When comprehending sentences, speakers use different types of information as cues to the correct interpretation. It is well known that cues may be used differently by nonnative speakers as compared to native speakers, but there is no consensus on how to best describe and explain these differences. In particular, some approaches predict that learners overrely on discourse-level cues compared to syntactic cues, and that this overreliance is independent of properties of the first language (L1; Clahsen & Felser, Reference Clahsen and Felser2006; Klein & Perdue, Reference Klein and Perdue1997). Others see the main difficulty of second language (L2) sentence comprehension in the need to integrate several cues, in particular when these cues differ from the cues relevant in the L1 (Sorace, Reference Sorace2000, Reference Sorace, Long and Doughty2003, Reference Sorace, Cornips and Corrigan2005). Against this background, the current study investigates the offline interpretation of different types of pronominal subjects as a test case for the use and integration of different types of cues in native and nonnative speakers.
The paper is structured as follows. We first summarize what is known about native speakers’ interpretation of different types of pronominals. We then present different assumptions about the use of cues in nonnative sentence comprehension and show that, although previous studies on nonnative pronoun resolution have contributed to testing these assumptions, there are still open questions. The French and Turkish structures tested in the current study are introduced. Finally, we present results from native speakers of French and Turkish and Turkish learners of French and conclude that, at least under certain conditions, L2 learners may overrely on discourse-level cues in the interpretation of pronominals, even if syntactic cues could be used in a similar way as in the L1.
BACKGROUND
The Interpretation of Ambiguous Pronominal Forms by Native Speakers
Pronominal forms are used to refer to previously introduced discourse referents. Very often there is more than one potential antecedent in the discourse context such that the pronoun is ambiguous. It has been shown that the interpretation of such ambiguous pronominals can, in most cases, not be explained by one factor (see, e.g., Kaiser & Trueswell, Reference Kaiser and Trueswell2008; Kehler & Rohde, Reference Kehler and Rohde2013). Rather, speakers use a range of different cues to determine the antecedent of an ambiguous pronoun. Cues are classified as syntactic when syntactic knowledge is necessary to use them, as is the case for the cue subjecthood. Other important cues for pronoun resolution are situated on the discourse level. Importantly, across languages, pronominal subjects of finite clauses are interpreted differently from implicit pronominal subjects of nonfinite clauses (PRO). Moreover, pronominal subjects of finite clauses are interpreted differently in pro-drop languages such as Turkish and non–pro-drop languages such as French. In the following, what is known about the cues used in these different configurations by native speakers is briefly summarized.
Ambiguous Pronominal Subjects of Finite Clauses in Pro-drop and Non–pro-drop Languages.
It has often been claimed that the choice and interpretation of pronominal forms that appear in finite clauses is directly linked to the salience level of the antecedents: The more reduced the form, the more likely it will be interpreted as referring to the most salient antecedent (Ariel, Reference Ariel1990; Givón, Reference Givón1983; Gundel, Hedberg, & Zacharski, Reference Gundel, Hedberg and Zacharski1993). What the most reduced form is thereby depends on the pronominal system of a language: In non–pro-drop languages, the most reduced forms are unstressed personal pronouns in most contexts, such as he in the English sentence in (1). In pro-drop languages, overt pronominal forms are also available but can be dropped when they occur in subject position such that the most reduced form in this position is a null pronoun, glossed as pro in the Turkish equivalent in (2).
(1) Erik stabbed Laurent when he worked in Rome.
(2) Ergin Latif’-i pro Roma’-da çalış-tığ-ı zaman bıçakla-dı.
Ergin Latif- acc pro Rome- locwork- nmlz-3sgtime stab- pst
It is interesting to note that the cues that speakers use to determine the most salient antecedent for the most reduced form seem to differ in these two types of languages. For unstressed personal pronouns in non–pro-drop languages, subjecthood has been shown to be influential such that subject antecedents are preferred over nonsubject antecedents (see, e.g., Crawley, Stevenson, & Kleiman, Reference Crawley, Stevenson and Kleinman1990; Frederiksen, Reference Frederiksen1981, for English subject pronouns). Evidence from word order variations in non–pro-drop Finnish suggests that subjecthood has an influence in addition to the discourse-level order of mention cue (Järvikivi, van Gompel, Hyönä, & Bertram, Reference Järvikivi, van Gompel, Hyönä and Bertram2005, but see Bouma & Hopp, Reference Bouma, Hopp, Artstein and Poesio2006, for conflicting evidence from German). Several other discourse-level cues influence the choice of antecedent as well. For instance, referents that are topicalized by repeated mention in the context are more accessible than nontopical antecedents (Cowles, Walenski, & Kluender, Reference Cowles, Walenski and Kluender2007), and implicit causal relations and other coherence relations also have a strong influence on interpretation (Garvey, Caramazza, & Yates, Reference Garvey, Caramazza and Yates1975; Kehler, Reference Kehler2002; McKoon, Greene, & Ratcliff, Reference McKoon, Greene and Ratcliff1993; Rohde, Kehler, & Elman, Reference Rohde, Kehler, Elman and Sun2006; Sanders & Noordman, Reference Sanders and Noordman2000).
Studies on the interpretation of null pronouns in pro-drop languages report that, as in non–pro-drop languages, subject antecedents are preferred over nonsubject antecedents (Carminati, Reference Carminati2002, for Italian; Alonso-Ovalle, Fernández-Solera, Frazier, & Clifton, Reference Alonso-Ovalle, Fernández-Solera, Frazier and Clifton2002; Gelormini-Lezama & Almor, Reference Gelormini-Lezama and Almor2011; Jegerski, VanPatten, & Keating, Reference Jegerski, VanPatten and Keating2011, for Spanish; Mayol, Reference Mayol2010, for Catalan). In contrast to non–pro-drop languages, however, there is evidence that discourse-level cues have little weight compared to subjecthood. Carminati (Reference Carminati2002) found that null subjects in Italian systematically prefer the subject of the antecedent sentence, situated in the specifier position of the inflectional phrase (specIP), independent of discourse-level properties of this antecedent. On the basis of this finding, she proposes the position of the antecedent hypothesis (PAH), according to which the syntactic position of the antecedent is the most important cue to the interpretation of ambiguous null pronouns in Italian. Conversely, overt pronouns should prefer antecedents that are not in specIP according to the PAH.Footnote 1 Converging evidence for the subject preference of null forms comes from Mayol (Reference Mayol2010), who found a preference for syntactic subjects for Catalan null pronouns but no influence of order of mention (first- or second-mentioned subjects), and Jegerski and colleagues (2011), who found that discourse structure influenced the interpretation of English overt pronouns but did not influence the interpretation of Spanish null pronouns. There is one study, however, that did find an influence of dislocation and clefting of antecedents on the interpretation of Spanish null pronouns (de la Fuente & Hemforth, Reference de la Fuente, Hemforth, Cabrelli Amaro, Lord, de Prada Pérez and Aaron2013), suggesting that other cues may play a role as well, at least in some cases. Moreover, there is no similarly converging evidence on the cues used for the interpretation of overt pronouns in pro-drop languages. Many studies report that these pronouns are highly ambiguous and, in particular, do not necessarily show the strong antisubject effect predicted by the PAH (Alonso-Ovalle et al., Reference Alonso-Ovalle, Fernández-Solera, Frazier and Clifton2002; Belletti, Bennati, & Sorace, Reference Belletti, Bennati and Sorace2007; Gelornini-Lezama & Almor, Reference Gelormini-Lezama and Almor2011; Jegerski et al., Reference Jegerski, VanPatten and Keating2011; Mayol, Reference Mayol2010). In sum, the available evidence suggests that subjecthood is a particularly important cue for null pronouns in pro-drop languages and that the interpretation of this type of pronoun seems to be less susceptible to discourse-level influences than that of overt pronouns in non–pro-drop languages.
