Linguistic Evaluability: Reconstructive Burden in Artificial Intelligence-Mediated Communication

Artificial intelligence increasingly generates, transforms, and recirculates the language through which judgments reach human evaluators. Yet communicative clarity does not guarantee that a representation carries enough of a judgment’s evaluative structure for another person to examine it. This paper develops linguistic evaluability as the degree to which a linguistic representation makes sufficient evaluative relationships recoverable for a particular form of independent examination. It argues that linguistic formulation allocates reconstructive burden between communicator and evaluator: attribution, evidence, criteria, context, and qualification may travel with a judgment or require recovery by whoever encounters it later.

The paper uses constrained reformulation as a philosophical stress test, with E-Prime providing one deliberately disruptive constraint. The analysis distinguishes grammatical resistance from reconstructive resistance and shows how reformulation can expose relational work hidden by otherwise successful communication. Linguistic evaluability remains scalar, relational, purpose-sensitive, and distinct from clarity, explicitness, argument quality, and epistemic compression. AI did not create the underlying problem, but it changes its stakes by enabling linguistic mediation at speeds, scales, and frequencies that human evaluative attention cannot match. As judgments travel farther and faster, the central question becomes not merely whether humans can understand what AI-mediated language tells them, but whether enough travels with those judgments for humans to examine what they are being asked to accept, use, and build upon.