Logical Satisfiability of Counterfactuals for Faithful Explanations in NLI

TL;DR

This paper proposes a method for evaluating the faithfulness of natural language inference explanations using logical counterfactuals, which does not require training separate verification models and effectively detects unfaithful explanations.

Contribution

It introduces a novel counterfactual-based metric for faithfulness evaluation in NLI that leverages logical predicates without needing additional explanation training.

Findings

The method effectively distinguishes human-model agreement and disagreement.

Counterfactual hypothesis generation is validated with few-shot priming.

The metric is sensitive to unfaithful explanations.

Abstract

Evaluating an explanation's faithfulness is desired for many reasons such as trust, interpretability and diagnosing the sources of model's errors. In this work, which focuses on the NLI task, we introduce the methodology of Faithfulness-through-Counterfactuals, which first generates a counterfactual hypothesis based on the logical predicates expressed in the explanation, and then evaluates if the model's prediction on the counterfactual is consistent with that expressed logic (i.e. if the new formula is \textit{logically satisfiable}). In contrast to existing approaches, this does not require any explanations for training a separate verification model. We first validate the efficacy of automatic counterfactual hypothesis generation, leveraging on the few-shot priming paradigm. Next, we show that our proposed metric distinguishes between human-model agreement and disagreement on new counterfactual input. In addition, we conduct a sensitivity analysis to validate that our metric is sensitive to unfaithful explanations.