One of the important issues in second language (L2) acquisition research is how individuals acquire and represent L2 syntactic structures in early and later stages of L2 acquisition. Many bilingual theories assume that bilingual speakers can, at least at some point during the learning trajectory, share certain linguistic representations across their languages (e.g., Dijkstra & van Heuven, 2002; Flege & Bohn, 2021; Kroll & Stewart, 1994; Nicoladis, 2006; Paradis & Navarro, 2003). This is also the case for the syntax, where structures that share similarities between languages may be merged into abstract, cross-linguistic syntactic representations (Hartsuiker et al., 2004). Evidence for such shared representations stems from the observation that bilingual speakers are influenced by recently encountered structures in one language when formulating or comprehending sentences in the other language. For instance, if a Dutch-English bilingual is confronted with a passive sentence in Dutch (e.g., “de kok wordt aangeraakt door de clown” [the cook is being touched by the clown]), he or she will be more likely to produce a sentence in the passive voice in English (e.g., “the lawyer was fired by the salesman”), instead of the highly preferred active voice (e.g., “the salesman fired the lawyer”). This phenomenon of repeating or predicting recently-encountered syntactic structures across sentences, regardless of differences in meaning, is referred to as syntactic or structural priming (Bock, 1986) and showcases prediction of the upcoming structure as an important mechanism in both language comprehension and production (Jaeger & Snider, 2013; Pickering & Garrod, 2007). Since such priming also occurs in the absence of lexical, phonological, or meaning overlap, it is often considered evidence for the existence of abstract syntactic representations (for a review see Pickering & Ferreira, 2008). Hence, when priming also occurs across languages, one could argue that these abstract representations are shared or at least connected between languages (Hartsuiker et al., 2004; van Gompel & Arai, 2018).
Interestingly, cross-linguistic structural priming has been observed with a wide variety of language pairs, such as English-Dutch (Bernolet et al., 2007, 2013; Desmet & Declercq, 2006; Schoonbaert et al., 2007), English-French (Hartsuiker et al., 2016), English-Polish (Fleischer et al., 2012), English-Mandarin (Chen et al., 2013; Huang et al., 2019), English-Korean (Hwang et al., 2018; Shin & Christianson, 2009; Song & Do, 2018), English-Irish (Favier et al., 2019), English-Swedish (Kantola & van Gompel, 2011), English-Spanish (Flett et al., 2013; Hartsuiker et al., 2004), Cantonese-Mandarin (Cai et al., 2011; Huang et al., 2019; Liu et al., 2022), Dutch-German (Bernolet et al., 2007), Dutch-French (Hartsuiker et al., 2016), and Spanish-Swedish (Montero-Melis & Jaeger, 2020) (see Muylle et al., 2023, for an overview of studies).
Some of these studies found that structural priming patterns may differ across L2 proficiency levels (e.g., Bernolet et al., 2013; Favier et al., 2019; Hartsuiker & Bernolet, 2017; Kim & McDonough, 2008; Montero-Melis & Jaeger, 2020), although the direction of the effect seems to depend on whether there is lexical overlap between prime-target pairs or not. For instance, Bernolet and colleagues (2013) correlated the magnitude of structural priming effects with L2 proficiency and observed that cross-linguistic priming effects became larger with increasing proficiency, whereas within-L2 priming effects in the presence of lexical overlap became smaller as proficiency increased. On the other hand, a recent study that investigated priming between Cantonese and Mandarin did not observe any effect of proficiency (Liu et al., 2022), which suggests that proficiency has no effect on the sharing of syntax between highly similar languages.
