Transformers Lack Explicit Representation for Linguistic Phase and Semantic Plasticity

Language involves a second parameter, phase, which determines how coactivated meanings combine or reverse contributions. Standard transformers lack explicit representations for this phase during frozen inference, as population-averaged training data deletes the history-indexed phase necessary for individual semantic interpretation.