Practical Benefits of Feature Feedback Under Distribution Shift

In attempts to develop sample-efficient and interpretable algorithms, researcher have explored myriad mechanisms for collecting and exploiting feature feedback, 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.

[1]  Greg Durrett,et al.  Model Agnostic Answer Reranking System for Adversarial Question Answering , 2021, EACL.

[2]  Graham Neubig,et al.  Weakly- and Semi-supervised Evidence Extraction , 2020, FINDINGS.

[3]  Samuel R. Bowman,et al.  Counterfactually-Augmented SNLI Training Data Does Not Yield Better Generalization Than Unaugmented Data , 2020, INSIGHTS.

[4]  Zachary Chase Lipton,et al.  Explaining The Efficacy of Counterfactually-Augmented Data , 2020, ICLR.

[5]  Eunsol Choi,et al.  QED: A Framework and Dataset for Explanations in Question Answering , 2020, Transactions of the Association for Computational Linguistics.

[6]  Byron C. Wallace,et al.  Learning to Faithfully Rationalize by Construction , 2020, ACL.

[7]  Quoc V. Le,et al.  ELECTRA: Pre-training Text Encoders as Discriminators Rather Than Generators , 2020, ICLR.

[8]  S. Dasgupta,et al.  Robust Learning from Discriminative Feature Feedback , 2020, AISTATS.

[9]  Byron C. Wallace,et al.  ERASER: A Benchmark to Evaluate Rationalized NLP Models , 2019, ACL.

[10]  Jianmo Ni,et al.  Justifying Recommendations using Distantly-Labeled Reviews and Fine-Grained Aspects , 2019, EMNLP.

[11]  Zachary Chase Lipton,et al.  Learning the Difference that Makes a Difference with Counterfactually-Augmented Data , 2019, ICLR.

[12]  Chris Callison-Burch,et al.  Seeing Things from a Different Angle:Discovering Diverse Perspectives about Claims , 2019, NAACL.

[13]  Regina Barzilay,et al.  Inferring Which Medical Treatments Work from Reports of Clinical Trials , 2019, NAACL.

[14]  Andrew Slavin Ross,et al.  Improving the Adversarial Robustness and Interpretability of Deep Neural Networks by Regularizing their Input Gradients , 2017, AAAI.

[15]  Frank Hutter,et al.  Decoupled Weight Decay Regularization , 2017, ICLR.

[16]  Samuel R. Bowman,et al.  A Broad-Coverage Challenge Corpus for Sentence Understanding through Inference , 2017, NAACL.

[17]  Sanjoy Dasgupta,et al.  Learning with Feature Feedback: from Theory to Practice , 2017, AISTATS.

[18]  Andrew Slavin Ross,et al.  Right for the Right Reasons: Training Differentiable Models by Constraining their Explanations , 2017, IJCAI.

[19]  Zeerak Waseem,et al.  Are You a Racist or Am I Seeing Things? Annotator Influence on Hate Speech Detection on Twitter , 2016, NLP+CSS@EMNLP.

[20]  Björn Ross,et al.  Measuring the Reliability of Hate Speech Annotations: The Case of the European Refugee Crisis , 2016, ArXiv.

[21]  Regina Barzilay,et al.  Rationalizing Neural Predictions , 2016, EMNLP.

[22]  Ye Zhang,et al.  Rationale-Augmented Convolutional Neural Networks for Text Classification , 2016, EMNLP.

[23]  Christopher Potts,et al.  Recursive Deep Models for Semantic Compositionality Over a Sentiment Treebank , 2013, EMNLP.

[24]  Christine D. Piatko,et al.  Using “Annotator Rationales” to Improve Machine Learning for Text Categorization , 2007, NAACL.

[25]  Andrew McCallum,et al.  Conditional Random Fields: Probabilistic Models for Segmenting and Labeling Sequence Data , 2001, ICML.

[26]  Marek Rei,et al.  Natural Language Inference with a Human Touch: Using Human Explanations to Guide Model Attention , 2021, ArXiv.

[27]  Ming-Wei Chang,et al.  BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding , 2019, NAACL.

[28]  Sanjoy Dasgupta,et al.  Learning from discriminative feature feedback , 2018, NeurIPS.

[29]  Preslav Nakov,et al.  SemEval-2017 Task 4: Sentiment Analysis in Twitter , 2017, *SEMEVAL.