Separating Reasoning from Interface: A Proposed Architecture for Large Semantic Models

This paper proposes an architecture that separates reasoning from language interface in AI systems, introducing the Large Semantic Model (LSM) and Language Translation Model (LTM) as distinct components. The LSM reasons in a compact semantic protocol while LTMs handle bidirectional conversion between the protocol and natural language, code, and other formal systems. The architecture argues for measurable gains in parameter efficiency, token efficiency, and system modularity. Published by independent researcher Aser Nasr, age 13, March 2026.