Since the sharing of syntax seems to be influenced by proficiency, there are two contrasting views on how L2 learners develop syntactic representations in the new language: either from separate to shared representations (e.g., Hartsuiker & Bernolet, 2017) or from shared to separate representations (e.g., De Bot, 1992; Montero-Melis & Jaeger, 2020). The first view assumes that L2 learners start with building language-specific syntactic representations in their L2, which then gradually become shared with the L1 when these are similar enough. The second view states that L2 learners use their L1 knowledge to organize the L2 and with increasing proficiency, the L2 representations will become separated from the L1, resulting in more native-like processing in the L2. When it comes to more abstract forms of priming (i.e., priming across languages and/or in the absence of lexical overlap), existing evidence seems to be more compatible with the view that learners start with separate, language-specific representations that become shared over time (see Muylle et al., 2023, for a discussion).
This idea was further elaborated in Hartsuiker and Bernolet's (2017) developmental theory of shared syntactic representations. According to this theory, L2 learners go through a number of different stages to establish shared syntactic representations across languages. At the start of L2 learning, the learner does not have any L2 syntactic representations yet, while some lexical representations may already be in place. In order to produce sentences, the learner will therefore use explicit strategies to combine the lexical items by either a) copying and editing utterances from a more proficient speaker, or b) transferring L1 syntax to the L2 without knowing whether this will result in a grammatical utterance. Hence, learners in the first stage may show within-language and cross-linguistic priming, but only when they are able to recall the prime sentence. In a second stage, which could already take place after limited exposure, the learner may develop item-specific syntactic representations. For instance, they encounter the verb "give" with a noun-phrase (NP) containing the object, followed by a prepositional phrase (PP) containing the recipient (e.g., "the cook gave a ball to the clown"), so they learn that "give" is complemented with NP + PP. Assuming that L2 syntactic knowledge is gathered through exposure (using a form of Hebbian learning), it is highly likely that the learner will first develop syntactic representations for frequent verbs and structures, and only later (i.e., in the third stage of the theory) for less frequent verbs and structures. Learners in the second and third stage will show item-specific priming, but only for the lexical items for which they already have L2 syntactic representations in place. Importantly, such priming would take place irrespective of whether the learner is able to recall the prime sentence, since this is no longer required to formulate sentences in the L2. In the fourth stage, the learner starts to generalize syntactic structures across lexical items (e.g., the verbs "give" and "send" can both be followed by NP + PP), which means that they will show priming within the L2, regardless of whether lexical items are repeated between prime and target. Finally, in the fifth stage, learners will merge the new L2 syntactic representations with the existing L1 representations. From that moment on, they will show priming from the L1 to the L2 and vice versa, both in presence and absence of meaning overlap between the prime and target.
1.1 The PP02 learning paradigm
The predictions of Hartsuiker and Bernolet’s (2017) developmental theory were tested in a series of experiments that used an artificial language (AL) learning paradigm to study structural priming within and across languages from the onset of L2 acquisition (Muylle et al., 2021a, 2021b, 2021c; Muylle, Bernolet, et al., 2020). In this paradigm, L1 Dutch speakers learned to produce intransitive, transitive, and ditransitive sentences in the AL (named “PP02”) by means of five different tasks that were administered sequentially. The first task was a vocabulary learning task in which participants learned the AL nouns to describe depicted human figurines (e.g., cook, witch) and objects (e.g., ball, hat). Next, they watched short action movie clips (taken from the normed stimulus set by Muylle, Wegner, et al., 2020) in which the human figurines interacted with each other and with objects, while listening to a sentence in the AL that described these actions. Following this exposure task, participants did a movie-sentence matching task in which they were asked to watch two movie clips on the screen and match a presented AL sentence with the correct movie clip. Next, there was a sentence production task in which the participants described movie clips in the AL and received feedback. Finally, intra-lingual and cross-linguistic priming effects were assessed by means of a structural priming task. This task consisted of two parts. First, participants had to judge whether a Dutch or AL sentence (i.e., the prime) matched a movie clip and after that, they saw a new movie and were asked to describe the action (i.e., the target) in Dutch or the AL, depending on a visual cue. Transitive primes could be in the active or passive voice and ditransitive primes could have a prepositional object (PO, e.g., "the cook is giving a book to the clown") or double-object (DO) dative structure (e.g., "the cook is giving the clown a book"). Intransitive sentences (e.g., the clown is waving") acted as fillers.
