Prepositions Matter in Quantifier Scope Disambiguation

Proceedings of the 29th International Conference on Computational Linguistics, pages 3960–3970 October 12–17, 2022. 3960 Prepositions Matter in Quantifier Scope Disambiguation Aleksander Leczkowski Faculty of Psychology University of Warsaw ale.leczkowski@gmail.com Justyna Grudzi´nska Faculty of Philosophy University of Warsaw j.grudzinska@uw.edu.pl Manuel Vargas Guzmán Faculty of Philosophy University of Warsaw m.vargas-guzman@uw.edu.pl Aleksander Wawer Institute of Computer Science Polish Academy of Sciences axw@ipipan.waw.pl Aleksandra Siemieniuk Faculty of Polish Studies University of Warsaw aleksandra.siemieniuk@uw.edu.pl Abstract Although it is widely agreed that world knowl- edge plays a significant role in quantifier scope disambiguation (QSD), there has been only very limited work on how to integrate this knowledge into a QSD model. This paper contributes to this scarce line of research by incorporating into a machine learning model our knowledge about relations, as conveyed by a manageable closed class of function words: prepositions. For data, we use a scope- disambiguated corpus created by AnderBois, Brasoveanu and Henderson, which is addition- ally annotated with prepositional senses using Schneider et al’s Semantic Network of Adpo- sition and Case Supersenses (SNACS) scheme. By applying Manshadi and Allen’s method to the corpus, we were able to inspect the informa- tion gain provided by prepositions for the QSD task. Statistical analysis of the performance of the classifiers, trained in scenarios with and without preposition information, supports the claim that prepositional senses have a strong positive impact on the learnability of automatic QSD systems. 1 Introduction QSD is a problem in natural language processing that arises in connection with sentences that contain multiple quantified NPs: (1) Every kid climbed a tree. Sentence (1) can be understood to mean that ev- ery kid climbed a possibly different tree. This is the so-called surface scope reading where the first quantified NP has wider scope than the second, corresponding to the surface ordering of the two NPs in the sentence: every kid > a tree. The other, and usually less preferred, reading is the one in which there is a single tree that all the kids climbed. This is the inverse scope reading where the second quantified NP has wider scope than the first, re- versing the order of the two NPs in the sentence: a tree > every kid. Many studies on quantifier scope have dealt with the issue of generating the set of possible scope readings for a sentence like (1), both from a theoretical perspective (May, 1978; Cooper, 1983; May, 1985; Hendriks, 1993; Steed- man, 2012; Barker and Shan, 2014) and computa- tionally (Woods, 1987; Hobbs and Shieber, 1987; Bos, 1996; Copestake et al., 2001; Egg et al., 2001; Bos et al., 2004; Koller et al., 2008; Evang and Bos, 2013; Sayeed, 2016). A much smaller num- ber of studies have focused on statistical and au- tomatic QSD and the problem of identifying the set of factors relevant to scope preferences (Hig- gins and Sadock, 2003; AnderBois et al., 2012; Manshadi and Allen, 2011; Manshadi et al., 2013). These studies have shown, mostly in line with what was proposed in the semantics …

Content truncated to 500 words