A Probabilistic Generative Grammar for Semantic Parsing

摘要

We present a framework that couples the syntax and semantics of naturallanguage sentences in a generative model, in order to develop a semantic parserthat jointly infers the syntactic, morphological, and semantic representationsof a given sentence under the guidance of background knowledge. To generate asentence in our framework, a semantic statement is first sampled from a prior,such as from a set of beliefs in a knowledge base. Given this semanticstatement, a grammar probabilistically generates the output sentence. A jointsemantic-syntactic parser is derived that returns the $k$-best semantic andsyntactic parses for a given sentence. The semantic prior is flexible, and canbe used to incorporate background knowledge during parsing, in ways unlikeprevious semantic parsing approaches. For example, semantic statementscorresponding to beliefs in a knowledge base can be given higher priorprobability, type-correct statements can be given somewhat lower probability,and beliefs outside the knowledge base can be given lower probability. Theconstruction of our grammar invokes a novel application of hierarchicalDirichlet processes (HDPs), which in turn, requires a novel and efficientinference approach. We present experimental results showing, for a simplegrammar, that our parser outperforms a state-of-the-art CCG semantic parser andscales to knowledge bases with millions of beliefs.