The Grow-Speech Rule

In the constrained span, each visible word must be legal at the point where it appears.

At the start, a visible word is legal if its spoken English form has one beat in the chosen pronunciation source. A core word with two or more beats becomes legal only after it has been defined with legal words.

A definition may use the new word only as the marked headword, as in “By define I mean to set down what a word will mean from then on.” The right-hand side of the definition must use only words that were legal before the definition began. The new word becomes legal after the defining sentence ends.

Details

• “Word” definition: A styled or hyphenated term may be introduced as one marked headword, as with “Grow-Speech” itself. Otherwise, hyphenated forms are read by their spoken parts.

  • Punctuation and markup syntax (eg. <em>) do not count as words.

  • Proper nouns, acronyms, link anchor text, code names, and foreign words receive no special pass; they count by their spoken English form or must be defined.

• Inflection versus derivation: A core word is the ordinary dictionary headword behind a word form.

  Grow-Speech constrains core words, not every surface form So inflected forms are legal if their core word is legal. Derived words and fixed compounds are new core words unless the writer has already grown them or unless the compound is merely a transparent phrase made of legal words. A legal word also licenses its ordinary grammatical forms for any later use, including in definitions: plural, possessive, tense, aspect, person, participle, and comparison forms.

  If you have defined “write”, you may then use “writes”, “wrote”, “written”, “writing” etc.; but if you have defined “nation”, you have not yet defined words such as “national”, “international”, “nationalism”. In close cases, use a normal dictionary headword test: if the form is merely listed under the same headword, it is the same core word; if it has its own headword or technical sense, treat it as a new word.

• Definition form: Introduce a multi-syllable word with a sentence like “When I say X, I mean …” or “Let X stand for …”—note that the framing words (“when”, “I”, “say”, “mean”, “is”, “let”, “stand”, “for”) are all monosyllabic, which is why this template works.

  Write the new word in bold. The right-hand side must use only words already permitted at that point. Once defined, the word stays available for the rest of the response. The headword may appear as the marked word being defined. The right-hand side of the definition must not use the headword unless it was already legal before the definition began. The new word becomes legal only after the defining sentence ends. The sole free use inside a definition is the marked headword itself.

  • Self-reference trap: The word “syllable” is itself 3 syllables, so a response that tries to “state” the rule has to first define it (eg. as “beat” or “the pulse of a word”). And likewise “monosyllable”, “definition”, “vocabulary”.

    An LLM that pattern-matches to “explaining the rules” without first growing the words to do so will fail on its own first sentence.

• Markup format: Grow-Speech applies to all text visible to the reader inside the constrained target range.

  Thus, markup punctuation is ignored. Machine-readable strings—URL targets, filenames, code-block contents—are also exempt as non-prose. But reader-visible text counts: headings, link anchor text or image captions all obey the rule unless explicitly marked inert.

• Proper nouns: A proper noun may be defined like any other word: “By Paris I mean the town in France.”

  A name made of several words is legal only if each word is legal or the full name is introduced as a marked term.

• Quoted matter: Quoted text may be marked as inert.

  Inert quoted text need not obey Grow-Speech, but none of its words become legal, and no definitions inside it count. The prose that introduces, frames, explains, or comments on the quote must still obey the rule.

  Quotation marks alone do not make text inert. A live prose cue such as “he said”, “she wrote”, “this raw name”, “this sign says”, or “this code” can mark the next quoted span as inert, if context makes clear that the span is outside text. (Even if the quoted text itself defines a word, that doesn’t count until the author defines it normally.) This is based on context; for a more rigorous treatment of raw text, you can use the Grow-Speech scansion DSL to mark a range as raw.

• Syllable counting: By spoken English pronunciation, not by spelling. Details:

  • Syllable examples: “through” = 1, “scythe” = 1, “strengths” = 1, “rhythm” = 2

  • Contractions count their spoken form: “don’t” = 1, “I’m” = 1, “wouldn’t” = 2, “doesn’t” = 2.

