Finite-state models, event logics and statistics in speech recognition

This paper presents a constraint–based approach to speech recognition which combines aspects of event logic with efficient processing strategies. Although stochastic approaches are currently at the forefront of speech–recognition applications, it has now been recognized that linguistic structure is required in order to deal with the problem of recognizing new words. The computational linguistic approach presented here offers solutions to the problems of how to process words which have not been heard before, and how to develop fine–grained knowledge representation and processing techniques for linguistic units smaller than the word. Furthermore, it is investigated how statistical data can be integrated into the phonological constraint model in order to minimize the discrepancy between expectations defined in the top–down constraints and the actual data.