Unsupervised sentence representations are dense numerical vectors that capture the semantic meaning of entire sentences without relying on human-annotated labels or task-specific supervision. In natural language processing, these embeddings are typically derived from pre-trained language models trained on raw, unlabelled text using self-supervised techniques such as contrastive learning or autoencoding. By learning to position semantically similar sentences close together and disperse unrelated sentences across a shared vector space, unsupervised sentence representations enable downstream applications like semantic search, text clustering, and semantic similarity assessment using standard distance metrics such as cosine similarity.
