Use this URL to cite or link to this record in EThOS: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.660062
Title: Robust lexical access using context sensitive dynamic programming and macro-substitutions
Author: Nowell, Peter
Awarding Body: University of Edinburgh
Current Institution: University of Edinburgh
Date of Award: 1991
Availability of Full Text:
Access from EThOS:
Full text unavailable from EThOS. Please try the link below.
Access from Institution:
Abstract:
This thesis presents a collection of techniques for automatically extracting and applying phonological knowledge. The phonological knowledge is extracted by statistically analysing the output of a lexical access algorithm. This knowledge is then incorporated into the lexical access algorithm at run time. The following items outline some of the more important and novel features of this approach. * We have adapted a classical DTW algorithm and chart parser for use in lexical access. This led to a number of improvements such as corrective training, generation of the N-best paths and beam width pruning. * Phonological knowledge is extracted by statistically analysing the output of the lexical access algorithm. This process is completely automatic and requires no human intervention. * The phonological knowledge is efficiently incorporated into the lexical access algorithm at run time. The rule application process is completely deterministic since the rules are implemented solely upon the basis of past substitutions. * The off-line storage is greatly reduced since each template in the lexicon contains a single pronunciation. The additional on-line storage is independent of the lexicon size. These techniques will therefore scale up to large vocabulary speaker independent speech recognition tasks. * The phonological rule base, lexicon, and phrase structure grammar rules are distinct entities. It is therefore possible to modify one without having to modify the others.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.660062  DOI: Not available
Share: