Abstract
The purpose of this study is to expand the testing ground of the Interface Hypothesis (Sorace, 2006) by investigating the degree to which second language (L2) learners of Bulgarian with English as their first language (L1) had acquired the pragmatic function of clitic doubling as a topicality marker. Advanced and intermediate L2 speakers of Bulgarian, as well as a control group of Bulgarian native speakers, participated in the experiment. The experimental materials included a proficiency test and a pragmatic felicity task. The results showed that the intermediate participants did not differentiate between the felicitous and the infelicitous options in the pragmatic felicity task in a target-like manner as their responses either did not exhibit a statistically significant difference or favored the response closest to the L1. However, the advanced L2 learners had successfully acquired the pragmatic meaning of clitic doubling in Bulgarian and performed in a native-like manner. The study highlights the fact that successful learning at the syntax–discourse interface cannot be excluded, and more research – exploring as many interface conditions as possible – needs to be carried out in order to validate the Interface Hypothesis as a legitimate constraint that permanently hinders native-like performance.
Keywords
I Introduction
Generative second language acquisition research at the earliest stages of its development was primarily concerned with issues related to UG (Universal Grammar) accessibility, second language (L2) syntactic knowledge, and the possibility for successful parameter resetting (White, 2009). The syntactic component of the interlanguage grammar received most of the attention and the principal debate was on the possibility for L2 acquisition in a UG-constrained manner parallel to what was conceived to be the case in the acquisition of one’s native language. Instances of successfully acquired constructions where poverty of the stimulus made it unlikely to ascribe target-like performance to the primary linguistic input and to the employment of basic cognitive capacities – e.g. successful acquisition of the Overt Pronoun Constraint as described in Pérez-Leroux and Glass (1997) – served as a solid argument for the availability of UG beyond the critical stage for language acquisition (White, 2003). However, based on the postulation of a modular architecture of linguistic systems, the emphasis since 2000 has seen a clear shift toward integrating the various grammar-internal modules (syntax, morphology, phonology and semantics), on the one hand, and the grammar-external domains (the conceptual-intentional and the articulatory-perceptual systems), on the other hand, in accounting for acquisition data. Cross-modular integration of grammar-internal and grammar-external domains is very often necessary for the construction of grammatically correct and pragmatically felicitous linguistic output. The need for such integration and the increased complexity associated with it is deemed likely to create acquisition problems and to be one of the primary loci of persistent non-target performance among L2 learners. This expectation is further reinforced by research that reveals interface coordination difficulties in areas outside L2 acquisition, namely first language (L1) acquisition (Schaeffer, 2000), bilingual acquisition (Haznedar, 2006; Serratrice et al., 2004), heritage language acquisition (Montrul, 2004) and L1 attrition (Tsimpli et al., 2004).
While accounting for experimental data on the L2 acquisition of interface conditions, White (2009) established a division between internal and external interfaces. The former refer to those instances whereby a particular linguistic structure abides by requirements set by more than one of the modules that constitute grammar. Syntax–semantics, syntax–morphology and phonology–morphology are interfaces that are internal to the grammatical system. On the other hand, in external interfaces, modules of the computational system interact with external domains (conceptual–intentional, articulatory–perceptual or sentence processing). As White (2009) pointed out, a closer look at the interfaces showed a dichotomy between the internal and the external interfaces with regard to the degree of fossilization and target-deviant behavior. The external interfaces seemed to be associated with a higher degree of severity when it came to persistent non-native-like performance and strong L1 transfer even at the stage of ultimate attainment (Tsimpli and Sorace, 2006), whereas the internal interfaces, problematic as they might be themselves, were often characterized by a successful acquisition outcome in spite of a usual delay (Slabakova, 2006).
In response to the growing interest in the L2 acquisition of external interfaces, this article presents an experimental study on the degree to which L2 learners of Bulgarian, with English as their native language, have acquired the pragmatic function of clitic doubling in Bulgarian, thus introducing another case of an external interface construction. The participants in the study were advanced speakers of L2 Bulgarian who had resided in Bulgaria long enough to attain a high level of proficiency and to be aware of the subtlety of a pragmatically felicitous instance of clitic doubling marking overt topicality. A group of intermediate L2 learners of Bulgarian was included in order to compare competence at the end-state with interlanguage grammar at an earlier stage in the acquisition process.
The organization of the article is as follows: first, Section II reviews previous research on the L2 acquisition of external interfaces and introduces the Interface Hypothesis. Next, Section III presents an overview of object clitics and clitic doubling conditions in Bulgarian. Section IV introduces the participants and explains the methodology of the experimental study on the L2 acquisition of the pragmatic effects of clitic doubling in Bulgarian. A detailed analysis of the results of the experiment is provided in Section V. A general discussion and conclusive remarks are presented in the last section.
II L2 acquisition of external interfaces and the Interface Hypothesis
A number of recent studies on end-state grammars have revealed vulnerability at the syntax–discourse interface where L1 interference is active even at advanced levels of L2 attainment. In a study by Belletti et al. (2007) on the production and interpretation of postverbal subjects, and null and overt pronominal subjects by English near-native L2 learners of Italian, the participants exhibited native-like syntactic proficiency but non-native-like performance in the use of postverbal subjects and overuse of overt pronominal subjects. Tsimpli and Sorace’s (2006) experiment on the use of null subject pronouns in L2 Greek by native speakers of Russian revealed significant overuse of overt subjects even among advanced L2 speakers. Sorace and Filiaci (2006), in an experiment on the interpretation of intrasentential anaphora in Italian by advanced English L2 speakers of Italian, showed that the L2 learners had acquired the syntactic constraints on pronominal subjects in Italian but displayed ‘residual indeterminacy’ with respect to the discourse strategies that were involved in the interpretation of those pronouns. Lozano (2006) presented evidence for deficient application of the discourse properties regulating subject–verb and verb–subject distribution in Spanish by L2 learners who were native speakers of Greek. Hopp (2004) examined scrambling in the L2 interlanguage of near-native English and Japanese learners of German. The results of that experiment showed that L2 learners exhibited native-like knowledge with respect to the UG-specified syntactic representations but occasionally failed in integrating discourse interpretation into the syntax.