Ambiguous Null Subjects (PRO) of Nonfinite Clauses.
Many syntactic theories assume that subjectless nonfinite dependent clauses contain a null subject that is often designated with the label PRO (following Chomsky, Reference Chomsky1981). PRO is controlled by an element of the matrix clause. How control is determined depends (along with other factors) on the grammatical function of the nonfinite clause. The current study is concerned with nonfinite clauses that are adverbial modifiers, as in the English and French present participle constructions exemplified in (3) (from Kortmann, Reference Kortmann, Haspelmath and König1995) and (4).
(3) Mary left PRO smiling.
(4) Pierre a giflé Jean PRO étant jeune.
“Peter slapped John PRO being young.”
In prescriptive grammars of several languages, it is stated that these kinds of modifiers should only be used when they refer to the subject of the matrix clause (e.g., Lynch, Reference Lynch2007, for English; Grévisse, Reference Grévisse1986, for French). Descriptive linguists note that subject control is indeed the standard case for PRO in nonfinite modifiers but point out that cases of object control seem to exist as well (see Haspelmath, Reference Haspelmath, Haspelmath and König1995; Haug, Fabricius-Hansen, Behrens, & Helland, Reference Haug, Fabricius-Hansen, Behrens, Helland, Fabricius-Hansen and Haug2012). Haspelmath (Reference Haspelmath, Haspelmath and König1995) proposes that object control is possible in most languages when the matrix object is more discourse prominent than the subject. We are not aware of experimental studies verifying this claim in potentially ambiguous cases of nonfinite modifiers, such as (4), and for the languages investigated in the current study. In sum, the prescriptive and descriptive literature suggests that subjecthood of the antecedent should be a particularly important cue to the interpretation of ambiguous PRO in nonfinite modifier clauses, but discourse-level cues have been suggested to be influential as well. Although the interpretation of PRO may vary between languages and constructions, we see no reason to assume a relation to the pro-drop property: We expect, thus, that speakers of pro-drop and non–pro-drop languages will be influenced primarily by subjecthood in this case.
The Interpretation of Ambiguous Pronominal Forms by Nonnative Speakers
Approaches to Nonnative Sentence Comprehension.
Many approaches to L2 acquisition assume that nonnative speakers use comprehension cues differently from native speakers. Theories make different predictions, however, as to the exact conditions under which deviations from nativelike cue use are assumed to occur.
The interface hypothesis predicts that difficulties will occur in particular when the interpretation of a given form is not determined by a purely syntactic mechanism, such that semantic or discourse-level cues have to be integrated for successful interpretation (Sorace, Reference Sorace2000, Reference Sorace, Long and Doughty2003, Reference Sorace, Cornips and Corrigan2005). According to Sorace and Filiaci (Reference Sorace and Filiaci2006), the resolution of pronouns in non–pro-drop as well as pro-drop languages presents such a case. In particular, they note that in pro-drop languages, null subject pronouns referring to object antecedents are not ungrammatical but, rather, inappropriate suggesting that the subject preference reflects the dominance of one particular cue rather than a purely syntactic mechanism. In addition, Sorace (Reference Sorace2000, Reference Sorace, Long and Doughty2003, Reference Sorace, Cornips and Corrigan2005) assumes that, among the cases in which several cues have to be combined, particular difficulties are to be expected when cues are used differently in the L1 as compared to the L2.
Other approaches assume that difficulties are not so much caused by the need to integrate different cues or by interference from the L1 but by the use of syntactic cues specifically. Clahsen and Felser (Reference Clahsen and Felser2006) claim that, at least during online processing, L2 learners may have difficulties with computing structural representations and, as a consequence, may overrely on semantic and discourse-level cues. The learner varieties approach to L2 acquisition predicts that the acquisition of syntactic knowledge in a L2 may be particularly difficult and that L2 learners at least at early stages of acquisition would therefore be influenced by semantic and discourse-level factors to a larger degree than native speakers in production (Dimroth, Gretsch, Jordens, Perdue, & Starren, Reference Dimroth, Gretsch, Jordens, Perdue, Starren, Dimroth and Starren2003; Klein & Perdue, Reference Klein and Perdue1997) as well as during online comprehension (Schimke, Reference Schimke2009; Verhagen, Reference Verhagen2009).
In sum, approaches differ in what they predict to be the main difficulty in L2 sentence comprehension. Whereas some approaches assume that the use of syntactic cues is particularly difficult, others see the main difficulty in the integration of different cues and predict this process to be vulnerable to L1 influence. The following section summarizes studies that have tested the interpretation of pronominal forms as a way to investigate the use of different cues in L2 sentence comprehension.
Evidence from Nonnative Pronoun Resolution.
Studies using online techniques have shown that learners might differ from native speakers during the process of pronoun resolution. More precisely, learners have been found (a) to be more strongly influenced by ambiguity than native speakers, leading to relatively longer processing times for ambiguous as compared to umambiguous pronouns (Roberts, Gullberg, & Indefrey, Reference Roberts, Gullberg and Indefrey2008), and (b) to consider more potential antecedents than native speakers when processing reflexive pronouns (Felser & Cunnings, Reference Felser and Cunnings2012). Other studies have shown that nonnative speakers may differ even in their final interpretations of pronominals. In two studies testing the interpretation of Italian null and overt pronouns by English near-native speakers of Italian, Belleti and colleagues (2007) and Sorace and Filiaci (Reference Sorace and Filiaci2006) found that both native and nonnative speakers interpreted null subject pronouns preferentially as referring to subject antecedents. The nonnative speakers differed from the natives, however, in more often interpreting overt pronouns as referring to subject antecedents as well. The results were interpreted as evidence in favor of the interface hypothesis. In particular, Belleti and colleagues (2007) and Sorace and Filiaci (Reference Sorace and Filiaci2006) assume that learners have difficulties with overt pronouns because of interference from English, in which overt pronouns are used in contexts where null pronouns would be used in Italian. Converging evidence comes from Gürel (Reference Gürel2002), who tested the interpretation of overt pronouns, null pronouns, and the reflexive pronoun kendisi “himself” as subjects of embedded clauses by near-native English learners of Turkish. She found that learners deviated from the target pattern in sometimes interpreting the overt subject o “he” of an embedded clause as referring to the matrix subject, an interpretation that is impossible for native speakers of Turkish (see Pronominal Forms in French and Turkish Subordinate Constructions section). Gürel (Reference Gürel2002) explains this finding as due to transfer from the English overt pronoun to the Turkish overt pronoun.
