Introduction
Over the past decades, research on bilingual or second language (L2) processing has made major strides in discovering an integrated, nonselective nature of bilingual memory representations (Dijkstra, Reference Dijkstra, Kroll and De Groot2005; Dijkstra & Van Heuven, Reference Dijkstra and Van Heuven2002; Dussias & Sagarra, Reference Dussias and Sagarra2007; Gollan et al., Reference Gollan, Forster and Frost1997; Sanoudaki & Thierry, Reference Sanoudaki and Thierry2015). Within this integrated system, bilinguals are presumed to have access to the knowledge of languages known to them at every level of representation. One notable characteristic of the integrated system of bilingual representations is the influence of knowledge of one language while processing another language, commonly referred to as crosslinguistic influence (CLI) in the field of bilingual processing (Jarvis & Pavlenko, Reference Jarvis and Pavlenko2008; Lago et al., Reference Lago, Mosca and Stutter Garcia2021; Kroll & Tokowicz, Reference Kroll, Tokowicz, Kroll, Annette and De Groot2005).
Contrary to the bulk of evidence supporting CLI in bilingual word processing, it remains less clear how bilinguals’ linguistic knowledge interferes across languages at the structural level, particularly in (morpho)syntactic processing (see Lago et al., Reference Lago, Mosca and Stutter Garcia2021, for review). One issue of ongoing debate concerns the extent to which bilinguals or L2 learners show native-like processing of L2 structures diverging from their L1 counterparts (Tolentino & Tokowicz, Reference Tolentino and Tokowicz2011; Trenkic et al., Reference Trenkic, Mirkovic and Altmann2014). Some researchers propose persistent difficulties when target structures lack overlapping properties with learners’ L1 (e.g., Andersson et al., Reference Andersson, Sayehli and Gullberg2019). In contrast, some theoretical models predict reduced processing difficulty when target structures are uniquely represented in the L2, causing less conflict between languages compared to situations where crosslinguistic competition arises from L1–L2 differences (e.g., MacWhinney, Reference MacWhinney, Robinson and Ellis2008, Reference MacWhinney, Gass and Mackey2013; Tokowicz & MacWhinney, Reference Tokowicz2005; Tolentino & Tokowicz, Reference Tolentino and Tokowicz2011). Since empirical support for each of these accounts is derived from discrete studies employing a variety of experimental designs and involving learner populations with diverse L1 backgrounds and proficiency levels, it is difficult to draw conclusive evidence as to the extent to which different types of crosslinguistic cues present processing challenges for L2 learners.
The goal of the current study is to address this unresolved issue by exploring the effect of CLI in L2 syntactic processing under various conditions involving both interlingual and intralingual contrasts. To this end, we compared two L2 groups whose L1s differ in terms of the target structures (e.g., Kim & Joo, Reference Kim and Joo2021; van Dijk et al., Reference van Dijk, Dijkstra and Unsworth2022). Specifically, we involved (Mandarin) Chinese- and Japanese-speaking learners of Korean with closely matched proficiency and compared their processing behaviors across two types of Korean morphological causative constructions. These constructions, derived from intransitive and transitive verbs, respectively, are expected to give rise to varying degrees of CLI for these learner groups.
This study goes beyond previous research by considering CLI in the course of integrating multiple types of information. During sentence processing in head-final languages like Korean and Japanese, comprehenders are required to construct a mental model that assigns syntactic and semantic representations to preverbal arguments, retain this information in memory until encountering the verb, and update their interpretations of the sentence by integrating the argument structure and verbal information (Kamide et al., Reference Kamide, Altmann and Haywood2003; Koh, Reference Koh1997). This process becomes particularly crucial in the processing of morphological causatives in Korean because, as will be discussed in detail below, different types of causative constructions require distinct case-marking sequences for the arguments (O’Grady, Reference O’Grady1991; Sohn, Reference Sohn2001). Therefore, for the successful processing of the causative constructions in Korean, comprehenders need to consider the compatibility of a causative predicate and the status of its argument structure encoded in case marking.
In two online processing experiments, we examined the effect of CLI on the integration of case-marking information and the causative verb under the following conditions: when the case-marking constraints in the target construction have complete overlap between the learner’s L1 and L2, when they lead to crosslinguistic competition, and when they constitute L2-unique information. By focusing on advanced-level learners, we aimed to ensure that participants had sufficient proficiency in Korean to process causative constructions in real time. This allowed us to investigate whether CLI persists at higher proficiency levels, even when learners have acquired considerable L2 competence. While previous research has shown that CLI effects may diminish with increasing proficiency (e.g., Hopp, Reference Hopp2017; Sagarra & Herschensohn, Reference Sagarra and Herschensohn2010), it remains unclear whether this applies equally to all types of crosslinguistic relationships (overlap, competition, and L2-unique information). By examining advanced learners’ processing of these different conditions, we aimed to contribute to a more detailed understanding of how proficiency interacts with various types of CLI in L2 sentence processing.
Before presenting our results, we will briefly overview the morphosyntactic constraints imposed on the formulation of morphological causative constructions in Korean and discuss how these constraints present CLI for Chinese and Japanese speakers. We will then review a number of theoretical accounts that provide different predictions on how such CLI effects influence the processing of the target constructions for Japanese- and Chinese-speaking participants.
