Explanations explained. Influence of Free-text Explanations on LLMs and the Role of Implicit Knowledge

ACL ARR 2025 February Submission7979 Authors

Abstract: In this work, we investigate the influence of different types of natural language explanations on LLMs' predictions, focusing on four different datasets presenting tasks that involve leveraging implicit knowledge. We conduct experiments with three SOTA LLMs on five types of explanations, either written by humans or machine-generated, through three generation methods: explain given the correct label (label-aware), explain and predict the label contextually (label-agnostic), and support the falseness of the correct label (label-contradicting). Our results demonstrate that providing explanations consistently improves the accuracy of LLM predictions, even when the models are not explicitly trained to take explanations as input, and pave the way to a study of the relationship between implicit content delivered by the explanation and its effectiveness.

Paper Type: Long

Research Area: Interpretability and Analysis of Models for NLP

Research Area Keywords: explanability, explanations, implicitness, generation, LLMs

Contribution Types: Model analysis & interpretability, Data resources, Data analysis

Languages Studied: English, Italian

Submission Number: 7979