Word embedding
Also: word vectors, word2vec, GloVe, static embeddings
A learned lookup table mapping each vocabulary word to a dense vector of typically 50–300 dimensions. Methods such as word2vec (skip-gram, CBOW) and GloVe train these from co-occurrence, following the distributional hypothesis; each word gets one vector regardless of context.
Example
Trained word vectors support analogy arithmetic: vec("king") − vec("man") + vec("woman") lands closest to vec("queen").