Related Experiment Videos
Bootstrapping Word Boundaries: A Bottom-up Corpus-Based Approach to Speech Segmentation
1Centre for Cognitive Science, University of Edinburgh, Edinburgh, United Kingdom
Cognitive Psychology
|July 1, 1997
Summary
This study shows that statistical patterns in speech sounds help segment words. This statistical learning may enable infants to develop sophisticated speech segmentation abilities from basic sound awareness.
Area of Science:
- Computational linguistics
- Developmental psycholinguistics
- Phonetics
Background:
- Speech segmentation is challenging due to continuous acoustic signals.
- Lexical access relies on identifying meaningful speech units.
- Infant speech perception development is not fully understood.
Purpose of the Study:
- To investigate the role of sequential phonological probabilities in speech segmentation.
- To model how infants might acquire speech segmentation skills.
- To connect computational models with psycholinguistic findings.
Main Methods:
- Utilized neural network models and statistical analysis.
- Analyzed idealized phonological transcriptions of the London-Lund Corpus.
- Examined distributional properties of phonetic segments in English.
Main Results:
- Phonetic segment distribution is a significant cue for speech segmentation.
- Statistical learning of subsegmental primitives can lead to complex segmentation abilities.
- Confirmed the effectiveness of the Metrical Segmentation Strategy.
Conclusions:
- Phonological probabilities are crucial for segmenting continuous speech.
- Infants may learn speech segmentation through statistical pattern detection.
- The Metrical Segmentation Strategy can be acquired without pre-defined linguistic categories.