Neural Enquirer: Learning to Query Tables in Natural Language

Proceedings of 2016 NAACL Human-Computer Question Answering Workshop, pages 29–35, San Diego, California, June 12-17, 2016. c⃝2016 Association for Computational Linguistics Neural Enquirer: Learning to Query Tables in Natural Language Pengcheng Yin The University of Hong Kong pcyin@cs.hku.hk Zhengdong Lu Huawei Noah’s Ark Lab Lu.Zhengdong@huawei.com Hang Li Huawei Noah’s Ark Lab HangLi.HL@huawei.com Ben Kao The University of Hong Kong kao@cs.hku.hk Abstract We propose NEURAL ENQUIRER — a neu- ral network architecture for answering natural language (NL) questions given a knowledge base (KB) table. Unlike previous work on end-to-end training of semantic parsers, NEU- RAL ENQUIRER is fully “neuralized”: it gives distributed representations of queries and KB tables, and executes queries through a series of differentiable operations. The model can be trained with gradient descent using both end- to-end and step-by-step supervision. During training the representations of queries and the KB table are jointly optimized with the query execution logic. Our experiments show that the model can learn to execute complex NL queries on KB tables with rich structures. 1 Introduction Natural language dialogue and question answering often involve querying a knowledge base (Wen et al., 2015; Berant et al., 2013). The traditional ap- proach involves two steps: First, a given query ˜Q is semantically parsed into an “executable” represen- tation, which is often expressed in certain logical form ˜Z (e.g., SQL-like queries). Second, the rep- resentation is executed against a KB from which an answer is obtained. For queries that involve com- plex semantics and logic (e.g., “Which city hosted the longest Olympic Games before the Games in Beijing?”), semantic parsing and query execution become extremely complex. For example, care- fully hand-crafted features and rules are needed to correctly parse a complex query into its logical form (see example in the lower-left corner of Fig- ure 1). To partially overcome this complexity, recent works (Clarke et al., 2010; Liang et al., 2011; Pa- supat and Liang, 2015) attempt to “backpropagate” query execution results to revise the semantic rep- resentation of a query. This approach, however, is greatly hindered by the fact that traditional seman- tic parsing mostly involves rule-based features and symbolic manipulation, and is subject to intractable search space incurred by the great flexibility of nat- ural language. In this paper we propose NEURAL ENQUIRER — a neural network system that learns to under- stand NL queries and execute them on a KB table from examples of queries and answers. Unlike sim- ilar efforts along this line of research (Neelakan- tan et al., 2015), NEURAL ENQUIRER is a fully neuralized, end-to-end differentiable network that jointly models semantic parsing and query execu- tion. It encodes queries and KB tables into dis- tributed representations, and executes compositional queries against the KB through a series of differen- tiable operations. The model is trained using query- answer pairs, where the distributed representations of queries and the KB are optimized together with the query execution logic in an end-to-end fash- ion. We demonstrate using a synthetic QA task that NEURAL ENQUIRER is capable of …

Content truncated to 500 words