Case-marking constraints in the morphological causative constructions in Korean
Our focus on the Korean morphological causative constructions is motivated by their distinct linguistic constraints on the preverbal arguments, giving rise to interlingual and intralingual contrasts for Mandarin Chinese (hereafter “Chinese”) and Japanese speakers. Korean causative constructions are broadly categorized into morphological and syntactic types (O’Grady, Reference O’Grady1991; Shibatani & Pardeshi, Reference Shibatani and Pardeshi2002; Sohn, Reference Sohn2001). While syntactic causatives convey a causative event via two predicates, one belonging to the matrix and another to the embedded clause, morphological causatives utilize verbal inflectional morphology to express causation within a single clause. Crucially, morphological causatives necessitate certain types of case markers for the causee (i.e., an entity undergoing the event initiated by the causer), depending on whether the causative predicate is derived from an intransitive or a transitive verb.
When the causative predicate is derived from an intransitive verb, thus denoting the causee’s intransitive action, the causee must be exclusively marked by the accusative marker –(l)ul, as illustrated in (1a). Because the causee in this construction bears the role of a direct object of the predicate, the dative- and nominative-marked causee renders the sentence ungrammatical (Shibatani & Chung, Reference Shibatani, Chung, Noriko and Strauss2001). This restriction is equally applicable to the Japanese translation counterpart (Shibatani & Chung, Reference Shibatani, Chung, Noriko and Strauss2001; Shibatani & Pardeshi, Reference Shibatani and Pardeshi2002), as shown in (1b). Consequently, for Japanese-speaking learners of Korean, the case-marking constraint in sentences like (1a) can be positively transferred from their L1. For Chinese speakers, these constraints represent L2-unique information because Chinese lacks a case-marking system, and the Chinese causative construction only instantiates the syntactic type (Huang et al., Reference Huang, Li and Li2009), as in (1c).
Unlike intransitive-based morphological causatives, transitive-based morphological causatives in Korean permit multiple types of case markers to be used with the causee. Specifically, the causative sentence (2a) allows the causee to be marked by either the accusative marker -(l)ul or the dative marker -eykey, while excluding the nominative marker -ka (for discussion, see O’Grady, Reference O’Grady1991; Yang, Reference Yang1994). For the Japanese translation counterpart (2b), in contrast, only the dative marker -ni is allowed for the causee, precluding the modification of the accusative marker -o and the nominative marker -ga (Shibatani & Chung, Reference Shibatani, Chung, Noriko and Strauss2001; Shibatani & Pardeshi, Reference Shibatani and Pardeshi2002). This discrepancy may induce crosslinguistic competition for Japanese speakers, particularly when encountering Korean transitive-based morphological causatives involving an accusative-marked causee, because this sentence form is acceptable in Korean but unacceptable in the Japanese translation counterpart. On the other hand, for Chinese speakers, this restriction constitutes L2-unique information that should be newly acquired because their L1 neither involves explicit case marking for the arguments nor includes morphological causatives, as illustrated in (2c).
In summary, the similarities and differences in the morphosyntactic operations involved in causative constructions across Korean, Japanese, and Chinese constitute crosslinguistic cues that are expected to present varying degrees of CLI effects for Chinese- and Japanese-speaking learners of Korean. Exploring how speakers of these languages process Korean morphological causative constructions can thus serve as an ideal diagnostic to assess the effects of different types of crosslinguistic cues in L2 sentence processing.
Theoretical models on crosslinguistic influence in L2 syntactic processing
While the role of CLI has been a key focus in research on bilingual processing, there is no general consensus on the extent to which crosslinguistic similarities and differences influence the processing of (morpho)syntactic information in bilingual sentence processing (Lago et al., Reference Lago, Mosca and Stutter Garcia2021; Roberts, Reference Roberts and van Gompel2013). One crucial yet largely unresolved issue concerns whether L2 learners can successfully process target structures that are uniquely represented in their L2.
Several studies have demonstrated that L2-unique information poses lingering processing difficulties for L2 learners (e.g., Chen et al., Reference Chen, Shu, Liu, Zhao and Li2007; Cho, Reference Cho2022; Grüter et al., Reference Grüter, Lew-Williams and Fernald2012; Ionin et al., Reference Ionin, Choi and Liu2021; Ojima et al., Reference Ojima, Nakata and Kakigi2005; Rossi et al., Reference Rossi, Gugler, Friederici and Hahne2006). The challenges associated with processing structures that do not overlap with the learners’ L1 are consistent with various theoretical positions. One such viewpoint is the blocking or overshadowing account, which emphasizes the significant role of prior knowledge in the acquisition of new information in an L2 (Ellis, Reference Ellis2006; Ellis & Sagarra, Reference Ellis and Sagarra2010, Reference Ellis and Sagarra2011; Ellis et al., Reference Ellis, Hafeez, Martin, Chen, Boland and Sagarra2014; Luk & Shirai, Reference Luk and Shirai2009; Paul & Grüter, Reference Paul and Grüter2016). The essential claim of this account is that previously acquired information in the L1 is deeply entrenched and thus occupies a privileged status in the learner’s mind. Consequently, the acquisition of new information in an L2 is subject to “interference, overshadowing and blocking, and perceptual learning, all biasing the ways in which learners selectively attend to their second language” (Ellis, Reference Ellis2006, p. 188). Extending this idea to L2 processing, it is hypothesized that an L2 structure containing no overlapping features with the L1 counterpart will present considerable difficulty because L2 learners are already tuned to L1-based parsing routines for the structure, which may overshadow the learning and deployment of new information during L2 processing.