The results of the studies listed above revealed the presence of a protracted divergence between L2 end-states and adult native grammars which was due not to the lack of knowledge of the syntactic representations that underlie particular structures but of their realization as triggered in the interplay with the specific discourse requirements. In view of those findings, Sorace and Filiaci (2006) proposed the Interface Hypothesis, which stated that ‘narrow syntactic properties are completely acquirable in a second language, even though they may exhibit significant developmental delays, whereas interface properties involving syntax and another cognitive domain may not be fully acquirable’ (Sorace and Filiaci, 2006: 340) and may trigger residual optionality effects.
Sorace (2006) viewed processing deficiency as a possible cause of L2 optionality besides lack of input frequency and L1 transfer. She argued that structures that involved the integration of purely syntactic knowledge with knowledge from other domains, such as discourse, were more complex than structures that involved syntactic knowledge alone. Therefore, the realization of those complex structures warranted additional processing effort, which made them more costly processing-wise.
In her most recent work on the Interface Hypothesis, Sorace (2011) maintains that resource allocation (‘the ability to flexibly direct attentional resources as a function of the complexity of the incoming material’; p. 23) exhibits a greater proneness to failure in discourse processing rather than in ‘the more automatic aspects of language processing’. Hence, resource allocation effects manifest themselves more distinctly at external interfaces than at internal interfaces. In addition, Sorace (2011) explored two potential accounts of the observed target divergence among L2 learners at the end-state: a representational account and a processing resources account. She assumed that ‘residual optionality may be related to access and integration of syntactic knowledge’ rather than to a representational deficit. The decreased integration abilities were attributed to ‘two non-linguistic factors’, namely input deficiency (both quantitative and qualitative) and executive control (resource allocation and inhibition of the language not in use) in bilinguals.
The predictions of the Interface Hypothesis were certainly supported by the examples of unsuccessful coordination of syntactic and pragmatic knowledge, and the end-state optionality effects revealed by the studies listed earlier in this section. However, recent research on the acquisition of external interfaces has produced results that have been far from conclusive. In contrast to the studies where the predictions of the Interface Hypothesis were borne out, there are abundant data that point to outcomes with highly advanced L2 learners that are at odds with those predictions. Next is a brief overview of some of those studies.
Valenzuela (2005), in a study similar to the one presented in this article, argued that only specific topics were involved in clitic doubling constructions such as Clitic Left Dislocation (CLLD) in Spanish. Non-specific topics, on the other hand, could be included only in contrastive left dislocation (CLD). Valenzuela’s near-native participants were 100% accurate in accepting specific dislocated topical objects doubled by a clitic, which indicated a successful integration of syntactic and discourse requirements. However, their accuracy substantially decreased with generic dislocates, which could not be doubled by a clitic despite being topics. In response to those results, Slabakova and Ivanov (2011) contended that the inferior accuracy on generic dislocates ‘might be due to some participants heeding discourse cues over semantic cues in processing those sentences’, and thus inappropriately providing clitic doubling with non-specific topics.
Rothman (2008) conducted a study on the L2 distribution of null and overt subject pronouns by English-speaking L2 learners of Spanish. The results of the study pointed to a delay in the acquisition of structures that involved the integration of syntactic and pragmatic knowledge but, crucially, not to inevitable L2 fossilization. The results of a study by Iverson et al. (2008) on the L2 acquisition of the discourse-dependent distinction between subjunctive and indicative complements with negated epistemic matrix predicates in Spanish also posed a challenge to the predictions of the Interface Hypothesis for obligatory residual optionality and L2 fossilization at the interfaces. Donaldson (2011) investigated the L2 acquisition of right dislocations by near-native L2 speakers of French with English as their L1. Right dislocation in French involves a particular syntactic structure that has a specific discourse function, namely to encode topic. The results of the study, which included production data, revealed target-like use of right dislocations by the near-native L2 speakers.
In view of the diverse findings with respect to the L2 acquisition of interface conditions, White (2011), after a comprehensive survey of studies on the interfaces, argued that they were not ‘monolithic’ as ‘it is not the case that all interfaces lead to difficulties, it is not the case that all phenomena at a particular interface are necessarily problematic, and it is not the case that acquisition failure is inevitable’ (White, 2011: 587). White also suggested that the efforts to pin down the locus of non-target behavior within either only representational deficiency, or only processing failure, might also be in the wrong direction as particular sources can be conducive to problems with particular interfaces, and in some cases target divergence could be due to multiple sources.
Considering the inconclusiveness of the experimental data as regards the L2 acquisition of external interfaces, the experiment presented in this article extends the array of investigated interface conditions. It also provides an empirical test of the Interface Hypothesis since mastery of clitic doubling marking overt topicality requires both the correct syntactic representation of clitic doubling and knowledge of the discourse interface conditions that govern the felicitous use of that construction. Before we proceed with the details of the experiment, next section offers a brief overview of Bulgarian object clitics and explains the discourse requirements for clitic doubling.
III Overview of Bulgarian object clitics
1 Bulgarian object clitic paradigm
Bulgarian has an intricate system of clitics, which includes accusative and dative pronominal clitics, possessive clitics, a future clitic, the present tense forms of the verb ‘to be’, accusative and dative reflexive clitics and an interrogative clitic. Bulgarian pronominal clitics appear in an 8-member paradigm as illustrated in Table 1, and they are marked for case, number and person as well as gender for 3rd person singular clitics.
Paradigm of Bulgarian accusative and dative clitics and corresponding strong pronouns
Since Bulgarian does not exhibit a case system, dative strong pronouns differ from accusative strong pronouns only by the addition of the preposition na. However, clitics have retained their case marking and with the exception of 1st and 2nd person plural, dative and accusative clitics have different forms.
Bulgarian object clitics are preverbal clitics, unlike other Slavic clitics (e.g. Serbian), which are second-position (Wackernagel) clitics (Franks and King, 2000). They invariably appear before the verb regardless of how much material precedes them as illustrated in (1) below.
