Transforming Meaning Representation Grammars to Improve Semantic Parsing

CoNLL 2008: Proceedings of the 12th Conference on Computational Natural Language Learning, pages 33–40 Manchester, August 2008 Transforming Meaning Representation Grammars to Improve Semantic Parsing Rohit J. Kate Department of Computer Sciences∗ The University of Texas at Austin 1 University Station C0500 Austin, TX 78712-0233, USA rjkate@cs.utexas.edu Abstract A semantic parser learning system learns to map natural language sentences into their domain-specific formal meaning rep- resentations, but if the constructs of the meaning representation language do not correspond well with the natural language then the system may not learn a good se- mantic parser. This paper presents ap- proaches for automatically transforming a meaning representation grammar (MRG) to conform it better with the natural lan- guage semantics. It introduces grammar transformation operators and meaning rep- resentation macros which are applied in an error-driven manner to transform an MRG while training a semantic parser learning system. Experimental results show that the automatically transformed MRGs lead to better learned semantic parsers which per- form comparable to the semantic parsers learned using manually engineered MRGs. 1 Introduction Semantic parsing is the task of converting natural language (NL) sentences into their meaning repre- sentations (MRs) which a computer program can execute to perform some domain-specific task, like controlling a robot, answering database queries etc. These MRs are expressed in a formal mean- ing representation language (MRL) unique to the domain to suit the application, like some specific command language to control a robot or some ∗Alumnus at the time of submission. ∗c⃝ 2008. Licensed under the Creative Commons Attribution-Noncommercial-Share Alike 3.0 Unported li- cense (http://creativecommons.org/licenses/by-nc-sa/3.0/). Some rights reserved. query language to execute database queries. A ma- chine learning system for semantic parsing takes NL sentences paired with their respective MRs as training data and induces a semantic parser which can then map novel NL sentences into their MRs. The grammar of an MRL, which we will call meaning representation grammar (MRG), is as- sumed to be deterministic and context-free which is true for grammars of almost all the computer executable languages. A semantic parsing learn- ing system typically exploits the given MRG of the MRL to learn a semantic parser (Kate and Mooney, 2006; Wong and Mooney, 2006). Although in dif- ferent ways, but the systems presented in these pa- pers learn how the NL phrases relate to the pro- ductions of the MRG, and using this information they parse a test sentence to compositionally gen- erate its best MR. In order to learn a good seman- tic parser, it is necessary that the productions of the MRG accurately represent the semantics be- ing expressed by the natural language. However, an MRL and its MRG are typically designed to best suit the application with little consideration for how well they correspond to the semantics of a natural language. Some other semantic parser learning systems which need MRL in the form of Prolog (Tang and Mooney, 2001) or λ-calculus (Zettlemoyer and Collins, 2007; Wong and Mooney, 2007) do not use productions of the MRG but …

Content truncated to 500 words