REV: Information-Theoretic Evaluation of Free-Text Rationales

Abstract

Free-text rationales are a promising step towards explainable AI, yet their evaluation remains an open research problem. While existing metrics have mostly focused on measuring the direct association between the rationale and a given label, we argue that an ideal metric should also be able to focus on the new information uniquely provided in the rationale that is otherwise not provided in the input or the label. We investigate this research problem from an information-theoretic perspective using the conditional V-information (Hewitt et al., 2021). More concretely, we propose a metric called REV (Rationale Evaluation with conditional V-information), that can quantify the new information in a rationale supporting a given label beyond the information already available in the input or the label. Experiments on reasoning tasks across four benchmarks, including few-shot prompting with GPT-3, demonstrate the effectiveness of REV in evaluating different types of rationale-label pairs, compared to existing metrics. Through several quantitative comparisons, we demonstrate the capability of REV in providing more sensitive measurements of new information in free-text rationales with respect to a label. Furthermore, REV is consistent with human judgments on rationale evaluations. Overall, when used alongside traditional performance metrics, REV provides deeper insights into a model's reasoning and prediction processes.

Measuring Pointwise 𝒱-Usable Information In-Context-ly-Usable Information In-Context-ly

“…PVI is used to select synthetic data as an augmentation to an intent detection classifier, which achieves state-of-the-art performance (Lin et al, 2023). Chen et al (2022a) and Prasad et al (2023) incorporate PVI into an informativeness metric to evaluate rationales, and find that it captures the expected flow of information in high-quality reasoning chains.…”

“…PVI measures the amount of usable information in an input for a given model, which reflects the ease with which a model can predict a certain label given an input. Though it is a recently proposed method, the effectiveness of PVI has been demonstrated in various NLP tasks (Chen et al, 2022a;Kulmizev and Nivre, 2023;Lin et al, 2023;Prasad et al, 2023).…”