Entropy-based Pruning of Backoff Language Models

TL;DR

This paper introduces an entropy-based criterion for pruning N-gram backoff language models, enabling significant size reduction while maintaining recognition accuracy, and compares it to heuristic methods.

Contribution

It develops an exact and efficient relative entropy measure for pruning N-grams in backoff models, improving model compression without accuracy loss.

Findings

Reduced Hub4 language model to 26% size without increasing error

Exact entropy criterion outperforms heuristic pruning slightly

High overlap (85%) with heuristic pruning in selected N-grams

Abstract

A criterion for pruning parameters from N-gram backoff language models is developed, based on the relative entropy between the original and the pruned model. It is shown that the relative entropy resulting from pruning a single N-gram can be computed exactly and efficiently for backoff models. The relative entropy measure can be expressed as a relative change in training set perplexity. This leads to a simple pruning criterion whereby all N-grams that change perplexity by less than a threshold are removed from the model. Experiments show that a production-quality Hub4 LM can be reduced to 26% its original size without increasing recognition error. We also compare the approach to a heuristic pruning criterion by Seymore and Rosenfeld (1996), and show that their approach can be interpreted as an approximation to the relative entropy criterion. Experimentally, both approaches select similar sets of N-grams (about 85% overlap), with the exact relative entropy criterion giving marginally better performance.