Exploring Distantly-Labeled Rationales in Neural Network Models

TL;DR

This paper introduces two auxiliary loss functions that enhance neural network models' ability to utilize distantly-labeled rationales by focusing on important words beyond labeled rationales and reducing emphasis on unhelpful rationales, leading to improved performance.

Contribution

It proposes novel auxiliary loss functions that improve the use of distantly-labeled rationales in neural networks by emphasizing important non-labeled words and reducing focus on irrelevant rationales.

Findings

Models effectively learn crucial clues from imperfect rationales.

Proposed methods outperform existing approaches.

Focus on important words beyond labeled rationales.

Abstract

Recent studies strive to incorporate various human rationales into neural networks to improve model performance, but few pay attention to the quality of the rationales. Most existing methods distribute their models' focus to distantly-labeled rationale words entirely and equally, while ignoring the potential important non-rationale words and not distinguishing the importance of different rationale words. In this paper, we propose two novel auxiliary loss functions to make better use of distantly-labeled rationales, which encourage models to maintain their focus on important words beyond labeled rationales (PINs) and alleviate redundant training on non-helpful rationales (NoIRs). Experiments on two representative classification tasks show that our proposed methods can push a classification model to effectively learn crucial clues from non-perfect rationales while maintaining the ability to spread its focus to other unlabeled important words, thus significantly outperform existing methods.