Overlap Analysis in Textual Entailment Recognition.
Ken Litkowski · 2009
CL Research participated in the main and search pilot tasks of the 2009 Recognizing Textual Entailment track (RTE-5). Our system was little changed from what was used in previous RTE exercises. As a result of an apparent increased complexity in the task, our scores have declined from those in previous years. We submitted one 2-way run in the main task, with an accuracy of 0.53, and two runs in the search pilot, with f-scores of 0.28 and 0.29. At present, our system consists solely of routines to examine the overlap of discourse entities between the texts and hypotheses. We present the algorithms used in the step. We examine the potential use of other resources, including WordNet, a Roget-style thesaurus, and FrameNet, but have not yet implemented methods for exploiting these resources.