(1) a. Toj sigurno ja poznava. he perhaps her-cl. know-3p.sg ‘Perhaps he knows her.’ b. * Toj ja sigurno poznava.
Their strictly preverbal position is violated only in cases when it will lead to their being a first element in the sentence. In those cases they appear post-verbally, which, in line with the Romance tradition, is referred to as the Tobler–Mussafia effect.
(2) a. Ø Vidjax go pro see-1p.sg him-cl. ‘I saw him.’ b. * Go vidjax.
2 Pragmatic effects of clitic doubling in Bulgarian
Bulgarian (alongside Macedonian) displays a phenomenon that is not found in other Slavic languages, namely clitic doubling, whereby a direct or indirect object DP and a coreferential clitic occur within the same clause (3).
(3) a. Ivan go vidja Maria. Ivan him-cl.ACC.masc. see-3p.sg Maria b. Maria go vidja Ivan. ‘Maria saw Ivan.’
Clitic doubling is considered one of the defining characteristics of the Balkan Sprachbund (Albanian, Bulgarian, Greek, Macedonia, Romanian). 1 As observed by Guentchéva (2008), in comparison to other Balkan languages, Bulgarian clitic doubling exhibits the slightest degree of grammaticalization and the highest degree of pragmatic significance. In Macedonian all definite direct and indirect objects need to be doubled by a co-referential clitic (Tomić, 2004). Albanian doubles all indirect objects (Kallulli, 2000). Romanian clitic doubling occurs with post-verbal direct objects and the presence of the special preposition pe. The function that is most consistently applicable to the usage of clitic doubling in Bulgarian is overt marking of topicality (Leafgren 1992, 1997). 2 In that regard, Lopašov (1978) establishes a grammaticalization continuum for clitic doubling in Balkan languages with Macedonian and Bulgarian at the two extremes and Albanian, Romanian and Greek coming in between those two in decreasing order of grammaticalization.
Doubling of topical objects in Bulgarian is pragmatically required and the lack of clitic doubling will significantly compromise the felicity of the utterance. Let us consider the short dialogue in (4).
(4) A: Njakoj viždal li e Ivan dnes? somebody seen Q is Ivan today ‘Has anybody seen Ivan today?’ B: a. Ivan #(go) vidjax sutrinta Ivan (him-cl.) saw-1p.sg in-the-morning ‘I saw Ivan in the morning.’ b. Sutrinta #(go) vidjax Ivan
The answer to a question as in (4) will be deemed pragmatically felicitous if the fronted topic is doubled by a preverbal agreeing clitic. 3 As shown in answer (4a), in instances where the object DP is extracted from its canonical object position and fronted for the purpose of revealing its topic nature, overt clitic doubling is strictly required and its absence would render the sentence pragmatically flawed and, to some extent, even grammatically incorrect unless some pair-list reading is imposed on it. Unlike the double topicality marking (fronting + clitic doubling) in (4a), (4b) has its topical object DP in its canonical position, and the only way its topical status could be expressed and the pragmatic felicitousness of the utterance saved is by the insertion of an agreeing clitic in preverbal position. If no clitic doubling is available, the clearly topic nature of the object DP would not be signaled in any way and that would render the sentence pragmatically odd. Moreover, Bulgarian may have its focused components in clause-final position, which could further diminish the pragmatic felicitousness of (4b) as it might not only be lacking in topicality marking, but also be potentially interpretable as a focal construction.
The pragmatic significance of a syntactic property such as clitic doubling, as shown in the above example, is a useful tool for measuring the ability of L2 learners to successfully coordinate syntactic and pragmatic knowledge. For that purpose it was used in the experimental study described in the next section.
IV An experimental study on the L2 acquisition of the pragmatic effects of clitic doubling in Bulgarian
1 Research question
The experimental study sought to establish the extent to which native speakers of English had learned a subtle property of Bulgarian grammar that does not occur in their L1, namely, marking of topical objects via clitic doubling. Since the construction investigated in this study presented a clear case of an interface condition and one of the two groups of participants consisted of very advanced L2 speakers of Bulgarian, it was an effective tool for testing the claims of Sorace’s (2006) Interface Hypothesis for end-state divergence with respect to L2 interface properties. The obtained data were investigated for traces of non-native optionality among the advanced learners at the end-state and more specifically for the overuse of L1 transferred constructions.
2 Participants
The participants in the study were 24 native speakers of English who were L2 learners of Bulgarian. They were divided into two groups: advanced and intermediate. This division was based on a proficiency cloze test. As the main experimental task was very demanding and time-consuming, it was imperative not to exhaust and discourage the informants with a detailed and lengthy proficiency test. For that reason we chose a cloze test along the lines of Chen (1996) and Slabakova (2001). It included a short adapted version of a popular fairy tale where words were deleted at regular intervals (every sixth or seventh word was deleted) until there were 40 blanks. The participants were asked to fill in each blank with a word that they thought would be the best meaningful fit. If a blank was filled with exactly the same word as the original, 1 point was given. If the supplied word was different from the original deleted word, no point was given, even if that word was meaningful in the particular context. The maximum total number of points for the cloze test was 40.
First, the score range of the control group was determined. Then, the raw scores of the L2 participants were compared with the control range, and those participants who fell in that range were considered advanced. Participants who fell below the control range were counted as non-advanced. The lowest score among the control participants (mean score: 29.4) was found to be 23, so that was taken to be the cut-off point and each participant whose score fell above it was considered advanced. Ten L2 learners were found to have scores that were above 23 and they formed the advanced group (mean score: 25.8). All other participants were significantly below the cut-off point and they were all placed into the intermediate (or non-advanced) group (mean score: 13.3).
Since the study was primarily interested in the performance of advanced L2 learners, the main purpose of the proficiency assessment was to separate the participants into advanced and non-advanced. Although the above procedure might seem quite simple, it was successful in separating the advanced learners from the total of 24 participants as was confirmed by the results in the subsequent experimental tasks. That division, as expected, was very strongly correlated with the number of years that the participants had learned Bulgarian, either in a classroom setting or in daily communication (in most cases, both). A further fine-grained division within the non-advanced group was not needed in view of the research questions posed by the study, so they were treated as intermediate learners, since they had studied Bulgarian for at least one year.
