Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Shared lexico-semantic representations in bilinguals
Poster Session B, Wednesday, September 30, 4:30 - 6:30 pm, Wangari Maathai
Xuanyi Jessica Chen1, Esti Blanco-Elorrieta1; 1New York University
The bilingual brain must simultaneously distinguish between languages while supporting meaning representations that can generalize across them. How these competing demands are resolved remains unclear. On the one hand, semantic representations are often assumed to be shared across languages, consistent with evidence for an amodal semantic system (Binder & Desai, 2011; Lambon Ralph et al., 2017) and prior demonstrations of cross-language decoding (Buchweitz et al., 2012; Correia et al., 2014; 2015). On the other hand, semantic access occurs through lexical systems that are strongly language-specific, raising the possibility that language context may fundamentally shape lexical-semantic representations during online processing. Here, we asked whether bilinguals maintain language-specific lexical-semantic representations or whether semantic and lexical information are represented in a shared representational space despite explicit language membership information. We recorded MEG data from 22 Spanish-English bilinguals during a phrase-completion task. Participants viewed visually presented nouns followed by auditory cues instructing them to complete a phrase aloud. To induce a strong language context, trials were organized into blocked English and Spanish contexts signaled before each block. Stimuli included non-cognate translation equivalents, cognates, and pseudowords. In addition, nouns appeared in singular or plural form, allowing us to test whether abstract semantic dimensions generalize across languages. We performed time-resolved sensor-level decoding with temporal generalization analyses during visual word presentation and assessed significance using cluster-based permutation tests. We first asked whether language membership itself is encoded during visual word processing. Language context could be decoded from non-cognate words beginning approximately 100 ms after stimulus onset, and this representation generalized to cognates and pseudowords between approximately 300–450 ms. Critically, language context remained decodable even when visual input was identical across languages, indicating that bilinguals rapidly encode context-relevant language-membership information during word processing. Having established that language context is represented, we next asked whether lexical-semantic representations nevertheless generalize across languages. We observed cross-language generalization for lexicality (real words vs. pseudowords) at approximately 150–200 ms, 300–350 ms, and 500–550 ms. Translation-equivalent words showed cross-language generalization of semantic information at approximately 100–200 ms, 300–350 ms, and 500–600 ms. Critically, we also found cross-language generalization for semantic number (singular vs. plural) from approximately 50–350 ms. Unlike translation-equivalent decoding, this analysis targeted an abstract semantic dimension that generalized across different lexical items and across languages. These findings therefore suggest that bilinguals do not merely share item-level representations across languages, but also represent higher-order semantic feature structure in a common representational space. Together, these findings suggest that bilingual word processing simultaneously encodes language-membership information while maintaining shared lexical-semantic representations across languages. Rather than relying on fully separate representational systems, bilingual language processing appears to involve integrated representational geometry that preserves language context while supporting a shared semantic organization across languages.
Topic Areas: Multilingualism, Meaning: Lexical Semantics