Decoding of lexical items and grammatical features in EEG: A cross-linguistic study

Area of Science:

• Neuroscience
• Psycholinguistics
• Cognitive Science

Background:

• Bilingualism research suggests language-invariant conceptual and grammatical representations.
• Previous studies often focus on typologically similar languages, leaving cross-linguistic morphosyntactic and lexical representation less understood.
• Investigating typologically different languages is crucial for understanding universal versus language-specific neural mechanisms.

Purpose of the Study:

• To investigate the neural representation of morphosyntactic features and lexical concepts in Korean-English bilinguals.
• To determine if these representations are language-invariant or language-specific across typologically different languages.
• To compare the effectiveness of event-related potentials (ERPs) and neural decoding for analyzing EEG data in bilinguals.

Main Methods:

• Utilized electroencephalography (EEG) data from eighteen Korean-English bilinguals.
• Employed machine learning for neural decoding to analyze brain activity while participants read singular/plural nouns and present/past tense verbs in both languages.
• Compared within-language and between-language decoding accuracy for lexical and morphosyntactic information.

Main Results:

• Event-related potentials (ERPs) showed limited ability to differentiate lexical, number, or tense information.
• Neural decoding successfully classified lexical and morphosyntactic information within each language (Korean and English).
• Decoding of lexical items and tense did not generalize between Korean and English; only number information showed cross-linguistic decoding.

Conclusions:

• Bilinguals exhibit robust, language-specific neural representations for lexical items and morphosyntactic features.
• Number information appears to have a shared neural representation pattern across typologically different languages (Korean and English).
• Neural decoding is a sensitive method for detecting subtle linguistic representations in EEG data, outperforming traditional ERP analysis in this context.