Large-scale Semantic Parsing via Schema Matching and Lexicon Extension

Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics, pages 423–433, Sofia, Bulgaria, August 4-9 2013. c⃝2013 Association for Computational Linguistics Large-scale Semantic Parsing via Schema Matching and Lexicon Extension Qingqing Cai Temple University Computer and Information Sciences qingqing.cai@temple.edu Alexander Yates Temple University Computer and Information Sciences yates@temple.edu Abstract Supervised training procedures for seman- tic parsers produce high-quality semantic parsers, but they have difficulty scaling to large databases because of the sheer number of logical constants for which they must see labeled training data. We present a technique for developing seman- tic parsers for large databases based on a reduction to standard supervised train- ing algorithms, schema matching, and pat- tern learning. Leveraging techniques from each of these areas, we develop a semantic parser for Freebase that is capable of pars- ing questions with an F1 that improves by 0.42 over a purely-supervised learning al- gorithm. 1 Introduction Semantic parsing is the task of translating natural language utterances to a formal meaning represen- tation language (Chen et al., 2010; Liang et al., 2009; Clarke et al., 2010; Liang et al., 2011; Artzi and Zettlemoyer, 2011). There has been recent in- terest in producing such semantic parsers for large, heterogeneous databases like Freebase (Krishna- murthy and Mitchell, 2012; Cai and Yates, 2013) and Yago2 (Yahya et al., 2012), which has driven the development of semi-supervised and distantly- supervised training methods for semantic parsing. Previous purely-supervised approaches have been limited to smaller domains and databases, such as the GeoQuery database, in part because of the cost of labeling enough samples to cover all of the log- ical constants involved in a domain. This paper investigates a reduction of the prob- lem of building a semantic parser to three stan- dard problems in semantics and machine learning: supervised training of a semantic parser, schema matching, and pattern learning. Figure 1 provides a visualization of our system architecture. We apply an existing supervised training algorithm for semantic parsing to a labeled data set. We (sentence, logical form) Training data Test questions Freebase Web Relations Extracted from Web Supervised Semantic Parser Learning MATCHER LEXTENDER Freebase PCCG Grammar and Lexicon (word, Freebase symbol) correspondences High-Coverage Freebase PCCG Grammar and Lexicon Figure 1: We reduce the task of learning a large- scale semantic parser to a combination of 1) a standard supervised algorithm for learning seman- tic parsers; 2) our MATCHER algorithm for find- ing correspondences between words and database symbols; and 3) our LEXTENDER algorithm for integrating (word, database symbol) matches into a semantic parsing lexicon. apply schema matching techniques to the prob- lem of finding correspondences between English words w and ontological symbols s. And we ap- ply pattern learning techniques to incorporate new (w, s) pairs into the lexicon of the trained seman- tic parser. This reduction allows us to apply standard tech- niques from each problem area, which in com- bination provide a large improvement over the purely-supervised approaches. On a dataset of 917 questions taken from 81 domains of …

Content truncated to 500 words