Similar predictions are made by the Morphological Congruency Hypothesis (Jiang et al., Reference Jiang, Novokshanova, Masuda and Wang2011) and the Feature Reassembly Hypothesis (Lardiere, Reference Lardiere2009). According to the Morphological Congruency Hypothesis, successful L2 processing depends primarily on the presence of morphological correspondence between the learner’s L1 and L2. When target morphological features are represented differently or are absent in the L2, learners may encounter substantial challenges in acquisition and processing. Similarly, the Feature Reassembly Hypothesis suggests that L2 learners have difficulties in reassembling linguistic features from their L1 to construct their L2 grammar, especially when the target structures lack corresponding features in their L1. Like the blocking or overshadowing account, both of these hypotheses predict persistent difficulties for L2 learners when processing structures that do not receive a converging characterization in their L1 and L2.
There is other evidence, however, suggesting that L2-unique information does not always lead to processing difficulty. One influential theoretical perspective in line with this evidence is the Competition Model (MacWhinney, Reference MacWhinney, Robinson and Ellis2008, Reference MacWhinney, Gass and Mackey2013), which assumes that the degree of difficulty associated with L2 processing is contingent on the extent to which L2 learners experience competition among cues. According to this model, sentence comprehension incurs processing costs arising from cue detection and selection, as comprehenders constantly evaluate the relative strength of available cues and select the most reliable ones. The model posits that these processing costs increase when conflicting information exists between a target structure in the L2 and a corresponding structure in the L1. Conversely, the model predicts less difficulty when L2 learners process target information that is uniquely represented in the L2, as little crosslinguistic competition is expected in this case. Evidence supporting the Competition Model comes from studies showing L2 learners’ convergence on native-like processing when they process L2-unique information, for example, the processing of English articles by learners from article-lacking L1 backgrounds (e.g., Trenkic et al., Reference Trenkic, Mirkovic and Altmann2014), the processing of English verb-particle constructions by learners from L1 backgrounds without phrasal verbs (e.g., Herbay et al., Reference Herbay, Gonnerman and Baum2018), and the processing of Korean honorific agreement by English speakers (e.g., Mueller & Jiang, Reference Mueller and Jiang2013).
The aforementioned theoretical models provide distinct predictions regarding the processing of Korean morphological causative constructions by Japanese and Chinese speakers. According to the blocking or overshadowing account, the Morphological Congruency Hypothesis, and the Feature Reassembly Hypothesis, Japanese speakers are predicted to experience processing difficulties with transitive-based causatives, which show crosslinguistic differences between Korean and Japanese, but less difficulty with intransitive-based ones, which align with the L1 counterpart. For Chinese speakers, these models predict overall difficulties with both intransitive- and transitive-based causatives because Chinese lacks corresponding features. Likewise, the Competition Model predicts that Japanese speakers will experience more processing difficulty with transitive-based causatives, which show L1–L2 contrast, compared to intransitive-based ones, which exhibit L1–L2 overlap. However, the model predicts that Chinese speakers will have less difficulty with both intransitive- and transitive-based causatives (compared to Japanese speakers) because the case-marking constraints in these constructions are unique to the L2 and, therefore, cause little crosslinguistic conflict.
Contrary to these predictions, it is also conceivable that no CLI effect emerges in the processing of causative constructions. Some studies suggest that L2 syntactic processing is guided in principle by the same parsing routines that operate in L1 processing (e.g., Omaki & Schulz, Reference Omaki and Schulz2011; Witzel et al., Reference Witzel, Witzel and Nicol2012). This account assumes that L2 learners can approximate native processing when they possess sufficient proficiency and language experience (see Cunnings, Reference Cunnings2017, for discussion). Extending this idea to the current study’s context, we might observe native-like performance among L2 learners, regardless of crosslinguistic similarities and differences, when they have high proficiency and substantial exposure to the L2.
On the other hand, certain accounts predict no effects of CLI based on the premise that L2 syntactic processing rests on fundamentally different parsing routines compared to the L1 counterpart (e.g., Clahsen & Felser, Reference Clahsen and Felser2006, Reference Clahsen and Felser2018). Additionally, numerous studies suggest that the integration of multiple information sources during L2 processing can pose lingering difficulties for L2 learners, resulting in divergent performance from native speakers (e.g., Sorace, Reference Sorace2011; Sorace & Serratrice, Reference Sorace and Serratrice2009). In the context of the current study, given the cognitively taxing process of computing the compatibility between case marking on preverbal arguments and the causative predicate, L2 learners might struggle to construct a precise and detailed syntactic representation during the processing of Korean causative constructions, thereby exhibiting different processing patterns from native speakers.
The current study aimed to test these predictions by conducting two self-paced reading experiments involving Japanese- and Chinese-speaking learners of Korean, whose proficiency levels were closely matched. This study is guided by the following research questions (RQs).
RQ1. How do L1–L2 overlapping cues (for Japanese speakers) and L2-unique cues (for Chinese speakers) affect Japanese- and Chinese-speaking learners’ sensitivity to case-marking violations in intransitive-based morphological causatives in Korean? (Experiment 1)
RQ2. How do L1–L2 contrasting cues (for Japanese speakers) and L2-unique cues (for Chinese speakers) affect Japanese- and Chinese-speaking learners’ sensitivity to case-marking violations in transitive-based morphological causatives in Korean? (Experiment 2)
Experiment 1: Intransitive-based morphological causatives
Participants
A total of 96 college students participated in this experiment, including 32 native speakers of Korean as a control group (NS group: 25 females and 7 males), 32 Chinese-speaking learners of Korean (Chinese group: 31 females and 1 male), and 32 Japanese-speaking learners of Korean (Japanese group: 26 females and 6 males). Participants in the Chinese and Japanese groups were recruited among international students enrolled at universities in Seoul, South Korea. They identified themselves as native speakers of the respective language (Chinese or Japanese) and reported growing up speaking the language since childhood.
