Finding Spanish Syllabification Rules with Decision Trees

摘要

Syllables have been proposed as a viable alternative to phonemes for automatic speech recognition, and for use in text-to-speech systems as a way to enhance the speech quality. The question then arises of how to obtain the correct syllabification rules for a particular language. Even for a language like Spanish, which has well defined syllabification rules, linguistic knowledge is often required to discover them. It is interesting to ask whether machine learning techniques can produce effective syllabification algorithms, and our aim here is to test the usefulness of classification trees for this task. Additionally, we would like to understand the sort of problems that arise in the process, with a view to applying it to other languages.