Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Decoding the incremental construction of syntactic structure
Poster Session E, Friday, October 2, 11:00 am - 1:00 pm, Wangari Maathai
This poster is part of the Sandbox Series.
Katarína Labancová1, Nina Kazanina1; 1Department of Basic Neurosciences, Faculty of Medicine, University of Geneva
Real-time language comprehension requires incrementally integrating lexical items into hierarchical phrase representations. Because linguistic input unfolds sequentially over time, earlier word representations may need to be maintained until the information required to form the complete concept becomes available. Prior work has shown that individual words are actively represented in neural activity until they can combine into a phrase (Desbordes et al., 2024). However, this study used simple phrase constructions, and it remains unclear how these dynamics extend to more complex syntactic configurations involving multiple phrases and delayed integration across longer temporal spans. Here, we investigate how long word-level representations remain actively represented in the neural signal when they must be bound to a phrase head, and how this timing depends on whether integration occurs immediately or is delayed by intervening material. This provides a temporal test of how lexical information is coordinated during incremental phrase construction. In an ongoing MEG study, healthy adult participants read sentences presented one word at a time. Using French language stimuli, where adjectives can precede or follow a noun that they modify, we manipulate when a complete phrase is formed. Compare “le beau petit verre” (the pretty little cup), where adjectives can only bind to the noun at the end of the phrase, and “le table rouge lourde” (the table red heavy) where each adjective can immediately bind to the noun, which is incrementally modified by them. At a higher hierarchical level, we manipulate whether two phrases are integrated into a single complex representation (PP-modification: “Look at the pretty little cup ON the heavy red table”) or describe two separate objects (conjunction: “Look at the pretty little cup AND the heavy red table”). After sentence presentation, participants answer a question (e.g. “What color is the table?”) probing one of the phrases. Multivariate decoding is used to identify neural activity patterns corresponding to individual words. We then use the temporal generalization method to track how long these word-level representations remain decodable during sentence processing and whether they re-emerge at retrieval. Crucially, this allows us to quantify when individual words are no longer independently decodable and how this timing relates to points of syntactic integration within the sentence when lexical information should be incorporated into higher-level representations. We test three main predictions. First, lexical items awaiting integration into phrase structure will remain decodable until the phrase head becomes available. Pre-nominal adjectives should therefore be decodable longer than post-nominal adjectives, which can immediately bind to the noun and do not need to be represented independently. Second, phrase heads are expected to remain decodable until the phrase is complete, both within a simple noun phrase (the noun should remain active throughout the presentation of post-nominal adjectives) and when they serve as anchors for larger hierarchical structures (i.e. “cup” should be active longer in the PP-modification condition compared to the conjunction condition), consistent with their role in organizing phrase-level representations . Data collection is ongoing, with pilot analyses currently underway and full within-subject results expected by the time of presentation.
Topic Areas: Syntax and Combinatorial Semantics, Computational Approaches