Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Neural Encoding of Semantic and Grammatical Dimensions of Parts of Speech: A Cross-linguistic MEG Study
Poster Session B, Wednesday, September 30, 4:30 - 6:30 pm, Wangari Maathai
This poster is part of the Sandbox Series.
Zirui Huang1, Natalia Bekemeier1, Nicola Molinaro3,4, James Magnuson3,4,5, Marianne Hundt1, Alexis Hervais-Adelman1,2; 1University of Zurich, 2University of Geneva, 3BCBL: Basque Center on Cognition, Brain and Language, 4Ikerbasque: Basque Foundation for Science, 5University of Connecticut
Parts of speech (PoS) are fundamental to linguistic theory, yet traditional PoS labels such as nouns, verbs, and adjectives conflate multiple linguistic levels: the lexical-semantic root, the grammatical function a word serves in context, and how root and function are combined. For example, a word with an action-denoting root like “sleep” can serve a predication-function in “she is sleeping” or a modification-function in “a sleeping baby”, entailing different linguistic operations to derive the intended word categories. Furthermore, PoS perception can also be affected by language-specific morphosyntactic cues, making it difficult to determine whether neural PoS effects reflect lexical semantics, grammatical function, or language-specific surface cues. This ongoing work adopts a decomposition of PoS, analysing lexical items typically assigned as nouns, verbs and adjectives in terms of their semantic root type (object, action, property) and grammatical function (reference, predication, modification). This parametric framework allows us to ask whether neural responses during naturalistic comprehension track these dimensions independently, and whether our brain encodes their interaction, recasting PoS as the relation between a word’s meaning and its in-context function. By applying this framework to data from three typologically distinct languages, English, Chinese, and Basque, we also ask which aspects of PoS representation are shared across languages and which are language-modulated. We collect and analyze source-localized MEG data from participants listening to naturalistic narratives. Each content word is assigned a semantic root and a grammatical function. Temporal response function encoding models are used to predict MEG responses across cortical parcels from these theoretical predictors, together with control factors including word frequency, surprisal, and acoustic features. Model comparisons assess the unique contribution of each theoretical predictor by comparing a full model against reduced models in which the relevant predictor set is reduced. Group-level effects are assessed across subjects using permutation-based statistics across source-space cortical parcels and time lags. Preliminary model-comparison results from Chinese (N = 24, 66 planned), English (N = 15, 66 planned), and Basque (N = 12, 66 planned) provide tentative evidence that decomposed PoS predictors explain additional variance in MEG responses beyond lexical and acoustic controls. Although the spatial patterns varied across languages, semantic root information tended to contribute in left and/or right temporal gyri, and unexpectedly also in occipital regions; grammatical function tended to contribute in frontal and parietal regions; and root-by-function interaction effects were associated with left and/or right frontal and temporal gyri. Together, these ongoing results provide initial support for decomposing conventional PoS categories into neurobiologically interpretable dimensions, suggesting that the brain tracks lexical semantics, grammatical function, as well as their context-dependent combination. Further analyses with larger samples will test the robustness of these effects and assess to what degree PoS-related cortical encoding patterns are cross-linguistically comparable.
Topic Areas: Syntax and Combinatorial Semantics, Meaning: Lexical Semantics