Sparse coding for discourse atoms

Classical (pre-neural-SAE) k-SVD-style sparse dictionary learning applied directly to a trained word-embedding matrix, recovering a small set of overcomplete 'atoms of discourse' such that each word vector is approximately a sparse linear combination of a few atoms.