The AL learning paradigm has proven to be successful in eliciting structural priming effects both within the AL and between the AL and Dutch from the first day of learning. In a multiple-session study, two experiments tested structural priming effects across five AL learning sessions (Muylle et al., 2021a). The first experiment used a relatively easy version of PP02 (e.g., no articles or verb conjugations), whereas in the second experiment, we administered a more difficult PP02 version (e.g., with articles and verb conjugations). Participants in both experiments were very fast in acquiring the AL and they already showed priming within the AL and from the AL to Dutch from the first day of learning. Priming from Dutch to the AL was present on the first day for transitives as well, but for ditransitives, there was only an effect from the second day onward. These results suggest that L2 learners may indeed start with L2-specific syntactic representations that are only later merged with L1 representations. However, since most priming effects were already present on Day 1 of testing, subsequent experiments using this paradigm only included one or two sessions.
A second series of studies used the PP02 learning paradigm to explore how similar the structures have to be across languages in order to be shared in the L2 learner's mind (Muylle et al., 2021c; Muylle, Bernolet, et al., 2020). One study (Muylle, Bernolet, et al., 2020) compared priming effects within the AL and between the AL and Dutch (that has SVO order and no overt case marking) across different versions of PP02 (i.e., a version with SVO order, one with SOV order, and one with case marking) in a between-subjects design and found similar effects across all versions, suggesting that differences in the presence of overt case marking or in word order across languages do not influence the extent to which learners share the representations of these languages. The other study (Muylle et al., 2021c) tested SVO and SOV PP02 structures in a within-subjects design and found that the presence of the SVO order in the AL hindered the learning and sharing of the SOV order.
Finally, Muylle et al. (2021b) studied the role of L1 and AL frequency distributions in priming from PP02 to L1 Dutch by comparing a PP02 version that had a DO bias versus one that had a PO bias in a between-subjects design. Here, immediate priming was not affected by the bias in the AL, but only by the Dutch bias (i.e., a PO bias). However, participants in the DO bias condition were more likely to produce DO structures in Dutch overall than participants in the PO condition, showing that the AL exerted an influence on the preference in Dutch, regardless of the immediate prime. Crucially, this experiment could not clarify whether learners first build L2 presentations for the most frequent structures and only later for less frequent structures, as predicted by the developmental theory.
In sum, the AL learning paradigm has proven to be a useful tool to mimic second language acquisition in a highly controlled context. Nevertheless, it is important to bear in mind that there are also important differences with natural language learning. For instance, since the participants need to be able to acquire the AL in the course of a few hours, the vocabulary size and number of different structures have to be sufficiently small. In addition, the participants' motivation and the learning context are different from learners' motivations and learning environments/contexts in real-life situations.
1.2 A different approach to proficiency
The multiple-session AL learning experiments (Muylle et al., 2021a) were not fully conclusive on whether L2 representations evolve gradually from being item-specific to abstract with increasing L2 proficiency as proposed by the developmental theory. An issue with these studies was that there was already evidence for the sharing of syntax across the L1 and the AL (assessed by structural priming) on the first day of AL learning. In other words, the sharing went so fast that it was not possible to observe the early stages of its development using an AL learning paradigm. These studies considered proficiency a feature that develops over time. Indeed, AL proficiency (i.e., average AL accuracy scores) increased from the first to the second session, but then reached a plateau.
