Two Missing Primitives in Contemporary Language Models: Strict Semantic Dominance and Revocable Semantic Dominance

This paper extends prior work on semantic instability in large language models by introducing controlled behavioral tests that isolate how semantic authority is assigned and withdrawn under ambiguity. Using a fixed set of polysemous passages, we compare model behavior under two conditions: strict semantic dominance, in which a single interpretation must be preserved globally, and revocable semantic dominance, in which interpretation may change with local context. Across multiple frontier models (GPT 5.2, Claude Sonnet 4.5, and Grok 4.1 Beta), strict dominance consistently produces hallucinated or distorted outputs when local context conflicts with the imposed meaning, while revocable dominance restores coherent interpretation without drift. The contrast demonstrates that contemporary models can execute meaning prioritization and revocation when explicitly instructed, but do not autonomously generate or govern these controls. The findings support the conclusion that semantic prioritization and revocation function as missing primitives in current language model architectures rather than as emergent, self-regulating processes.