skip to main content
Language:
Search Limited to: Search Limited to: Resource type Show Results with: Show Results with: Search type Index

Using naturalistic speech EEG recordings to test the cognitive reality of representations suggested by a Computational model of early phonetic learning

info:eu-repo/semantics/OpenAccess

Digital Resources/Online E-Resources

Citations Cited by
  • Title:
    Using naturalistic speech EEG recordings to test the cognitive reality of representations suggested by a Computational model of early phonetic learning
  • Author: Balleroy, Ambre
  • Subjects: [SCCO]Cognitive science ; Continuous speech processing ; Early phonetic learning ; Encoding ; Perceptual space learning ; Temporal Response Function (TRF)
  • Description: A recent computational model of early phonetic learning predicts that infants may not be learning phonetic categories but finer-grained units. While this model successfully predicts some behavioral effects, this does not guarantee that the model’s internal representations provide a faithful equivalent of human cognitive representations. We test the cognitive reality of the units learned by the model using a continuous speech encoding approach (the multivariate Temporal Response Function) on an openly-available electroencephalogram dataset. Brain activity recorded while participants were listening to an audiobook was predicted significantly above chance level using features representing the units onsets, but not the units entropy. However, the onsets had no added values when combined with established acoustic baselines (a combination of a spectrogram and of acoustic onsets). Since the information contained in the units learned by the tested model is either not significantly tracked by the brain or redundant with low-level acoustic information, this is not in favor of their cognitive reality. Further work is needed, however, before a more definitive conclusion can be reached and we discuss what the next steps may be, including finding ways to summarize the units that would preserve their language-dependent characteristics and replicating the current analysis using infant data, to test whether the units may be transitory during development.
  • Publisher: HAL CCSD
  • Creation Date: 2023
  • Language: English
  • Source: Dumas (Dépôt Universitaire de Mémoires Après Soutenance)

Searching Remote Databases, Please Wait