Human Interpretation of Saliency-based Explanation Over Text

TL;DR

This study investigates how people interpret saliency-based explanations of text models, revealing common misunderstandings and proposing methods to improve explanation comprehension through adjusted saliency and visualization changes.

Contribution

It provides empirical insights into laypeople's interpretation of saliency explanations and introduces methods to enhance understanding by correcting perceptual distortions.

Findings

People often misinterpret explanations due to superficial factors like word length.

Adjusting saliency scores based on perception estimates improves understanding.

Using bar charts instead of heatmaps can reduce misinterpretation.

Abstract

While a lot of research in explainable AI focuses on producing effective explanations, less work is devoted to the question of how people understand and interpret the explanation. In this work, we focus on this question through a study of saliency-based explanations over textual data. Feature-attribution explanations of text models aim to communicate which parts of the input text were more influential than others towards the model decision. Many current explanation methods, such as gradient-based or Shapley value-based methods, provide measures of importance which are well-understood mathematically. But how does a person receiving the explanation (the explainee) comprehend it? And does their understanding match what the explanation attempted to communicate? We empirically investigate the effect of various factors of the input, the feature-attribution explanation, and visualization procedure, on laypeople's interpretation of the explanation. We query crowdworkers for their interpretation on tasks in English and German, and fit a GAMM model to their responses considering the factors of interest. We find that people often mis-interpret the explanations: superficial and unrelated factors, such as word length, influence the explainees' importance assignment despite the explanation communicating importance directly. We then show that some of this distortion can be attenuated: we propose a method to adjust saliencies based on model estimates of over- and under-perception, and explore bar charts as an alternative to heatmap saliency visualization. We find that both approaches can attenuate the distorting effect of specific factors, leading to better-calibrated understanding of the explanation.