Another way to deal with proficiency is by comparing priming effects in participants with high AL accuracy scores with those in participants that have low scores. In other words, the effect of AL proficiency on priming can also be studied by testing individual differences in learning on the first day of AL acquisition. Although several studies used the AL learning paradigm (Muylle et al., 2021a, 2021b, 2021c; Muylle, Bernolet, et al., 2020), the influence of AL accuracy on priming was not addressed in the individual studies because of the limited number of participants. However, in the current study, we combined the data from these methodologically similar studies into one big data frame. This allowed us to test Hartsuiker and Bernolet’s hypotheses on a sample of 336 participants. In addition, we investigated whether other individual differences such as L1 proficiency and working memory (WM) play a role in the learning of the AL and in the sharing of syntax across the AL and L1.
1.3 Working memory
One important variable to consider is WM capacity, since explicit memory mechanisms (which are restricted by WM capacity) play a crucial role in the first stage of the developmental theory. In this stage, learners have no L2 syntactic representations in place yet on which they can rely to formulate sentences. Therefore, they may imitate more proficient L2 speakers by copying and editing their utterances. Generally speaking, the developmental theory assumes a more important role for explicit memory processes in early vs. later stages of L2 acquisition, based on the finding that the magnitude of L2 priming with lexical overlap decreases with increasing proficiency (Bernolet et al., 2013; Hartsuiker & Bernolet's 2017 re-analysis of Schoonbaert et al., 2007). Coumel and colleagues (2024) recently tested this prediction by studying L2 priming with repeated verbs across low and high proficiency L2 speakers of French in a dual-task design (i.e., a letter series recall task between prime and target). In both groups, priming effects became smaller when the intervening task was more difficult (i.e., when letter series became longer). Interestingly, an exploratory comparison across proficiency levels suggested that less proficient speakers relied more on explicit memory of the prime than more proficient speakers, who fell back to their own preferences to formulate sentences. Apart from the developmental theory, the well-established lexical boost effect to priming (i.e., larger priming effects when there is lexical overlap between prime and target compared to when there is no such overlap) is often explained in terms of explicit memory mechanisms and hence also depends on WM capacity (e.g., Bock & Griffin, 2000; Chang et al., 2006, 2012; Reitter et al., 2011; Zhang et al., 2020).
To assess WM capacity in the AL learning studies, we mostly administered the forward and backward digit span tests (WAIS-IV subtests; Wechsler, 2008), except for the first multiple-session experiment, where we used the Operation Span (OSPAN) test (Oswald et al., 2015). In the digit span tests, participants repeat increasingly longer number sequences (either forward in the forward digit span task or backward in the backward digit span task) for as long as they can correctly recall them. The results of these tests provide an approximation of how many items can be retained in WM in the correct order. Because the backward digit span test requires transformation of the retained information, it may be more informative about the processing capacity of the participant’s WM than the forward digit span (Alloway et al., 2006). In addition, the backward digit span may involve visuo-spatial processing (e.g., Li & Lewandowsky, 1995; St Clair-Thompson & Allen, 2013).
In addition, we inspected whether digit span scores predicted priming. Participants with a high forward digit span score tended to show more priming than those with a lower score, but this effect was only present in ditransitive and not in transitive structures. Since ditransitive sentences contain more phrasal constituents than transitive sentences, it is not surprising that processing ditransitives requires more WM capacity, and hence, learners with higher WM capacity have more resources left to maintain the prime structure in WM than learners with lower WM capacity.
The backward digit span, on the other hand, showed a less straightforward relation with priming. In general, participants with high backward digit span scores showed smaller priming effects than participants with lower scores, which seems opposite to the effect of the forward digit span. However, a closer inspection of the data revealed that backward digit span scores mainly affected priming within L1 Dutch or from the AL to Dutch when there was lexical overlap. One explanation could be that participants with lower span scores rely more on explicit memory when formulating a new sentence and hence show more priming in conditions with lexical overlap. Participants with higher span scores, however, may have a better memory of sentences preceding the prime, which enables them to develop abstract syntactic representations at a faster pace. As such they rely less on explicit memory of the prime to formulate sentences and fall back on more implicit priming mechanisms (cf. Coumel et al., 2023). Indeed, priming effects that are supported by explicit memory of the prime tend to be larger than those that are not (e.g., Hartsuiker et al., 2008).
