In 1968, Knuth showed that combining formal structure with natural language communicates algorithms better than either alone. Pseudocode became standard practice. This paper generalizes the insight. Knowledge representation faces the same forced choice today: formal systems (OWL, RDF) that lose meaning, or natural language that loses structure. The paper argues the generalization was blocked by a single condition: no reader existed that could hold richer formal systems (modal logic, type theory, probability theory, lambda calculus, relational algebra) alongside multilingual natural language simultaneously. AI systems (c. 2024) satisfy that condition. Companion paper realizing Leibniz's characteristica universalis: https://doi.org/10.5281/zenodo.18733511 Spec (AI-facing): https://lingenic.ai Human-facing: https://lingenic.com