Painted in Words

A sentence that compiles to a picture

The line in the bench above — Gules, on a bend Argent three mullets Sable — is not a description of a shield. It is the shield, in the only form heraldry treats as authoritative. A coat of arms is legally and historically defined by its blazon: the words. The painting is a performance of the words, the way a score is performed. Two artists, five centuries and a thousand miles apart, working only from that one sentence, will produce arms that any herald counts as identical — same field, same charges, same tinctures, same arrangement — even if no two brushstrokes match.

That is a strange and specific kind of object. It is a controlled language: a small, rule-bound subset of a natural language (here, an Anglo-Norman-flavoured English) honed over centuries so that an expression has, for its content, essentially one reading. The grammar is rigid. Word order carries meaning — the field is named first, always; what sits on a charge is named after it; tincture follows the thing it colours. There is a controlled vocabulary of perhaps a few hundred terms. And from a well-formed sentence you can mechanically recover the arms — which is exactly what the bench does: it parses the words and draws one faithful version of them.

The two verbs say it outright. To blazon a shield is to put it into words; to emblazon it is to paint those words back into an image. Encode and decode. The bench is an emblazoner: a small, deterministic decoder for the part of the language that is geometric enough to render honestly in a browser.

Read the parse

Under the input, the bench shows its working — because a claim that the words determine the picture is only worth anything if you can watch it happen. Each token is coloured by the role the grammar gives it: field, division, ordinary, charge, number, connective. Below that is the sequence of drawing operations the sentence compiled to. Change a word and both change. Type something the parser cannot place and it is struck through, in red, and the shield refuses to invent it — the honest failure mode of a real decoder.

This is the whole conceit made operable. A herald reading Or, three chevrons Gules aloud is, functionally, calling a program: field gold; draw three red chevrons, stacked. The grammar's rigidity is not pedantry. It is what lets a description travel — across a battlefield, across a century, across a language — and arrive as the same image.

The rule of tincture

Heraldry's vocabulary of colour is the tinctures: two metals — Or (gold) and Argent (silver) — and the colours, chiefly Gules (red), Azure (blue), Sable (black), Vert (green) and Purpure (purple), plus the patterned furs. (Note the swatches under the bench: no shade is canonical — a medieval Gules and a modern one differ, and the convention is the name, not a hex value. The bench picks one defensible shade per name and says so.)

Governing them is one famous constraint, the rule of tincture: do not put metal on metal, nor colour on colour. A red charge may sit on a gold or silver field; it should not sit on a blue one. The reason is purely functional — contrast, legibility at a distance, on a banner in motion. The bench keeps a lamp lit for it: load real arms and it stays green; build a colour-on-colour shield and it warns.

But here the page must be careful, because the truth is a tension, not a slogan. Some authorities call the rule nearly inviolable; others note the exceptions are so many that calling it a rule overstates it. Both are quoted, verbatim and sourced, in the apparatus. The honest summary: it is a strong, near-universal convention with real, named exceptions — and the most celebrated exception proves it was a convention people knew they were breaking.

Load Jerusalem in the bench. Argent, a cross potent between four crosses Or: silver field, five gold crosses — metal on metal. It is among the most famous coats in Europe, and the traditional story is that the rule was broken on purpose, Jerusalem being too holy for ordinary constraint. (Honestly: whether such violations were deliberate statements or early mistakes later rationalised is disputed, and Jerusalem is the best-known breach but not the only one.) A coat that flouts the rule was given its own name — armes à enquérir, "arms to enquire about," whose very wrongness was meant to make you ask. The lamp goes amber, and the page tells you why instead of hiding it.

Where the language is honestly free

It would be a lie to claim a blazon fixes everything. It does not, and the slack is the interesting part. A blazon pins the content — which charges, what tinctures, how arranged, in what order — and leaves the style to the artist: the exact shape of the shield, the posture and ferocity of a lion, the number of points on an "engrailed" line, the precise proportion of a fess. Two emblazonments of one blazon can look strikingly different and both be correct. The language specifies an equivalence class of pictures, not a single picture.

So the bench is honest about its own scope. It renders the part of blazon that is genuinely geometric and determinate — fields, partitions, the ordinaries, and simple repeatable charges (roundels, mullets, lozenges, crosses, and their kin). It does not draw lions, eagles, or the menagerie of beasts and artefacts that fill the richest arms — those are the language's open lexicon, and rendering them faithfully is an art, not a parse. The grammar is the claim here; the beasts are vocabulary. Where the bench cannot draw a charge, it says so rather than faking it.

There is also genuine dialect. English blazon and French blazon order and abbreviate differently; default numbers and positions are conventions a reader supplies, not marks on the page; and across centuries the vocabulary drifted. A controlled language, yes — but a living, human one, with the ambiguities that implies. The marvel is not that it is perfectly formal. It is how close to formal a community of medieval heralds drove an ordinary language, and how long it has held.

A controlled language, six centuries early

The systematic language took shape with the rolls of arms — long parchment registers blazoning hundreds of coats in terse, patterned phrases. Scholars place the birth of "classic blazon" around 1250 (a best estimate, not a hard date), building on an armorial practice already a century old; by the time John Guillim's A Display of Heraldrie was printed in 1610, the vocabulary and word order were codified enough to teach from a book. What these heralds built fits, almost exactly, the modern definition of a controlled natural language: a restricted subset of an ordinary tongue — here Anglo-Norman-inflected English — with its grammar and vocabulary deliberately pared down to cut ambiguity. They were doing this six to seven centuries before there was any theory of such things: Chomsky's work on formal grammars is from 1956–57.

It would be a flourish too far to say the heralds "invented formal grammar" — they wrote no production rules and their conventions are, in practice, often bent. But the language they left is parseable enough that people have since written real parsers for it. DrawShield, Karl Wilcox's blazon engine, runs a Flex/Bison grammar that turns a British blazon into a syntax tree and draws it; a 2023 Edinburgh dissertation built a context-free grammar for blazon and measured exactly how ambiguous it remains; and a 2025 ACM paper, Type-Safe Blazon, gives it a type system that flags ill-formed and convention-breaking arms. Modern computer science keeps rediscovering that the medieval heralds had built it a working specification language to formalise.

It belongs with the other times a craft reached a deep idea early, pressed there by a practical need rather than a proof. Bell-ringers generated Gray codes in stone towers because a half-tonne bell can only swap with its neighbour. Heralds built a controlled language because an identity had to survive being shouted across a field and copied, faithfully, by a stranger who had never seen it. The need came first; the formalism we'd later name was the shape the need pressed the practice into.

The apparatus — every claim, sourced