WT5?! Training Text-to-Text Models to Explain their Predictions

TL;DR

This paper introduces a method to train text-to-text models to generate natural language explanations alongside predictions, achieving state-of-the-art explainability results without altering training procedures.

Contribution

It demonstrates a simple approach to enhance model interpretability by training models to produce explanations without changing existing training or decoding methods.

Findings

Achieves state-of-the-art explainability benchmark results

Enables learning from limited labeled explanations

Allows transfer of rationalization abilities across datasets

Abstract

Neural networks have recently achieved human-level performance on various challenging natural language processing (NLP) tasks, but it is notoriously difficult to understand why a neural network produced a particular prediction. In this paper, we leverage the text-to-text framework proposed by Raffel et al.(2019) to train language models to output a natural text explanation alongside their prediction. Crucially, this requires no modifications to the loss function or training and decoding procedures -- we simply train the model to output the explanation after generating the (natural text) prediction. We show that this approach not only obtains state-of-the-art results on explainability benchmarks, but also permits learning from a limited set of labeled explanations and transferring rationalization abilities across datasets. To facilitate reproducibility and future work, we release our code use to train the models.