Item

Abstract

INTRODUCTION Humans need to link the words and morphemes of a sentence to comprehend it. However, this ability is constrained in time by the capacity of our working memory. Recently, it has also been suggested that the length of the underlying neural processing time windows also poses a constraint. In particular, cycles of neural oscillations in the delta band (< 4 Hertz) have been discussed to serve and constrain the formation of multi-word chunks (Henke & Meyer 2021; Henke & Meyer 2023; Lo et al. 2023). But if we sample and process speech chunk by chunk, how can we link words and morphemes that belong to separate chunks? That is, how can we establish dependencies across different memory units or different processing cycles? METHODS During EEG recording, native German speakers listen to trials consisting of six 10-syllable sentences that contain an agreement dependency. Eight conditions were created by manipulating three factors: (i) whether there is gender agreement or not (i.e., agreement violation); (ii) gender (female/male); and (iii) agreement within a single chunk or across a chunk boundary. We employed a frequency-tagging paradigm to assess neural segmentation. In addition, we analyzed ERPs to the second element of the dependency. PREDICTION Data collection is still in progress. We expect a peak at the rate of sentences in both the within-chunk and across-chunk conditions in the power spectrum, suggesting the active segmentation of continuous speech into multi-word memory chunks. Additionally, if dependency processing across chunks is more difficult, the difference between the agreement and violation conditions will be smaller in the across-chunk condition than in the within-chunk condition. The results help us better understand how we integrate the time-constrained chunk-by-chunk sampling of speech and the processing of non-adjacent dependencies and whether the sampling of chunks in some way blocks the formation of non-adjacent dependencies in natural language comprehension.