We investigate how explicit feature decorrelation affects monosemantic representations in language models. The study disentangles when monosemanticity should be encouraged or suppressed to support downstream interpretability objectives.