required to a greater extent than the native speakers. We also predicted that they would interpret null subject pronouns in a native-like way.

These predictions were borne out in the study. In the case of null pronouns, the L2 learners preferred the same antecedent (i.e. the subject), and to the same degree, as the native speakers. This pattern held for both forward and backward anaphora. In the case of overt pronouns, the L2 learners preferred the same antecedent (i.e. the object), and to a similar degree, as the native speakers in forward anaphora, but they differed from the native speakers in backward anaphora. In backward anaphora, the L2 learners preferred the two non-subject antecedents (the extra-sentential one more strongly), whereas the preferences of the native speakers were split between the subject and the object antecedent. In both conditions with overt pronouns the L2 learners chose the subject antecedent less often than the native speakers, especially in backward anaphora. So, rather surprisingly, it was the native speakers and not the L2 learners who showed a tendency to overgeneralise overt pronouns to null pronoun contexts.

One possible explanation for this unexpected finding is that the L2 learners are experiencing acceleration in the acquisition of the discourse-pragmatic constraints on the use of overt subject pronouns in Italian due to the influence of their L1. Recall that even though 13–15-year-old Croatian monolinguals in Kraš (in preparation) interpreted the overt pronoun as coreferential with the subject antecedent more often than the adults in both forward and backward anaphora, they preferred a non-subject antecedent for the pronoun in both conditions. Importantly, the distribution of the three responses in backward anaphora for monolingual Croatian 13–15-year-olds was very similar to that for the L2 learners in the present study: the choice of the extra-sentential, object and subject antecedent accounted for 46.18%, 33.68% and 20.14% of responses, respectively (cf. Figure 3). The 13–15-year-old Croatian monolinguals in Kraš (in preparation) are thus interpreting the overt pronoun in backward anaphora in a more target-like way than the Italian monolinguals of the same age in the present study, the responses of monolingual adults in Kraš (2008b) and Kraš (2008a) being considered as the target for Croatian and Italian, respectively. The acquisition of the discourse-pragmatic constraints on the use of overt pronouns in Italian thus seems to be delayed compared to the acquisition of the same property in Croatian, especially in backward anaphora. This delay might be due to possible cross-linguistic differences between Italian and Croatian concerning some (as yet unspecified) properties of the linguistic system that are related to the distribution of overt pronouns in discourse. These properties may affect the rate of acquisition of the relevant discourse-pragmatic principles given that these principles are not acquired in isolation, but as part of the whole linguistic system. Even though this hypothesis requires further elaboration and empirical validation, we can certainly speculate that if Croatian monolinguals indeed acquire the discourse-pragmatic constraints on the use of overt pronouns earlier than Italian monolinguals (which in itself needs to be confirmed by additional studies), then the acquisition of these constraints in Italian might be accelerated in Croatian-speaking L2 learners compared to monolingual native speakers due to cross-linguistic influence in the form of L1 transfer. Note that cases of acceleration have been documented in simultaneous bilingualism (e.g. Patuto, Repetto, & Müller, 2011) and in bidialectal acquisition (e.g. Cornips & Hulk, 2006), and that cross-linguistic influence has been proposed as a possible cause of this phenomenon in some of these cases (e.g. Cornips & Hulk, 2006).

One potential problem for the hypothesis that the discourse-pragmatic constraints on the use of overt subject pronouns are acquired earlier in Croatian than in Italian is the findings of Serratrice (2007). In Serratrice’s study, 6–9-year-old Italian monolinguals interpreted the Italian overt pronoun in backward anaphora in a more target-like way than the 13–14-year-old monolinguals in our study, i.e. by choosing coreference with the object antecedent in most cases. However, given that the tasks used by Serratrice (2007) and the present study were not identical, the findings of thetwo studies cannot be compared directly and any inconsistencies between them should be treated with caution. In contrast, the present study, Kraš (2008a), Kraš (2008b) and Kraš (in preparation) all used the same task, only in two different languages, which makes a direct comparison of the findings of these studies more valid.

