A Supervised Semantic Parsing with Lexicon Extension and Syntactic Constraint

Proceedings of Recent Advances in Natural Language Processing, pages 362–370, Hissar, Bulgaria, Sep 7–9 2015. A Supervised Semantic Parsing with Lexical Extension and Syntactic Constraint Zhihua Liao Foreign Studies College Hunan Normal University Changsha, China liao.zhihua61@gmail.com Qixian Zeng Foreign Studies College Hunan Normal University Changsha, China wanyouto@qq.com Qiyun Wang Foreign Studies College Hunan Normal University Changsha, China qywang@qq.com Abstract Existing semantic parsing research has steadily improved accuracy on a few do- mains and their corresponding meaning representations. In this paper, we present a novel supervised semantic parsing algo- rithm, which includes the lexicon exten- sion and the syntactic supervision. This algorithm adopts a large-scale knowl- edge base from the open-domain Free- base to construct efficient, rich Combi- natory Categorial Grammar (CCG) lexi- con in order to supplement the inadequacy of its manually-annotated training dataset in the small closed-domain while allows for the syntactic supervision from the dependency-parsed sentences to penalize the ungrammatical semantic parses. Eval- uations on both benchmark closed-domain datasets demonstrate that this approach learns highly accurate parser, whose pars- ing performance benefits greatly from the open-domain CCG lexicon and syntactic constraint. 1 Introduction Semantic parsers convert natural language sen- tences to logical forms through a meaning repre- sentation language. Recent research has focused on learning such parsers directly from corpora made up of sentences paired with logical meaning representations (Artzi and Zettlemoyer, 2011; Lu et al., 2008; Lu and Tou, 2011; Liao and Zhang, 2013; Kwiatkowski et al., 2010; Kwiatkowski et al., 2011; Kwiatkowski et al., 2012; Zettlemoyer and Collins, 2005; Zettlemoyer and Collins, 2007; Zettlemoyer and Collins, 2009; Zettlemoyer and Collins, 2012). And its goal is to learn a gram- mar that can map new, unseen sentences onto their corresponding meanings, or logical expressions. For decades there have been many algorithms that learn probabilistic CCG grammars. These grammars are well suited to the semantic parsing because of the close linking with syntactic and se- mantic information. Thus, they are used to model a wide range of complex linguistic phenomena and are strongly lexicalized, which store all language- specific grammatical information directly with the words and the CCG lexicon. This CCG lexicon is useful for learning parser. However, it often suffers from the sparsity and the diversity in the training and testing datasets. Consequently, we hold that a large-scale knowledge base should play a key role in the semantic parsing. That is, it might be quite favorable in training such parser and resolving these syntactic ambiguities. Using the knowledge base which contains rich semantic information from the open-domain such as Free- base, can improve efficiently the parser’s ability to solve complex syntactic parsing problem and benefit the accuracy. Besides, many previous ap- proaches do not involve the syntactic constraint to penalize the ungrammatical parses when semantic parsing. This paper presents a supervised approach to learn semantic parsing task using a large-scale open-domain knowledge base and syntactic con- straint. The semantic parser is trained to learn parsing via a large-scale open-domain CCG lex- icon while simultaneously producing parses that …

Content truncated to 500 words