Possibilities of Neural Natural Language Inference Models

Abstract

Natural language inference (NLI) is one of the fundamental tasks for natural language understanding. As with other NLP tasks, recent studies show the remarkable impact of incorporating deep neural networks in NLI. However, it remains unclear to what extent such DNN-based models are capable of learning the systematicity underlying NLI from given labeled training instances. In this paper, we investigate the capability of recent DNN-based NLI models in learning the inferential systematicity. Experiments showed that the generalization ability of current neural models is limited to the case where the syntactic structures are nearly the same as those in the training set.