Let us now relate the results of the present study to the debate regarding the sources of instability at the discourse–syntax interface in bilingual grammars. Recall that based on the native-like patterns of interpretation of Italian null and overt subject pronouns in Croatian-speaking adult L2 learners, who had in fact acquired Italian as their L3, Kraš (2008a) argued that instability is primarily due to cross-linguistic influence and not the bilinguals’ suboptimal processing abilities. However, Sorace (2011) argued that the native-like behaviour of these learners might be explained under the hypothesis that L3 learners experience less resource competition in resolving anaphoric dependencies than L2 learners. In the present study we tested “true” L2 learners of Italian with Croatian as their L1 and found no evidence of greater instability at the discourse–syntax interface in these learners than in native-speaking controls. As a matter of fact, the learners demonstrated less instability in their interpretation of overt subject pronouns, especially in backward anaphora, than the controls. We have proposed that this might be due to the possibility that the L2 learners are transferring the discourse-pragmatic constraints on the use of overt pronouns (which are supposedly the same in their L1 and L2) from their L1, where these are acquired earlier than in their L2. We believe that L1 transfer is operative in the learners’ interpretation of null subject pronouns as well, but in this case it results in the absence of difference in the responses of the L2 learners and the controls under the assumption that the use of null pronouns in discourse is regulated in the same way in Croatian and Italian and that it is acquired at roughly the same age (or at least by the age of 13) in the monolingual development of these languages.

We believe that our finding, that a particular group of L2 learners whose L1 presumably does not differ from the L2 in the interface domain under investigation did not exhibit greater instability at the discourse–syntax interface compared with a group of native-speaking controls, strengthens our position that cross-linguistic influence might be the main reason for the non-target-like behaviour of the L2 learners whose L1 does differ from the L2 in the relevant domain. If instability was caused by the learners’ suboptimal processing abilities, then it should be manifested by all L2 learners, regardless of the properties of their L1. Note that we do not deny the possibility that some sort of instability might be present in the on-line behaviour of all L2 learners regardless of their L1 background.¹¹ However, evidence concerning the on-line behaviour of L2 learners in relation to anaphora resolution need not have a direct bearing on explaining the patterns discovered in the learners’ off-line behaviour.

Conclusion

This study has provided experimental data on the abilities of two groups of 13–14-year-olds, monolingual native speakers and L1 Croatian highly proficient child L2 learners, in the domain of intra-sentential forward and backward anaphora resolution in Italian. The data show that both groups of speakers obey the discourse-pragmatic constraints on the resolution of this kind of anaphora with null subject pronouns. The data also show that, possibly as a result of L1-driven acceleration, the L2 learners obey these constraints even more consistently than the native speakers when it comes to overt subject pronouns. The patterns of antecedent assignment shown by the child L2 learners are fully consistent with those displayed by the adult native speakers and by L1 Croatian near-natives who have acquired Italian as the L3 in Kraš (2008a). Target-like behaviour of Croatian-speaking L2 learners in these two studies stands in sharp contrast to that exhibited by L1 English near-natives in Belletti et al. (2007) and Sorace and Filiaci (2006), who showed a tendency to overgeneralise overt subject pronouns to contexts where null subject pronouns are required. Based on this discrepancy, we argue that instability at the discourse–syntax interface demonstrated by English-Italian bilinguals is primarily due to cross-linguistic influence and not their hypothesised suboptimal processing abilities.

We do not, however, believe that our findings can settle the issue concerning the sources of instability at the discourse–syntax interface in bilingual grammars. Our argumentation in the present paper is primarily based on the data from highly proficient bilinguals of two language combinations, English-Italian and Croatian-Italian, whose abilities were tested in relation to the same interface phenomenon and by means of very similar methodology. To gain a deeper insight into this issue we need data on a wider range of phenomena, coming from bilinguals of a larger number of language combinations and obtained by a greater variety of methodologies, both experimental and naturalistic.¹² Considering the data available, there is still a long way to go.

Footnotes

Acknowledgements

I wish to thank Maja Miličević for helping me with the statistical analysis, Dario Ban for his help with the data collection, and Catherine Dickie and Madeleine Beveridge for proofreading the paper. I am also grateful to three anonymous reviewers and the editors for their very helpful comments and suggestions. Special thanks are due to the children and staff of “Belvedere”, “Dolac”, “Gelsi” and “San Nicolò” primary schools in Rijeka and the “Carlo Stuparich” lower secondary school in Trieste, without whose collaboration this study would not have been possible.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Notes

Author biography

Tihana Kraš is Assistant Professor in the Department of English Language and Literature in the Faculty of Humanities and Social Sciences at the University of Rijeka. Her research focuses on second language acquisition, first language acquisition (including simultaneous bilingualism and atypical language development) and first and second language processing.

