More than a mere sequence: predictive processing of wh-dependencies in early bilinguals

1 Introduction

The idea that information processing is predictive rather than merely integrative is not new and so far it has generated important insights on various topics in cognitive neuroscience, including vision, attention, motor control, motor imagery, action understanding, music, language processing, emotional processing, executive functions, the theory of mind and so forth [1–4]. Growing evidence suggests that even though incoming stimuli are typically impoverished, noisy, and partial, they are highly structured: the brain relies on these structures to form models for making predictions, which are then matched against the incoming stimuli.

Research on sentence comprehension in adult monolingual speakers has suggested that, in addition to the immediate integration of incoming words into a syntactic structure as a sentence unfolds, predictive processing also takes place, with predictions on incoming information forming even before the information appears in the input [4–6]. According to this view, sentence processing is “driven by the predictive relationships between verbs, their syntactic arguments, and the realworld contexts in which they occur” [5, p. 247]. As for bilingual speakers, predictive processing in the second language (L2) appears to take place in highly proficient bilinguals [7]. So far, research on processing strategies in bilinguals has been largely focused on typologically non-distant languages and to some extent on typologically distant languages, but the variation in processing patterns found in bilingual speakers of distant languages has not been fully explored [8–12]. Yet, geographically close languages, even when typologically distant as for instance Spanish and Basque, may structurally influence each other [13], which in turn may have a wider effect on processing strategies in bilinguals (the author is grateful to an anonymous reviewer for this suggestion). Thus, it remains unclear whether in early bilinguals, who use typologically distant languages such as Basque and Spanish, predictive processing mechanisms in L2 operate as efficiently as in native speakers.

One goal of the present study was to determine whether young, cognitively healthy native speakers of Basque, who acquired Spanish by the age of 5, rely on the same processing strategies in comprehension of Spanish wh-dependencies as young, cognitively healthy native Spanish speakers. Assuming comparable accuracy scores between the two groups, we wanted to determine if there would be considerable differences in their response times (RTs). Considerable differences in comprehension RTs would indicate a difference in the ease of processing, which in turn may be due to a reliance on different comprehension strategies based on syntactic cueing [14]. Growing evidence suggests that syntactic cueing affects comprehension in cognitively healthy monolingual and bilingual persons, as well as in aphasic patients [15–18].

The research question investigated in the present study has merit, because Basque and Spanish are structurally different languages (section 2), and thus different syntactic cues may underpin different strategies in the comprehension of wh-dependencies in the two languages. If syntactic cueing strategies in L1 affect comprehension of wh-dependencies in L2 in early Basque-Spanish bilinguals, then predictions based on L1 will interfere with the predictions that would be made solely on the basis of L2 cues, which would result in longer RTs in bilinguals. Thus, compared to native Spanish speakers, early Basque-Spanish bilinguals may not simply “need more time” to integrate incoming information as a specific type of Spanish wh-dependencies unfolds, but instead they may rely on different, more time-costly comprehension strategies due to interference of predictions based on L1. The question whether predictive processing mechanisms in L1 and L2 operate on different timescales in early bilinguals is not only theoretically relevant, but also important for educational assessments as well as neurocognitive testing of populations with brain disorders.

2 Present study: linguistic background

Wh-dependencies are dependency relations between a wh-word, such as what, who, which, why, how etc. and another element in a sentence [19]. In syntax, a dependency relation is defined by the elements that form dependency as well as by the distance between them, which in turn depends on their syntactic hierarchy. The present study focused on direct and embedded questions introduced by interrogative pronouns quién (“who”) and qué (“what, which”), and relative clauses introduced by que. Linguistic theory postulates that these syntactic structures involve wh-movement, i.e. a syntactic operation that moves a wh-word to a position different from the one in which it originated, leaving behind a co-indexed trace (t) or gap [20]. The trace contains important information on semantic (“thematic”) roles, i.e. information on who did what to whom in a sentence (for example, the boy in The boy kissed the girl is assigned the role of agent, because the boy performed the action of kissing, whereas the girl is assigned the role of patient, meaning that the girl underwent the action of kissing). Crucially, the moved wh-phrase receives a thematic role via its trace.

In Spanish, like in English, the distance between a moved element and its gap is longer in object wh-structures, as in (2), than in subject wh-structures, as in (1):

1. ¿Quién_it_i comió una naranja?

   who ate an orange

   “Who ate an orange?”

