Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Decoding sentence planning across languages with EEG
Poster Session D, Thursday, October 1, 4:30 - 6:30 pm, Wangari Maathai
This poster is part of the Sandbox Series.
Xueyang Huang1, Jonathan R. Brennan1; 1University of Michigan
How bilingual speakers build syntactic structure during sentence production across languages remains an open question. Current psycho- and neurolinguistic models propose that sentence generation proceeds incrementally, either through hierarchical (Momma & Phillips, 2018; Giglio et al., 2024) or linear planning, but it remains unclear how these strategies manifest in bilinguals. The present study uses time-resolved neural decoding to track the dynamics of sentence structure planning during bilingual sentence production for constructions that differ in surface word order but share similar underlying structure. We focus on the production of verbs and perfective aspectual markers in Mandarin and English (have in English vs. verbal-le in Mandarin), which differ in surface order (aspectual marker preceding the verb in English, vice versa in Mandarin). Under a hierarchical account (Huang & Li, 2009), there should be an earlier neural activation of the perfective aspect than verbs in both languages. Under a linear account, activations of these two syntactic components should emerge in different orders between Mandarin and English. We will record EEG from 30 Mandarin-English bilingual adults performing two picture description tasks: phrase production and sentence production. In both tasks, participants describe simple cartoon scenes in either language using matched materials. In the phrase production task, we manipulate the presentation order of the verb cue and aspectual marker cue to create conditions that are either congruent or incongruent with the canonical word order of each language, while keeping the intended utterance constant across cue-order conditions. This tests whether neural representations during phrasal production are shaped more strongly by hierarchical relations or linear sequences. We also include (1) sentences with neutral aspect marking and (2) with progressive aspect markers which preserve the same surface order and underlying structure across languages. Analyses will use time-resolved multivariate decoding (Gwilliams et al., 2022; Zhao et al., 2025) to examine the aspectual representations during sentence planning and test whether decoders trained on phrase production generalize to sentence production, probing whether shared neural representations across phrasal and sentential structure building are driven primarily by hierarchical planning or linear sequencing. The hierarchical account predicts that decoding accuracy for aspectual markers will peak earlier than verb decoding in both languages and all constructions, despite their differing surface word orders. The linear account predicts that the relative timing of decoding peaks should reverse across languages for the perfective constructions only. In the phrase-to-sentence analysis, successful generalization would be reflected by above-chance decoding accuracy when classifiers trained on phrase production are tested on sentence production. We further predict higher RSA similarity between phrase and sentence planning periods when phrase-level ordering aligns with the underlying syntactic structure of the target sentence, even when surface word order is incongruent, such as Mandarin perfective marking. Such findings would suggest not only that bilingual speakers recruit shared, structure-sensitive neural mechanisms during sentence planning across languages, but also that human language production more generally relies on common abstract planning operations that are preserved despite cross-linguistic word order differences.
Topic Areas: Language Production, Multilingualism