Observer-Dependent Entropy Retrieval in Linguistic Comprehension: A Bounded-Access Theory of Semantic Timing, Reanalysis, and Collapse

Language comprehension theories can model prediction, processing cost, and interpretive outcome, but they do not yet formalize access: the point at which linguistic structure becomes retrievable to a bounded observer. This paper develops Observer-Dependent Entropy Retrieval in Linguistic Comprehension, the linguistic-domain formulation of ODER, as a bounded-access theory of semantic timing, reanalysis, and collapse. The framework separates availability, prediction, retrieval, and comprehension as distinct operations. Meaning becomes available when linguistic structure is retrieved and stabilized under finite attention, working memory, contextual familiarity, and interference constraints, not when a signal is merely present or statistically predictable. The framework constructs an observable linguistic retrieval manifold, defined by task-defined retrievable structure S_obs(τ) over observer-indexed retrieval time, and imposes reconstructibility: a law of access must be identifiable from the observable retrieval trajectory. S_obs(τ) does not denote total linguistic meaning, but the structure made retrievable under a specified task, observer, and measurement operator. Reconstructibility restricts admissible access laws to first-order, observable-only dynamics. Bounded access further forces gap form, where retrieval depends on the remaining stabilizable structure, S_max−S_obs(τ). Within the spectrally admissible smooth-access regime, the hyperbolic tangent is selected as the conditional extremal baseline, not as a universal empirical fit. Linguistic ODER derives a ranked measurement family for observer-indexed timing, retrieval rate divergence, temporal localization, reanalysis pressure, ambiguity interference, compression thresholds, and aggregation loss. Controlled linguistic and synthetic trace testbeds instantiate these dynamics under known structural conditions and identify the boundary of the constant rate smooth-access baseline. Prior constant-rate convergence results are treated here as benchmarked evidence for a smooth-access subset, not as a ceiling on the theory’s scope. ERP and reading-time analyses provide natural measurement surfaces for the theory’s access-layer timing predictions, while full saturation-specific validation of the tanh-regulated smooth-access law remains a separate empirical target. The framework supplies language processing with a formal access variable and a disciplined program for studying when meaning becomes available to bounded observers.