tailieunhanh - Báo cáo khoa học: "Efficient Search for Transformation-based Inference"

This paper addresses the search problem in textual inference, where systems need to infer one piece of text from another. A prominent approach to this task is attempts to transform one text into the other through a sequence of inference-preserving transformations, . a proof, while estimating the proof’s validity. | Efficient Search for Transformation-based Inference Asher Stern Roni Stern- Ido Dagan Ariel Felner Computer Science Department Bar-Ilan University ị Information Systems Engineering Ben Gurion University astern7@ dagan@ felner@ Abstract This paper addresses the search problem in textual inference where systems need to infer one piece of text from another. A prominent approach to this task is attempts to transform one text into the other through a sequence of inference-preserving transformations . a proof while estimating the proof s validity. This raises a search challenge of finding the best possible proof. We explore this challenge through a comprehensive investigation of prominent search algorithms and propose two novel algorithmic components specifically designed for textual inference a gradient-style evaluation function and a locallookahead node expansion method. Evaluations using the open-source system B IUTEE show the contribution of these ideas to search efficiency and proof quality. 1 Introduction In many NLP settings it is necessary to identify that a certain semantic inference relation holds between two pieces of text. For example in paraphrase recognition it is necessary to identify that the meanings of two text fragments are roughly equivalent. In passage retrieval for question answering it is needed to detect text passages from which a satisfying answer can be inferred. A generic formulation for the inference relation between two texts is given by the Recognizing Textual Entailment RTE paradigm Dagan et al. 2005 which is adapted here for our investigation. In this setting a system is given two text fragments termed text T and hy 283 pothesis H and has to recognize whether the hypothesis is entailed by inferred from the text. An appealing approach to such textual inferences is to explicitly transform T into H using a sequence of transformations Bar-Haim et al. 2007 Harmeling 2009 Mehdad 2009 Wang .