To ensure that L2 participants had sufficient proficiency for processing the target constructions, we established inclusion criteria that restricted participation to highly advanced learners who had achieved a level within the two top tiers (Level 5 and Level 6) in the Test of Proficiency in Korean (TOPIK), an official Korean proficiency test. According to the test developers’ level descriptors, these two levels indicate language skills required for using the language in professional or academic fields. Thus, we assumed that our L2 participants would have little difficulty understanding our linguistic stimuli.
In addition to the L2 learners’ TOPIK levels, we also assessed their current Korean proficiency using a C-test (Lee-Ellis 2009). In the C-test, participants were asked to read passages containing 50 blanks and provide a word in each blank based on the first few letters provided. Descriptive statistics on the proficiency scores are presented in Table 1. The C-test scores between the two learner groups did not significantly differ, t(62) = 0.977, p = .332, Cohen’s d = 0.244, confirming that the two learner groups were closely matched in their Korean proficiency. However, given individual variability, we included the proficiency scores in our analysis models as additional factors to investigate whether individuals’ proficiency interacts with CLI effects in influencing L2 sentence processing.
Table 1. Experiment 1: summary of participant information
Note. The values in the parentheses indicate standard deviations
Despite the comparable proficiency of the two groups, we observed group differences in certain areas, indicating differences in their L2 learning experiences (see Table 1). Specifically, the Chinese group spent a longer time studying Korean than the Japanese group, t(62) = 2.716, p = .009, Cohen’s d = 0.679, while the Japanese group began Korean learning at an earlier age than the Chinese group, t(62) = 4.554, p < .001, Cohen’s d = 1.139. However, the two groups did not significantly differ in the length of residence in Korea, t(62) = 0.684, p = .497, Cohen’s d = 0.171. To account for individual variability in L2 learning experiences, we included each of these indices as an interactive variable in our analyses.
All participants provided informed written consent prior to the main experiment and received monetary compensation for their participation.
Materials
The experimental items consisted of 12 Korean intransitive-based morphological causative sentences, each presented on a word-by-word basis across five regions (Rs), as illustrated in (3).
To test participants’ sensitivity to case-marking errors, the case marker for the causee in Region 2 (R2) was systematically manipulated. Half of the experimental items were felicitous, with an accusative-marked causee (e.g., aki-lul ‘baby-Acc’). The other half were infelicitous, with a dative-marked causee (e.g., aki-eykey ‘baby-Dat’). Region 3 (R3), containing the causative predicate (e.g., cay-we-se ‘sleep-Caus-and’), was identified as the earliest possible point for detecting syntactic anomalies. This region served as the critical focus for our analysis. To capture any delayed processing effects, Region 4 (R4), which included the following locative phrase (e.g., chimtay-ey ‘bed-Loc’), was analyzed as a spillover region. Finally, the verb phrase in Region 5 (R5, e.g., noha-sse-yo ‘lay-Past-Decl’) marked the conclusion of each sentence.
Based on the assumption that syntactic ungrammaticality disrupts processing and increases reading times (RTs) (e.g., Kaan et al., Reference Kaan, Ballantyne and Wijnen2015), we predicted that any sensitivity to the syntactic violation in our experimental stimuli would lead to increased RTs at the critical and/or the spillover region in the ungrammatical sentences (i.e., dative condition) compared to the grammatical counterparts (i.e., accusative condition).
The experimental items were counterbalanced for the case-marking type in R2 across two lists, and each participant encountered only one type of each item (6 items per condition). The stimuli were interspersed with 56 fillers containing a variety of structures. Among them, 12 items included transitive-based morphological causatives with either an accusative- or a dative-marked causee (counterbalanced across two lists), as in (4). The inclusion of transitive-based morphological causatives as fillers was crucial because participants needed to be alerted to the presence of causative sentences fully compatible with a dative-marked causee (in contrast to the experimental stimuli where a dative causee leads to a grammatical violation). These fillers served to prevent participants from strategically generating expectations about the grammaticality of the critical sentence prior to the causative verb based on the case-marking sequence.
The words used in the experimental and filler items were taken from the vocabulary list designed for intermediate-level Korean learners, as suggested by the International Standard Curriculum of Korean Language (Kim et al., Reference Kim, Kim, Kang, Kim, Kim and Lee2011). The whole list of experimental items appears in Appendix A, which is accessible through the OSF repository.
Procedure
Each participant completed the task individually in a quiet lab. Prior to the experiment, a research assistant provided guidance for accessing PCIbex Farm (https://farm.pcibex.net/) using a lab computer. Stimuli were displayed on a computer screen, region-by-region, employing the non-cumulative moving-window format (Just et al., Reference Just, Carpenter and Woolley1982). Before the main experiment, participants completed the language background questionnaire, received oral and written instructions for the task, and practiced with five sample items. In each trial, a series of dashes appeared on the screen, indicating the position of each word in the target sentence. Participants disclosed each word by pressing the spacebar at their own pace. The per-word RT, operationalized as the duration between the initial and the next bar press, was automatically recorded by the program. After reading the target sentence, participants were prompted to answer a statement verification question by clicking on “yes” or “no” presented on the screen. After responding, they proceeded to the next trial without feedback on their answer. The task took approximately 20–30 minutes.
