A hierarchical lexical representation for bi-directional spelling-to-pronunciation/pronunciation-to-spelling generation

Abstract

We propose a hierarchical framework for integrating a variety of linguistic knowledge sources of lexical representation in English, in order to facilitate their concurrent utilization in language applications. Our unified lexical representation encompasses information including morphology, stress, syllabification, phonemics and graphemics. Each linguistic knowledge source occupies a distinct stratum in the hierarchy. The merits of the proposed framework is demonstrated on the test bed of bi-directional spelling-to-pronunciation/pronunciation-to-spelling generation. Constraints from the multiple linguistic knowledge sources are administered in parallel during generation, by means of a probabilistic parsing paradigm. This paper extends the previous work on spelling-to-pronunciation generation as reported in ( Meng et al., 1996), by presenting our full results on bi-directional generation which includes pronunciation-to-spelling generation. We will also introduce a robust parsing technique which is aimed for maximizing the coverage of our parser for generation. We believe that our formalism will be especially applicable for augmenting the vocabulary of existing speech recognition and synthesis systems. This work is also the precursor to the ANGIE system ( Lau and Seneff, 1997; Seneff et al., 1996), which extends our lexical representation to the phonetic level, and applies successfully in speech recognition, word spotting and durational modeling ( Chung and Seneff, 1997).