2. ¿A qui“n. mordió_j el perrito t_jt_i?

   to whom bit the puppy

   “Whom did the puppy bite?”

The preposition a, as in example (2), marks animate objects in Spanish [21] and therefore it could serve as a processing cue [22, 23]. Since it appears before the moved wh-word, it signals an object structure, allowing the parser, i.e. sentence analyzer, to assign a temporary thematic role of patient even before encountering the gap (t_i). Therefore, reliance on this cue would facilitate comprehension of object wh-dependencies, cancelling out the processing differences between subject and more demanding object wh-dependencies [24].

Unlike Spanish, which overtly marks object position in a sentence and the role of patient, as shown in (2), Basque overtly marks subject position and the role of agent, as shown in (3):

1. Gizon-a -k emakume-a -ri liburu-a

   Man the woman the book the

   eman dio.

   give has

   “The man has given the book to the woman.” [25, p. 72].

The subject position in (3) is morphologically marked by ergative case (-k). Linguistic theory characterizes Basque as an ergative language and Spanish as an accusative language [26]. This means that in Spanish the subjects of transitive and intransitive verbs bear the nominative case and the object of transitive verbs bears the accusative case, whereas in Basque subjects of transitive verbs are assigned the ergative case, and subjects of intransitive and objects of transitive verbs are assigned the absolutive case [26].

Even though there is no unlimited variation in language structure, some structural differences between Spanish and Basque may be relevant to the processing question under investigation. For example, word orders differ in Spanish and Basque: Spanish is head-initial and its canonical word order is subject-verb-object (SVO), whereas Basque is head-final and its canonical word order is subject-object-verb (SOV) [26]. Furthermore, the difference in overtly marking agents (Basque) vs. patients (Spanish) may be critical in processing wh-dependencies in L2. In other words, the mismatch in overt marking of agents vs. patients in the two languages may affect L2 processing in early Basque-Spanish bilinguals and reveal predominance of the default L1 processing strategies.

In readily observable linear terms, the two languages share the order of appearance of these two roles in wh-questions, i.e. agent first, then patient in subject wh-questions, and patient first, then agent in object wh-questions. However, Spanish and Basque differ in the order of appearance of these roles in relative clauses (RCs): Spanish retains the same order in subject and object RCs as in subject and object wh-questions, whereas Basque reverses the order, having overtly marked agents appearing before unmarked patients in object RCs. Thus, if only the order of appearance affects comprehension strategies in bilinguals, one would expect to find group differences in RCs, but not in wh-questions.

In addition, since Basque does not employ the same cue in syntactic marking of animate objects that Spanish does, there is a possibility that early Basque-Spanish bilinguals do not utilize this cue in comprehension of Spanish object wh-dependencies. This would be indicated by their systematically slower comprehension of all object wh-dependencies relative to native Spanish speakers. On the other hand, early Basque-Spanish bilinguals may differ from native Spanish speakers in the comprehension of Spanish wh-dependencies regardless of syntactic cueing, showing generally longer response times. In this case, the bilingual group would be generally slower in all conditions compared to the native group. This would indicate that the impact of L1 syntax on comprehension of Spanish wh-dependencies in early Basque-Spanish bilinguals may be more complex than postulated in the previous hypothesis, involving for instance a combination of preferences in overt thematic role marking and specifics of word order.

3 Methods

Data analyzed in the present study was collected within a larger study on comprehension cuing strategies in elderly Spanish speakers [24]. Unlike the present study, which is focused on the comprehension of wh-dependencies in young Basque-Spanish bilinguals, the previous study on elderly persons’ comprehension of wh-dependencies did not investigate the potential effects of bilingualism.

4 Results

5 Discussion

Empirical testing of how an early age of acquisition of a language typologically distant from the native language interacts with the processing of specific principles of L2 syntax is yet to provide conclusive evidence on the extent of transfer from the L1 into L2 [31]. The purpose of the present study was to determine whether early bilingual Basque-Spanish speakers who acquired Spanish by the age of five comprehend wh-dependences in L2 with the same ease as native Spanish speakers. The main finding of the present study is that overall, the early Basque-Spanish bilinguals needed more time for the comprehension of Spanish wh-dependencies relative to native Spanish speakers. This is indicated by statistically significant group differences in RTs in 9 out of 10 conditions across three types of wh-dependencies. Since the tests allowed only a limited amount of time for answers, i.e., 5000 ms, the need for more time also affected accuracy, although in only one condition-object who embedded questions-resulting in a significantly lower score in the bilingual group. Given the lack of statistically significant group differences on the test of verbal working memory capacity (section 3.2), the longer RTs in bilinguals cannot be explained in terms of limited verbal working memory capacity that would affect syntactic analysis of the input in L2. Instead, the group differences appear to be due to differences in processing strategies.

The present findings clearly rule out the notion that early Basque-Spanish bilinguals comprehend Spanish wh-dependencies with the same ease as native Spanish speakers and they mostly agree with previous studies that indicate the effects of early setting of syntactic parameters in L1 on L2. Furthermore, the present study results argue against the hypothesis that early Basque-Spanish bilinguals use completely different processing strategies relative to native Spanish speakers because, if true, it would yield considerable group differences on all conditions, with no exemptions. Instead, the present data suggests that L1 strategies may interfere to some degree with the comprehension of Spanish wh-dependencies in early Basque-Spanish bilinguals. The interference appears to reflect competing comprehension strategies, motivated by the differences in native and nonnative morphosyntax. Since Basque and are structurally different languages, with different word orders (SVO, SOV) and differences in overt marking of thematic roles, it is not surprising that predictive processing strategies from L1 interfere with predictive processing strategies in L2, slowing down comprehension in L2 and, in paradigms with sufficiently limited time to respond (as in the present study), also affecting accuracy. Our data indicates a salient role of overt marking of thematic roles in these strategies and no apparent role of their linear order.

Regardless, the differential overt marking of agents vs. patients cannot explain a small portion of our data, i.e. lack of statistically significant group differences in RTs on embedded who subject questions: Basque systematically marks agents and Spanish systematically marks patients in this type of wh-dependencies as well. The result cannot be explained in terms of discourse linking [32], which assumes systematic differences between discourse-linked which questions and non-discourse-linked who questions, because of slower processing times of all other who questions and all which questions in early Basque-Spanish bilinguals. In addition, the relatively small number of participants per group indicates caution in the interpretation of present findings. Thus, the actual extent of L1 interference with the comprehension of Spanish wh-dependencies in early Basque-Spanish bilinguals requires further studies with larger samples, and the use of more nuanced experimental paradigms.

For example, mainstream psycholinguistics holds that bilingual speakers, even when they use only one language, retain simultaneous activation of both languages [33–34]. This evidence comes mostly from research investigating representations up to the lexical level. These representations contribute to higher levels of processing, including syntactic computations and representations. Yet, the simultaneous and continuous activation of two languages is cognitively costly and it violates the most basic principle of cognitive economy. This is particularly relevant for studies investigating syntactic representations in bilinguals. Some recently put forth proposals postulate that syntactic representations in L1 and L2 are shared between languages, providing evidence, for instance, from syntactic priming in speech production [35] and connectionist modeling of word-level comprehension [36]. However, it remains unclear how exactly coactivated or shared syntactic representations from L1 and L2 interact during spoken L2 comprehension to yield considerably slower sentence-level processing in L2 in early bilinguals. To tease apart the aspects of this interaction, future studies will focus on determining electrophysiological correlates of overt thematic role marking and other relevant predictive processes’ timescales in early Basque-Spanish bilinguals.

In conclusion, the purpose of the present study was to determine whether early bilingual Basque-Spanish speakers comprehend wh-dependences in L2, which they had acquired by the age of 5, with the same ease as native Spanish speakers. Early exposure to L2 appears to affect a person’s capacity to master an L2 [37], structural plasticity of the bilingual brain [38] and the neural correlates of L2 grammar processing [39]. The present study finding that syntax in L2 may not be processed with native-like ease, even when L2 is acquired before the age of 5, is consistent with the notion that differences in processing patterns [12, 40] may be more emphasized in speakers of typologically more distant languages (Indo-European such as Spanish vs. non-Indo-European such as Basque), affecting predictive processing in L2. The finding is also highly relevant for the discussion on whether syntax is separate or shared between languages in bilinguals, because it raises questions on how specific features of grammar, such as overt marking of thematic roles, co-operate and compete in typologically distant languages.