Learning Explanations from Language Data

TL;DR

This paper demonstrates that PatternAttribution, originally used for explaining neural networks in vision, can also produce meaningful interpretations for language classification models.

Contribution

It extends the application of PatternAttribution to the language domain, showing its versatility in generating explanations across modalities.

Findings

PatternAttribution produces interpretable explanations in language models.

The method is effective in understanding language classification decisions.

It bridges the gap between vision and language interpretability techniques.

Abstract

PatternAttribution is a recent method, introduced in the vision domain, that explains classifications of deep neural networks. We demonstrate that it also generates meaningful interpretations in the language domain.