Related Experiment Videos
The possible-word constraint in the segmentation of continuous speech
D Norris1, J M McQueen, A Cutler
1MRC Applied Psychology Unit, Cambridge, United Kingdom. dennis.norris@mrcapu.cam.ac.uk
Cognitive Psychology
|February 19, 1998
Summary
The Possible-Word Constraint (PWC) impacts continuous speech recognition by limiting word candidates that would create impossible adjacent word segments. This finding aids in understanding speech segmentation and word detection.
Area of Science:
- Psycholinguistics
- Computational Linguistics
- Auditory Perception
Background:
- Continuous speech recognition involves identifying word boundaries.
- Listeners' ability to detect words is influenced by surrounding phonetic information.
Purpose of the Study:
- To investigate the role of linguistic constraints in continuous speech word recognition.
- To propose and model the Possible-Word Constraint (PWC).
Main Methods:
- Two word-spotting experiments were conducted with human listeners.
- A competition-based computational model of speech recognition was developed and tested.
Main Results:
- Listeners had significantly more difficulty detecting a target word when it was preceded by an impossible word segment (e.g., 'fapple').
- Detection accuracy improved when the preceding segment could form a valid word (e.g., 'vuffapple').
- The implemented PWC in the computational model accurately simulated experimental findings.
Conclusions:
- The Possible-Word Constraint (PWC) acts as a crucial factor in continuous speech recognition.
- The PWC effectively reduces the activation of word candidates that violate language-specific word formation rules.
- Modeling the PWC enhances the accuracy of speech segmentation simulations.