Universal Sentence Representations Learning with Conditional Masked Language Model

  • ,
  • Yinfei Yang ,
  • Daniel Cer ,
  • Jax Law ,
  • Eric Darve

EMNLP 2021 |

This paper presents a novel training method, Conditional Masked Language Modeling (CMLM), to effectively learn sentence representations on large scale unlabeled corpora. CMLM integrates sentence representation learning into MLM training by conditioning on the encoded vectors of adjacent sentences. Our English CMLM model achieves state-of-the-art performance on SentEval, even outperforming models learned using (semi-)supervised signals. As a fully unsupervised learning method, CMLM can be conveniently extended to a broad range of languages and domains. We find that a multilingual CMLM model co-trained with bitext retrieval~(BR) and natural language inference~(NLI) tasks outperforms the previous state-of-the-art multilingual models by a large margin. We explore the same language bias of the learned representations, and propose a principle component based approach to remove the language identifying information from the representation while still retaining sentence semantics.