Knowledge-Based Question Answering as Machine Translation

Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics, pages 967–976, Baltimore, Maryland, USA, June 23-25 2014. c⃝2014 Association for Computational Linguistics Knowledge-Based Question Answering as Machine Translation Junwei Bao† ∗, Nan Duan‡ , Ming Zhou‡ , Tiejun Zhao† †Harbin Institute of Technology ‡Microsoft Research baojunwei001@gmail.com {nanduan, mingzhou}@microsoft.com tjzhao@hit.edu.cn Abstract A typical knowledge-based question an- swering (KB-QA) system faces two chal- lenges: one is to transform natural lan- guage questions into their meaning repre- sentations (MRs); the other is to retrieve answers from knowledge bases (KBs) us- ing generated MRs. Unlike previous meth- ods which treat them in a cascaded man- ner, we present a translation-based ap- proach to solve these two tasks in one u- nified framework. We translate questions to answers based on CYK parsing. An- swers as translations of the span covered by each CYK cell are obtained by a ques- tion translation method, which first gener- ates formal triple queries as MRs for the span based on question patterns and re- lation expressions, and then retrieves an- swers from a given KB based on triple queries generated. A linear model is de- fined over derivations, and minimum er- ror rate training is used to tune feature weights based on a set of question-answer pairs. Compared to a KB-QA system us- ing a state-of-the-art semantic parser, our method achieves better results. 1 Introduction Knowledge-based question answering (KB-QA) computes answers to natural language (NL) ques- tions based on existing knowledge bases (KBs). Most previous systems tackle this task in a cas- caded manner: First, the input question is trans- formed into its meaning representation (MR) by an independent semantic parser (Zettlemoyer and Collins, 2005; Mooney, 2007; Artzi and Zettle- moyer, 2011; Liang et al., 2011; Cai and Yates, ∗This work was finished while the author was visiting Mi- crosoft Research Asia. 2013; Poon, 2013; Artzi et al., 2013; Kwiatkowski et al., 2013; Berant et al., 2013); Then, the answer- s are retrieved from existing KBs using generated MRs as queries. Unlike existing KB-QA systems which treat se- mantic parsing and answer retrieval as two cas- caded tasks, this paper presents a unified frame- work that can integrate semantic parsing into the question answering procedure directly. Borrow- ing ideas from machine translation (MT), we treat the QA task as a translation procedure. Like MT, CYK parsing is used to parse each input question, and answers of the span covered by each CYK cel- l are considered the translations of that cell; un- like MT, which uses offline-generated translation tables to translate source phrases into target trans- lations, a semantic parsing-based question trans- lation method is used to translate each span into its answers on-the-fly, based on question patterns and relation expressions. The final answers can be obtained from the root cell. Derivations generated during such a translation procedure are modeled by a linear model, and minimum error rate train- ing (MERT) (Och, 2003) is used to tune feature weights based on a set of question-answer pairs. …

Content truncated to 500 words