Trigger-Pair Predictors in Parsing and Tagging

Trigger-Pair Predictors in Parsing and Tagging Ezra Black, Andrew Finch, Hideki Kashioka ATR Interpreting Telecommunications Laboratories 2-2 Hikaridai, Seika-cho Soraku-gun, Kyoto, Japan 619-02 {black,finch ,kashioka}~atr. itl. co. jp Abstract In this article, we apply to natural language parsing and tagging the device of trigger- pair predictors, previously employed exclu- sively within the field of language mod- elling for speech recognition. Given the task of predicting the correct rule to as- sociate with a parse-tree node, or the cor- recl lag to associate with a word of text, and assuming a particular class of pars- ing or tagging model, we quantify the in- formation gain realized by taking account of rule or tag trigger-pair predictors, i.e. pairs consisting of a "triggering' rule or tag which has already occurred in the docu- ment being processed, together with a spe- cific "triggered" rule or tag whose proba- bility of occurrence within the current sen- tence we wish to estimate. This informa- tion gain is shown to be subslantial. Fur- ther, by utilizing trigger pairs taken from the same general sort of doculnent as is be- ing processed (e.g. same subject matter or same discourse type) as opposed to pre- dictors derived from a comprehensive gen- eral set of English texts we can signifi- cantly increase this information gain. 1 Introduction If a person or device wished to predict which words or grammatical constructions were about to occur in some document, intuitively one of the most helpful things to know would seem to be which words and constructions occurred within the last half-dozen or dozen sentences of the document. Other things be- ing equal, a text that has so far been larded with, say, mountaineering terms, is a good bet to continue featuring them. An author with the habit of ending sentences with adverbial clauses of confirmation, e.g. "as we all know", will probably keep up that habit as the discourse progresses. Within the field of language modelling for speech recognition, maintaining a cache of words that have occurred so far within a document, and using this information to alter probabilities of occurrence of particular choices for the word being predicted, has proved a winning strategy (Kuhn et al., 1990). Mod- els using trigger pairs of words, i.e. pairs consist- ing of a "triggering" word which has already oc- curred in the document being processed, plus a spe- cific "triggered" word whose probability of occur- rence as the next word of the document needs to be estimated, have yielded perplexity 1 reductions of 29 38% over the baseline trigram model, for a 5 million-word Wall Street Journal training corpus (Rosenfeld, 1996). This paper introduces the idea of using trigger pair techniques to ~ssist in the prediction of rule and tag occurrences, within the context of natural language parsing and tagging. Given the task of predicting the correct rule to associate with a parse- tree node, or the correct tag to associate with a word of text, and assuming a particular class of parsing or …

Content truncated to 500 words