Rationalizing Text Matching: Learning Sparse Alignments via Optimal Transport
This work addresses the need for self-explaining models in text matching, offering a method to improve interpretability without alignment annotations, though it is incremental in extending rationalization techniques to alignment tasks.
The paper tackles the problem of generating interpretable alignments for text matching by introducing constrained optimal transport variants that produce sparse alignments, achieving high sparsity and fidelity while maintaining prediction accuracy on multiple datasets.
Selecting input features of top relevance has become a popular method for building self-explaining models. In this work, we extend this selective rationalization approach to text matching, where the goal is to jointly select and align text pieces, such as tokens or sentences, as a justification for the downstream prediction. Our approach employs optimal transport (OT) to find a minimal cost alignment between the inputs. However, directly applying OT often produces dense and therefore uninterpretable alignments. To overcome this limitation, we introduce novel constrained variants of the OT problem that result in highly sparse alignments with controllable sparsity. Our model is end-to-end differentiable using the Sinkhorn algorithm for OT and can be trained without any alignment annotations. We evaluate our model on the StackExchange, MultiNews, e-SNLI, and MultiRC datasets. Our model achieves very sparse rationale selections with high fidelity while preserving prediction accuracy compared to strong attention baseline models.