Semiotic Density: From Ideographic Writing to Framework Injection — Why Two Words Outperform Fifty Instructions in Human-AI Interaction

Ask an AI model to generate "a selfie of a giraffe." No instruction specifies the camera angle, the framing, the expression, the savanna backdrop, or the comic tension of a non-human animal performing a quintessentially human act. Yet the model produces all of it. Two words carry more implicit semantic information than fifty words of explicit instruction could encode. This paper identifies the mechanism behind this phenomenon: semiotic density, the ratio of implicit semantic field activated by a term to its explicit surface form. We trace this principle across three timescales: ideographic writing systems (three millennia of density engineering), Chain of Density prompting (Adams et al., 2023), and Framework Injection (Gomes, 2026). We ground these observations in six theoretical frameworks: Frame Semantics (Fillmore, 1982), Prototype Theory (Rosch, 1975), Conceptual Blending (Fauconnier & Turner, 2002), Spreading Activation, Uniform Information Density (Frank & Jaeger, 2008), and Construction Grammar (Goldberg, 1995). We present empirical evidence from twelve experiments (n > 1,000 evaluations, 6 domains, 5 model providers). Key findings: (i) capacity-compensatory effect — FI benefit is largest for smaller models (d=1.92, Haiku 4.5) and smallest for larger models (d=0.69, Opus 4.6); (ii) frame dilution in multi-turn conversations — FI maintains advantage across 20 turns (d=0.121, p=0.017) following logarithmic decay (r=-0.78), with finance domain exception (d=0.630); (iii) coercion is robust to typographic noise but sensitive to register shift (10-12% reduction across 5/5 models). We formalize five Semiotic Density Engineering operations (Densify, Rarefy, Rotate, Subtract, Blend) and derive new design principles for human-AI interaction. Part of the Artisanal Intelligence Program.