Universal Translation by Constraint Closure: An Auditable Architecture for Grounded Decipherment, Active Semantic Learning, and Constrained AI Completion

Abstract

This paper presents Universal Translation by Constraint Closure (UTCC), a general architecture for deciphering initially unknown communication systems without assuming a pre-existing dictionary, grammar, or fixed semantic mapping. UTCC treats translation as a problem of constrained scientific inference: competing interpretations are generated, tested against grounded evidence, actively discriminated through informative interventions, and retained only when warranted by diagnostic closure. Unresolved meanings remain explicitly undefined rather than being completed by guesswork, while generative AI may assist communication without contaminating the evidential record. The framework is derived from the Necessity Method and combines cultural immersion, active probing, provenance control, reverse generation, and continual learning. Proof-of-principle experiments include exact synthetic decipherment and partial real-language decipherment in structured semantic domains using independently selected Japanese and Turkish data. The results support UTCC as a reproducible methodology for progressively learning unknown communication systems while preserving uncertainty and evidential accountability.