conceptnet

This paper explores interpretable Natural Language Inference (NLI) using graph-based representations of atomic propositions, introducing a pipeline that achieves 89.7% accuracy on SNLI with a trade-off in performance for interpretability.