Practical Benefits of Feature Feedback Under Distribution Shift

TL;DR

This paper investigates how feature feedback, like rationales, can improve out-of-domain performance in NLP tasks, showing significant benefits in sentiment analysis but not in natural language inference.

Contribution

It demonstrates that feature feedback enhances out-of-domain robustness in sentiment analysis, revealing benefits beyond traditional in-sample metrics.

Findings

Feature feedback improves out-of-domain sentiment analysis performance.

No significant in-domain performance difference with feature feedback.

Limited benefits of feature feedback in natural language inference.

Abstract

In attempts to develop sample-efficient and interpretable algorithms, researcher have explored myriad mechanisms for collecting and exploiting feature feedback (or rationales) auxiliary annotations provided for training (but not test) instances that highlight salient evidence. Examples include bounding boxes around objects and salient spans in text. Despite its intuitive appeal, feature feedback has not delivered significant gains in practical problems as assessed on iid holdout sets. However, recent works on counterfactually augmented data suggest an alternative benefit of supplemental annotations, beyond interpretability: lessening sensitivity to spurious patterns and consequently delivering gains in out-of-domain evaluations. We speculate that while existing methods for incorporating feature feedback have delivered negligible in-sample performance gains, they may nevertheless provide out-of-domain benefits. Our experiments addressing sentiment analysis, show that feature feedback methods perform significantly better on various natural out-of-domain datasets despite comparable in-domain evaluations. By contrast, performance on natural language inference remains comparable. Finally, we compare those tasks where feature feedback does (and does not) help.