Abstract
A major challenge of semantic parsing is the vocabulary mismatch problembetween natural language and target ontology. In this paper, we propose asentence rewriting based semantic parsing method, which can effectively resolvethe mismatch problem by rewriting a sentence into a new form which has the samestructure with its target logical form. Specifically, we propose twosentence-rewriting methods for two common types of mismatch: a dictionary-basedmethod for 1-N mismatch and a template-based method for N-1 mismatch. Weevaluate our entence rewriting based semantic parser on the benchmark semanticparsing dataset -- WEBQUESTIONS. Experimental results show that our systemoutperforms the base system with a 3.4% gain in F1, and generates logical formsmore accurately and parses sentences more robustly.