Recognition Performance of a Structured Language Model

TL;DR

This paper introduces a structured language model that builds hierarchical representations of word sequences to improve speech recognition accuracy, demonstrating better perplexity and word error rates than traditional trigram models.

Contribution

The paper presents a novel hierarchical language model inspired by linguistic analysis, enhancing speech recognition by capturing long-distance dependencies.

Findings

Improved perplexity over trigram models

Reduced word error rate in speech recognition

Effective use of hierarchical structure in language modeling

Abstract

A new language model for speech recognition inspired by linguistic analysis is presented. The model develops hidden hierarchical structure incrementally and uses it to extract meaningful information from the word history - thus enabling the use of extended distance dependencies - in an attempt to complement the locality of currently used trigram models. The structured language model, its probabilistic parameterization and performance in a two-pass speech recognizer are presented. Experiments on the SWITCHBOARD corpus show an improvement in both perplexity and word error rate over conventional trigram models.