Prepositional Phrase Attachment through a Backed-Off Model

TL;DR

This paper applies backed-off n-gram language modeling techniques to resolve prepositional phrase attachment ambiguity, achieving notable accuracy improvements and highlighting the significance of low-frequency events.

Contribution

It introduces a novel application of backed-off language models to prepositional phrase attachment, demonstrating effectiveness on Wall Street Journal data.

Findings

Achieved 84.5% accuracy on WSJ data.

Ignoring low-count events reduces accuracy to 81.6%.

Backed-off models are effective for syntactic disambiguation.

Abstract

Recent work has considered corpus-based or statistical approaches to the problem of prepositional phrase attachment ambiguity. Typically, ambiguous verb phrases of the form {v np1 p np2} are resolved through a model which considers values of the four head words (v, n1, p and n2). This paper shows that the problem is analogous to n-gram language models in speech recognition, and that one of the most common methods for language modeling, the backed-off estimate, is applicable. Results on Wall Street Journal data of 84.5% accuracy are obtained using this method. A surprising result is the importance of low-count events - ignoring events which occur less than 5 times in training data reduces performance to 81.6%.