    Don’t smuggle in illegal words via contraction. Common contraction traps: “doesn’t” (2), “wouldn’t” (2), “couldn’t’ve” (3), “shouldn’t’ve” (3). When in doubt, expand to monosyllables: “did not”, “would not”, “could not have”.

  • Compound and hyphenated words sum their parts: “blackboard” = 2, “well-known” = 2.

    A hyphenated form is treated as its spoken form; when in doubt, count the parts.

  • Acronyms read as letters cost 1–3 syllable per letter (“LLM” = 3, “FBI” = 3); acronyms read as words count by pronunciation (“NASA” = 2, “RAM” = 1).

    Acronyms read as individual spoken letters must count as the sum of the syllable counts of the spoken letter names. Thus “LLM” is 3 beats (‘L-L-M’) and “WWW” is not 3 but 9, but ‘RAM’ (pronounced ‘ram’) is just 1.

  • Numerals: digits and digit strings and Roman numerals are exempt, the same way URL targets and code spans are exempt.

    So 1998, Page 47, v3.2.1, c. 1300 BC all pass without growth. If a number is spelled out in words (nineteen ninety-eight, forty-seven), the words count by their spoken form: one/two/three/eight/ten/twelve = 1, seven = 2, eleven = 3, etc. Prefer digits for years, dates, page references, versions, and citations; reach for word form only when the number is a true lexical part of the sentence (a third of the time, four years ago). Because spelled out words are clunky compared to digits, we do not exempt them from definition. (An author who wants them for some reason, such as esthetics, can make an explicit exception here.)

  • Proper nouns get no free pass—either they fit (“Bach”, “York”, “Smith”, “Marx”) or they must be defined.

    • Citation exception: we exempt surnames in formal citations to avoid an explosion of biographical details (which may be quite burdensome for an author to even learn).

      Thus, it is fine to say “Schmidhuber 1991” without any definition, but if one talks about “Schmidhuber”, they should define him (“Schmidhuber is a Swiss AI researcher…”).

  • Foreign words count by their English pronunciation.

  • Ambiguity favors the author: for example, words with variable syllabification (“fire”, “hour”, “real”) count as one syllable if a current monosyllabic pronunciation exists in standard speech.

  • Checking: If still in doubt, you may consult the CMU Pronouncing Dictionary (cmudict).

    If a word has a variant with 1 beat in cmudict, it is legal as 1-beat. If all variants have 2+ beats, it requires definition.

• Recursive consequences: A definition must not use the headword in its right-hand side unless the headword was already legal.

  After the defining sentence ends, the new word may of course be used in later prose, including later explanations that state recursive rules.

Process Discipline

1. Before emitting any multi-syllable word, scan the prior text to confirm it was defined.

2. If not, define it first or paraphrase it away.

   When asked to write about X in Grow-Speech, do not immediately try to define X. First, ask: can I say this with monosyllables? If yes, do so. If no, can I approximate? Only if the concept is central and recurring should you invest the syllables to define it.

   The whole point of the form is to feel the cost of each abstraction, so when in doubt, paraphrase rather than define—only define a term when you’ll reuse it enough to earn its weight. (Heuristic: prefer paraphrasing for any word used <4×.)

3. You may want to write multiple drafts sequentially to clean up definitions.

• In author mode, the author’s dialectal English one-beat pronunciation should be used.

• In audit mode, a word is legal only if the chosen pronunciation source gives a one-beat form, or if the word has already been defined.

  To audit: Before you give the final text, check each word from left to right. Keep a list of grown words. When you find a word with more than one beat, either show that it was grown before this point, define it before first use, or replace it with short words.

