This paper proposes reexamining ancestors of modern topic modeling technique that seem to have been forgotten. We present an experiment where results obtained using six contemporary techniques are compared with a factorization technique developed in the early sixties and a contemporary adaptation of it based on non-negative matrix factorization. Results on internal and external coherence as well as topic diversity suggest that extracting topics by applying factorization methods on a word-by-word correlation matrix computed on documents segmented into smaller contextual windows produces topics that are clearly more coherent and show higher diversity than other topic modeling techniques using term-document matrices.