All participants were native speakers of British English or American English and they had started studying Bulgarian long after the critical period of acquisition. The advanced participants (n = 10) had a mean age of 39.2 years and their mean number of years of exposure to Bulgarian was 12.7 (only one advanced L2 learner had studied Bulgarian for less than 10 years). All 10 participants in the advanced group had lived in Bulgaria for a number of years and all but one of them were still living there at the time of the experiment.
The participants in the advanced group came from various occupational backgrounds: a Bulgarian–English translator, three journalists (one was a New York Times correspondent and two of them worked for Bulgarian magazines), a non-fictional writer, an artist, a PhD student of Bulgarian folk music, a Bulgarian National Television employee, a member of a religious mission and the Canadian wife of a Bulgarian linguist. Four of those advanced learners had Bulgarian spouses and two of them had children who were native speakers of Bulgarian. All of them used Bulgarian on a daily basis and for most of them native-like proficiency in the language was crucial in their professional environment.
The intermediate participants (n = 14) had a mean age of 31.9 years and they had been exposed to Bulgarian for an average of 2.6 years. Five of them were employees at the US Embassy in Bulgaria; there were also college professors, ESL teachers, a scientist, members of religious missions in Bulgaria and Peace Corps volunteers. At the time of the experiment, most of them were residing in Bulgaria. Many of them were taking formal classes in Bulgarian or had a Bulgarian tutor. Most intermediate learners claimed that they used Bulgarian on a daily basis and that their understanding of the language was essential in their workplace. Nevertheless, some admitted that they were still struggling with Bulgarian but were generally satisfied with their progress and were hoping to attain a higher level of proficiency.
Finally, there was a control group (n =16) of monolingual native Bulgarian speakers. Their mean age was 30.7 years and they were all living in Bulgaria when the experiment was conducted. Ten of the participants in the control group lived in the capital of Bulgaria, Sofia, and six came from other regions in the country. Since there are no dialectal distinctions in the use of Bulgarian clitics, the regional variety of the participants in the control group could not have had any effect on their results. All control participants had attended higher education institutions in Bulgaria and held university degrees. Although they had an excellent command of Standard Bulgarian, none of them had studied Bulgarian linguistics.
3 Methodology
The study included a context sentence evaluation task whereby a particular situation was described in English and then followed by a short dialogue. The dialogue consisted of a question and four answer options for which the participants had to provide an appropriateness evaluation on a scale from 1 to 5 (5: perfectly acceptable; 1: totally unacceptable). Since placing the dialogue within a specific context and the participants’ understanding of that context was crucial for the experiment, the description of the context was presented in English. Thus we eliminated a possible misunderstanding of the context, especially by the intermediate participants, which would have lead to them either not being focused enough on the main part of each test item (the question and the answer with or without clitic doubling) or would have discouraged them from participating altogether. 4
The experiment included four conditions in a 2×2 design: Topic × Focus 5 and Accusative × Dative; see Table 2. The sentences in the topic condition were aimed at directly testing the knowledge with regard to clitic doubling as an overt marker of topicality. The significance of the focus condition, on the other hand, was to ascertain whether in responses to wh-questions the participants recognized the infelicitousness of clitic doubling with focal direct and indirect objects as opposed to its felicitousness with topical objects. Low evaluation of clitic doubling with focal objects would reinforce the presence of knowledge as to its correct pragmatic function, namely marking topicality.
Experimental conditions
Each of the four experimental conditions was represented by 10 short dialogues. The dialogues were situated within a context described in English. Each dialogue consisted of a question followed by four answer options. In the topic conditions, the question introduced an entity that was repeated in the answer as a topic. All topics were definite and specific, which made them eligible for clitic doubling. Thus this study tested knowledge of the syntax–discourse interface without mixing in knowledge of semantic properties as in Valenzuela (2005). 6
In the focus conditions, a wh-question asked for an object that appeared in the answer as a new, focal element and was therefore incompatible with clitic doubling. Two of the answer options were pragmatically felicitous and two were infelicitous. Although the pragmatic infelicity of some of the sentences considerably degraded them, they were all grammatical sentences if taken on their own and outside of the particular context. The four options (sentence types) that were to be evaluated were as follows:
Option 1: [+Object fronting] [+Clitic doubling]
Option 2: [–Object fronting] [+Clitic doubling]
Option 3: [+Object fronting] [–Clitic doubling]
Option 4: [–Object fronting] [–Clitic doubling]
The first two options, which involved clitic doubling, were felicitous in the topic conditions and were expected to receive higher evaluations in those conditions, whereas the latter two options were considered infelicitous in topic contexts and lower evaluations were expected for them. The opposite held in the focus conditions where doubling of a focal object as in Options 1 and 2 was deemed infelicitous. Options 3 and 4 involved object fronting (with no doubling) and neutral SVO word order, respectively. Both were typical word orders for presenting focal objects in Bulgarian and were therefore pragmatically acceptable in the Focus conditions.
