Driving Semantic Parsing from the World’s Response

Proceedings of the Fourteenth Conference on Computational Natural Language Learning, pages 18–27, Uppsala, Sweden, 15-16 July 2010. c⃝2010 Association for Computational Linguistics Driving Semantic Parsing from the World’s Response James Clarke Dan Goldwasser Ming-Wei Chang Dan Roth Department of Computer Science University of Illinois Urbana, IL 61820 {clarkeje,goldwas1,mchang21,danr}@illinois.edu Abstract Current approaches to semantic parsing, the task of converting text to a formal meaning representation, rely on annotated training data mapping sentences to logi- cal forms. Providing this supervision is a major bottleneck in scaling semantic parsers. This paper presents a new learn- ing paradigm aimed at alleviating the su- pervision burden. We develop two novel learning algorithms capable of predicting complex structures which only rely on a binary feedback signal based on the con- text of an external world. In addition we reformulate the semantic parsing problem to reduce the dependency of the model on syntactic patterns, thus allowing our parser to scale better using less supervision. Our results surprisingly show that without us- ing any annotated meaning representations learning with a weak feedback signal is ca- pable of producing a parser that is compet- itive with fully supervised parsers. 1 Introduction Semantic Parsing, the process of converting text into a formal meaning representation (MR), is one of the key challenges in natural language process- ing. Unlike shallow approaches for semantic in- terpretation (e.g., semantic role labeling and in- formation extraction) which often result in an in- complete or ambiguous interpretation of the natu- ral language (NL) input, the output of a semantic parser is a complete meaning representation that can be executed directly by a computer program. Semantic parsing has mainly been studied in the context of providing natural language interfaces to computer systems. In these settings the target meaning representation is defined by the seman- tics of the underlying task. For example, provid- ing access to databases: a question posed in nat- ural language is converted into a formal database query that can be executed to retrieve information. Example 1 shows a NL input query and its corre- sponding meaning representation. Example 1 Geoquery input text and output MR “What is the largest state that borders Texas?” largest(state(next to(const(texas)))) Previous works (Zelle and Mooney, 1996; Tang and Mooney, 2001; Zettlemoyer and Collins, 2005; Ge and Mooney, 2005; Zettlemoyer and Collins, 2007; Wong and Mooney, 2007) employ machine learning techniques to construct a seman- tic parser. The learning algorithm is given a set of input sentences and their corresponding meaning representations, and learns a statistical semantic parser — a set of rules mapping lexical items and syntactic patterns to their meaning representation and a score associated with each rule. Given a sen- tence, these rules are applied recursively to derive the most probable meaning representation. Since semantic interpretation is limited to syntactic pat- terns identified in the training data, the learning algorithm requires considerable amounts of anno- tated data to account for the syntactic variations associated with the meaning representation. An- notating sentences with their MR is a difficult, …

Content truncated to 500 words