Lexical Generalization in CCG Grammar Induction for Semantic Parsing

Proceedings of the 2011 Conference on Empirical Methods in Natural Language Processing, pages 1512–1523, Edinburgh, Scotland, UK, July 27–31, 2011. c⃝2011 Association for Computational Linguistics Lexical Generalization in CCG Grammar Induction for Semantic Parsing Tom Kwiatkowski∗ t.m.kwiatkowksi@sms.ed.ac.uk Luke Zettlemoyer† lsz@cs.washington.edu Sharon Goldwater∗ sgwater@inf.ed.ac.uk Mark Steedman∗ steedman@inf.ed.ac.uk ∗School of Informatics University of Edinburgh Edinburgh, EH8 9AB, UK †Computer Science & Engineering University of Washington Seattle, WA 98195 Abstract We consider the problem of learning fac- tored probabilistic CCG grammars for seman- tic parsing from data containing sentences paired with logical-form meaning representa- tions. Traditional CCG lexicons list lexical items that pair words and phrases with syntac- tic and semantic content. Such lexicons can be inefficient when words appear repeatedly with closely related lexical content. In this paper, we introduce factored lexicons, which include both lexemes to model word meaning and templates to model systematic variation in word usage. We also present an algorithm for learning factored CCG lexicons, along with a probabilistic parse-selection model. Evalua- tions on benchmark datasets demonstrate that the approach learns highly accurate parsers, whose generalization performance benefits greatly from the lexical factoring. 1 Introduction Semantic parsers automatically recover representa- tions of meaning from natural language sentences. Recent work has focused on learning such parsers directly from corpora made up of sentences paired with logical meaning representations (Kate et al., 2005; Kate and Mooney, 2006; Wong and Mooney, 2006, 2007; Zettlemoyer and Collins, 2005, 2007; Lu et al., 2008; Kwiatkowski et al., 2010). For example, in a flight booking domain we might have access to training examples such as: Sentence: I want flights from Boston Meaning: λx.flight(x)∧ from(x,bos) and the goal is to learn a grammar that can map new, unseen, sentences onto their corresponding mean- ings, or logical forms. One approach to this problem has developed al- gorithms for leaning probabilistic CCG grammars (Zettlemoyer and Collins, 2005, 2007; Kwiatkowski et al., 2010). These grammars are well-suited to the task of semantic parsing, as they closely link syn- tax and semantics. They can be used to model a wide range of complex linguistic phenomena and are strongly lexicalized, storing all language-specific grammatical information directly with the words in the lexicon. For example, a typical learned lexicon might include entries such as: (1) flight ⊢ N :λx.flight(x) (2) flight ⊢ N/(S|NP):λ fλx.flight(x)∧ f(x) (3) flight ⊢ N\N :λ fλx.flight(x)∧ f(x) (4) fare ⊢ N :λx.cost(x) (5) fare ⊢ N/(S|NP):λ fλx.cost(x)∧ f(x) (6) fare ⊢ N\N :λ fλx.cost(x)∧ f(x) (7) Boston ⊢ NP:bos (8) Boston ⊢ N\N :λ fλx.from(x,bos)∧ f(x) (9) New York ⊢ NP:nyc (10) New York ⊢ N\N :λ fλx.from(x,nyc)∧ f(x) Although lexicalization of this kind is useful for learning, as we will see, these grammars can also suffer from sparsity in the training data, since closely related entries must be repeatedly learned for all members of a certain class of words. For exam- ple, the list above shows a selection of lexical items that would have to be learned separately. In this list, the word “flight” is …

Content truncated to 500 words