Generating Hypothetical Events for Abductive Inference

Proceedings of the 10th Conference on Lexical and Computational Semantics, pages 67–77 August 5–6, 2021, Bangkok, Thailand (online) ©2021 Association for Computational Linguistics 67 Generating Hypothetical Events for Abductive Inference Debjit Paul Research Training Group AIPHES Institute for Computational Linguistics Heidelberg University paul@cl.uni-heidelberg.de Anette Frank Research Training Group AIPHES Institute for Computational Linguistics Heidelberg University frank@cl.uni-heidelberg.de Abstract Abductive reasoning starts from some observa- tions and aims at finding the most plausible ex- planation for these observations. To perform abduction, humans often make use of temporal and causal inferences, and knowledge about how some hypothetical situation can result in different outcomes. This work offers the first study of how such knowledge impacts the Ab- ductive αNLI task – which consists in choos- ing the more likely explanation for given ob- servations. We train a specialized language model LMI that is tasked to generate what could happen next from a hypothetical sce- nario that evolves from a given event. We then propose a multi-task model MT L to solve the αNLI task, which predicts a plausible ex- planation by a) considering different possible events emerging from candidate hypotheses – events generated by LMI – and b) selecting the one that is most similar to the observed outcome. We show that our MT L model im- proves over prior vanilla pre-trained LMs fine- tuned on αNLI. Our manual evaluation and analysis suggest that learning about possible next events from different hypothetical scenar- ios supports abductive inference. 1 Introduction Abductive reasoning (AR) is inference to the best explanation. It typically starts from an incomplete set of observations about everyday situations and comes up with what can be considered the most likely possible explanation given these observa- tions (Pople, 1973; Douven, 2017). One of the key characteristics that make abductive reasoning more challenging and distinct from other types of reason- ing is its non-monotonic character (Strasser and An- tonelli, 2019) i.e., even the most likely explanations are not necessarily correct. For example, in Figure 1, the most likely explanation for Observation 1: “wet grass outside my house” is that “it has been The grass outside my house is wet The sprinkler outside was switched on Observation 2 It rained last night Sprinkler made the grass wet Plausible Explanation Observation 1 : : Figure 1: Motivational example illustrating Abductive Reasoning and its non-monotonic character. raining”. However, when a new piece of informa- tion (observation or evidence) becomes available, the explanation must possibly be retracted, show- ing the defeasible character of abduction. With the new observation (“the sprinkler was switched on”) the most plausible explanation changes to “Sprin- kler caused the grass to be wet”. Humans, in such situations, could induce or validate such abductive inferences by performing hypothetical reasoning (such as “What would happen if the sprinkler was switched on?”) to arrive at a plausible explanation for “wet grass outside my house”. In this work, we focus on the αNLI task (Bhaga- vatula et al., 2020), where given two observations (O1 at time t1, O2 at …

Content truncated to 500 words