Using contradictions improves question answering systems
This work addresses safety-critical domains like medicine and science by enhancing QA reliability, though it is incremental as it builds on existing NLI methods.
The paper tackled the problem of improving question answering systems by incorporating contradiction detection alongside entailment, finding that combining contradiction, entailment, and confidence scores yields the best performance, with slight improvements over entailment-only systems in specific datasets.
This work examines the use of contradiction in natural language inference (NLI) for question answering (QA). Typically, NLI systems help answer questions by determining if a potential answer is \emph{entailed} (supported) by some background context. But is it useful to also determine if an answer contradicts the context? We test this in two settings, multiple choice and extractive QA, and find that systems that incorporate contradiction can do slightly better than entailment-only systems on certain datasets. However, the best performances come from using contradiction, entailment, and QA model confidence scores together. This has implications for the deployment of QA systems in domains such as medicine and science where safety is an issue.