Results
We first checked participants’ accuracy in answering the verification questions. All three groups demonstrated a high accuracy rate: 91.8% (SD = 3.7) in the NS group, 85.1% (SD = 9.9) in the Chinese group, and 89.1% (SD = 8.0) in the Japanese group. These results indicate that participants generally paid attention to the sentence meaning during the task.Footnote 2
Before analyzing the RT data, we removed RTs shorter than 100 milliseconds (ms) and longer than 5000 ms as outliers, which affected 1.6% of the data (0.2% in the NS group, 0.7% in the Chinese group, and 3.7% in the Japanese group). We also eliminated RTs that deviated beyond three standard deviations from the grand mean (2.3% in total; 2.4% in the NS group, 2.4% in the Chinese group, and 2.3% in the Japanese group). The remaining RTs were log-transformed for data normalization (Ratcliff, Reference Ratcliff1993) and then converted to residual RTs to adjust for differences in individuals’ reading speed (Trueswell et al., Reference Trueswell, Tanenhaus and Garnsey1994).
The residual RT profiles after data trimming are presented in Figure 1 for the NS group, Figure 2 for the Chinese group, and Figure 3 for the Japanese group. Raw RTs can be found in Appendix B. A visual examination of the graphs reveals that all three groups showed increased RTs in the dative condition compared to the accusative condition in the critical (R3) and following regions (R4 and R5).
Figure 1. Experiment 1: Native speaker group’s reading time profile. Error bars indicate 95% confidence intervals.
Figure 2. Experiment 1: Chinese group’s reading time profile. Error bars indicate 95% confidence intervals.
Figure 3. Experiment 1: Japanese group’s reading time profile. Error bars indicate 95% confidence intervals.
To analyze these patterns in detail, we fit linear mixed-effects regression models to the residual RTs in the critical (R4) and spillover regions (R5). The analyses were conducted using the lme4 package (Bates et al., Reference Bates, Maechler, Bolker and Walker2015) in R version 4.2.1 (R Core Team, 2022). Each model included fixed effects of group (NS, Chinese, Japanese) and case marking (accusative, dative). The fixed effect of group was Helmert coded, with one contrast comparing the NS group to the two learner groups and another contrast comparing the Chinese group to the Japanese group. The effect of case marking was centered around the mean and contrast-coded by assigning −.5 to the accusative condition and .5 to the dative condition. The initial models included the maximal random-effects structure allowed by the design, with random intercepts for participants and items, as well as a by-participant random slope for case marking and a by-item random slope for group. In cases where the model did not converge, we simplified it by excluding either the by-item or the by-participant slope. The outcomes of the models are presented in Table 2.
Table 2. Experiment 1: model summary for region 3 and region 4
*** : p < .001.
In the critical region (R3), there was a significant effect of group (NS vs. L2 groups), with longer RTs in the learner groups than in the NS group. Critically, the model returned a significant effect of case marking, with increased RTs in the dative than the accusative condition. This effect did not interact with group, suggesting that all three groups performed equally. This result was further supported by separate analyses for each group. We conducted separate linear mixed-effects regression analyses for each group, including the fixed effect of case marking (centered and contrast-coded), with an adjusted alpha level of .017 (.05/3) after Bonferroni correction. The results showed a robust effect of case marking for the NS group (β = 0.118, SE = 0.039, p = .004), the Chinese group (β = 0.119, SE = 0.046, p = .015), and the Japanese group (β = 0.111, SE = 0.042, p = .016), all demonstrating a clear tendency toward longer RTs in the dative condition than the accusative condition.
To examine whether the observed sensitivity was related to any processing difficulties in the previous region, not as a result of participants’ sensitivity to the grammatical violations, we fitted another mixed-effects model to the RT data in the second region (R2). The results revealed no main effects of group, case marking, or their interaction (all ps > .1). These results suggest that the effect of case marking observed in R3 was specifically linked to the participants’ grammatical sensitivity, rather than being a consequence of processing patterns carried over from the previous region.
We further investigated whether the effect of case marking interacted with the learner-specific variables, including C-test scores, onset of L2 learning, duration of L2 learning, and duration of residence in Korea. In these analyses, we added each of these variables (z-score transformed and centered around the mean) to each learner group’s model. However, the results indicated that none of these variables exhibited a significant interaction with case-marking condition, both for the Chinese and Japanese groups (all ps > .05).
Turning to the model for the spillover region (R4), the only significant effect observed was for case marking, indicating longer RTs in the dative condition compared to the accusative condition. Further separate by-group analyses by group revealed a significant effect of case marking for the NS group (β = 0.188, SE = 0.041, p < .001), the Chinese group (β = 0.110, SE = 0.035, p = .003), and the Japanese group (β = 0.100, SE = 0.036, p = .010) at the adjusted alpha level. When including the learner-related variables (proficiency, onset of L2 learning, duration of L2 learning, and duration of residence in Korea) in each learner model, no significant interaction was found between any of these variables and case marking (all ps > .05).
Discussion
Experiment 1 aimed to investigate the effects of L1–L2 overlapping cues and L2-unique cues in the L2 processing of Korean intransitive-based morphological causatives. The analyses of RT data showed that both the Chinese and Japanese groups exhibited sensitivity to case-marking violations in the target sentences, as reflected by their increased RTs in the dative condition compared to the accusative condition at the critical and spillover regions. These processing patterns were comparable to the native speaker performance. The proficiency and L2 learning experience of the learner groups did not significantly impact these outcomes.