Although all of these results are in line with the interface hypothesis, they are also compatible with other approaches. In particular, the preference that learners exhibited for first-mentioned subject antecedents could be due to a general preference for discourse-prominent antecedents. The study by Jegerski and colleagues (2011), mentioned previously, provides evidence for a strong influence of a discourse-level cue in a L2 group. Similar to Belleti and colleagues (2007) and Sorace and Filiaci (Reference Sorace and Filiaci2006), Jegerski and colleagues studied the resolution of null and overt pronouns in a pro-drop language (Spanish), by learners coming from a non–pro-drop language (English). Contrary to the earlier studies, however, Jegerski and colleagues manipulated discourse structure in that the clause that contained the ambiguous pronoun was either in a subordinate or coordinate discourse relationship with the clause containing the possible antecedents (following segmented discourse representation theory, as presented in Asher & Vieu, Reference Asher and Vieu2005). They showed that native speakers of English resolved overt English pronouns in coordinated segments more often to the previous subject than was the case for pronouns in subordinated segments. No such effect appeared for Spanish native speakers for Spanish null or overt pronouns. Interestingly, the English learners of Spanish showed the same discourse-structure effect that had been observed in native English when resolving Spanish null and overt pronouns. This suggests an overreliance on discourse-level cues. It may be the case, however, that the non–pro-drop source language, English, has introduced a bias toward using discourse-level cues in pronoun interpretation in the L2. Finally, Okuma (Reference Okuma, Herschensohn and Tanner2011) tested the interpretation of null and overt Japanese subject and object pronouns by English learners. In Japanese, topichood can be marked by a particle, making it possible to test whether topic marking of the subject or the object influences interpretation preferences of native and nonnative speakers. For null pronouns, Japanese native speakers showed a subject preference that was not influenced by topic marking, in line with the predictions of the PAH for Italian. Learners, however, showed a less robust subject preference and evidence for an influence of discourse-level cues. More precisely, they interpreted null pronouns more often as referring to the subject when the subject was topic marked than when it was not topic marked and more often as referring to the object when the object was first mentioned than when it was second mentioned. Although these results are in line with the idea that learners make more use of discourse-level cues than native speakers, the author reports statistical tests only for between-group differences and does not report whether the aforementioned effects were statistically reliable. Moreover, the possibility again cannot be excluded that the non–pro-drop source language, English, introduced a bias toward the use of discourse-level cues.
In sum, although the available evidence suggests that learners sometimes interpret pronouns in a nonnativelike way, the origins of this target-deviant behavior are still poorly understood. In particular, the available results do not allow for dissociating a potential overreliance of learners on discourse-level cues from transfer effects from the L1. Such dissociation has not been possible because studies so far have concentrated on learners coming from non–pro-drop languages, in which discourse-level cues play an important role in pronoun interpretation. The current study therefore aims to complement the available evidence by testing whether learners from a pro-drop language (Turkish) are sensitive to discourse-level effects in their non–pro-drop L2 (French). We target (a) pronominal subjects of finite clauses, a structure for which French differs from Turkish in that discourse-level cues are expected to play a greater role, and (b) structures containing cases of PRO, for which subjecthood is expected to be the dominant cue in both languages.
THE CURRENT STUDY
In what follows, the relevant structures of the two languages under investigation as well as our experimental manipulations are introduced before results from native and nonnative speakers are presented.
Pronominal Forms in French and Turkish Subordinate Constructions
The current study tests pronominal forms in finite and nonfinite subordinate clauses. The finite subordinate structure for French is illustrated in (5). As mentioned previously, French is a non–pro-drop language in which clitic personal pronouns, as in (5), are the most common form of pronominal reference. There is a series of strong pronouns that, contrary to the clitics, can receive stress, but these pronouns are used rarely in written language except for in specific positions, such as following prepositions.
(5) Eric a poignardé Laurent quand il travaillait à Rome.
“Eric stabbed Laurent when he worked in Rome.”
In pro-drop languages, null subject pronouns of finite clauses (pro) alternate with overt pronouns. In Turkish main clauses, the overt pronoun o “he” can thus occur instead of the dropped pronoun (pro) in certain contexts (Enç, Reference Enç, Slobin and Zimmer1986). The current study tested null subjects of subordinate clauses, such as in (2), repeated in (6).
(6) Ergin Latif’-i pro Roma’-da çalış-tığ-ı zaman bıçakla-dı.
Ergin Latif- acc pro Rome- locwork- nmlz-3sgtime stab- pst
“Erik stabbed Laurent when he worked in Rome.”
In subordinate clauses, there is no similar alternation between the dropped and overt pronoun in most cases because the use of o is not possible when reference to an intrasentential antecedent is intended (Kornfilt, Reference Kornfilt1997, p. 135). There is a potential alternative overt form, the reflexive pronoun kendisi, which can be used instead of the null form in (6), but this does not lead to a different interpretation of the sentence in that both pro and kendisi usually corefer with the matrix subject (Gürel, Reference Gürel2002). Finally, note that Turkish subordinate clauses are nominalized. Nevertheless, in line with Gürel (Reference Gürel2002), we assume the null form in (6) to represent a case of a dropped pronoun (pro) and not of an infinitival PRO because, in spite of nominalization, the subordinate verb is marked for subject-verb agreement.
The second type of structure investigated in the current study involves nonfinite modifier clauses, such as in (4), repeated in (7) in French.
(7) Pierre a giflé Jean PRO étant jeune.
“Peter slapped John PRO being young.”
The Turkish equivalent is given in (8).
(8) Sait Can’-ı PRO genç-ken tokatla-dı.
Sait Can- acc PRO young-when slap- pst
In (8), the clause consisting of the implicit subject and the predicative “young” is subordinated by means of the suffix -(y)ken. The element y that surfaces in this suffix whenever the preceding element ends with a vowel has been analyzed as originating from a copula form (Kornfilt, Reference Kornfilt1997, p. 80). Despite this verbal origin, the suffix -(y)ken cannot be marked for subject-verb agreement. For this reason, we consider the clause to be nonfinite and its implicit subject to represent a case of PRO.
On the basis of the evidence summarized in the introduction, we predict, for native speakers, discourse-level cues to have a stronger influence on the interpretation of (5) than of its equivalent in (6) but do not expect a different influence in (7) compared to (8), in which subjecthood is expected to be an important cue in both languages. As for L2 learners, if they were influenced by subjecthood and not by discourse-level cues in the interpretation of pronominal subjects of both finite and nonfinite clauses, this would provide evidence against an overreliance on discourse-level cues in L2 learners and might plausibly be explained by L1 influence. If, in contrast, learners rely on discourse-level cues in the L2, even in cases in which these cues play no role in parallel structures in the L1, this would provide evidence for a general overreliance on this type of cue in L2 learners.
Manipulations
We investigated the influence of discourse-level cues on the resolution of ambiguous pronouns in structures of the type illustrated in (5) through (8) by manipulating two different factors. The manipulations always concerned the first-mentioned referent of the matrix clause (abbreviated N1). The first factor was whether N1 was the agent or the patient of the transitive action described in the sentence. We assume that agent-initial structures are canonical structures for the agentive verbs used in our stimuli sentences. By contrast, we consider the patient-initial sentences to be marked and assume that this word order is used in cases in which the patient has an important role in the context in which the sentence is uttered (Comrie, Reference Comrie, Cole and Sadock1977; Desclés, Guentcheva, & Shaumyan, Reference Desclés, Guentcheva and Shaumyan1985; Givón, Reference Givón1979; Prévost, Reference Prévost, Apothéloz, Combettes and Neveu2009; Shibitani, Reference Shibitani1985; Tannenbaum & Williams, Reference Tannenbaum and Williams1968; Thompson, Reference Thompson, Channon and Shockey1978, among others). As a consequence, if participants resolve pronouns more toward the first-mentioned antecedent when its appearance in this position is marked (patient-initial sentences) than when it is unmarked (agent-initial sentences), this would show that they are sensitive to discourse prominence. Second, we manipulated whether N1 was dislocated from the rest of the sentence. Such dislocation structures are considered markers of topicality and are particularly frequent in French but also appear in other languages (Degand & Simon, Reference Degand, Simon, Stede, Chiarcos, Grabski and Lagerwerf2005; Lacheret-Dujour & François, Reference Lacheret-Dujour and François2004; Lambrecht Reference Lambrecht1981; Prévost, Reference Prévost, Apothéloz, Combettes and Neveu2009; Shaer, Cook, Frey, & Maienborn, Reference Shaer, Cook, Frey and Maienborn2008, among others). Again, if participants resolve the pronoun more toward N1 when it is dislocated than when it is not dislocated, this would show that they are sensitive to discourse prominence. Alternatively, if participants do not react to our manipulations and instead always prefer a referent with a particular grammatical role, this would provide evidence against reliance on discourse prominence and in favor of a strong influence of a syntactic cue on resolution preferences.
