Promoting multiword expressions in A* TAG parsing

Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers, pages 429–439, Osaka, Japan, December 11-17 2016. Promoting multiword expressions in A⋆ TAG parsing Jakub Waszczuk and Agata Savary Université François-Rabelais Tours, 3 place Jean-Jaurès, 41000 Blois, France first.last@univ-tours.fr Yannick Parmentier LIFO - Université d’Orléans, 6, rue Léonard de Vinci, 45067 Orléans, France first.last@univ-orleans.fr Abstract Multiword expressions (MWEs) are pervasive in natural languages and often have both idiomatic and compositional readings, which leads to high syntactic ambiguity. We show that for some MWE types idiomatic readings are usually the correct ones. We propose a heuristic for an A⋆ parser for Tree Adjoining Grammars which benefits from this knowledge by promoting MWE- oriented analyses. This strategy leads to a substantial reduction in the parsing search space in case of true positive MWE occurrences, while avoiding parsing failures in case of false positives. 1 Introduction Multiword expressions (MWEs), e.g. by and large, red tape, and to pull one’s socks up ’to correct one’s work or behavior’, are linguistic objects containing two or more words and showing idiosyncratic behavior at different levels. Notably, their meaning is often not deducible from the meanings of their components and from their syntactic structure in a fully compositional way. Thus, interpretation-oriented NLP tasks, such as semantic calculus or translation, call for MWE-dedicated procedures. Syntactic parsing often underlies such tasks, and the crucial issue is at which point the MWE identification should take place: before (Nivre and Nilsson, 2004), after (Constant et al., 2012) or during parsing (Wehrli et al., 2010; Green et al., 2013; Candito and Constant, 2014; Nasr et al., 2015; Constant and Nivre, 2016). The last, joint, approach proves the most efficient due to at least two reasons. Firstly, some MWEs coincide with word combinations that cross phrase boundaries, which is hard to detect prior to parsing, as in example (1). Secondly, while most MWEs have both an idiomatic and a compositional reading, as in examples (2)–(3), the former occurs much more frequently than the latter for large classes of MWEs. In Sec. 6 we show that, indeed, the idiomaticity rate, i.e. the ratio of occurrences with idiomatic reading to all occurrences in a corpus, exceeds 0.95 for verbal MWEs and compounds. This suggests that promoting MWE-oriented analyses in parsing might lead to rapidly achieved correct parses. (Wehrli, 2014) shows that, indeed, the quality of symbolic parsing significantly increases if an occurrence of a MWE is admitted as soon as the necessary syntactic constraints are fulfilled. Our goal is to apply a similar strategy, i.e. to systematically promote MWE-oriented interpretations, while parsing with Tree Adjoining Grammars (TAGs). (1) After all the preparations we finally left. (2) After being criticized, she pulled her socks up. (3) When the kid shivered with cold, she pulled its socks up. (4) Acid rains in Ghana are equally grim. Consider the sentence in example (4). At least two competing analyses are syntactically valid for the first 4 words: rains is (a) a verb with the subject acid, …

Content truncated to 500 words