5th International Conference on Spoken Language Processing

Sydney, Australia
November 30 - December 4, 1998

A Unified Framework for Sublexical and Linguistic Modelling Supporting Flexible Vocabulary Speech Understanding

Raymond Lau, Stephanie Seneff

MIT Laboratory for Computer Science, USA

Previously, we introduced the ANGIE framework for modelling speech where morphological and phonological substructures of words are jointly characterized by a context-free grammar and represented in a multi-layered hierarchical structure. We also demonstrated a phonetic recognizer and a word-spotter based on ANGIE. In this work, we extend ANGIE to a competitive continuous speech recognition system. Furthermore, given that ANGIE is based on a context-free framework, we have decided to combine ANGIE with TINA, a context-free based framework for natural language understanding, into an integrated system. The integration led to a 21.7% reduction in word error rate compared to a baseline word bigram recognizer on ATIS. We also examined the addition of new words to the vocabulary, an area we believe will benefit from both ANGIE and the ANGIE-plus-TINA integration. The combination reduced error rate by 20.8% over the baseline and outperformed several other configurations tested not involving an integrated ANGIE-plus-TINA.

Full Paper

Bibliographic reference.  Lau, Raymond / Seneff, Stephanie (1998): "A unified framework for sublexical and linguistic modelling supporting flexible vocabulary speech understanding", In ICSLP-1998, paper 0053.