Related to this is factor analysis, the technique predominantly used in psychology to extract meaningful factors (analogue of components in principal component analysis).

Unlike PCA, it assumes meaningful "latent" factors (such as "royalty" above) and tries to find a rotation which best loads these onto the data. To achieve this, it doesn't attempt to encode the data perfectly but leaves room for error in the reduction to factors.