An Experimental Technique for Establishing Lexical Variants by Rule in Automatic Recognition of Continuous Speech

Publisher Website
Google Scholar
Abstract
The primary task in phonetic-to-lexical translation is to develop procedures which, from a pragmatic point of view, provide the necessary paths for changing an acoustic-continuum reality into a series of discrete standard-orthographical forms, which, at the present time, exist only as theoretical abstractions, Implicit in this problem is the assumption that words exist acoustically, even though complexly, in a many-to-one mapping. For automatic recognition, a technique is required that will provide the variety that is to be accounted for in this many-to-one mapping. Since words are known to be highly variable with regard to their phonetic constituents, it is necessary to tie a representation system to some unit which is smaller than the word itself, but larger than the individual phonetic unit. In the present system, it was tentatively decided to attempt variant generation based on units generally corresponding to the syllable. While present results are not conclusive, it appears viable to continue the development of this approach. Whether or not a within-word unit can be used to characterize all interword phenomena has yet to be determined.

This publication has 0 references indexed in Scilit: