Weakly- and Semi-supervised Evidence Extraction
This addresses the need for verifiable predictions in tasks like classification, but it is incremental as it builds on existing interpretability methods.
The paper tackles the problem of extracting supporting evidence for predictions when only a few evidence annotations are available, by combining few strong semi-supervision annotations with abundant weak supervision from document-level labels. The result shows that their methods outperform baselines, achieving substantial gains with as few as a hundred evidence annotations.
For many prediction tasks, stakeholders desire not only predictions but also supporting evidence that a human can use to verify its correctness. However, in practice, additional annotations marking supporting evidence may only be available for a minority of training examples (if available at all). In this paper, we propose new methods to combine few evidence annotations (strong semi-supervision) with abundant document-level labels (weak supervision) for the task of evidence extraction. Evaluating on two classification tasks that feature evidence annotations, we find that our methods outperform baselines adapted from the interpretability literature to our task. Our approach yields substantial gains with as few as hundred evidence annotations. Code and datasets to reproduce our work are available at https://github.com/danishpruthi/evidence-extraction.