Presentation

Search Abstracts | Symposia | Slide Sessions | Poster Sessions | Poster Slams

Delta-band Neural Tracking Reflects Speech Chunking Instead of Semantic Relatedness between Words

Poster D30 in Poster Session D with Social Hour, Friday, October 7, 5:30 - 7:15 pm EDT, Millennium Hall

Yuhan Lu¹, Nai Ding², Xing Tian³; ¹East China Normal University, ²Zhejiang University, ³New York University Shanghai

Chunking is an important mechanism for language comprehension: the brain applies a set of syntactic rules to recursively integrate words into a large linguistic structure, e.g., phrase and sentence, and then derives meaning of the structure. Delta-band cortical activity tracking the time course of different levels of linguistic hierarchy (e.g., syllable, word, phrase, and sentence) has been viewed as evidence that the brain applied syntactic rule to group linguistic units into superordinate chunks. Challenging this position, it has been argued that the brain does not construct multi-word chunks at all but simply combines semantic features of individual words via statistical and semantic analysis. For example, the context-dependent neural response is generally expected to be high at the beginning of a sentence and to be low at the end of a sentence, forming an apparent neural tracking of sentences. It remains elusive, therefore, whether delta-band neural tracking of sentences truly reflects the mental representation of sentences or is merely caused by the neural encoding of semantic relatedness between words. The current study investigates to what extent the delta-band neural responses to speech can be explained by semantic relatedness between words. We used word sequences and sentence sequences to dissociate semantic relatedness from sentential structure, and asked listeners to attend to semantic features of words while cortical activity was recorded using magnetoencephalography (MEG). Semantic relatedness between adjacent words, simulated using a word2vec-based model, predicted a weaker 1-Hz response to word sequence than sentence. Consistent with the model prediction, empirical neural activity showed a weaker 1-Hz response to word sequence than sentence, when participants performed a word-level task that did not require sentential processing. However, when listeners performed a chunk-level task that explicitly parsed the word sequences into multi-word chunks repeating at 1 Hz, 1-Hz response to word sequence was stronger than the response during word-level task. These results suggest that delta-band neural activity primarily reflects the mental representation of multi-word chunks defined by either implicit syntactic rule or explicit chunking rule, instead of neural encoding of semantic relatedness between words.

Topic Areas: Syntax, Speech Perception