A major problem of lexical access is that of lexical segmentation: how is fluent, continuous speech, which has few reliable cues to word boundaries, mapped onto a lexicon consisting of discrete entries? A plausible segmentation process must be efficient, generating small sets of word candidates, and reliable over different types of input: citation-form speech; speech in noise; and natural fluent speech, where phonological processes have operated. A computational tool for assessing the performance of various segmentation strategies is described. Input to a machine-readable dictionary is based on a featural representation, allowing flexibility in the degree of specification of the input string. The segmentation strategies can therefore be tested on input which represents the variability inherent in continuous natural speech. The tool can also be used to generate psycholinguistic predictions.
Cite as: McQueen, J.M., Briscoe, E.J. (1991) A computational tool for examining lexical segmentation in continuous speech. Proc. 2nd European Conference on Speech Communication and Technology (Eurospeech 1991), 697-700, doi: 10.21437/Eurospeech.1991-184
@inproceedings{mcqueen91_eurospeech,
author={J. M. McQueen and E. J. Briscoe},
title={{A computational tool for examining lexical segmentation in continuous speech}},
year=1991,
booktitle={Proc. 2nd European Conference on Speech Communication and Technology (Eurospeech 1991)},
pages={697--700},
doi={10.21437/Eurospeech.1991-184}
}