Using Interpretation Methods for Model Enhancement

TL;DR

This paper introduces a versatile framework that uses various interpretation methods and gold rationales to improve neural NLP models, especially in low-resource scenarios, demonstrating the effectiveness of novel erasure and extractor-based methods.

Contribution

The paper proposes a general framework for model enhancement using multiple interpretation methods, including two novel approaches, outperforming gradient-based methods in many settings.

Findings

Framework improves models with interpretation methods.

Novel erasure/replace and extractor-based methods outperform gradient-based methods.

Effective especially in low-resource settings.

Abstract

In the age of neural natural language processing, there are plenty of works trying to derive interpretations of neural models. Intuitively, when gold rationales exist during training, one can additionally train the model to match its interpretation with the rationales. However, this intuitive idea has not been fully explored. In this paper, we propose a framework of utilizing interpretation methods and gold rationales to enhance models. Our framework is very general in the sense that it can incorporate various interpretation methods. Previously proposed gradient-based methods can be shown as an instance of our framework. We also propose two novel instances utilizing two other types of interpretation methods, erasure/replace-based and extractor-based methods, for model enhancement. We conduct comprehensive experiments on a variety of tasks. Experimental results show that our framework is effective especially in low-resource settings in enhancing models with various interpretation methods, and our two newly-proposed methods outperform gradient-based methods in most settings. Code is available at https://github.com/Chord-Chen-30/UIMER.