Presentation
Search Abstracts | Symposia | Slide Sessions | Poster Sessions | Lightning Talks
Distributed Neural Representations for Semantic Structures During Sentence Production
There is a Poster PDF for this presentation, but you must be a current member or registered to attend SNL 2023 to view it. Please go to your Account Home page to register.
Poster B9 in Poster Session B, Tuesday, October 24, 3:30 - 5:15 pm CEST, Espace Vieux-Port
Laura Giglio1,2, Peter Hagoort1,2, Markus Ostarek1; 1Max Planck Institute for Psycholinguistics, 2Donders Institute for Brain, Cognition and Behaviour
When speaking and listening, we dynamically compose concepts to build event representations. The neural representations for compositional processing have so far been mostly studied during sentence comprehension. A recent study found that sentence compositional processes are realized via two pathways (Frankland & Greene, 2020). Abstract thematic relations (‘man-as-agent’) were encoded by the left mid superior temporal cortex (lmSTC), while specific event representations (‘man-as-attacker’) were encoded by the anterior medial prefrontal cortex (amPFC) and the hippocampus. In an fMRI study (n=38), we asked first whether we could find brain representations for compositional processing during sentence production. We then investigated whether the same regions were involved in encoding semantic structures at different levels of specificity. We used a rapid serial visual presentation sentence recall paradigm to elicit sentence production from the conceptual memory of an event. The stimuli consisted of sentences such as “the musician attacks the athlete”, with nouns (athletes vs. musicians) and verbs (contact vs. perception events) distinguishable by semantic categories. With voxel-wise encoding models, we probed the specificity of the semantic structure built during the production of each sentence in three ROIs (lmSTC, amPFC and hippocampus) and in the whole-brain. We compared an unstructured model of word meaning without relational information (‘bag-of-nouns’) with a model that encodes abstract thematic relations (‘musician-as-agent’) and a model encoding event-specific relational structure (‘musician-as-attacker’). Whole-brain results confirmed that we could find brain representations for compositional structures in sentence production. ROI analyses in the superior temporal cortex (STC) indicated that verb-specific event representations were encoded in the mid-anterior STC. The posterior STC instead encoded sentence meaning abstracting over events, as well as the ordinal and syntactic structure of the sentence. The amPFC and the hippocampus were not found to encode semantic structures at any level of specificity, in contrast with Frankland & Greene’s comprehension results (2020). Whole-brain analyses revealed that the encoding of sentence meaning at different levels of specificity was by no means limited to the STC and instead highlighted a large left fronto-parieto-temporal network. Our results therefore suggest more distributed representations for compositional processing than previously shown by Frankland & Greene (2020). Further analyses revealed that the different results could not be a consequence of modality differences (i.e., production vs. comprehension), since the same analysis over the comprehension of the sentence to be recalled in this experiment showed equally distributed brain representations. Therefore, this study suggests that the encoding of specific as well as abstract semantic structures during sentence production is realized in a distributed network of left-lateralized brain regions.
Topic Areas: Language Production, Syntax and Combinatorial Semantics