Below are two sample test items that illustrate the topic dative and focus accusative conditions:
Topic Dative Condition
Ivan is an exchange student at the University of Iowa. His friends Peter and Maria promised him that they would write emails to him at least once a week, so that he didn’t feel lonely and isolated. One day Maria asked Peter:
A: Pisa li wrote-2p.sg Q to Ivan this week ‘Did you write to Ivan this week?’ B: a. to Ivan him-cl.DAT wrote-1p.sg three times this week ‘I wrote to Ivan three times this week.’ b. Tazi sedmica tri pâti c. # d. # Tazi sedmica tri pâti pisax
Focus Accusative Condition
On her way to work this morning, Petya saw Ivan upset and talking to their neighbor Peter about having lost something while he was jogging in the park. She was in a hurry and didn’t have time to find out what made him so upset, but in the evening she asked him:
A: Kakvo zagubi dokato bjagaše tazi sutrin? what lost-2p.sg while ran-2p.sg this morning ‘What did you lose while you were running this morning?’ B: a. # keys for car-def. them-cl.ACC lost-1p. sg ‘I lost the car keys.’ b. # Zagubix c. d. Zagubix
The task was presented to the participants both in writing and in spoken language recorded by two native speakers. Natural intonation was used in the recordings of the evaluated sentences (non-focus intonation in the topic conditions and focus intonation in the focus conditions) to ensure that the learners and the native speakers were not imposing their own intonation on the test sentences, which would have been the case in a written only test. The bimodal presentation of the test items also made sure that the participants analysed the sentences as being part of a dialogue rather than as stand-alone sentences. We were concerned that if the participants had been given the test items only in writing, some of them would have looked only at the sentences that were supposed to be evaluated and would have given them very high evaluation scores since out of context all those sentences were grammatically correct. In addition, the test items of all four conditions as well as the answer options for each test item were randomized in order to ensure the absence of pattern awareness among the participants.
The next section provides a detailed account of the group and individual results in all four experimental conditions as well as the results of a parallel study that targeted the purely syntactic properties of Bulgarian object clitics.
V Results
1 Group results
a Topic Accusative Condition:
Figure 1 shows the distribution of the evaluation means for all groups in the Topic Accusative Condition. An evaluation of 5 on the acceptability scale means that the sentence is perfectly acceptable whereas an evaluation of 1 means totally unacceptable. As expected, the control group gave the highest appropriateness evaluation (4.67) to Option 1 (fronting + doubling) followed by the other pragmatically felicitous option, which included doubling with no fronting (4.26). The evaluations of the infelicitous options – Option 3 (fronting with no doubling) and Option 4 (SVO word order) – received evaluations of 2.13 and 2.89, respectively.
Topic Accusative Condition
It was interesting to observe that the evaluation means of the advanced group for the felicitous options were higher in absolute values than those of the control group (4.75 vs. 4.67 for Option 1, and 4.34 vs. 4.26 for Option 2, respectively), although those numbers are very close and by no means statistically different. The evaluation means of the advanced group for the infelicitous options were higher than those of the control group (2.73 and 3.29).
In order to establish the statistical significance of the observed values a two-way repeated-measures ANOVA was performed with evaluation option as a within-participants variable and group as a between-participants variable. The results showed a significant main effect of evaluation option (F 3,111 = 44.857, p < 0.001), a significant main effect of group (F 2,37 = 3.317, p = 0.047) and significant interaction (F 6,111 = 25.791, p < 0.001).
The results of a one-way repeated measures ANOVA within each of the groups showed a significant main effect for both the control and the advanced group of evaluation options (control: F 3,45 = 77.945, p < 0.001; advanced: F 3,27 = 38.464, p = 0.00). A post-hoc Tukey HSD comparison revealed statistically significant differences between the felicitous options (Option 1 and Option 2) and the infelicitous options (Option 3 and Option 4) for both the control and the advanced group. There was no statistically significant difference between the evaluations of the two felicitous options for either group. While there was no statistically significant difference between the evaluations of the infelicitous options for the advanced group, the control group evaluated the infelicitous Option 3 significantly lower than the other infelicitous option, which included neutral SVO word order. To a certain extent, although not relevant to the main issue investigated by this study, this was expected. The word order of Option 3 is used in Bulgarian most often for assigning focus on a fronted element, which is not the case in the context of the Topic Accusative condition and therefore the lower evaluations. The neutral SVO word order received an evaluation around 3, which was to some extent expected as a lot of the participants felt they had to give credit to the grammatical correctness of the sentence. Most important, however, is the fact that those evaluations were consistently lower than the evaluations of the felicitous options, and the subtle distinction in the interpretative appropriateness of the sentences was recognized by the participants in both the control and the advanced group.
A comparison between the ratings of the control group and the intermediate group revealed a dissimilarity, which was not observed in the comparison of the control group with the advanced group. The intermediate group valued the infelicitous SVO option (Option 4), which was similar to what their L1 used in a similar context, as highly appropriate. The second highest evaluation was given to the other infelicitous option (Option 3), which was also similar to one of the L1 options of marking topicality, namely a fronted object as in ‘Sushi, I like.’
A one-way repeated-measures ANOVA showed a significant main effect of evaluation option for the intermediate group (F 3,39 = 5.309, p = 0.004). The post-hoc Tukey HSD comparison revealed the source of that main effect, which was found to be in the statistically significant difference between the infelicitous SVO option, which had received the highest mean evaluation (4.33) and the felicitous Option 2, with the lowest mean evaluation (3.29). What was of greater importance, however, was that in most of the cases there were no statistically significant differences, which was indicative of the lack of knowledge on the part of the intermediate L2 learners as a group as to the pragmatic requirement for clitic doubling in topic constructions. In the single instance when their evaluations reached statistical significance, it occurred in favor of one of the infelicitous options, which was the opposite of what was observed in the data of the control group.
b Topic Dative Condition:
Figure 2 shows the distribution of the evaluation means of the control group and the advanced group in the Topic Dative Condition. The most important observation with regard to those evaluations is that both groups gave the highest scores to the felicitous Option 1 (4.61 and 4.69) as they did in the Topic Accusative Condition. Similarly, the infelicitous options(Option 3 and Option 4) received the lowest scores (3.18 and 3.32 for Option 3, and 3.21 and 3.05 for Option 4).
Topic Dative Condition
However, there were two main points in the above data that presented a significant departure from what was observed in the Topic Accusative Condition. First, the evaluation of the control group for the infelicitous Option 3 in the Topic Dative Condition was significantly higher than in the Topic Accusative Condition (3.18 vs. 2.13). 7 This ‘milder’ degree of infelicity appears to be also reflected in the evaluations of the advanced group when compared to their evaluations for the infelicitous Option 3 in the Topic Accusative Condition (3.32 vs. 2.73).
