Question Answering over Structured Data: an Entailment-Based Approach to Question Analysis

International Conference RANLP 2009 - Borovets, Bulgaria, pages 305–311 Question Answering over Structured Data: an Entailment-Based Approach to Question Analysis Matteo Negri and Milen Kouylekov FBK-irst - Fondazione Bruno Kessler Via Sommarive 18, 38100 Povo (TN), Italy negri,kouylekov@fbk.eu Abstract This paper addresses question analysis in the framework of Question Answering over struc- tured data. The problem is set as a relation extraction task, where all the relations of inter- est in a given domain have to be extracted from natural language questions. The proposed ap- proach applies the notion of Textual Entailment to compare the input questions with a reposi- tory of relational textual patterns. The under- lying assumption is that a question expresses a certain relation if a pattern for that relation is entailed by the question. We report on a number of experiments, testing different simple distance- based entailment algorithms over a dataset of 1487 English questions covering the domain of cultural events in a town, and 75 relations that are relevant in this domain. The positive results obtained demonstrate the feasibility of the over- all approach, and its effectiveness in the proposed QA scenario. Keywords Restricted-Domain Question Answering, Textual Entailment, Relation Extraction. 1 Introduction Question analysis is the Question Answering (QA) subtask that consists in analysing a natural language question in order to identify all the relevant informa- tion needed to extract the correct answer from a given data source. Depending on the QA application, rele- vant information may include the identification of: i) the Expected Answer Type (i.e. the semantic cate- gory of the sought-after answer), ii) the word sense of the question terms, iii) the most important keywords, iv) named entities, and v) relations between entities. In the framework of QA over structured data, extract- ing from an input question all the relevant relations in a given domain becomes crucial, as it allows to fully capture the context in which the request has to be in- terpreted and, in turn, to determine the constraints on the database query. For instance, given the question “What movie can I see today at cinema Astra?”, an effective database query will select a concept of type Movie, with specific relations with a Date (e.g. Has- Date(Movie:?, Date:“today”)) and a Cinema (e.g. HasMovieSite(Movie:?, Site:“Astra”)). Successful answer retrieval depends on capturing all and only the relations expressed in the question: unrecognized rela- tions will determine underspecified queries (often lead- ing to redundant answers), while spuriously recognized relations will determine overspecified queries (leading to answer extraction failures). While in open-domain QA any type of relation is potentially relevant, making their automatic identifi- cation unfeasible in an exhaustive manner, QA over structured data in a restricted domain presents a rea- sonable setting to address the task. In this paper we investigate the applicability of Textual Entailment (TE) as a possible solution to the problem. TE has been recently proposed as a comprehensive framework for applied semantics [4], where the mapping between linguistic objects is carried out by means of semantic inferences at the …

Content truncated to 500 words