References

Arppe

(2008). Univariate, bivariate and multivariate methods in corpus-based lexicography: A study of synonymy. Unpublished doctoral dissertation, University of Helsinki.

Arppe

(2013). Package ‘polytomous’: Polytomous logistic regression for fixed and mixed effects (version 0.1.4). The R Project for Statistical Computing. Retrieved from http://cran.r-project.org/web/packages/polytomous/index.html

Belletti

Bennati

Sorace

(2007). Theoretical and developmental issues in the syntax of subjects: Evidence from near-native Italian. Natural Language and Linguistic Theory, 25, 657–689.

Carminati

M. N.

(2002). The processing of Italian subject pronouns. Unpublished doctoral dissertation, University of Massachusetts at Amherst.

Cornips

Hulk

(2006). External and internal factors in bilingual and bidialectal language development: Grammatical gender of the Dutch definite determiner. In Claire

White

Jourdan

(Eds.), L2 acquisition and creole genesis: Dialogues (pp. 355–378). Amsterdam, The Netherlands/Philadelphia, PA: John Benjamins.

Costa

Santesteban

(2004). Lexical access in bilingual speech production: Evidence from language switching in highly proficient bilinguals and L2 learners. Journal of Memory and Language, 50, 491–511.

Han

Arppe

Newman

(2013). Topic marking in a Shanghainese corpus: From observation to prediction. Corpus Linguistics and Linguistic Theory. Epub ahead of print 5 June 2013. doi:101515/cllt-2013-0014.

Kraš

(2008a). Anaphora resolution in near-native Italian grammars: Evidence from native speakers of Croatian. EUROSLA Yearbook, 8, 107–134.

Kraš

(2008b). Anaphora resolution in Croatian: Psycholinguistic evidence from native speakers. In Tadić

Dimitrova-Vulchanova

Koeva

(Eds.), Proceedings of the Sixth International Conference “Formal Approaches to South Slavic and Balkan Languages” (pp. 67–72). Zagreb, Croatia: Croatian Language Technologies Society – Faculty of Humanities and Social Sciences.

10.

Kraš

(in preparation). Resolution of forward and backward anaphora with null and overt subject pronouns in Croatian by monolingual children.

11.

Kraš

Stipeć

(2013). Interpretation of ambiguous subject pronouns in Croatian by people with Down syndrome and typically developing children. In Baiz

Goldman

Hawkes

(Eds.), Proceedings of the 37th annual Boston University Conference on Language Development (pp. 178–190). Somerville, MA: Cascadilla Press.

12.

Kupisch

(2014). Adjective placement in simultaneous bilinguals (German–Italian) and the concept of cross-linguistic overcorrection. Bilingualism: Language and Cognition, 17, 222–233.

13.

Lozano

(2009). Selective deficits at the syntax-discourse interface: Evidence from the CEDEL2 corpus. In Snape

Leung

Y. I.

Sharwood Smith

(Eds.), Representational deficits in SLA: Studies in honor of Roger Hawkins (pp. 128–166), Amsterdam, The Netherlands/Philadelphia, PA: John Benjamins.

14.

Meisel

J. M.

(2009). Second language acquisition in early childhood. Zeitschrift für Sprachwissenschaft, 28, 5–34.

15.

Montrul

(2004). Subject and object expression in Spanish heritage speakers: A case of morphosyntactic convergence. Bilingualism: Language and Cognition, 7, 125–142.

16.

Montrul

Dias

Thomé-Williams

(2008). Subject expression in the non-native acquisition of Brazilian Portuguese. In Pires

Rothman

(Eds.), Minimalist enquiries into child and adult language acquisition: Case studies across Portuguese (pp. 301–325). Berlin, Germany/New York, NY: Mouton de Gruyter.

17.

Montrul

Rodríguez Louro

(2006). Beyond the syntax of the null subject parameter: A look at the discourse-pragmatic distribution of null and overt subjects by L2 learners of Spanish. In Torrens

Escobar

(Eds.), The acquisition of syntax in Romance languages (pp. 401–418). Amsterdam, The Netherlands/Philadelphia, PA: John Benjamins.

