Semantic parsing has emerged as a significant and powerful paradigm fornatural language interface and question answering systems. Traditional methodsof building a semantic parser rely on high-quality lexicons, hand-craftedgrammars and linguistic features which are limited by applied domain orrepresentation. In this paper, we propose a general approach to learn fromdenotations based on Seq2Seq model augmented with attention mechanism. Weencode input sequence into vectors and use dynamic programming to infercandidate logical forms. We utilize the fact that similar utterances shouldhave similar logical forms to help reduce the searching space. Under ourlearning policy, the Seq2Seq model can learn mappings gradually with noises.Curriculum learning is adopted to make the learning smoother. We test ourmethod on the arithmetic domain which shows our model can successfully inferthe correct logical forms and learn the word meanings, compositionality andoperation orders simultaneously.
展开▼
