“I’m Not Mad”: Commonsense Implications of Negation and Contradiction

Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 4380–4397 June 6–11, 2021. ©2021 Association for Computational Linguistics 4380 “I’m Not Mad”: Commonsense Implications of Negation and Contradiction Liwei Jiang1 Antoine Bosselut2,3 Chandra Bhagavatula2 Yejin Choi1,2 1Paul G. Allen School of Computer Science & Engineering, University of Washington 2Allen Institute for Artificial Intelligence 3Stanford University {lwjiang,yejin}@cs.washington.edu,{chandrab}@allenai.org Abstract Natural language inference requires reason- ing about contradictions, negations, and their commonsense implications. Given a simple premise (e.g., “I’m mad at you”), humans can reason about the varying shades of contradic- tory statements ranging from straightforward negations (“I’m not mad at you”) to common- sense contradictions (“I’m happy”). Moreover, these negated or contradictory statements shift the commonsense implications of the origi- nal premise in nontrivial ways. For example, while “I’m mad” implies “I’m unhappy about something,” negating the premise (i.e., “I’m not mad”) does not necessarily negate the cor- responding commonsense implications. In this paper, we present the first comprehen- sive study focusing on commonsense impli- cations of negated statements and contradic- tions. We introduce ANION1, a new com- monsense knowledge graph with 624K if-then rules focusing on negated and contradictory events. We then present joint generative and discriminative inference models for this new resource, providing novel empirical insights on how logical negations and commonsense contradictions reshape the commonsense im- plications of their original premises. 1 Introduction Humans reason about underlying causes and ef- fects of events described in text. For example, in Figure 1, the event “X wears a mask” is as- sociated with many causal inferences such as “X is seen as responsible,” or “Others get protected.” Hypothesizing and reasoning about commonsense inferences is used for understanding complex sit- uations encountered in everyday life (Sap et al., 2019; Bisk et al., 2020; Bhagavatula et al., 2020; Sakaguchi et al., 2020). This ability eludes AI sys- tems, and has motivated the design of a wealth of 1Data and code available at https://github.com/ liweijiang/anion X takes his mask off Original Commonsense Contradiction X wears a mask X loses his mask responsible get protected to throw the used mask away to have a mask to breath freely nervous to avoid X X is seen as… As a result, others want… As a result, X feels… As a result, X wants… Before, X needed… As a result, others want… As a result, others then… Because X wanted… Because X wanted… Before, X needed… saves money from buying a mask As a result, X then… X doesn’t wear a mask Logical Negation carefree X is seen as… X is seen as… Figure 1: Commonsense inferences for the event “X wears a mask,” its logical negation and commonsense contradiction events, and their associated inferences. commonsense knowledge resources, such as Cyc (Lenat, 1995), ConceptNet (Speer et al., 2017), and ATOMIC (Sap et al., 2020; Hwang et al., 2020), to provide structured reasoning capabilities to AI systems (Lin et al., 2019; Feng et al., 2020). However, reasoning about …

Content truncated to 500 words