Second, the evaluation of the advanced group for the felicitous Option 2 (clitic doubling with no fronting) in the Topic Dative Condition is much lower than in the Topic Accusative Condition (3.70 vs. 4.34). A closer look at the individual results, however, shows that this is to a certain extent due to the very low evaluations (1.6 and 2.9) of two of the participants.
To establish the statistical significance of the observed values across the three groups, a two-way repeated-measures ANOVA between all four evaluation options and three groups was performed. The results showed no significant main effect of group (F 2,37 = 0.590, p = 0.560), but significant main effect of condition (F 3,111 = 19.021, p < 0.001) and significant interaction (F 6,111 = 16.220, p < 0.001).
The results of the one-way repeated measures ANOVAs within each of the groups showed a significant main effect of evaluation option for both the control and the advanced group (control: F3,45 = 44.325, p < 0.001; advanced: F 3,27 = 11.438, p < 0.001). A post-hoc Tukey HSD comparison revealed a statistically significant difference between the felicitous options (Option 1 and Option 2) and the infelicitous options (Option 3 and Option 4) for the control group. There were no statistically significant differences between the evaluations of the two felicitous options, on the one hand, and the two infelicitous options, on the other hand.
The only point of statistical significance in the evaluations of the advanced group was between the felicitous Option 1 and all other options, including the other felicitous option, Option 2. In this respect their results differed from the results of the control group whose evaluations of the two felicitous options were not statistically significant.
The comparison between the results of the control group and the intermediate group in the Topic Dative Condition showed the same pattern as in the Topic Accusative Condition. Once again, the intermediate group considered the infelicitous options, Option 3 and Option 4, highly appropriate. This was in line with the expectations for L1 transfer and preference for the two options that were similar to what was used in the L1 (English) in similar contexts.
A repeated-measures one-way ANOVA showed a significant main effect of evaluation option in the intermediate group (F 3,39 = 5.637, p = 0.003). The following Tukey HSD post-hoc comparison identified statistically significant differences between the infelicitous SVO option (Option 4) and the felicitous Option 1 and Option 2. Those, however, were in the wrong direction as the L1-like SVO option received the highest evaluation among all four options.
c Focus Accusative Condition:
The focus conditions reversed the felicity of the four answer options. Now, the clitic doubling options Option 1 (with fronting) and Option 2 (without fronting) were infelicitous since the introduction of new information as a response to a wh-question involved focal constructions, which were not consistent with the pragmatic function of clitic doubling in Bulgarian, namely to mark topicality.
Figure 3 shows the mean ratings of the control and advanced groups in the Focus Accusative Condition. Both the control and the advanced group gave very high ratings to the felicitous options, Option 3 and Option 4. On the other hand, the evaluations of the infelicitous, clitic doubling options, Option 1 and Option 2 received significantly lower evaluations. Although the evaluations of the advanced group for the infelicitous options were consistently lower than the evaluations for the highly rated Options 3 and 4, they were not as low as those of the control group (1.86 vs. 2.82, 1.43 vs. 2.09). This, however, will be a moot point if those evaluations are shown to be statistically different from the felicitous options, which can be the best indicator of the ability to perceive discourse-conditioned differences. 8
Focus Accusative Condition
A two-way repeated-measures ANOVA between all four evaluation options and three groups showed a significant main effect of group (F 2,37 = 20.146, p < 0.001), a significant main effect of evaluation option (F3,111 = 190.272, p < 0.001) and significant interaction (F 6,111 = 19.052, p < 0.001).
The one-way repeated-measures ANOVAs within each group revealed a significant main effect of evaluation options for both the control and the advanced group (control: F 3,45 = 438.885), p < 0.001; advanced: F 3,27 = 44.053, p < 0.001). The Tukey HSD post-hoc comparison identified the points of statistical significance, the most important of which were found to be between the felicitous options, Option 3 and Option 4, and the infelicitous ones, Option 1 and Option 2. In addition, there were statistically significant differences between the two felicitous options in the evaluations of the control group as they preferred the SVO order for focal constructions rather than the option of fronting with no doubling. The differences in the evaluations of the advanced group for the felicitous options were not statistically significant.
When compared with the evaluations of the control group, those of the intermediate group were also very high. However, the evaluation means of the intermediate group for the infelicitous options were much higher than those of the control group. The one-way repeated-measures ANOVA showed a significant main effect of the evaluation options of the intermediate group (intermediate: F 3,39 = 10.530, p < 0.001). The following Tukey HSD comparison revealed the points of significant differences between the evaluation means. The rating of the felicitous Option 4 was significantly higher than the evaluations of both infelicitous options, Option 1 and Option 2. This ‘native-like’ behavior in the evaluation of the felicitous options should not necessarily be taken as an indicator of knowledge on the part of the intermediate learners of the pragmatic conditions regulating the felicitous use of clitic doubling constructions.
A closer look at the evaluations of the intermediate group in the topic conditions (see Figures 1 and 2) shows a distinct preference for the infelicitous no-doubling SVO option. Considering their strongly non-native-like performance in the topic conditions, it was most likely that the ‘native-like’ performance of the intermediate learners in the focus conditions was due to their general preference for constructions without clitics and the L1-like SVO option rather than to any particular knowledge of the infelicity of clitic doubling with focal direct and indirect objects.
d Focus Dative Condition:
The results of the control and the advanced group in the Focus Dative Condition were very similar to their results in the Focus Accusative Condition (see Figure 4). Once again they both gave the highest evaluations to the felicitous options Option 3 and Option 4, and significantly lower evaluations to the infelicitous options Option 1 and Option 2. The evaluations of the advanced group for the infelicitous options were again higher than those of the control group but, most importantly, significantly lower than the evaluations of the felicitous options.
Focus Dative Condition
A two-way repeated-measures ANOVA between all four evaluation options and three groups showed a significant main effect of group (F 2,37 = 23.080, p < 0.001), a significant main effect of evaluation option (F 3,111 = 344.482, p < 0.001) and significant interaction (F 6,111 = 51.423, p < 0.001).
