LSA uses singular value decomposition (SVD), which is a mathematical generalization similar to factor analysis

Three matrices:
1) orthogonal factor value 2) original column entities 3) diagonal scaling values

A perfect overlap of all the sentences produces LSA= 1.0 (Nestojko, 2014)

LSA is practical only when computational power and algorithm efficacy are advanced enough to support SVD that contains thousands-of-words-by-thousands-of-contexts metrics (equivalent to adult’s language exposure)

LSA is not a fixed mechanism but an evolving approximation (Landauer, 1998)