Towards Consistent Natural-Language Explanations via Explanation-Consistency Finetuning
This addresses the issue of unreliable explanations from LLMs for users needing to understand model decisions, though it is incremental as it builds on existing finetuning methods.
The paper tackles the problem of large language models generating inconsistent natural-language explanations across related inputs, proposing explanation-consistency finetuning to improve consistency, resulting in a 10.0% relative improvement on finetuning datasets and generalization to out-of-distribution datasets with a 4.5% relative gain.
Large language models (LLMs) often generate convincing, fluent explanations. However, different from humans, they often generate inconsistent explanations on different inputs. For example, an LLM may generate the explanation "all birds can fly" when answering the question "Can sparrows fly?" but meanwhile answer "no" to the related question "Can penguins fly?". Explanations should be consistent across related examples so that they allow a human to simulate the LLM's decision process on multiple examples. We propose explanation-consistency finetuning (EC-finetuning), a method that adapts LLMs to generate more consistent natural-language explanations on related examples. EC-finetuning involves finetuning LLMs on synthetic data that is carefully constructed to contain consistent explanations. Across a variety of question-answering datasets in various domains, EC-finetuning yields a 10.0% relative explanation consistency improvement on four finetuning datasets, and generalizes to seven out-of-distribution datasets not seen during finetuning (+4.5% relative). Code is available at https://github.com/yandachen/explanation-consistency-finetuning .