WORK IN PROGRESS — 2026
Reading Machine
Interpretation is the production rule.
Geometry is the phenotype.
A reading machine. Its rules are borrowed from L-systems — a rewriting system biology once used to simulate how plants grow.
It does not parse. It reads a text, derives a grammar from the text's own letter transitions, then lets that grammar rewrite itself. Each time it reads, it grows a new set of shapes, and a new set of explanations.
THE MACHINE
HOW IT READS
01 — Grammar extraction
The seed text is scanned for character transitions. For each character the machine keeps its three most frequent successors, weighted by count. One symbol therefore carries several competing productions — a stochastic grammar, not a single rule. The text writes its own probabilities.
02 — Rewriting
Starting from the text's most frequent character as axiom, productions are applied in parallel. At each application a rule is sampled from its distribution. VARIANCE is the temperature: at 0 the machine always takes the likeliest reading, at 1 it reads loosely. SEED fixes which loose reading you get.
03 — Turtle interpretation
The final string is read left to right by a turtle. In Latin, vowels draw and consonants turn. Chinese has no vowel letters, so the machine uses the direct analogue: the characters carrying the first half of the text's token mass draw, the long tail turns. Segment length decays with depth.
CONTEXT
Lindenmayer's rewriting system was built to describe how a plant repeats itself into complexity. Language does the same — small rules, iterated, produce form.
The machine is deliberately transparent, against the black box. Every rule it uses is printed above; every generation of the string is visible; the geometry is a direct reading of the last line. Nothing is learned, nothing is hidden. Output is vector — it exports to SVG and goes to the riso press.