Data Analyses
Data from all experiments were analyzed with logistic mixed-effect models using R (R Development Core Team, 2012) and the R packages lme 4 (Bates, Maechler, & Bolker, Reference Bates, Maechler and Bolker2012) and languageR (Baayen, Reference Baayen2011). The dependent variable was whether participants chose the first (N1) or the second noun (N2) mentioned as the antecedent of the ambiguous pronominal form. For all datasets, we first computed a maximal model including word order (agent-initial, patient-initial), dislocation (dislocation, no dislocation), and their interaction as fixed effects and participants and items as random effects (this will be referred to as the full model in what follows). Random slopes for the experimental factors had to be excluded from all models to avoid convergence failure. Second, we computed a simpler model (referred to as the simple model in what follows) with only the main effects and no interaction and compared the two models by log-likelihood ratio tests (see, e.g., Barr, Levy, Scheepers, & Tily, Reference Barr, Levy, Scheepers and Tily2013). This comparison tested whether the full model represented a significantly better fit to the data than the simple model. From the six experiments to be reported, the full model was selected as the best fit to the data for Experiments 1, 3, and 5. For Experiments 4 and 6, the simple model was selected because the full model was not a significantly better data fit. Finally, we also selected the simple model for Experiment 2, for which the full model did not converge. We report the estimate (ß), standard error, z value, and p value for all fixed-effect predictors included in the selected model for the respective dataset. We report results first for native speakers’ interpretation of pronominal subjects of finite clauses, then for native speakers’ interpretation of PRO, and finally for the Turkish learners’ interpretation of pronominal forms in these two cases in French.
Native Speakers’ Interpretation of Pronominal Subjects of Finite Clauses
Experiments 1 and 2 tested how native speakers of French (Experiment 1) and Turkish (Experiment 2) interpret subject pronouns in finite subordinate clauses.
Experiment 1
Experimental Factors and Predictions.
The two experimental factors that we manipulated resulted in four different conditions, illustrated in (9a–d). As both subjects and objects can be dislocated, we could manipulate whether the first-mentioned noun was the agent or the patient in simple active sentences in the dislocation conditions (i.e., the conditions shown in [9c] and [9d]). As word order cannot be changed in French simple clauses, such as in the condition illustrated in (9a), we resorted to a passive sentence to achieve a patient-initial structure in sentence (9b). This means that, although the thematic role changes in a systematic way throughout our manipulations, there are three conditions in which N1 is the syntactic subject and one condition where N1 is the syntactic object.
(9) a. N1 agent, not dislocated (SagentVO)Footnote 2
Eric a poignardé Laurent quand il travaillait à Rome.
………………………………..travaillait à Rome.
“Eric stabbed Laurent when he worked in Rome.”
………………………………..worked in Rome.
b. N1 patient, not dislocated (SpatientVO)
Eric a été poignardé par Laurent quand il travaillait à Rome.
“Eric has been stabbed by Laurent when he worked in Rome.”
c. N1 agent, dislocated (Sagent, sVO)
Eric, il a poignardé Laurent quand il travaillait à Rome.
“Eric, he stabbed Laurent when he worked in Rome.”
d. N1 patient, dislocated (Opatient, SoV)
Eric, Laurent l’a poignardé quand il travaillait à Rome.
“Eric, Laurent stabbed him when he worked in Rome.”
If there are more N1 interpretations in the conditions illustrated in (9b) and (9d) as compared to those in (9a) and (9c) (patient-initial compared to agent-initial conditions) and in the conditions in (9c) and (9d) compared to those in (9a) and (9b) (dislocation conditions compared to no dislocation conditions), this would be evidence for an influence of discourse-level cues on French native pronoun resolution. If, however, preferences depend on the grammatical role, there should be a difference between the conditions in (9a) through (9c), in which N1 is the subject, and the condition shown in (9d), in which N1 is the object.
Participants.
Sixty-one undergraduate students of psychology at the University of Paris XIII-Villetaneuse participated. All participants were French native speakers, and none had been bilingual as children. Foreign languages learned later in life were English (all participants), Spanish (n = 37), German (n = 13), and Italian (n = 5).
Materials.
We constructed 12 experimental sentences, all containing a main clause followed by a subordinate clause starting with quand “when”. The main clause contained an activity verb. In each sentence, the subject and the object were proper names of the same gender. Half of the proper names were female. As illustrated in (9a–d), the subordinate clause contained an ambiguous subject pronoun that could potentially refer to the subject or the object of the main clause. Each sentence was followed by a paraphrase of the subordinate clause starting with a gap (as in Example [9a]). Interpretational preferences for the ambiguous pronoun were measured by asking participants to fill this gap, which necessitated the resolution of the pronoun. Four versions of the questionnaire were created, such that each item appeared in a different condition in each version, following a Latin square design. This assured that we could compare results from the same item in different conditions to each other and seems particularly important given the relatively low number of items in the current study. There were three items per condition in each version. The 12 experimental items were interspersed with 40 filler items of various syntactic structures as well as with the 12 experimental items of Experiment 3. The filler items were also followed by a paraphrase with a gap but contained no ambiguous pronouns.
Procedure.
The questionnaire consisted of eight pages and a cover sheet, which informed the participants that there were no incorrect responses to the questions and that they were supposed to answer spontaneously. Participants were instructed to read each sentence and then fill the gap in the following sentence. Answers were coded according to whether participants chose N1 or N2 as an antecedent for the pronoun.
Results.
Figure 1 presents the percentage of N1 interpretations in the four conditions.
Percentage of N1 interpretations in the four conditions of Experiment 1 (native speakers of French interpreting clitic subjects of finite clauses). Error bars indicate the standard error of the mean.
We found a main effect of word order, due to the fact that N1 interpretations were more frequent in the patient-initial than the agent-initial conditions (78% vs. 59%; ß = –.53, SE = .09, z = –5.81, p < .001). The effect of dislocation did not reach significance (ß = –.18, SE = .09, z = –1.94, p = .052). The Word Order × Dislocation interaction was significant (ß = .22, SE = .09, z = 2.45, p = .014). This reflects the fact that, in the agent-initial sentences, it did not make a difference whether the subject was dislocated or not. In the patient-initial sentences, passivation (and, thus, no dislocation) led to an even higher rate of attachment to N1 than dislocation of the object.
Discussion.
