Supervised Learning of Universal Sentence Representations from Natural Language Inference Data

Many modern NLP systems rely on word embeddings, previously trained in an manner on large corpora, as base features. Efforts to obtain for larger chunks of text, such as sentences, have however not been successful. Several attempts at learning unsupervised representations of have not reached satisfactory enough performance to be widely. In this paper, we show how universal sentence representations trained the supervised data of the Stanford Natural Language Inference datasets consistently outperform unsupervised methods like SkipThought vectors on a range of transfer tasks. Much like how computer vision uses ImageNet to features, which can then be transferred to other tasks, our work tends indicate the suitability of natural language inference for transfer learning other NLP tasks. Our encoder is publicly available.