Towards Consistent Natural-Language Explanations via Explanation-Consistency Finetuning

TL;DR

This paper introduces explanation-consistency finetuning (EC-finetuning), a method to improve the consistency of natural-language explanations generated by large language models across related examples, enhancing their reliability.

Contribution

The paper proposes EC-finetuning, a novel finetuning approach using synthetic data to significantly improve explanation consistency in LLMs across multiple datasets and domains.

Findings

10.0% relative improvement in explanation consistency on finetuning datasets

Generalizes with +4.5% relative consistency on out-of-distribution datasets

Effective across various question-answering domains

Abstract

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 .