A Finite State and Data-Oriented Method for Grapheme to Phoneme Conversion

TL;DR

This paper introduces a finite-state and data-driven approach for converting written Dutch words into phonemes, achieving high accuracy through rule-based and machine learning enhancements.

Contribution

It presents a novel finite-state method combined with transformation-based learning to improve grapheme-to-phoneme conversion accuracy.

Findings

Over 93% phoneme accuracy with hand-crafted rules

Improved to 99% accuracy using transformation-based learning

Effective on a dataset of 40,000 words

Abstract

A finite-state method, based on leftmost longest-match replacement, is presented for segmenting words into graphemes, and for converting graphemes into phonemes. A small set of hand-crafted conversion rules for Dutch achieves a phoneme accuracy of over 93%. The accuracy of the system is further improved by using transformation-based learning. The phoneme accuracy of the best system (using a large set of rule templates and a `lazy' variant of Brill's algoritm), trained on only 40K words, reaches 99% accuracy.