A Survey of Semantic Representation Methods for Natural Language Understanding
Semantic is the core research object of NLU studies.There are multiple approaches to capture semantics,leading to various semantic representations.These approaches are scattered across domains and tasks,such as the query graphs and SparQL for knowledge graph question answering,the SQL for table question answering,and the Frame and AMR for parsing sentence semantics.This paper reviews common semantic representations and categorizes them into two types:External World Semantics(EWS)and the Language Interior Semantics(LIS),following the notable world-language relation.For the former,this paper discusses the research trendings,the design of new representations,and the comparison metrics.For the latter,this paper briefly summarizes the debate around whether LIS truly contains semantics.Finally,this paper introduces researches combining both types,revealing their strong potential.
semantic representationsemantic parsingnatural language understanding
NETL
NSTL
万方数据
