UMIACS Computational Linguistics Colloquium Series, October 16, 1997

Statistical Language Modeling:
The New Wave in Computational Linguistics?

Ralph Weischedel
BBN Technologies

The traditional approach to performance (as opposed to competence) in computational linguistics has been to write detailed syntactic, semantic, and discourse rules reflecting the structure of language and "tweak" those rules in a limited domain to achieve adequate performance.

By contrast, a new approach, growing in acceptance, is to apply statistical learning algorithms to capture linguistic details automatically from examples.

In this talk we first overview the new approach and state-of-the-art performance in both parsing and language understanding. In most cases, statistical language modeling has equaled or outperformed traditional approaches. Second, we will detail a recent success in applying statistical language models to information extraction from text or speech.

Return to the UMD Computational Linguistics Colloquium Series.