A Probabilistic Generative Grammar for Semantic Parsing

Proceedings of the 21st Conference on Computational Natural Language Learning (CoNLL 2017), pages 248–259, Vancouver, Canada, August 3 - August 4, 2017. c⃝2017 Association for Computational Linguistics A Probabilistic Generative Grammar for Semantic Parsing Abulhair Saparov Carnegie Mellon University Machine Learning Department Pittsburgh, P.A. asaparov@cs.cmu.edu Vijay Saraswat IBM T.J. Watson Research Center Yorktown Heights, N.Y. vijay@saraswat.org Tom M. Mitchell Carnegie Mellon University Machine Learning Department Pittsburgh, P.A. tom.mitchell@cmu.edu Abstract We present a generative model of nat- ural language sentences and demon- strate its application to semantic pars- ing. In the generative process, a logical form sampled from a prior, and condi- tioned on this logical form, a grammar probabilistically generates the output sentence. Grammar induction using MCMC is applied to learn the gram- mar given a set of labeled sentences with corresponding logical forms. We develop a semantic parser that finds the logical form with the highest posterior probability exactly. We obtain strong results on the GeoQuery dataset and achieve state-of-the-art F1 on Jobs. 1 Introduction Accurate and efficient semantic parsing is a long-standing goal in natural language pro- cessing. Existing approaches are quite suc- cessful in particular domains (Zettlemoyer and Collins, 2005, 2007; Wong and Mooney, 2007; Liang et al., 2011; Kwiatkowski et al., 2010, 2011, 2013; Li et al., 2013; Zhao and Huang, 2014; Dong and Lapata, 2016). However, they are largely domain-specific, relying on addi- tional supervision such as a lexicon that pro- vides the semantics or the type of each to- ken in a set (Zettlemoyer and Collins, 2005, 2007; Kwiatkowski et al., 2010, 2011; Liang et al., 2011; Zhao and Huang, 2014; Dong and Lapata, 2016), or a set of initial syn- chronous context-free grammar rules (Wong and Mooney, 2007; Li et al., 2013). To ap- ply the above systems to a new domain, addi- tional supervision is necessary. When begin- ning to read text from a new domain, humans do not need to re-learn basic English gram- Semantic prior Logical form turn on device(Ada,gpu cluster) Generative semantic grammar Parser S VP NP NP P NP V NP “Ada started the machine with the GPU” Figure 1: High-level illustration of the setting in which our grammar is applied in this paper. The dark arrows outline the generative process. During parsing, the in- put is the observed sentence, and we wish to find the most probable logical form and derivation given the training data under the semantic prior. mar. Rather, they may encounter novel ter- minology. With this in mind, our approach is akin to that of (Kwiatkowski et al., 2013) where we provide domain-independent super- vision to train a semantic parser on a new do- main. More specifically, we restrict the rules that may be learned during training to a set that characterizes the general syntax of En- glish. While we do not explicitly present and evaluate an open-domain semantic parser, we hope our work provides a step in that direc- tion. Knowledge plays a critical role in natural language understanding. Even seemingly triv- ial …

Content truncated to 500 words