Related Experiment Videos
Phonological underspecification and mapping mechanisms in the speech recognition lexicon.
Linda Wheeldon1, Rachelle Waksler
1Department of Psychology, University of Birmingham, Birmingham, UK. L.R.WHEELDON@Bham.ac.uk
Brain and Language
|June 3, 2004
Summary
Speech recognition models benefit from underspecified lexical entries, supporting models that allow for phonological variations. Findings favor context-independent mapping for speech processing.
Area of Science:
- Cognitive Science
- Psycholinguistics
- Speech Processing
Background:
- Speech recognition models face challenges with phonological variations.
- Debates exist on phonological underspecification in the mental lexicon and speech-to-lexical mapping mechanisms.
Purpose of the Study:
- To investigate speech recognition models' tolerance for phonological mismatch.
- To test predictions of models with underspecified versus fully specified lexical entries.
- To differentiate between context-dependent and context-independent mapping mechanisms.
Main Methods:
- Cross-modal repetition priming experiments in English.
- Analysis of speech signal to abstract lexical entry mapping.
- Comparison of model predictions regarding phonological underspecification and context effects.
Main Results:
- Evidence supports phonological underspecification, favoring models with underspecified lexical entries.
- No contextual effects were observed, supporting context-independent mapping mechanisms.
- Findings challenge models relying on fully specified lexical entries and context-dependent mapping.
Conclusions:
- Speech recognition models should incorporate underspecified lexical entries for better phonological variation tolerance.
- Context-independent mapping mechanisms are more effective for speech processing.
- The study provides empirical support for specific speech recognition model architectures.