Using Supertag in MUC-7 Template Relation Task

The Template Relation (TR) task is an information extraction problem introduced in the 7th Message Understanding Conference (MUC-7). In this paper, we have proposed an approach to convert this problem into a discriminative one. We obtain F-Measure of 78% on sentence-level relation which is comparable to the best system presented in MUC-7, while almost no extra annotation work is required. In our approach, we first use Supertagger [Joshi 1994] and Lightweight Dependency Analyzer [Srinivas 1997] to do shallow parsing on the text. Then features are abstracted from the shallow parsing result. We use these features as weak hypotheses, and combine them into the final one with the boosting algorithm [Schapire 2000].