Presentation

Search Abstracts | Symposia | Slide Sessions | Poster Sessions | Poster Slams

Using multivariate pattern analyses of EEG to examine linguistic transfer in second language learning

Poster C45 in Poster Session C, Friday, October 7, 10:15 am - 12:00 pm EDT, Millennium Hall

Victoria Ogunniyi¹, David Abugaber², Irene Finestrat³, Alicia Luque⁴, Kara Morgan-Short^5,6; ¹Honors College, University of Illinois Chicago, ²Department of Linguistics, University of Michigan, ³Department of Spanish and Portuguese, Northwestern University, ⁴Department of Language and Culture (ISK), UiT The Arctic University of Norway, ⁵Department of Hispanic and Italian Studies, University of Illinois Chicago, ⁶Department of Psychology, University of Illinois Chicago

Grammar learning is an important aspect of learning a second language (L2). However, theories disagree about the role of grammatical transfer between a learner’s L2 and their native language (L1) (e.g., Tsimpli & Dimitrakopoulou, 2007; Schwartz & Sprouse, 1996). Event-related potential (ERP) research on this topic has produced conflicting results, with some studies finding native-like ERPs to L2-specific grammar features (e.g., Alemán Bañón et al., 2014), whereas other studies show only trends toward sensitivity (e.g., Gabriele et al., 2021). This disparity may be attributed to variability during language processing resulting in qualitatively different ERP responses among individuals, in which negative and positive ERP components associated with linguistic processing (i.e., N400 and P600) may cancel each other out, leading to null results in group-averaged analyses even when individual participants show neural evidence of native-like grammar processing (Tanner, Goldshtein, & Weissman, 2018). Our study addresses this limitation in previous ERP examinations of linguistic transfer for different domains of L2 grammar via multivariate pattern analysis (MVPA), a machine learning method that accounts for individual differences in neural activity by calculating, for each participant, patterns in correlations between electrodes that best distinguish trial conditions (Fahrenfort et al., 2018). In our experiment, 52 native English speakers were recruited from second-year university Spanish courses who evidenced low-intermediate Spanish proficiency. Participants read Spanish sentences while performing a grammaticality judgment task during simultaneous electrophysiological recording (EEG). Sentences consisted of three conditions that tested either grammatical features shared between L1 English and L2 Spanish (30 grammatical/30 ungrammatical trials for subject-verb number agreement; 31 grammatical/31 ungrammatical trials for determiner-noun number agreement) or a feature unique to L2 Spanish (31 grammatical/31 ungrammatical trials for determiner-noun gender agreement). For each of these three conditions, MVPA was performed to distinguish grammatical vs. ungrammatical trials on -200 ms to 1200 ms relative to the critical word in each sentence. Analyses were performed via the ADAM toolbox (Fahrenfort et al., 2018) using a backward decoding model based on a linear discriminant analysis classifier with cross-class balancing of sample sizes and five-fold cross-validation and included correction for multiple comparisons using cluster-based permutation testing. Behavioral results showed above-chance accuracy in grammaticality judgments for each of the three conditions (subject-verb agreement d′=1.14, determiner-noun number agreement d′=0.95, determiner-noun gender agreement d′=0.93). For our EEG results, our MVPA decoding yielded above-chance trial classification accuracy to features shared between their L1 and the L2 (subject-verb number agreement, determiner-noun number agreement), but only in learners with higher proficiency. Above-chance trial classification accuracy was not found for features unique to the L2 (determiner-noun gender agreement). Our finding of significant MVPA decoding only for L2 grammatical features that are shared with the L1 is most consistent with transfer theories that posit differences for L2-unique grammatical features, although future studies should examine high proficiency L2 learners to fully test the predictions. More generally, this study represents (to our knowledge) the first use of MVPA to analyze L2 grammar processing, demonstrating how this technique can complement the ERP method to inform theoretical questions about learning L2s.

Topic Areas: Multilingualism, Morphology