Statistical Machine Translation for Query Expansion in Answer Retrieval

Proceedings of the 45th Annual Meeting of the Association of Computational Linguistics, pages 464–471, Prague, Czech Republic, June 2007. c⃝2007 Association for Computational Linguistics Statistical Machine Translation for Query Expansion in Answer Retrieval Stefan Riezler, Alexander Vasserman, Ioannis Tsochantaridis, Vibhu Mittal and Yi Liu Google Inc., 1600 Amphitheatre Parkway, Mountain View, CA 94043 {riezler|avasserm|ioannis|vibhu|yliu}@google.com Abstract We present an approach to query expan- sion in answer retrieval that uses Statisti- cal Machine Translation (SMT) techniques to bridge the lexical gap between ques- tions and answers. SMT-based query ex- pansion is done by i) using a full-sentence paraphraser to introduce synonyms in con- text of the entire query, and ii) by trans- lating query terms into answer terms us- ing a full-sentence SMT model trained on question-answer pairs. We evaluate these global, context-aware query expansion tech- niques on tfidf retrieval from 10 million question-answer pairs extracted from FAQ pages. Experimental results show that SMT- based expansion improves retrieval perfor- mance over local expansion and over re- trieval without expansion. 1 Introduction One of the fundamental problems in Question An- swering (QA) has been recognized to be the “lexi- cal chasm” (Berger et al., 2000) between question strings and answer strings. This problem is mani- fested in a mismatch between question and answer vocabularies, and is aggravated by the inherent am- biguity of natural language. Several approaches have been presented that apply natural language process- ing technology to close this gap. For example, syn- tactic information has been deployed to reformu- late questions (Hermjakob et al., 2002) or to re- place questions by syntactically similar ones (Lin and Pantel, 2001); lexical ontologies such as Word- net1 have been used to find synonyms for question words (Burke et al., 1997; Hovy et al., 2000; Prager et al., 2001; Harabagiu et al., 2001), and statisti- cal machine translation (SMT) models trained on question-answer pairs have been used to rank can- didate answers according to their translation prob- abilities (Berger et al., 2000; Echihabi and Marcu, 2003; Soricut and Brill, 2006). Information retrieval (IR) is faced by a similar fundamental problem of “term mismatch” between queries and documents. A standard IR solution, query expansion, attempts to increase the chances of matching words in relevant documents by adding terms with similar statistical properties to those in the original query (Voorhees, 1994; Qiu and Frei, 1993; Xu and Croft, 1996). In this paper we will concentrate on the task of answer retrieval from FAQ pages, i.e., an IR prob- lem where user queries are matched against docu- ments consisting of question-answer pairs found in FAQ pages. Equivalently, this is a QA problem that concentrates on finding answers given FAQ docu- ments that are known to contain the answers. Our approach to close the lexical gap in this setting at- tempts to marry QA and IR technology by deploy- ing SMT methods for query expansion in answer retrieval. We present two approaches to SMT-based query expansion, both of which are implemented in the framework of phrase-based SMT (Och and …

Content truncated to 500 words