A Causal Lens for Evaluating Faithfulness Metrics

TL;DR

This paper introduces Causal Diagnosticity, a framework for evaluating faithfulness metrics of natural language explanations in LLMs, revealing variability in metric performance across tasks and models.

Contribution

It proposes a principled benchmark using model editing to generate explanation pairs, enabling systematic comparison of faithfulness metrics.

Findings

Filler Tokens metric performs best overall

Continuous metrics are more diagnostic than binary ones

Performance of metrics varies across tasks and models

Abstract

Large Language Models (LLMs) offer natural language explanations as an alternative to feature attribution methods for model interpretability. However, despite their plausibility, they may not reflect the model's true reasoning faithfully. While several faithfulness metrics have been proposed, they are often evaluated in isolation, making principled comparisons between them difficult. We present Causal Diagnosticity, a testbed framework for evaluating faithfulness metrics for natural language explanations. We use the concept of diagnosticity, and employ model-editing methods to generate faithful-unfaithful explanation pairs. Our benchmark includes four tasks: fact-checking, analogy, object counting, and multi-hop reasoning. We evaluate prominent faithfulness metrics, including post-hoc explanation and chain-of-thought methods. Diagnostic performance varies across tasks and models, with Filler Tokens performing best overall. Additionally, continuous metrics are generally more diagnostic than binary ones but can be sensitive to noise and model choice. Our results highlight the need for more robust faithfulness metrics.