Since the analyses of the multiple-session study indicated a rather complicated relation between WM capacity and priming, we did not further test the relation between digit span scores and priming in the subsequent AL learning studies. Instead, these data were mainly used to control for individual differences between groups in between-subjects designs. Therefore, the current re-analysis also aimed to further examine the effects of the forward and backward digit span score. In addition, we studied the effect of the digit span scores on accuracy in the AL.
1.4 L1 proficiency
Another factor that may influence the learning of the AL is the proficiency in the L1. Many studies have shown that there is a strong relationship between L1 and foreign language learning ability (e.g., Cummins, 1984; Dufva & Voeten, 1999; Hulstijn & Bossers, 1992; Sparks & Ganschow, 1993). Concretely, it has been argued that mastering a certain skill (e.g., phonology) in the L1 serves as a basis for learning that same skill in the L2, and that there is a basic language learning mechanism that underlies learning in both L1 and L2 (e.g., Sparks, 1995; Sparks & Ganschow, 1993). As such, it may be interesting to study whether L1 proficiency also predicts accuracy in the AL. A positive relationship would further support the AL learning paradigm as an appropriate method to study natural L2 acquisition. In all studies using this paradigm, we measured the participant's score on the Dutch LexTALE test, a short vocabulary test that consists of 60 lexical decision trials (Lemhöfer & Broersma, 2012). Although the LexTALE test provides an estimation of the vocabulary size in particular, the scores correlate well with other language proficiency measures (Lemhöfer & Broersma, 2012). In the current re-analysis, we therefore also included the Dutch LexTALE score as a covariate in our models.
1.5 Research questions and hypotheses
In this re-analysis, we took together all (priming) results from the AL studies to answer the following three research questions:
1) How do individual differences influence accuracy in the AL?
For this, AL accuracy was taken as a dependent variable and we tested whether forward digit span, backward digit span, and LexTALE Dutch score influenced accuracy scores in the various AL tasks. Since L1 learning abilities are related to foreign language learning abilities (e.g., Cummins, 1984; Dufva & Voeten, 1999; Hulstijn & Bossers, 1992; Sparks & Ganschow, 1993), we expected an effect of the L1 LexTALE score on AL accuracy.
2) How do individual differences influence structural priming in general?
Here, we tested whether both digit span scores, the LexTALE Dutch score, and AL accuracy predicted priming effects overall. Since several theories assume a role for WM in priming (e.g., Bock & Griffin, 2000; Chang et al., 2006, 2012; Hartsuiker & Bernolet, 2017; Reitter et al., 2011; Zhang et al., 2020), we expected that both digit span scores would affect priming. For AL accuracy, we also predicted a positive effect, since Hartsuiker and Bernolet (2017) predicted more priming with increasing proficiency in most of the conditions.
3) How does AL accuracy influence structural priming in the different priming conditions?
To investigate the assumption of Hartsuiker and Bernolet's developmental theory that L2 syntactic representations become more abstract with increasing proficiency and finally become shared with existing L1 representations, we tested whether priming effects became larger with increasing PP02 proficiency. Concretely, we expected more priming for high vs. low proficiency participants in all priming conditions, except for the related PP02-PP02 condition. In other words, we predicted an interaction between Prime structure, Relatedness, Target Language, and AL accuracy. On the one hand, priming effects should be larger within the AL than from the AL to Dutch and should be larger for related prime-target pairs (i.e., with verb overlap) compared to unrelated pairs (i.e., without verb overlap). Moreover, this effect of relatedness is thought to be larger within the AL (i.e., the lexical boost effect) than between languages (i.e., the translation equivalent boost effect). On the other hand, priming effects should be larger with increasing proficiency, but only for the more abstract types of priming and not for related PP02-PP02 priming.