These findings are consistent with the Competition Model, which predicted native-like processing for both Chinese and Japanese groups by virtue of the absence of expected crosslinguistic competition. Moreover, the processing patterns of the L2 groups could be interpreted in line with theoretical perspectives proposing identical mechanisms underlying L1 and L2 processing (e.g., Omaki & Schulz, Reference Omaki and Schulz2011; Witzel et al., Reference Witzel, Witzel and Nicol2012) and the idea that sufficient proficiency and L2 experience allow learners to achieve native-like processing (e.g., Cunnings, Reference Cunnings2017; Dussias & Piñar, Reference Dussias and Piñar2010).
To further investigate the effects of L1–L2 competing information and L2-unique cues, we conducted another self-paced reading task (Experiment 2), focusing on transitive-based morphological causatives. These constructions introduced crosslinguistic competition for Japanese speakers while representing L2-unique information for Chinese speakers.
Experiment 2: Transitive-based morphological causatives
Participants
Because intransitive- and transitive-based morphological causatives are closely associated both structurally and semantically, we recruited new groups of participants for Experiment 2, who did not participate in Experiment 1. They included 38 native Korean speakers as a control group (NS group: 25 females and 13 males), 38 Chinese-speaking L2 learners (Chinese group: 30 females and 8 males), and 38 Japanese-speaking L2 learners (Japanese group: 36 females and 2 males). They were drawn from among graduate and undergraduate student pools at universities in Seoul. Participant details are presented in Table 3.
Table 3. Experiment 2: summary of participant information
Note. The values in the parentheses indicate standard deviations.
As in Experiment 1, we only included L2 participants who achieved at least Level 5 on TOPIK. To ensure comparable proficiency across participants in the two experiments, the new learner groups’ Korean proficiency was assessed using the C-test in the same manner as in Experiment 1. The C-test scores of the four learner groups (Chinese and Japanese speakers from both experiments) did not significantly differ, F(3, 136) = 0.719, p = .542, η 2 = 0.016, confirming that the learner groups across both experiments had similar proficiency levels.
When comparing learner-related variables across the four learner groups, we observed some differences. Specifically, the Chinese group from Experiment 2 had a significantly earlier onset of L2 learning compared to the Japanese group from Experiment 1 (p = .039). In addition, the Chinese group from Experiment 1 had a longer duration of studying Korean than the Chinese group from Experiment 2 (p = .020). However, there were no significant differences in the duration of residence in Korea across the four groups (p = .076). Similar to Experiment 1, these learner-related variables, along with the proficiency scores, were included in the analysis models to examine their potential interaction with CLI effects.
All participants provided written consent and received monetary compensation for their participation.
Materials
The experimental stimuli for Experiment 2 included 12 Korean transitive-based morphological causatives, each presented word-by-word across eight regions, as illustrated in (5) (see Appendix C for the list of items). Participants were randomly assigned to one of two lists counterbalanced for the case-marking type for the causee (accusative, nominative). Assuming that sensitivity to the ungrammaticality of sentences with a nominative-marked causee (nominative condition) leads to increased RTs at the causative verb, we analyzed R4 as the critical region and R5 as a spillover region.
The experimental items were intertwined with 60 fillers containing various types of structures. Twelve fillers were constructed analogously to the experimental sentences but involved a syntactic causative predicate, as shown in (6). Unlike morphological causatives, syntactic causatives in Korean allow the causee to be modified by the nominative, as well as the accusative and dative markers. Therefore, the inclusion of syntactic causatives as fillers served to prevent any strategic processing that relies solely on preverbal arguments.
Procedure
The procedure was identical to that of Experiment 1.
Results
All three groups showed high accuracy rates in the comprehension check-up questions: 91.4% (SD = 4.3) in the NS group, 88.8% (SD = 3.4) in the Chinese group, and 89.3% (SD = 3.1) in the Japanese group. These results confirmed the participants’ close attention to the task.Footnote 3
Prior to data analysis, we trimmed RTs in the same manner as Experiment 1. We first eliminated RTs shorter than 100 ms and longer than 5000 ms (0.4% in total; 0.1% in the NS group, 0.4% in the Chinese group, and 0.7% in the Japanese group). Also removed were RTs beyond 3 standard deviations from the mean (2.6% in total; 2.2% in the NS group, 2.5% in the Chinese group, and 2.6% in the Japanese group). As in Experiment 1, the trimmed RTs were converted to log-transformed residual RTs and then were converted to log-transformed residual RTs. Raw RT results are presented in Appendix D.
The residual RT profiles for each group are presented in Figure 4 (NS group), Figure 5 (Chinese group), and Figure 6 (Japanese group). Notably, the three groups showed varying degrees of RT differences between ungrammatical (nominative condition) and grammatical sentences (accusative condition) in both the critical and subsequent regions.
Figure 4. Experiment 2: Native speaker group’s reading time profile. Error bars indicate 95% confidence intervals.
Figure 5. Experiment 2: Chinese group’s reading time profile. Error bars indicate 95% confidence intervals.
Figure 6. Experiment 2: Japanese group’s reading time profile. Error bars indicate 95% confidence intervals.
To assess the statistical significance of these differences, we conducted linear mixed-effects regression in the same manner as in Experiment 1. In both the critical region (R4) and the spillover region (R5), no significant effects associated with group, condition, or their interaction were observed (all ps > .05). Although the NS group displayed a tendency towards prolonged processing time for sentences in the nominative condition compared to the accusative condition in the critical region (as visible in Figure 4), this difference achieved only marginal significance (p = .063). These results suggest that none of the groups consistently demonstrated sensitivity to the grammatical violation in the critical and spillover regions. However, upon careful inspection of Figures 4, 5, and 6, noticeable differences in RTs between the two conditions emerged after the spillover region, particularly for the NS and Chinese groups. Therefore, we decided to conduct an additional analysis to explore the processing patterns of each group in Region 6.
