Explanation Generation for Contradiction Reconciliation with LLMs

Jason Chan Zhixue Zhao Robert Gaizauskas · 2026

Read Explanation Generation for Contradiction Reconciliation with LLMs by Jason Chan Zhixue Zhao Robert Gaizauskas (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-22735.txt.

Description

Chan, Zhao, and Gaizauskas set a new task for language models: explaining away apparent contradictions. Most work treats contradictions as errors to settle by picking a winner. Human reasoners instead often invent a story that makes both claims fit. The authors reuse inference datasets to build the task and its quality metrics. Tests on 13 models show limited success. Longer reasoning chains link to worse results, not better. A training proof of concept improves accuracy on two story reasoning benchmarks in non-thinking mode.

Catalog record

CREATORJason Chan Zhixue Zhao Robert Gaizauskas
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE52507
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.22735

Browse the full library · About the Tower of Babel