Answer Sentence Retrieval by Matching Dependency Paths acquired from Question/Answer Sentence Pairs

Proceedings of the 13th Conference of the European Chapter of the Association for Computational Linguistics, pages 88–98, Avignon, France, April 23 - 27 2012. c⃝2012 Association for Computational Linguistics Answer Sentence Retrieval by Matching Dependency Paths Acquired from Question/Answer Sentence Pairs Michael Kaisser AGT Group (R&D) GmbH J¨agerstr. 41, 10117 Berlin, Germany mkaisser@agtgermany.com Abstract In Information Retrieval (IR) in general and Question Answering (QA) in particu- lar, queries and relevant textual content of- ten significantly differ in their properties and are therefore difficult to relate with tra- ditional IR methods, e.g. key-word match- ing. In this paper we describe an algorithm that addresses this problem, but rather than looking at it on a term matching/term re- formulation level, we focus on the syntac- tic differences between questions and rele- vant text passages. To this end we propose a novel algorithm that analyzes dependency structures of queries and known relevant text passages and acquires transformational patterns that can be used to retrieve rele- vant textual content. We evaluate our algo- rithm in a QA setting, and show that it out- performs a baseline that uses only depen- dency information contained in the ques- tions by 300% and that it also improves per- formance of a state of the art QA system significantly. 1 Introduction It is a well known problem in Information Re- trieval (IR) and Question Answering (QA) that queries and relevant textual content often signif- icantly differ in their properties, and are therefore difficult to match with traditional IR methods. A common example is a user entering words to de- scribe their information need that do not match the words used in the most relevant indexed doc- uments. This work addresses this problem, but shifts focus from words to syntactic structures of questions and relevant pieces of text. To this end, we present a novel algorithm that analyses the de- pendency structures of known valid answer sen- tence and from these acquires patterns that can be used to more precisely retrieve relevant text pas- sages from the underlying document collection. To achieve this, the position of key phrases in the answer sentence relative to the answer itself is an- alyzed and linked to a certain syntactic question type. Unlike most previous work that uses depen- dency paths for QA (see Section 2), our approach does not require a candidate sentence to be similar to the question in any respect. We learn valid de- pendency structures from the known answer sen- tences alone, and therefore are able to link a much wider spectrum of answer sentences to the ques- tion. The work in this paper is presented and eval- uated in a classical factoid Question Answering (QA) setting. The main reason for this is that in QA suitable training and test data is available in the public domain, e.g. via the Text REtrieval Conference (TREC), see for example (Voorhees, 1999). The methods described in this paper how- ever can also be applied to other IR scenarios, e.g. web search. The …

Content truncated to 500 words