The results show that when N1 is the patient of the sentence, it is chosen significantly more often as an antecedent of the pronoun than when it is the agent. This pattern cannot be explained by a preference for a referent with a specific syntactic role and shows, thus, that French readers are sensitive to discourse prominence when it consists of placing the patient in initial position. Participants did not show any sensitivity to topicalization by dislocation, however. Moreover, as evidenced by the significant interaction between word order and dislocation, the effect of the two types of patient-initial structures is not equal but is stronger for initial patients in passive sentences than for initial patients in object-left dislocation structures, a difference to which we will return in the general discussion. In sum, there is no evidence that French native speakers are guided by syntactic role as the single or predominant cue in the resolution of intrasentential overt pronouns.Footnote 3 In contrast, there clearly is an influence of discourse prominence, but the strength of this effect is not equal across constructions.
Experiment 2
Experimental Factors and Predictions.
Experiment 2 aimed at testing the factors investigated in Experiment 1 for native speakers of Turkish in parallel constructions. The pronominal forms in the two languages are parallel in the sense that they are the most reduced forms possible in this context (clitic pronoun in French, pro in Turkish). As for the word order variation, because patient-initial structures are possible in Turkish simple active sentences, we decided to not resort to a passive structure but to directly compare active SOV to active OSV sentences in the conditions shown in (10a) and (10b). Determining preferences for passive sentences in Turkish would evidently have been interesting as well. The present choice allowed us, however, to test the influence of word order and dislocation in a more systematic way than was possible in French. Finally, whereas, in French, the dislocated referent was taken up by a clitic pronoun, in Turkish, the first mentioned referent was separated by a comma and not taken up by an overt form. The resulting four conditions are illustrated in (10a–d).
(10) a. N1 agent and not dislocated (SOV)
Ergin Latif’-i pro Roma’-da çalış-tığ-ı zaman bıçakla-dı.
Ergin Latif- acc pro Rome- locwork- nmlz-3sgtime stab- pst
b. N1 patient and not dislocated (OSV)
Ergin’-i Latif pro Roma’-da çalış-tığ-ı zaman bıçakla-dı.
Ergin- accLatif pro Rome- locwork- nmlz-3sgtime stab- pst
c. N1 agent and dislocated (S, OV)
Ergin, Latif’-i pro Roma’-da çalış-tığ-ı zaman bıçakla-dı.
Ergin, Latif- acc pro Rome- locwork- nmlz-3sgtime stab- pst
d. N1 patient and dislocated (O, SV)
Ergin’-i, Latif pro Roma’-da çalış-tığ-ı zaman bıçakla-dı.
Ergin- acc, Latif pro Rome- locwork- nmlz-3sgtime stab- pst
Although Turkish has a more flexible word order than French, OSV sentences are nevertheless marked compared to SOV sentences and are often analyzed as a means of topicalizing the object (Ergüvanlı, Reference Ergüvanlı1984; Kornfilt, Reference Kornfilt1997, p. 200). If Turkish readers take the marked nature of patient-initial sentences into account in a similar way as we found French readers to do, they should resolve pro more often toward N1 when it is the patient of the sentence (10b and 10d) than when it is the agent. Furthermore, if they are sensitive to the impact of dislocation, they should resolve pro more often toward N1 when there is dislocation (10c and 10d), than when there is no dislocation. Finally, if preferences are guided by a preference for the subject, as predicted by previous work on the interpretation of null subject pronouns in pro-drop languages (e.g., Carminati, Reference Carminati2002; Jegerski et al., Reference Jegerski, VanPatten and Keating2011; Mayol, Reference Mayol2010), participants should choose N1 as the preferred antecedent more often in sentences of type (10a) and (10c), in which N1 is the subject, than of type (10b) and (10d), in which N1 is the object.
Participants.
Sixty-four undergraduate students of mathematics at the University of Istanbul volunteered to complete a written questionnaire. All participants were Turkish native speakers, and none had been bilingual as children. Foreign languages learned later in life were English (all participants), German (n = 6), Russian (n = 2), Italian (n = 1), and Arabic (n = 1).
Materials and Procedure.
We constructed 12 experimental sentences, all containing a main clause in which a subordinate clause was embedded. All sentences were translation equivalents of the respective French sentences. The subordinate clause of time contained an ambiguous null subject pronoun that could refer to the subject or the object of the main clause. Each sentence was followed by a paraphrase of the subordinate clause starting with a gap, and interpretational preferences for the ambiguous pronoun were measured by asking participants to fill this gap. Four versions of the questionnaire were created such that each item appeared in a different condition in each version. There were three items per condition in each version. As in Experiment 1, the 12 experimental items were interspersed with 40 filler items of various syntactic structures as well as with the 12 experimental items of Experiment 4. The filler items were also followed by a paraphrase with a gap but contained no ambiguous pronouns. The procedure was the same as in Experiment 1.
Results.
Figure 2 presents the percentage of N1 interpretations in the four conditions. Six out of 756 (.8%) answers in the questionnaires mentioned neither N1 nor N2 and were thus discarded from further analysis.
Percentage of N1 interpretations in the four conditions of Experiment 2 (native speakers of Turkish interpreting null subjects [pro] of finite clauses). Error bars indicate the standard error of the mean.
We found a main effect of word order, due to the fact that N1 interpretations were more frequent when N1 was the subject than when it was the object (57.0% vs. 30.5%; ß = .62, SE = .08, z = 7.62, p < .001). There was no effect of dislocation (ß = .04, SE = .08, z = .51, p = .610).
Discussion.
The results are in line with previous work on pro-drop languages (Jegerski et al., Reference Jegerski, VanPatten and Keating2011; Mayol, Reference Mayol2010) and contrast with the results for French. Neither patient-initial structures nor left dislocations led to more N1 interpretations. Instead, we found that N1 was chosen more often as the antecedent of the pronoun when it was the subject than when it was the object. This is compatible with a general subject preference for Turkish null pronouns. This preference is modulated by word order: When the subject appears in second position (patient-initial conditions), there are more interpretations toward it than when it appears in first position (69.5% vs. 57.0%). There is no obvious discourse-level reason why the subject in second position should be chosen more often as the antecedent. In Turkish, prominent positions are the initial (topical) position as well as the immediately preverbal (focus) position, which in our materials is occupied by the subordinate clause. It is possible that linear proximity of a referent fosters interpretations toward it, independent of syntactic and discourse-level effects. Such an effect is also suggested by Beletti and colleagues (2007) and Sorace and Filiaci (Reference Sorace and Filiaci2006). In these studies, the order of the subordinate clause containing a subject null pronoun and the main clause containing a subject and an object referent was varied. When the subordinate clause preceded the main clause, the ambiguous null subject was linearly closer to the subject of the matrix clause than when the subordinate clause followed the main clause. The authors observed a much stronger subject preference for preposed than for postposed subordinate clauses and suggested that proximity plays a role in the resolution of Italian null pronouns in addition to subjecthood. Note that the stronger preference for second-mentioned subjects in our data means that the subject preference for Turkish null pronouns cannot be due to the topical status of initial subject referents.
In sum, the results from the first two experiments confirm previous observations concerning the differences between pro-drop and non–pro-drop languages for pronominal subjects of finite clauses. There is a strong contrast between the results for French, for which an initial position of the patient clearly increased the accessibility of this referent, and the results for Turkish, for which neither an initial position of the patient nor dislocation of N1 increased accessibility, but, instead, subjecthood and linear proximity were influential. The following two experiments test whether, as expected, this picture changes when null subjects of nonfinite clauses are concerned.
Native Speakers’ Interpretation of Null Subjects of Nonfinite Clauses (PRO)
In the following two experiments, we tested native speakers’ interpretation of PRO. The materials for Experiments 3 (French) and 4 (Turkish) were included in the same questionnaires as those from Experiments 1 and 2, respectively. This means that the experimental items from Experiment 1 and 3 (French) and 2 and 4 (Turkish) acted as filler items to each other, along with 40 other filler items.
