Can Language Models perform Abductive Commonsense Reasoning?

TL;DR

This paper reviews methodologies for abductive commonsense reasoning, re-implements baseline models, and analyzes current approaches' weaknesses, providing insights into how language models handle this challenging task.

Contribution

It offers a comprehensive review, re-implementation of baseline models, and critical analysis of existing approaches to abductive reasoning in language models.

Findings

Current models have notable weaknesses in abductive reasoning.

Re-implemented baselines provide a benchmark for future research.

Analysis highlights areas for improvement in model reasoning capabilities.

Abstract

Abductive Reasoning is a task of inferring the most plausible hypothesis given a set of observations. In literature, the community has approached to solve this challenge by classifying/generating a likely hypothesis that does not contradict with a past observation and future observation. Some of the most well-known benchmarks that tackle this problem are aNLI and aNLG (pronounced as alpha-NLI and alpha-NLG). In this report, I review over some of the methodologies that were attempted to solve this challenge, re-implement the baseline models, and analyze some of the weaknesses that current approaches have. The code and the re-implemented results are available at this link.