Auditory Word Comprehension Is Less Incremental in Isolated Words

Abstract

Partial speech input is often understood to trigger rapid and automatic activation of successively higher-level representations of words, from sound to meaning. Here we show evidence from magnetoencephalography that this type of incremental processing is limited when words are heard in isolation as compared to continuous speech. This suggests a less unified and automatic word recognition process than is often assumed. We present evidence from isolated words that neural effects of phoneme probability, quantified by phoneme surprisal, are significantly stronger than (statistically null) effects of phoneme-by-phoneme lexical uncertainty, quantified by cohort entropy. In contrast, we find robust effects of both cohort entropy and phoneme surprisal during perception of connected speech, with a significant interaction between the contexts. This dissociation rules out models of word recognition in which phoneme surprisal and cohort entropy are common indicators of a uniform process, even though these closely related information-theoretic measures both arise from the probability distribution of wordforms consistent with the input. We propose that phoneme surprisal effects reflect automatic access of a lower level of representation of the auditory input (e.g., wordforms) while the occurrence of cohort entropy effects is task-sensitive, driven by a competition process or a higher-level representation that is engaged late (or not at all) during the processing of single words.

Effect of contextual diversity on word recognition in different semantic contexts

“…The sentence disappeared by pressing the SPACE bar after participants understood; then, a true or false judgment question was presented and participants were asked to make a judgment based on the sentences they have just read. As in previous studies (Gaston et al, 2023), good performance on the task requires lexical-syntactic and conceptual information access to have occurred for most stimuli because the probe words were unpredictable. Thus, the question in our study will change the following areas of the original sentence: (1) the animate noun, (2) the verb, (3) the quantity word, (4) the target noun, and (5) the rest of the sentence.…”

Neural Dynamics of the Processing of Speech Features: Evidence for a Progression of Features from Acoustic to Sentential Processing

“…Before starting, we first analyzed how the phonetic features, phoneme onset, phoneme surprisal and cohort entropy should best be modeled, since different previous studies have used used different approaches: modeling word-initial phonemes as separate features 1 ; including word-initial phonemes only in phoneme surprisal and cohort entropy 85 ; and including word-initial phoneme only in phoneme onset 2 . We compared models with and without word-initial phoneme onset on a base model with envelope spectrogram, envelope onset and word onset.…”

“…These three TRFs were averaged to generate one average TRF per testing fold, which was then used to compute the prediction accuracy against the testing set. The TRFs and corresponding prediction accuracies from each of the testing folds were further averaged to generate a single TRF and single prediction accuracy per source dipole.Phonetic feature modellingBefore starting, we first analyzed how the phonetic features, phoneme onset, phoneme surprisal and cohort entropy should best be modeled, since different previous studies have used used different approaches: modeling word-initial phonemes as separate features 1 ; including wordinitial phonemes only in phoneme surprisal and cohort entropy85 ; and including word-initial phoneme only in phoneme onset 2 . We compared models with and without word-initial phoneme onset on a base model with envelope spectrogram, envelope onset and word onset.…”

Recurrent neural networks as neuro-computational models of human speech recognition

“…The stimuli were based on a set of 1000 target words used in a MEG experiment (Gaston et al, 2023), spoken by a human male speaker for the massive auditory lexical decision (MALD) database (Tucker et al, 2019). To simulate realistic lexical neighborhoods for those target words, additional words were added to the lexicon if they 1) differed from a target word by a single phoneme, 2) had a frequency count of at least 1000 in the SUBTLEX-US corpus (Brysbaert and New, 2009), and 3) were included in the MALD database.…”

“…Auditory baseline model A model of auditory processing was generated based on gammatone spectrograms as in previous experiments (Brodbeck et al, 2020;Gaston et al, 2023). First, high resolution gammatone spectrograms were generated for all word stimuli with 256 frequency bands in equivalent rectangular bandwidth (ERB) space between 20 and 5000 Hz.…”