Combining Event Semantics and Degree Semantics for Natural Language Inference

Proceedings of the 28th International Conference on Computational Linguistics, pages 1758–1764 Barcelona, Spain (Online), December 8-13, 2020 1758 Combining Event Semantics and Degree Semantics for Natural Language Inference Izumi Haruta1 haruta.izumi@is.ocha.ac.jp Koji Mineshima2 minesima@abelard.flet.keio.ac.jp Daisuke Bekki1 bekki@is.ocha.ac.jp 1Ochanomizu University, Tokyo, Japan 2Keio University, Tokyo, Japan Abstract In formal semantics, there are two well-developed semantic frameworks: event semantics, which treats verbs and adverbial modifiers using the notion of event, and degree semantics, which ana- lyzes adjectives and comparatives using the notion of degree. However, it is not obvious whether these frameworks can be combined to handle cases in which the phenomena in question are inter- acting with each other. Here, we study this issue by focusing on natural language inference (NLI). We implement a logic-based NLI system that combines event semantics and degree semantics and their interaction with lexical knowledge. We evaluate the system on various NLI datasets containing linguistically challenging problems. The results show that the system achieves high accuracies on these datasets in comparison with previous logic-based systems and deep-learning- based systems. This suggests that the two semantic frameworks can be combined consistently to handle various combinations of linguistic phenomena without compromising the advantage of either framework. 1 Introduction Since Montague (1970), formal compositional semantics has provided successful accounts of linguistic phenomena using logical expressions along with syntactic structures. In recent years, with the devel- opment of wide-coverage parsers such as Combinatory Categorial Grammar (CCG) parsers (Clark and Curran, 2007), some of the formal theories have been implemented as robust computational seman- tics (Bos, 2008; Mineshima et al., 2015; Abzianidze, 2016) and have been applied to natural language inference (NLI), which is the task of determining whether a text entails a hypothesis. This paper attempts to push forward this paradigm by focusing on the interaction of two semantic frameworks, namely, event semantics and degree semantics. Generally speaking, research in linguistic formal semantics has tended to focus on creating an in- depth theory for a variety of linguistic phenomena, such as quantifiers, adjectives, comparatives, and tense. However, it is often not obvious whether these independent theories can be combined and ex- tended to cases in which the phenomena in question interact. Thus, event semantics has been developed largely to account for the semantics of verb phrases and adverbial modifiers (Davidson, 1967; Parsons, 1990). In contrast, degree semantics provides an analysis of gradable expressions such as adjectives and comparatives, using the notion of degree (Cresswell, 1976; Stechow, 1984; Kennedy, 1997). Although each theory has been elaborated for its own sake, it is not clear how to combine these two and handle expressions that require the application of both theories, such as the comparative form of adverbs, though it is a necessary step for analyzing real texts. The computational modeling of compositional semantic theories mentioned above enables us to pre- cisely compute the predictions of each theory. In addition, their application to NLI tasks provides a systematic way of evaluating a formal semantic theory. Building on the previous logic-based approaches to NLI, …

Content truncated to 500 words