Experiment 3
Experimental Factors and Predictions.
The same factors that were manipulated in Experiment 1 for overt pronouns were applied to PRO structures, such as in (11), leading to the four conditions illustrated in (11a–d).
(11) a. N1 agent, not dislocated (SagentVO)
Pierre a giflé Jean PRO étant jeune.
“Peter slapped John being young.”
b. N1 patient, not dislocated (SpatientVO)
Pierre a été giflé par Jean PRO étant jeune.
“Peter was slapped by John being young.”
c. N1 agent, dislocated (Sagent, sVO)
Pierre, il a giflé Jean PRO étant jeune.
“Peter, he slapped John being young.”
d. N1 patient, dislocated (Opatient, SoV)
Pierre, Jean l’a giflé PRO étant jeune.
“Peter, John slapped him being young.”
We predict that the interpretation of PRO should mainly be influenced by subjecthood such that N1 should be preferred more strongly in sentences of type (11a–c), where N1 is the subject, than in sentences of type (11d), in which N1 is the object.
Participants, Materials, and Procedure.
The participants were the same as in Experiment 1. With regard to materials, we constructed 12 experimental sentences, all containing a main clause followed by a nonfinite subordinate clause including the present participle étant “being” and a nonverbal predicative. In all other respects, the sentences were constructed exactly as the sentences described in Experiment 1. As for the procedure, the questionnaire was the same as the one used in Experiment 1, and the answers were coded according to the same principles.
Results.
Figure 3 presents the percentage of N1 interpretations in the four conditions.
Percentage of N1 interpretations in the four conditions of Experiment 3 (native speakers of French interpreting implicit subjects [PRO] of nonfinite clauses). Error bars indicate the standard error of the mean.
There was a main effect of word order due to a higher number of N1 interpretations in agent-initial sentences (ß = .21, SE = .08, z = 2.46, p = .014). There also was a main effect of dislocation, due to a higher proportion of N1 interpretations in sentences without dislocation (ß = -.18, SE = .08, z = –2.19, p = .029). These main effects cannot be interpreted by themselves, however, as they were strongly modulated by a Word Order × Dislocation interaction (ß = .25, SE = .09, z = 2.97, p = .003). This interaction reflects the fact that there is one condition in which there are fewer N1 interpretations than in the other three conditions—namely, the only condition in which the first-mentioned referent was the object (11d). Put differently, the effects appearing in this experiment can be explained by a moderate subject preference that was present to a similar degree in all four conditions. There is no evidence of an influence of word order or dislocation in addition to this effect.
Discussion.
The results for PRO contrast strongly with the results for overt clitic pronouns reported in Experiment 1. As expected, the resolution of null subjects in nonfinite constructions in native French is influenced by the syntactic role of the antecedent, whereas the discourse-level factors that we manipulated did not have any effect. It should be noted, however, that the subject preference is less strong than would have been expected if a reference to the object was ungrammatical, as suggested by prescriptive grammar.
Experiment 4
Experimental Factors and Predictions.
The same factors were manipulated as in Experiment 2, leading to the four conditions illustrated in (12a–d).
(12) a. N1 agent and not dislocated (SOV)
Sait Can’-ı PRO genç-ken tokatla-dı.
Sait Can- acc PRO young-when slap- pst
b. N1 patient and not dislocated (OSV)
Sait’-i Can PRO genç-ken tokatla-dı.
Sait- accCan PRO young-when slap- pst
c. N1 agent and dislocated (S, OV)
Sait, Can’-ı PRO genç-ken tokatla-dı.
Sait, Can- acc PRO young-when slap- pst
d. N1 patient and dislocated (O, SV)
Sait’-i, Can PRO genç-ken tokatla-dı.
Sait- acc,Can PRO young-when slap- pst
We expect Turkish speakers to be mainly influenced by subjecthood and, thus, to interpret PRO toward the N1 more in the conditions shown in (a) and (c), in which N1 is subject, than in (b) and (d), in which it is object.
Participants, Materials, and Procedure.
The participants were the same as in Experiment 2. With regard to the materials, we constructed 12 experimental sentences, all containing a main clause and a clause consisting in a nonverbal predicative that was subordinated by the suffix -(y)ken. In all other respects, the sentences were constructed exactly as the sentences described in Experiment 2. As for the procedure, the questionnaire was the same as the one used in Experiment 2, and the answers were coded according to the same principles.
Results.
Figure 4 presents the percentage of N1 interpretations in the four conditions. Eight out of 756 (1.06%) answers in the questionnaires mentioned neither N1 nor N2 and were thus discarded from further analysis.
Percentage of N1 interpretations in the four conditions of Experiment 4 (native speakers of Turkish interpreting implicit subjects [PRO] of nonfinite clauses). Error bars indicate the standard error of the mean.
As in Experiment 2, we observed a main effect of word order, due to the fact that N1 interpretations were more frequent when N1 was the subject than when it was the object (57% vs. 32.5%; ß = .53, SE = .08, z = 6.74, p < .001). There was no significant effect of dislocation (ß = .07, SE = .08, z = .88, p = .379).
Discussion.
The data replicate the pattern observed in Experiment 2 and show a preference for the syntactic subject that is particularly strong when the subject is noninitial. There was no influence of the discourse-level factors that we manipulated.
In sum, the results of Experiments 1–4 confirm previous observations in the literature. Native speakers of French were influenced by a discourse-level cue, the initial position of the patient, when interpreting a clitic subject pronoun of a finite clause. By contrast, Turkish native speakers were mainly influenced by subjecthood when interpreting a null subject pronoun of a finite clause. When interpreting PRO, native speakers of both French and Turkish were mainly influenced by subjecthood. In the following two experiments, we tested how Turkish learners of French interpret French overt pronouns and PRO.
Nonnative Interpretation of Overt Pronouns and PRO
Experiments 5 and 6 were conducted with Turkish learners of French at an intermediate level of proficiency. As was done for Experiments 1 and 3 as well as for Experiments 2 and 4, the materials of both Experiments 5 and 6 were included in the same questionnaire, such that the experimental items from the two experiments acted as filler items to each other.
Experiment 5
Experimental Factors and Predictions.
Experiment 5 used the same materials as Experiment 1. If learners are influenced by subjecthood, they should interpret the pronoun as referring to the N1 more often in sentences of type (9a–c), in which N1 is subject, than in sentences of type (9d), in which N1 is object. If, however, there are more N1 interpretations in the conditions shown in (9b) and (9d) (patient-initial) compared to (9a) and (9c) (agent-initial), or in the conditions in (9c) and (9d) (dislocation) compared to (9a) and (9b) (no dislocation), this would be evidence for an influence of discourse-level cues.
Participants and Procedure.
Eighty-three Turkish learners of French participated. All participants were native speakers of Turkish who were not bilingual as children and who started learning French after puberty. Knowledge of foreign languages other than French was reported as follows: English (n = 67), German (n = 21), and Italian (n = 4). Fifty-three of the participants were students at the French-speaking University of Galatasaray and 30 at the University of Eskisehir. The students at the University of Galatasaray were first-year students in humanities. The students at the University of Eskisehir were fourth-year students of French. All participants were at a proficiency level of B1 or B2, according to the Common European Framework of Reference (Council of Europe, 2001). The procedure for this experiment was the same as in the previous experiments.