The one-way repeated-measures ANOVAs revealed a significant main effect in the evaluations of both the control and the advanced group (control: F 3,45 = 685.891, p < 0.001; advanced: F 3,27 = 56.115, p < 0.001). The Tukey HSD post-hoc showed statistically significant difference between the felicitous and the infelicitous options.
Similarly to what was observed in the Focus Accusative Condition, the evaluations of the intermediate group showed preference for the felicitous non-doubling options Option 3 and Option 4. However, their evaluations of the infelicitous options were much higher than those of the control and the advanced group. The repeated-measures ANOVA showed a significant main effect of evaluation options in the results of the intermediate group (intermediate: F 3,39 = 21.422, p < 0.001). However, the Tukey HSD post-hoc comparison revealed the lack of any significant difference between the two felicitous options and one of the infelicitous options (Option 1).
The results in the Focus Dative Condition were in support of the claim that the intermediate L2 learners were not aware of the pragmatic significance of clitic doubling, or the lack thereof, and the only reason they gave ‘native-like’, high evaluations to the felicitous non-doubling options in the focus conditions was their general preference for constructions with no clitics. The individual results of the L2 intermediate learners revealed two main evaluation patterns: they either gave high evaluations to all four options in all four conditions or gave consistently higher evaluations to the options with no clitic doubling (Option 3 and Option 4), which simply happened to be felicitous in the focus conditions.
2 Individual results
For each of the four conditions (topic accusative, topic dative, focus accusative and focus dative), a repeated-measures one-way ANOVA was performed on the individual evaluations of each participant followed by a Tukey HSD post-hoc comparison to determine whether the differences between the evaluations of the four options (fronting/doubling; fronting/no doubling; no fronting/doubling; no fronting/no doubling) were statistically significant. A close look at the individual results revealed that the evaluations fell into five main patterns, which could be categorized as felicitous or infelicitous.
Felicitous patterns:
the evaluations of the two felicitous options are higher and statistically different from the evaluations of both infelicitous options the evaluation of one of the felicitous options is higher and statistically different from the evaluations of both infelicitous options
Infelicitous patterns:
the evaluation of one (or both) of the felicitous options is higher but not statistically different from the evaluation of one (or both) of the infelicitous options the evaluation of one (or both) of the infelicitous options is higher but not statistically different from the evaluation of one (or both) of the felicitous options the evaluation of one (or both) of the infelicitous options is higher and statistically different from the evaluation of one (or both) of the felicitous options
a Control group:
The individual results of the control group in the topic conditions fell predominantly into the felicitous patters with the exception of one participant in the Topic Accusative Condition and two participants in the Topic Dative Condition. As observed earlier in this section, fronting with no clitic doubling (Option 3) received much higher evaluations in the Topic Dative Condition than in the Topic Accusative Condition. However, for the majority of native participants the difference in the evaluations of Option 3 and the evaluations of at least one (in most cases both) of the felicitous Options 1 and 2 (clitic doubling with fronting and clitic doubling without fronting, respectively) was statistically significant.
For 11 native participants (69%) in the Topic Accusative Condition and eight (50%) in the Topic Dative Condition, the results fell into the pattern whereby both felicitous options differed significantly from both infelicitous options. Four participants in the Topic Accusative Condition and six in the Topic Dative Condition evaluated only one of the felicitous options significantly higher than both infelicitous options. For only one control participant in the Accusative Topic condition and two control participants in the Dative Topic condition the differences in the evaluations of the felicitous and the infelicitous options were not statistically significant. Therefore, 94% of the control participants (15 out of 16) had a statistically significant preference for one or both of the felicitous options in the Accusative Topic condition, and 88% of the control participants (14 out of 16) had a statistically significant preference for one or both of the felicitous options in the Dative Topic condition.
In both focus conditions, the participants in the control group invariably gave evaluations of the felicitous options Option 3 and Option 4 that exhibited a statistically significant difference from the evaluations of both infelicitous options, Option 1 and Option 2.
b Advanced group:
With the exception of two advanced participants, the results of the advanced group in the Topic Accusative condition were native-like as most of them (8 out of 10) evaluated both felicitous options significantly higher than the infelicitous ones. In the Topic Dative condition, the same two participants performed within one of the infelicitous patterns. However, in the Topic Dative condition, there were also two participants who gave highest evaluations to one of the felicitous options (Option 1) but it did not reach statistical significance in comparison with one of the highly evaluated infelicitous options (Option 3).
In the focus conditions, all but one advanced learner assigned very high evaluations to the felicitous options (Option 3 and Option 4) as in almost all cases both of them differed significantly from the evaluations of the two infelicitous options.
c Intermediate group:
The majority of the 14 intermediate learners exhibited strong non-native-like behavior in their evaluations in the topic conditions. In most cases they assigned the highest evaluation scores to the infelicitous Option 4, which was not surprising considering the fact that SVO word order was what their L1 used in similar contexts. However, for the most part, the difference between their evaluations did not attain statistical significance. This was indicative of their lack of knowledge as to the pragmatic requirement for clitic doubling in topical constructions, which made them incapable of discriminating between the felicitous and the infelicitous options as they gave relatively high evaluations to all of them.
In the focus conditions, the results of some of the intermediate participants converged with the results of the control group. This, however, should not necessarily be taken as indication of any pragmatic awareness of the incompatibility between clitic doubling and focal constructions. It should rather be considered as a coincidental occurrence due to the intermediate L2 learners’ preference for the L1-like clitic-free SVO constructions, which just happened to be felicitous in the focus conditions.
However, the results of two participants presented a notable departure from the pattern observed among the intermediate learners. They assigned native-like evaluations not only in the focus conditions but also in the topic conditions.
3 Results from a study on the L2 acquisition of the syntax of Bulgarian clitics
In order to ascertain the degree to which the purely syntactic properties of Bulgarian clitics were acquired by the same participants, a study (Ivanov, 2011) parallel to the one described above was conducted. The study included a grammaticality judgment task investigating six syntactic properties of Bulgarian object clitics: pre-verbal (non-argument position), pre-verbal (non-Wackernagel) position, Tobler–Mussafia effect, object clitic clustering in ditransitive constructions, lack of coordination with full DPs, obligatory clitic doubling with specific predicates (see note 2). Participants whose grammaticality judgments on all six of those properties did not reveal statistically significant difference from the control group were assigned [+syntax] status. The results of that study revealed native-like performance on the part of all advanced L2 learners and some of the intermediate participants, and confirmed the delayed acquisition of properties at the discourse–syntax interface vis-à-vis properties in the domain of narrow syntax.
Acquisition outcomes from the two experiments (narrow syntax and syntax–discourse) were combined in a contingency table (see Table 3). The table was based on individual results and each cell in the table represented the number of successful learners. A contingency estimation of the acquisition of syntax and acquisition of pragmatics revealed a significance at χ2 = 14.505, p < 0.001. Therefore, the two types of properties were shown to be actually related in the acquisition process. Those results were in line with the findings of other studies (Iverson et al., 2008; Rothman, 2008), which showed that acquisition of properties related to the interface between discourse and syntax was preceded by acquisition of the narrow syntax.
Contingency of acquisition of syntax and acquisition of pragmatics
Note: χ 2 = 14.505, p < 0.001
The results of the experimental studies on the acquisition of Bulgarian clitic doubling as an overt marker of topicality and on the syntax of Bulgarian clitics, indeed, revealed an asymmetry between the acquisition of syntactic properties and properties contingent upon the integration of syntactic structures with discourse requirements. However, that asymmetry was not along the line of learnability (as suggested by the Interface Hypothesis) but antecedence. The appropriate L2 syntax was successfully acquired by L2 learners before the interface conditions that were based on syntactic properties. This delay in the acquisition of interface properties was reflected in the performance of the L2 learners in the two studies. Out of 13 participants who displayed native-like knowledge of the syntactic properties of Bulgarian clitics, 10 were also native-like in their treatment of topic-marking clitic doubling. Three of the participants with native-like syntax, however, were still not native-like with respect to the interface property. Eleven of the participants had acquired neither the syntax, nor the pragmatics of Bulgarian clitics and their responses showed strong evidence of L1 transfer. As expected, none of the participants in the studies had learned the discourse property before acquiring the relevant syntax.
VI Conclusions
In the experimental study of this article the advanced learners of Bulgarian with English as their native language performed like the native controls in both the Topic and the Focus condition of the context sentence evaluation task, which can be interpreted as successful acquisition of the pragmatic meaning of clitic doubling in Bulgarian.
Misanalysis of clitics as pronouns, and a subsequent L1 transfer, could arguably have been the cause of the more native-like performance of the participants in the Focus conditions. 9 However, we believe that the inclusion of the Focus condition provided further support as to whether the participants distinguished between clitic doubling in topic conditions and no clitic doubling in focus conditions. Establishing whether the participants performed differently in those two conditions with respect to the use of a doubling clitic was important in view of the fact that they could have interpreted clitic doubling as universally applicable with definite objects in Bulgarian (regardless of information structure) as is the case in Macedonian and Albanian. If the performance of the advanced group in the Focus condition were accounted for by L1 transfer alone, how would we account for the fact that their evaluations of the un-English-like focal construction with fronting (and obviously no clitic doubling), Option 3, received the same evaluation as the English-like SVO order (4.15 vs. 4.40 in the Accusative condition and 4.50 vs. 4.45 in the Dative condition were certainly not significantly different)?
English does not allow resumptive pronouns in either topic or focus sentences. Bulgarian requires resumptive clitics in topic sentences, but does not allow them in focus sentences. If the advanced participants had adhered to their L1 transfer strategy, they would have utilized that strategy in the Topic condition as well, which they did not. It is hard to believe that they did L1 transfer in the Focus condition but abandoned that strategy and opted for native-likeness in the Topic condition.
The results of the intermediate learners showed strong preference for the infelicitous non-clitic-doubling options in the topic conditions. Their preference for an SVO (English-like) word order as a response to most of the test items in the accusative and dative topic conditions could serve as an argument in favor of a strong L1 influence. This might be indicative of the fact that they were still unaware of the pragmatic significance of clitic doubling, although with further exposure to Bulgarian they were expected to perform like the advanced group.
The claims of the Interface Hypothesis for permanent L2 deficiency in interface coordination were not borne out by this study as learnability of interface properties by advanced L2 learners was shown to be possible. Here, it is important to note that in evaluating the results of the context sentence evaluation task as described in this article, the author adopted an interpretation of the Interface Hypothesis which contends for optionality effects and permanent L2 fossilization with regard to external interfaces. The optionality predicted by the Interface Hypothesis was not supported by the results of the advanced participants. What could be interpreted as a manifestation of an optionality effect were the multiple cases with intermediate participants who provided evaluations on the test items that did not exhibit statistically significant differences. However, an important point regarding the formulation of the Interface Hypothesis by Sorace and Filiaci’s (2006: 340) is its elusiveness:
narrow syntactic properties are completely acquirable in a second language, even though they may exhibit significant developmental delays, whereas interface properties involving syntax and another cognitive domain may not be fully acquirable.
That formulation makes a significant use of the word may, which detracts from the merit of the hypothesis by weakening its predictive power. Furthermore, it makes the hypothesis virtually unfalsifiable.
In order to validate a certain hypothesis, empirical tests need to be conducted and the results are compared to the predictions of the given hypothesis. Therefore, a firmer stance as regards the acquisition outcomes with external interface conditions needs to be taken, which will be either corroborated or refuted by experimental data. This issue was explored in more detail in Slabakova and Rothman (2011: 572), who also concluded that:
because the IH has been so important to the development of acquisition theory and an emerging conduit of cross-disciplinary collaborations, it is our hope that the IH will offer stronger positions once again and thus continue to meet the scientific burden of a testable hypothesis.
In its present form, the Interface Hypothesis will be able to accommodate all kinds of results as they will invariably fall into either the ‘may’ or the ‘may not’ condition.
Footnotes
Acknowledgements
This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors. The author is grateful to Roumyana Slabakova and Jason Rothman for their significant contribution to this work. All inconsistencies and errors are mine alone.