Semantic Parsing via Paraphrasing

Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics, pages 1415–1425, Baltimore, Maryland, USA, June 23-25 2014. c⃝2014 Association for Computational Linguistics Semantic Parsing via Paraphrasing Jonathan Berant Stanford University joberant@stanford.edu Percy Liang Stanford University pliang@cs.stanford.edu Abstract A central challenge in semantic parsing is handling the myriad ways in which knowl- edge base predicates can be expressed. Traditionally, semantic parsers are trained primarily from text paired with knowledge base information. Our goal is to exploit the much larger amounts of raw text not tied to any knowledge base. In this pa- per, we turn semantic parsing on its head. Given an input utterance, we first use a simple method to deterministically gener- ate a set of candidate logical forms with a canonical realization in natural language for each. Then, we use a paraphrase model to choose the realization that best para- phrases the input, and output the corre- sponding logical form. We present two simple paraphrase models, an association model and a vector space model, and train them jointly from question-answer pairs. Our system PARASEMPRE improves state- of-the-art accuracies on two recently re- leased question-answering datasets. 1 Introduction We consider the semantic parsing problem of map- ping natural language utterances into logical forms to be executed on a knowledge base (KB) (Zelle and Mooney, 1996; Zettlemoyer and Collins, 2005; Wong and Mooney, 2007; Kwiatkowski et al., 2010). Scaling semantic parsers to large knowledge bases has attracted substantial atten- tion recently (Cai and Yates, 2013; Berant et al., 2013; Kwiatkowski et al., 2013), since it drives applications such as question answering (QA) and information extraction (IE). Semantic parsers need to somehow associate natural language phrases with logical predicates, e.g., they must learn that the constructions “What What party did Clay establish? paraphrase model What political party founded by Henry Clay? ... What event involved the people Henry Clay? Type.PoliticalParty ⊓ Founder.HenryClay ... Type.Event ⊓ Involved.HenryClay Whig Party Figure 1: Semantic parsing via paraphrasing: For each candidate logical form (in red), we generate canonical utter- ances (in purple). The model is trained to paraphrase the in- put utterance (in green) into the canonical utterances associ- ated with the correct denotation (in blue). does X do for a living?”, “What is X’s profes- sion?”, and “Who is X?”, should all map to the logical predicate Profession. To learn these map- pings, traditional semantic parsers use data which pairs natural language with the KB. However, this leaves untapped a vast amount of text not related to the KB. For instance, the utterances “Where is ACL in 2014?” and “What is the location of ACL 2014?” cannot be used in traditional semantic parsing methods, since the KB does not contain an entity ACL2014, but this pair clearly contains valuable linguistic information. As another refer- ence point, out of 500,000 relations extracted by the ReVerb Open IE system (Fader et al., 2011), only about 10,000 can be aligned to Freebase (Be- rant et al., 2013). In this paper, we present a novel approach …

Content truncated to 500 words