18.

Paradis

Genesee

Crago

(2011). Dual language development and disorders: A handbook on bilingualism and second language learning. Baltimore, MD: Brookes Publishing.

19.

Paradis

Navarro

(2003). Subject realization and cross-linguistic interference in the bilingual acquisition of Spanish and English: What is the role of input? Journal of Child Language, 30, 1–23.

20.

Patuto

Repetto

Müller

(2011). Delay and acceleration in bilingual first language acquisition: The same or different. In Rinke

Kupisch

(Eds.) The development of grammar: Language acquisition and diachronic change; In honour of Jürgen M. Meisel (pp. 231–261). Amsterdam, The Netherlands/Philadelphia, PA: John Benjamins.

21.

Pérez-Leroux

A. T.

Glass

W. R.

(1999). Null anaphora in Spanish second language acquisition: Probabilistic versus generative approaches. Second Language Research, 15, 220–249.

22.

R Core Team. (2012). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. Retrieved from http://www.R-project.org/

23.

Roberts

Gullberg

Indefrey

(2008). On-line pronoun resolution in L2 discourse: L1 influence and general learner effects. Studies in Second Language Acquisition, 30, 333–357.

24.

Rothman

(2009). Pragmatic deficits with syntactic consequences: L2 pronominal subjects and the syntax-pragmatics interface. Journal of Pragmatics, 41, 951–973.

25.

Schwartz

B. D.

(2004). Why child L2 acquisition? In Van Kampen

Baauw

(Eds.), Proceedings of generative approaches to language acquisition 2003 (pp. 47–66). Utrecht, The Netherlands: LOT Occasional Series.

26.

Serratrice

(2007). Cross-linguistic influence in the interpretation of anaphoric and cataphoric pronouns in English-Italian bilingual children. Bilingualism: Language and Cognition, 10, 225–238.

27.

Serratrice

Sorace

Paoli

(2004). Subjects and objects in Italian-English bilingual and monolingual acquisition. Bilingualism: Language and Cognition, 7, 183–206.

28.

Sorace

(2000). Differential effects of attrition in the L1 syntax of near-native L2 speakers. In Howell

S. C.

Fish

S. A.

Keith-Lucas

(Eds.), Proceedings of the 24th annual Boston University Conference on Language Development (pp. 719–725). Somerville, MA: Cascadilla Press.

29.

Sorace

(2011). Pinning down the concept of “interface” in bilingualism. Linguistic Approaches to Bilingualism, 1, 1–33.

30.

Sorace

Filiaci

(2006). Anaphora resolution in near-native speakers of Italian. Second Language Research, 22, 339–368.

31.

Sorace

Serratrice

Filiaci

Baldo

(2009). Discourse conditions on subject pronoun realization: Testing the linguistic intuitions of older bilingual children. Lingua, 119, 460–477.

32.

Tsimpli

I. M.

Sorace

(2006). Differentiating interfaces: L2 performance in syntax-semantics and syntax-discourse phenomena. In Bamman

Magnitskaia

Zaller

(Eds.), Proceedings of the 30th annual Boston University Conference on Language Development (pp. 653–664). Somerville, MA: Cascadilla Press.

33.

Tsimpli

I. M.

Sorace

Heycock

Filiaci

(2004). First language attrition and syntactic subjects: A study of Greek and Italian near-native speakers of English. International Journal of Bilingualism, 8, 257–277.

34.

Unsworth

(2005). Child L2, adult L2, child L1: Differences and similarities. A study on the acquisition of direct object scrambling in Dutch. Unpublished doctoral dissertation, Utrecht University, The Netherlands.

35.

Unsworth

(2013). Assessing age of onset effects in (early) child L2 acquisition. Language Acquisition, 20, 74–92.

36.

Wilson

Sorace

Keller

(2009). Antecedent preferences for anaphoric demonstratives in L2 German. In Chandlee

Franchini

Lord

Rheiner

G.-M.

(Eds.), Proceedings of the 33rd annual Boston University Conference on Language Development (pp. 634–645). Somerville, MA: Cascadilla Press.