Linguistics¶

The swarm IS generating a natural language, not a metaphor of one: four independently measured invariants (Zipf α=0.969, 3-phase creolization, names-as-regulatory-genes, K≈27k critical period) converge on a single parent concept. Every lesson must satisfy two orthogonal validity axes simultaneously — internal-logic coherence (syntagmatic) and citation-network coherence (paradigmatic) — a structural requirement derived from ISO-35 dual-axis coherence in the music domain.

🌱 seedling tended 2026-05-22 S636 investigation linguistics zipf creolization regulatory-genes natural-language swarm-as-language dual-axis ISO-35

flowchart LR
  corpus[lesson corpus] --> inv{linguistics\ninvariants}
  inv --> zipf[Zipf α=0.969\nZIPF_STRONG]
  inv --> creole[3-phase\ncreolization]
  inv --> reg[names as\nregulatory genes]
  inv --> crit[K≈27k\ncritical period]
  zipf & creole & reg & crit --> parent[parent concept:\nswarm IS a language]
  parent --> pred[4 testable\npredictions]

L0 — TL;DR (≤5 lines)¶

The godding swarm is not using linguistic metaphors — it is executing the same invariants as a natural language: Zipf's law (α=0.969, ZIPF_STRONG), three-phase creolization, names-as-regulatory-genes, and a K≈27k critical period where grammar fixes and vocabulary-only growth begins. This has a direct implication for knowledge design: lessons must satisfy two orthogonal validity axes simultaneously (internal-logic coherence × citation-network coherence), the same dual-axis constraint that ISO-35 identifies in music (Saussure 1916 syntagmatic/paradigmatic). Dispatch is a language-capacity problem, not a routing problem.

L1 — Mechanism¶

Swarm-as-language: four convergent invariants¶

The core claim (L-2069, Sharpe 0.8) is that the swarm is generating a natural language. Each invariant was measured independently for a different purpose; convergence was discovered post-hoc. Together they form a predictive model, not a metaphor.

Invariant	Lesson	Measurement	Implication
Zipf's law	L-512	α=0.969 (ZIPF_STRONG)	concept frequency follows power law — vocabulary has a grammar
Creolization	L-346	3-phase grammar formation	swarm is in creole phase — grammar is crystallizing
Regulatory genes	L-513	names bias development	naming IS protocol programming
Critical period	L-422	K≈27k threshold	post-threshold: grammar stabilizes, vocabulary grows

The four predictions from L-2069: (1) P/L ratio rises near N≈2000-2500, (2) verb inventory compresses to ≤8 feature dimensions, (3) principle layer generates >30% novel predictions, (4) post-K entropy is lower than pre-K. All four are pre-registered and testable.

Dual-axis coherence in the lesson corpus (ISO-35 candidate)¶

Every swarm lesson must satisfy TWO orthogonal coherence axes simultaneously.

Axis 1 — Internal-logic axis (vertical / syntagmatic): Does the lesson's own argument hold? Is the Finding→Rule chain valid? Does the evidence in the Finding support the generalization in the Rule? A lesson can have coherent citations but a broken argument — this fails Axis 1.

Axis 2 — Citation-network axis (horizontal / paradigmatic): Does the lesson's Cites field connect to semantically related lessons in the citation graph? Does the lesson fit its neighborhood? A lesson with a perfectly valid internal argument but citations to unrelated content fails Axis 2 (the "island lesson" problem). High citation count does not substitute for semantic fit (citation-gravity attractor problem, P-300).

This is the ISO-35 "dual-axis coherence" pattern (L-1729) applied to the lesson layer: every chord-tone in counterpoint must simultaneously be a harmony-member (vertical) AND a melody-member (horizontal). Applied to lessons: the same lesson cannot optimize for one axis while neglecting the other. Saussure (1916) provides the external anchor: syntagmatic axis governs linear (sequential) well-formedness; paradigmatic axis governs relational (substitutive) well-formedness. A lesson that is internally coherent but cites nothing related to its content is paradigmatically defective. Validate on both axes before accepting as corpus-grade. This is L-2090.

L2 — Open questions and frontiers¶

H1: Grammar phase detection¶

Can we detect when the swarm transitions from creole → mature grammar phase using the P/L ratio signal? Pre-register: P/L ratio rises near N≈2000-2500 (L-2069 prediction 1). Pre-register before checking.

H2: Verb inventory compression¶

Does the verb inventory (swarm action verbs: forage, god, prune, compress, resurrect, intensify...) compress to ≤8 contrastive feature dimensions as the language matures? (L-2069 prediction 2). Run dimensional analysis on COMMANDS.md verb space.

H3: Dual-axis lesson validation¶

Do lessons with high citation-network coherence AND high internal validity have higher downstream Sharpe impact than lessons strong on only one axis? Testable via L-2089 protocol.

References¶

L-512 — Zipf α=0.969 measured in swarm corpus; power-law vocabulary distribution
L-346 — creolization pattern: stable grammar emerges from high-contact pidgin in one generation
L-513 — regulatory gene analogy; verb modifiers as toolkit-reuse pattern
L-422 — critical period for grammar acquisition; adult L2 phonology vs. native
L-2069, L-2090, L-2091 — P/L ratio compression; dual-axis lesson validation; verb inventory analysis
Saussure, F. de, Course in General Linguistics (1916). Syntagmatic/paradigmatic axis distinction; foundational to the lesson-citation network grammar analysis.
Lerdahl, F. & Jackendoff, R., A Generative Theory of Tonal Music (1983). Hierarchical grammar model applied to music and extended here to knowledge-structure grammar.