Table 4 presents the model outcomes in this region. The model revealed a significant effect of group (NS vs. L2 groups), with longer RTs in the learner groups compared to the NS group. We also found a main effect of case marking, driven by longer RTs in the nominative condition compared to the accusative condition. This effect demonstrated a significant interaction with group when comparing the NS and L2 groups, and a marginal interaction when comparing the Chinese and Japanese groups. These interactions suggest that the effect of case marking emerged to different degrees across the groups.
Table 4. Experiment 2: model summary for region 6
*** : p < .001, **: p < .01, *: p < .05.
To scrutinize these interactions in detail, we constructed separate linear mixed-effects regression models for each group, each including the fixed effect of case marking (contrast-coded and centered), and the random effects of participants and items. The alpha level for these analyses was adjusted to .017 after the Bonferroni correction. The results revealed a significant effect of case marking for the NS group (β = 0.159, SE = 0.054, p < .001) and the Chinese group (β = 0.086, SE = 0.031, p = .005), but not for the Japanese group (β = –0.040, SE = 0.029, p = .166). When adding the learner-specific variables (proficiency, onset of L2 learning, L2 learning duration, residence in Korea) to each L2 group’s model, no significant interactions were found between any of these variables and case marking (all ps > .05).
Discussion
In Experiment 2, we investigated the effects of L1–L2 competing cues and L2-unique cues in the L2 processing of Korean transitive-based morphological causatives. The results showed grammatical sensitivity in Region 6 among the NS group and the Chinese group, but not the Japanese group. The delayed effect of case marking observed in the NS and Chinese groups may be attributed to the linguistic complexity of the target structure. The transitive-based causatives involved an additional argument marked with an accusative case, which was inserted between the theme and a causative verb. This additional argument could potentially introduce agreement attraction (e.g., Kwon & Sturt, Reference Kwon and Sturt2019), serving as a distractor that interferes with the process of checking compatibility between the case marker attached to the theme and the verb. As a result, the detection of grammatical errors might have been delayed, leading to the delayed effect of case marking for the NS and Chinese groups.
Despite the processing delay, the Chinese group converged on native-like processing by successfully detecting the agreement violation. In contrast, the proficiency-matched Japanese group did not show such sensitivity. These findings suggest that L2-unique cues, as experienced by the Chinese group, present fewer processing difficulties compared to the L1–L2 competing cues encountered by the Japanese group. In the following section, we discuss these findings in conjunction with the outcomes from Experiment 1 within the framework of the theoretical approaches reviewed earlier.
General discussion
The objective of this study was to explore whether L2 learners can engage in native-like processing when presented with morphosyntactic cues that manifest varying degrees of CLI effects. In two self-paced reading experiments, we found that L1–L2 contrasting cues posed greater challenges for L2 learners in comparison to L1–L2 overlapping and L2-unique cues. The proficiency scores of the L2 participants, along with their onset of L2 learning, duration of L2 learning, and duration of residence in Korea, did not influence their processing behaviors.
The overall findings of this study highlight the significant role of crosslinguistic competition as a substantial obstacle in L2 sentence processing. Our statistical models revealed no significant differences in overall RTs between Japanese and Chinese speakers in both Experiment 1 (see Table 2) and Experiment 2 (see Table 4). This lack of significant group effects suggests that the two groups exhibited comparable processing speeds, which aligns with our efforts to match the proficiency levels of the participants. However, despite similar overall RTs, the groups displayed distinct patterns in their sensitivity to grammatical violations under specific conditions across experiments. In Experiment 1, both the Japanese and Chinese groups successfully detected case-marking errors in intransitive-based morphological causatives, where the case-marking constraints presented L1–L2 overlapping cues for the Japanese group and L2-unique cues for the Chinese group. However, in Experiment 2, the Japanese group failed to exhibit sensitivity to case-marking errors in transitive-based morphological causatives, where the target sentences contained features that conflicted with their L1 counterparts. This finding contrasted with the Chinese group, who successfully detected the violation. Given that the target morphosyntactic constraints were absent in the learner’s L1, thus constituting L2-unique cues, this outcome aligns with previous research indicating that L2 learners can effectively utilize information specific to the target language to engage in native-like processing (e.g., Herbay et al., Reference Herbay, Gonnerman and Baum2018; Kim, Reference Kim2025; Tokowicz & MacWhinney, Reference Tokowicz2005; Trenkic et al., Reference Trenkic, Mirkovic and Altmann2014). Consequently, we conclude that L1–L2 contrasting cues, rather than L1–L2 overlapping or L2-specific cues, pose specific challenges for L2 learners.
The findings from both experiments provide support for the Competition Model of L2 processing (MacWhinney, Reference MacWhinney, Robinson and Ellis2008, Reference MacWhinney, Gass and Mackey2013). Consistent with this model, the processing challenges observed among L2 participants with Korean morphological causatives varied contingent on the presence of crosslinguistic competition. The Competition Model ascribes difficulties arising from crosslinguistic competition to the processing costs associated with computing the relative weight of available cues from both L1 and L2 (MacWhinney, Reference MacWhinney, Gass and Mackey2013). Applying this framework to our study, we can hypothesize that the Japanese group encountered processing difficulties with L1–L2 contrasting cues because a substantial amount of their memory resources had to be allocated to suppress the influence of the competing information from their L1. In contrast, the absence of such competition for the Chinese group may have allowed them to more effectively use their freed-up memory resources, contributing to their successful detection of grammatical errors in the target sentences.
