Instance-Based Learning of Span Representations: A Case Study through Named Entity Recognition
This addresses the need for interpretable rationales in practical applications like named entity recognition, though it is incremental as it builds on existing instance-based approaches.
The authors tackled the problem of creating interpretable models for structured prediction by developing an instance-based learning method that classifies spans based on similarities to training examples, achieving high interpretability without performance loss in named entity recognition.
Interpretable rationales for model predictions play a critical role in practical applications. In this study, we develop models possessing interpretable inference process for structured prediction. Specifically, we present a method of instance-based learning that learns similarities between spans. At inference time, each span is assigned a class label based on its similar spans in the training set, where it is easy to understand how much each training instance contributes to the predictions. Through empirical analysis on named entity recognition, we demonstrate that our method enables to build models that have high interpretability without sacrificing performance.