Results.
Fourteen out of 996 (1.4%) answers in the questionnaires mentioned neither N1 nor N2 and were thus discarded from further analysis. Figure 5 presents the percentage of N1 interpretations in the four conditions.
Percentage of N1 interpretations in the four conditions of Experiment 5 (Turkish learners of French interpreting clitic subjects of finite clauses). Error bars indicate the standard error of the mean.
There was a main effect of word order, due to the fact that N1 interpretations were more frequent for patients (73.5% vs. 66%; ß = –.21, SE = .07, z = –2.75, p = .006). There was no effect of dislocation (ß = –.04, SE = .08, z = .54, p = .590). The interaction was significant (ß = .33, SE = .08, z = 4.29, p < .001). This was the case because, as with the native speakers, learners reacted more strongly to initial patients by passivation than to initial patients by dislocation.
Discussion.
The pattern of results shows similarities with the results for the French native speakers (Experiment 1) and a striking dissimilarity with the results for parallel structures in Turkish (Experiment 2). Like the native French speakers, Turkish learners prove to be sensitive to the enhanced prominence of patient referents in initial position. Moreover, like the native speakers, learners react more strongly to an initial position induced by passivation than to an initial position induced by dislocation. There is no evidence that syntactic role would be a predominant cue. They differ from the native speakers of French in that dislocation seems to have an effect as well. This can be seen in the agent-initial conditions, in which the number of N1 interpretation increases when the initial subject is dislocated (60% vs. 72%). In the patient-initial conditions, the effect of dislocation cannot be easily assessed because of the impossibility to use a nondislocated initial object in French active sentences. The passive sentence mode used in the condition shown in (9b) may in itself be a stronger cue increasing the salience of the N1.
The fact that learners reacted to discourse-level cues is in line with theories assuming a particular reliance of L2 learners on this type of cue (Clahsen & Felser, Reference Clahsen and Felser2006; Klein & Perdue, Reference Klein and Perdue1997). As native speakers were not sensitive to these same cues in translation equivalents of the current sentences (Experiment 2), L1 influence seems to be an unlikely explanation for these effects, but it cannot be dismissed completely that the L1 played a role. More precisely, Turkish learners may have perceived the French overt pronoun as similar to the Turkish overt pronoun o, and this pronoun may in turn be sensitive to discourse-level effects. O cannot be used in subordinate clauses when reference to an intrasentential antecedent is intended, but it is conceivable that learners transferred interpretation strategies from other contexts. Thus, it seems important to test whether learners continue to show overreliance on discourse-level cues in cases in which the L1 and the L2 have similar structures and in which neither L1 nor L2 native speakers are influenced by this type of cue. This is the case for the PRO structures tested in Experiment 6.
Experiment 6
Experimental Factors and Predictions.
Experiment 6 used the materials of Experiment 3 (French sentences containing PRO). If learners pattern with native speakers of both French and Turkish, they should show a general subject preference in the interpretation of PRO. If, instead, learners continue to show a strong reliance on discourse-level cues, the results should show an effect of word order and potentially dislocation, similar to Experiment 5.
Participants, Materials, and Procedure.
The participants were the same as in Experiment 5. The materials were the same as in Experiment 3. The procedure was the same as in the previous experiments.
Results.
Nine out of 996 (.9%) answers in the questionnaires mentioned neither N1 nor N2 and were thus discarded from further analysis. Figure 6 presents the percentage of N1 interpretations in the four conditions.
Percentage of N1 interpretations in the four conditions of Experiment 6 (Turkish learners of French interpreting implicit subjects [PRO] of nonfinite clauses). Error bars indicate the standard error of the mean.
There was a main effect of word order, due to the fact that N1 interpretations were more frequent for patient-initial sentences (57% vs. 47%; ß = –.27, SE = .07, z = –3.67, p < .001). The factor dislocation was significant as well: There were more N1 choices when there was dislocation than when there was no dislocation (57.5% vs. 46.5%; ß = .29, SE = .07, z = 3.95, p < .001).
Discussion.
Contrary to native speakers of French as well as native speakers of Turkish in Turkish parallel constructions, Turkish-speaking learners of French exhibited no sign of a subject preference in these constructions in French. By contrast, both discourse-level factors that we manipulated showed an effect. There were more N1 interpretations when N1 was a patient compared to when it was an agent and when it was dislocated compared to when it was not dislocated. Comparing the results of Experiments 5 and 6, discourse-level cues, thus, had a similar influence, despite the very different constructions tested. We conclude that, at least in the construction tested here, learners do not transfer the use of a syntactic cue from their L1 to interpret the ambiguous PRO. Instead, in this construction, they resort to discourse-level cues to a greater extent than native speakers of both Turkish and French.
GENERAL DISCUSSION
To summarize our experimental results, we have shown that a discourse-level cue—the initial position of the patient—influenced the resolution of overt subject pronouns of French finite clauses (Experiment 1). A syntactic cue—subjecthood—influenced resolution preferences for null subjects of finite clauses (pro) in native speakers of Turkish (Experiment 2) and for null subjects of nonfinite clauses (PRO) in native speakers of both French (Experiment 3) and Turkish (Experiment 4). By contrast, we did not find any influence of subjecthood on resolution preferences for French overt pronouns or PRO by Turkish learners of French (Experiments 5 and 6). Instead, learners tended to resolve ambiguous overt pronouns as well as PRO more toward the N1 when it was the patient than when it was the agent and when it was dislocated than when it was not dislocated. In what follows, we discuss these results, in turn, for native speakers and learners.
Native Interpretation of Overt Pronouns, pro, and PRO
The results for the overt pronouns in French native speakers (Experiment 1) support previous research suggesting that discourse-level cues have a strong influence on the resolution of overt pronouns (e.g., Cowles et al., Reference Cowles, Walenski and Kluender2007; Kehler, Reference Kehler2002). Patient-initial structures led to a higher percentage of N1 interpretations than agent-initial structures, in particular in passive sentences. We assume that patient-initial structures are marked as compared to agent-initial structures (Comrie, Reference Comrie, Cole and Sadock1977; Desclés et al., Reference Desclés, Guentcheva and Shaumyan1985; Givón, Reference Givón1979; Prévost, Reference Prévost, Apothéloz, Combettes and Neveu2009; Shibitani, Reference Shibitani1985; Tannenbaum & Williams, Reference Tannenbaum and Williams1968; Thompson, Reference Thompson, Channon and Shockey1978, among others) and that this marked order increases the discourse prominence of the first-mentioned antecedent. Following this line of reasoning, the increased number of N1 interpretations in these conditions is thus evidence for a reliance on discourse-level factors for the interpretation of overt pronouns in native speakers of French. According to the same reasoning, one might have expected that dislocated antecedents are likewise more discourse prominent and should, thus, also increase the number of N1 interpretations. In our data, however, dislocated referents were not systematically preferred over nondislocated ones. This replicates previous results from Colonna, Schimke, and Hemforth (Reference Colonna, Schimke and Hemforth2012), who also found that topicalization by dislocation led to a slight numerical, but statistically unreliable, increase in N1 interpretations as compared to the same sentence without dislocation. Apparently, if there is an effect of topicalization by dislocation, it is not very strong. This may be so for several reasons. First, the first-mentioned antecedent could be considered a topic by default (e.g., Cowles et al., Reference Cowles, Walenski and Kluender2007; Gernsbacher Reference Gernsbacher1990) such that additional topicalization cannot considerably reinforce a N1 preference. Second, topicalization structures may be so frequent in French that their effect is weaker as compared to, for example, passivation. Third, it could also be that, although dislocations are frequent, they are also a device that is characteristic of spoken discourse rather than of written sentences presented in isolation. Participants may have had difficulties interpreting the dislocation in this unusual context, weakening its effect. Together, it can be concluded that not every type of discourse prominence necessarily leads to the same effect on pronoun resolution (see also Colonna et al., Reference Colonna, Schimke and Hemforth2012) but also that we clearly have provided further evidence that discourse-level cues in general can influence the resolution of overt pronouns. The results for Turkish null subject pronouns of finite clauses (Experiment 2) confirm previous findings that null pronouns in finite clauses (pro) seem to be less prone to an influence of discourse-level cues across languages (Jegerski et al., Reference Jegerski, VanPatten and Keating2011). The strong reliance on subjecthood is predicted by the PAH (Carminati, Reference Carminati2002). The PAH has been proposed specifically for null subject pronouns in Italian subordinate clauses, in which, in contrast to Turkish subordinate clauses, both the null and the overt pronoun are possible. The fact that we nevertheless replicated a similar subject preference as in Italian may indicate that either the presence of two competing forms is not crucial for the mechanism underlying the PAH or that Turkish native speakers may generalize preferences from main clauses, in which both forms are possible, to subordinate clauses.
