Disassociating lexical certainty and information gain in the ERP signal - a spatiotemporally unbiased approach

Activity: Talk or presentationPresentation

Description

In this seminar, I will present the methodology and results of a new study (Hjortdal, Frid, Novén & Roll, submitted). The study is concerned with the influence of suprasegmental word accent and stød cues in lexical access. According to most models of spoken word recognition, lexical access is driven by an interplay between forward- and backward-looking processes. Listeners estimate probabilities of the lexical identity of an unfolding word based on the currently available acoustic input. Concurrently, lexical hypotheses are disconfirmed as the speech signal unfolds and listeners gain information and can disconfirm a subset of lexical hypotheses. In this study, we combined corpus studies and temporally fine-grained analysis of event-related potentials to disassociate the neurophysiological correlates of lexical certainty and information gain. In addition to presenting results, I will focus on methodological aspects of the study, including how measures of lexical certainty (cohort entropy) and information gain (phoneme surprisal) were calculated in Python based on Danish and Swedish corpora and pronunciation lexica. I will also focus on the event-related potential analysis which took a spatiotemporally unbiased approach without hypotheses about spatiotemporal regions and included fixed continuous effects of cohort entropy and phoneme surprisal. Finally, we compared the suprasegmental models to segmental models which did not take prosodic word accent cues into account to investigate in which spatiotemporal regions suprasegmental cues modulated lexical access over and above segmental cues.
Period2022 Dec 15
Held atGeneral Linguistics
Degree of RecognitionLocal

Free keywords

  • erp
  • statistics
  • Computational tools
  • prediction
  • entropy
  • informationsteknologi