Presentation
Search Abstracts | Symposia | Slide Sessions | Poster Sessions | Poster Slams
Periodic Chunking of Language: Rhythmic Neuronal Processing Mirrored in Self-Paced Reading?
Poster A44 in Poster Session A, Thursday, October 6, 10:15 am - 12:00 pm EDT, Millennium Hall
Chia-Wen Lo1, Mark Anderson3, Lena Henke1, Lars Meyer1,2; 1Max Planck Institute for Human Cognition and Brain Sciences, 2University Clinic Münster, 3Cardiff University
INTRODUCTION: Humans segment verbal stimuli into multi-word chunks (Fodor and Bever 1965). Chunks are limited in duration due to cognitive limitations (e.g., memory constraints; Christiansen and Chater 2016). Recently, it has been suggested that the duration and pace of chunking might also be affected by the period of the underlying rhythmic neuronal processes (Meyer et al., 2020; Henke and Meyer, 2021). Periodic electrophysiological activity may relate to phrase-level language processing (Ding et al., 2016). In particular, neural oscillations in the delta band (< 4 Hz) have been found to synchronize with multi-word chunks (Meyer et al. 2017; Henke and Meyer 2021). It remains unclear whether this apparent role of rhythmic activity is behaviorally relevant for language comprehension. It is also unclear what type of cognitive units rhythmic activity may relate to (for discussion, see Kazanina and Tavano 2021). We test here whether behavioral responses during language comprehension show periodic patterns within the range of the delta band. We also assess whether this hypothetical behavioral periodicity links to chunks as formalized through natural language processing (NLP). METHODS: We analyzed self-paced reading data from 180 participants (Futrell et al., 2021). Participants read 10 stories from the National Stories Corpus word by word, advancing through button press. In the first step, we performed frequency analysis on word-by-word reading times (RT) to assess periodicity. RTs were converted into a time-series. To highlight chunking, we performed differencing of the time series, highlighting abrupt changes from slow to fast RTs. Such changes have been related to chunking previously (Tosatto et al., 2021). In the second step, we statistically predicted the slow–fast changes from a model that defines and outputs chunks as saturated local syntactic dependency graphs (Anderson et al., 2019). RESULTS: Frequency analysis showed periodicity of differenced RTs at a frequency of ~2 Hertz, suggesting that slowdown–speedup transitions occur periodically with a period of 0.5 seconds. Moreover, they cluster at sentence boundaries (Just and Carpenter, 1980; Rayner et al., 2000), but also at the boundaries of the cognitive units predicted by our NLP chunker. This was substantiated by regression of reading times on word positions within chunks, which show a continuous increase from chunk on– to offset. Our results provide the first evidence that endogenous electrophysiological rhythms in the delta band are behaviorally relevant for the segmentation of verbal stimuli into multi-word chunks. In other words: Chunking is a periodic behavior, possibly related to the periodicity of the underlying neuronal processes.
Topic Areas: , Computational Approaches