Concerning the interpretation of PRO, our results from Experiments 3 and 4 confirm the expectation that, in both languages investigated, subjecthood of the antecedent was the dominant cue to their interpretation. The discourse-level cues we manipulated did not influence interpretations in either language, so it remains an open question to what degree and under which conditions the interpretation of PRO can be influenced by discourse prominence in native speakers, as suggested by Haspelmath (Reference Haspelmath, Haspelmath and König1995). In sum, native speaker results confirmed the predictions in that a discourse-level cue was of influence for French overt pronouns only. Against this background, we wanted to know whether learners coming from a pro-drop source language would rely on discourse-level factors in the interpretation of the overt pronoun and of PRO in a non–pro-drop target language.
Nonnative Interpretation of Overt Pronouns and PRO
The Turkish learners tested in the current study were influenced by discourse-level cues (initial position of the patient and dislocation) when interpreting overt pronouns and PRO in French and were not influenced by subjecthood in either case. Concerning overt pronouns, there are several factors that may have contributed to this effect. First, as suggested previously, L2 learners may generally overrely on discourse-level cues (Clahsen & Felser, Reference Clahsen and Felser2006; Klein & Perdue, Reference Klein and Perdue1997) and, therefore, have reacted to these cues in the interpretation of the pronoun. As the resulting interpretations are close to native speakers’ interpretations, it may also be the case that learners have acquired nativelike interpretations, independent of any overreliance on discourse-level cues. A point in favor of an overreliance on discourse-level cues is the observation that learners differed from native speakers of French in reacting more strongly to dislocation, thereby showing a stronger reaction to a discourse-level cue than the native speakers. Finally, as discussed in more detail in the preceding text, an influence of the L1 cannot be excluded. In particular, even though discourse-level cues were found to not influence Turkish native speakers in parallel constructions (Experiment 2), learners may have been influenced by other L1 constructions when interpreting pronouns in the L2. In light of this, the results from Experiment 6, in which we found that learners were influenced by discourse-level cues when interpreting PRO, are particularly telling.
In this case, learners made use of discourse-level cues in a construction where a null form (PRO) was possible both in the L1 and the L2 and where native speakers of both languages were not influenced by discourse-level cues. The possibility can thus be excluded that learners have acquired the pattern of the L2, or were influenced by the L1, leaving an overreliance on discourse-level cues as a plausible explanation. We cannot say, however, what the influence of the fact that French is not a pro-drop language is. In particular, learners could have learned from the overt pronoun contexts that discourse-level principles play a particular role in pronoun interpretation in French and then have overgeneralized these principles to the interpretation of PRO. This explanation would leave open why transfer from French overt pronouns to French PRO should be more likely than from Turkish PRO to French PRO. Nevertheless, future studies should test whether the current result generalizes to other contexts. In particular, it would be telling to find out whether learners coming from a pro-drop language show an overreliance on discourse-level cues when interpreting dropped subject pronouns (pro) in a pro-drop target language.
Despite the restriction to a specific construction, our data show that learners may overrely on discourse-level cues at least in one specific case. We suspect, in line with previous research on production (Dimroth et al., Reference Dimroth, Gretsch, Jordens, Perdue, Starren, Dimroth and Starren2003; Meisel, Reference Meisel and Eubank1991, among others), that learners may find syntactic as opposed to other types of knowledge particularly hard to use in a L2. However, we observed the lack of an effect of subjecthood in offline data only. As noted by a reviewer, it cannot be excluded that participants relied on subjecthood during the course of processing. This seems unlikely in light of the fact that previous studies have found that it is often precisely during online processing that learners are more prone to relying on discourse-level cues and have particular difficulties with the use of structural information (Clahsen & Felser, Reference Clahsen and Felser2006, for an overview; see also Felser & Cunnings, Reference Felser and Cunnings2012; Pan & Felser, Reference Pan and Felser2011; Pan, Schimke, & Felser, Reference Pan, Schimke and Felser2015; Roberts & Felser, Reference Roberts and Felser2011). The present study, along with studies by Jegerski and colleagues (2011) and Okuma (Reference Okuma, Herschensohn and Tanner2011), extends the evidence from online processing and from production studies (such as Klein & Perdue, Reference Klein and Perdue1997) in showing that learners can differ from native speakers in an overreliance on discourse-level cues even in offline final interpretation preferences.
Finally, our results open up new questions, which we briefly discuss in what follows. First, even though an overreliance on discourse-level cues may be a property of L2 sentence processing in general, independent of the L1, it would be worthwhile to investigate whether specific cues are used differently depending on their use in the L1. As we have tested learners from one L1 only, we do not know whether the sensitivity to word order and dislocation we found is influenced by the use of word order and dislocation in Turkish. For word order, this seems a plausible hypothesis, as this is a frequently exploited means of information structure marking in Turkish. This is not the case for dislocation (Kornfilt, Reference Kornfilt1997, p. 205), but learners may have taken the comma following the N1 as a written translation of intonational prominence, which is used in Turkish (cf. Kornfilt, Reference Kornfilt1997, pp. 506–507). Second, questions are also raised by the fact that even though the learners reacted to discourse-level cues when interpreting the overt pronoun as well as PRO, there are differences between the results in these two cases that we have not addressed so far. In particular, there were, on average, 70% N1 interpretations in the four conditions of Experiment 5 (overt pronoun) but only 52% N1 interpretations in the four conditions of Experiment 6 (PRO), suggesting that PRO is, in general, resolved more locally than an overt pronoun. A reviewer suggested that learners may have been uncertain as to how to interpret the relatively rare present participle structure in French and, therefore, have chosen the nearest possible antecedent. It is possible that linear proximity did indeed play a role, either due to uncertainty or to transfer from Turkish, for which we likewise found an influence of this factor. If this was the only mechanism at work, however, we should have found no differences between experimental conditions. Finally, we want to stress that our results were obtained with a relatively small number of items and with specific constructions and are thus certainly in need of replication with different items and constructions.
In sum, the current results support the view that learners may overrely on discourse-level cues when interpreting L2 pronominal forms. Future studies could test whether this result generalizes across languages, constructions, and learner populations.