Example

Worked opening hypothetical:

What I do here is a word stunt. I call this kind of word stunt: Grow-Speech. It was done first by a man, Guy L. Steele, in his talk where he said he was “Growing a Language”. (That is a raw quote; its words do not grow my word list.) By language I mean the words and ways by which we speak. He would speak and show us how he could grow a language with new words, right then and there.

How did Steele write Grow-Speech? Like this!

Each word I write must be a word of one beat. By syllable I mean the beat in a word. By define I mean to set down what a word will mean from then on. The one law: I may use a word with more than one syllable if I define it first. Once I define it, it is mine to use from then on.

So now I can tell you that Grow-Speech is a language which defines all words using one-syllable words or words defined in the past.

After that opening the writer has earned “Grow-Speech”, “language”, “syllable”, and “define” (and, by the same trick, can earn “primitive”, “abstraction”, etc. one at a time). That ratchet is the whole game.

Scansion

To help write a proper Grow-Speech, we can add hidden markup using HTML comments to encode a micro-DSL, in the style of my poetry metadata. Comments do not have to comply with Grow-Speech nor do they count as new definitions (they are only for audit), and may include additional commentary or summary or scaffolding. To write a properly marked up passage, I suggest an iterative process of a first draft writing in “draft mode”, freely, and then rewrite a second draft with full scansion added and corrections made, and then triple-checking it with a third final clean draft.

A Grow-Speech scansion header block should define the used format, and list any exceptions or changed rules. Most authors would use Grow-Speech (standard), but an author might choose to use many spelled-out numbers: Grow-Speech (exception: all spelled-out number legal). It should list all new words in a block at front with their syllable count and definitions, in order of definition; and likewise for “raw” text. Finally, it should explicitly list any errors, exceptions, violations, ambiguous, or miscellaneous notes, particularly for dialectical words where an American or British reader might be surprised.

The Grow-Speech as a whole should be surrounded by comment tags, to allow switching. Whether definitions should persist between switching is left up to the author of a document; they may want each Grow-Speech to be independent, or to accumulate. By default, definitions do not persist across separate BEGIN Grow-Speech/END Grow-Speech spans. If an author wants cumulative growth across spans, mark this in the header: PERSIST: yes; otherwise, the default is PERSIST: no.

Inside the Grow-Speech text proper, each defined word should be marked inline as a new numbered definition at the defining use (and only that first use), and raw quotes with illegal words should be marked inline as a new numbered raw quote. If it is unclear what the scope of a newly defined term or a raw quote would be, it should be surrounded with an opening and closing comment.

<!-- Grow-Speech (standard)
PERSIST: no

DEFINITIONS:
1. Grow-Speech (2): this kind of word stunt
2. language (2): the words and ways by which we speak
3. syllable (3): the beat in a word
4. define (2): to set down what a word will mean from then on

RAW:
1. title from Steele: “Growing a Language” ('growing' is legal, but not yet 'language')

AMBIGUOUS/EXCEPTIONS:
N/A
-->

<!-- BEGIN Grow-Speech -->
What I do here is a word stunt.
I call this kind of word stunt: <!-- DEFINE:1 --> **Grow-Speech**<!-- /DEFINE:1 -->.
It was done first by a man, Guy L. Steele, in his talk
where he said he was <!-- RAW:1 -->“Growing a Language”<!-- /RAW:1 -->.
(That is a raw quote; its words do not grow my word list.)
By <!-- DEFINE:2 --> **language** I mean the words and ways by which we speak.
He would speak and show us how he could grow a language with new words,
right then and there.

How did Steele write Grow-Speech? Like this!

Each word I write must be a word of one beat.
By <!-- DEFINE:3 --> **syllable** I mean the beat in a word.
By <!-- DEFINE:4 --> **define** I mean to set down what
a word will mean from then on.
The one law: I may use a word with more than one syllable if I define it first.
Once I define it, it is mine to use from then on.

So now I can tell you that Grow-Speech is a language which
defines all words using one-syllable words or words defined in the past.
<!-- END Grow-Speech -->

Appendix