The current study also provides a unique insight into the differential effects of crosslinguistic cues in the integration of multiple types of information. As previously noted, the real-time processing of Korean morphological causatives requires comprehenders to incrementally construct syntactic representations that integrate previously encoded argument structure information with a causative verb. Although integrating multiple types of cues during processing might impose cognitive demands on L2 memory resources (Sorace, Reference Sorace2011; Sorace & Serratrice, Reference Sorace and Serratrice2009), our L2 participants had less difficulty in this integration when the target structures shared overlapping features with their L1 or when they presented L2-unique information. These findings suggest that the presence of crosslinguistic interactions may play a more prominent role in L2 syntactic processing than the sheer number of information types that need to be integrated.
Along with these implications, some caveats must be acknowledged when arguing for the role of CLI. Each experiment involved causatives with differing structures—intransitive causatives in Experiment 1 and transitive causatives in Experiment 2. As a reviewer pointed out, differences in the structural complexity of these sentences could present a potential confound in comparing the results between the two experiments. While distinct outcomes were obtained within each experiment, the findings cannot be directly applied to a single group’s processing behavior across experiments due to the varying complexity of the target structures. For example, the Japanese group’s insensitivity to the grammatical violations in Experiment 2, but not in Experiment 1, may not solely result from the L1–L2 contrasting features in transitive causatives. Instead, it could also reflect the greater structural complexity of transitive compared to intransitive causatives: Transitive causatives involve longer sentences and additional arguments, thus imposing increased processing demands. To address this concern, future research should aim to disentangle these effects by using matched structures with comparable sentence lengths and argument structures while systematically varying degrees of crosslinguistic competition. Such a design would allow for a more precise examination of the interaction between CLI and structural complexity in L2 processing.
Another important limitation pertains to the delayed effect of grammatical sensitivity observed among the native speaker group and the Chinese groups in Experiment 2. This finding raises the possibility that structural complexity may have influenced the results, confounding the interpretation of grammatical sensitivity. As previously noted, more complex structures likely require greater cognitive resources, which could have independently impacted the participants’ performance, alongside the effects of CLI. While this possibility does not diminish the observed group differences, it points to the need for further research to separate the effects of structural complexity from those of grammatical processing in the context of CLI.
In contrast to the pronounced effects of crosslinguistic cues, we did not find any interacting effects of learner-related variables, including proficiency, onset of L2 learning, duration of L2 learning, and duration of residence. It is important to note that our participants were highly advanced, with substantial knowledge of the target structures. While some studies indicate that with increasing proficiency, L2 learners progress from exhibiting L1 transfer to converge on native-like processing (e.g., Hopp, Reference Hopp2017; Sagarra & Herschensohn, Reference Sagarra and Herschensohn2010), our findings suggest that CLI effects persist even among highly proficient L2 learners. These findings indicate that CLI may be a robust influence in L2 syntactic processing, resistant to the moderating effects of proficiency.
Such results align with previous research demonstrating the robust effects of L1 across learners with diverse proficiency levels and learning experiences (e.g., Andersson et al., Reference Andersson, Sayehli and Gullberg2019; Hopp & Grüter, Reference Hopp and Grüter2023; Mickan & Lemhöfer, Reference Mickan and Lemhöfer2020; Rankin, Reference Rankin2014). These findings lend support to the theoretical perspective that bilinguals or L2 learners possess an integrated system of syntactic representations shared between their L1 and L2, operating in a parallel and non-language-selective manner (Bernolet et al., Reference Bernolet, Hartsuiker and Pickering2009; Hartsuiker et al., Reference Hartsuiker, Pickering and Veltkamp2004; Serratrice, Reference Serratrice2016). The notion of integrated, shared syntactic representations in L2 learners was well reflected in the consistent effects of CLI observed in the current study.
Alternatively, the absence of effects from the L2-related variables in this study might be attributed to their limited variability. As previously mentioned, we recruited highly proficient learners for both experiments, resulting in L2 groups with participants who shared similar L2 learning experiences and prolonged naturalistic exposure to L2 Korean. This homogeneity might have constrained the variability of learner-related factors, potentially obscuring their interaction with CLI effects. To address this limitation, future research should involve learners with diverse proficiency levels and L2 learning experiences. If those learners consistently demonstrate processing patterns influenced by the manipulation of crosslinguistic cues, as witnessed in our participants, it would provide compelling evidence for the persistent effects of CLI in L2 syntactic processing.
Conclusion
The current study furnishes evidence that L1–L2 contrasting cues, but not L1–L2 overlapping and L2-unique cues, disrupt L2 learners’ ability to detect morphosyntactic violations during the processing of Korean morphological causatives. These findings suggest that crosslinguistic competition constitutes a major source of difficulty during L2 processing, resonating with the Competition Model (MacWhinney, Reference MacWhinney, Robinson and Ellis2008, Reference MacWhinney, Gass and Mackey2013) and non-selective, integrated models of bilingual syntactic representations (e.g., Hartsuiker et al., Reference Hartsuiker, Pickering and Veltkamp2004). This study extends previous research by directly comparing the effects of L2-unique cues with those of L1–L2 converging and L1–L2 contrasting cues using the same experimental setup. To advance our understanding of the mechanisms involved in bilingual or L2 syntactic processing, future research should investigate the interaction between different types of crosslinguistic cues across a wide range of syntactic phenomena and among learners with varying proficiency levels and L2 learning experiences.
Replication package
The data that support the findings of this study are openly available in the OSF repository, at https://osf.io/736hb/ .
Competing interests
The author(s) declare none.
Open access
