期刊名称：Bulletin of the Technical Committee on Data Engineering
出版年度：2016
卷号：39
期号：3
页码：63
出版社：IEEE Computer Society
摘要：We propose NEURAL ENQUIRER — a neural network architecture for answering natural language(NL) questions based on a knowledge base (KB) table. Unlike existing work on end-to-end training ofsemantic parsers [13, 12], NEURAL ENQUIRER is fully “neuralized”: it finds distributed representationsof queries and KB tables, and executes queries through a series of neural network components called“executors”. Executors model query operations and compute intermediate execution results in the formof table annotations at different levels. NEURAL ENQUIRER can be trained with gradient descent, withwhich the representations of queries and the KB table are jointly optimized with the query executionlogic. The training can be done in an end-to-end fashion, and it can also be carried out with strongerguidance, e.g., step-by-step supervision for complex queries. NEURAL ENQUIRER is one step towardsbuilding neural network systems that can understand natural language in real-world tasks. As a proof-of-concept, we conduct experiments on a synthetic QA task, and demonstrate that the model can learnto execute reasonably complex NL queries on small-scale KB tables.
