← Playground

Playground · Constructed language

Information Machines

What would a language look like if the way information enters, connects, changes, and leaves a conversation were written directly into its grammar?

Information Machines is a small constructed language I build on the side. Most of my research approaches meaning from the outside — with analysis, diagnostics, and evidence; this is the inside-out version. Rather than describe how some existing language handles information, the project designs one whose grammar is about nothing else, and watches what breaks. It doesn’t have to serve the research or defend a claim — it’s an object about language that can actually be run.

The operators

The first version doesn’t need a world, a vocabulary, or a sound system. It needs a handful of moves for changing an information state, and rules for combining them. These are the early candidates — the grammar’s verbs, so to speak:

ADD Put a new piece of information into the current state.
LINK Relate it to what came before — cause, explanation, contrast, example.
DERIVE Mark that this follows from information already present.
REVISE Change something that was asserted earlier.
RETRACT Withdraw a previous commitment.
QUERY Point at a gap and ask for it to be filled.
PRESERVE Deliberately keep several readings open, without choosing yet.

A second axis is where information comes from — witnessed, reported, inferred, remembered, hypothesized, imagined. Whether the grammar should force you to mark the source, the way some natural languages do, is exactly the kind of question the project exists to play with. Nothing is decided in advance.

Unit tests for a language

The project doesn’t depend on anyone else adopting it, so it needs its own way of making progress. It borrows one from software: every version moves through the same loop.

  1. Add or change one rule.
  2. Use it to express a fixed set of texts.
  3. Write down what can’t be said, comes out clumsy, or turns unexpectedly ambiguous.
  4. Revise the design — keeping the old version, and the reason for the change.
  5. Re-run the old tests: did the change break something that used to work?

Over time that produces the real artifacts — a grammar, a growing dictionary, a corpus of test sentences, a changelog, a graveyard of abandoned rules, and the occasional short text written in the language itself.

The test suite

The tests are just things people routinely do with language. A rule earns its place only if the language can still say all of these afterward:

  • Describe something that happened yesterday.
  • Pass along a rumor you can’t confirm.
  • Correct something you just said wrong.
  • Write a recipe, or a set of instructions.
  • Explain why something happened.
  • Describe a dream.
  • Give a stranger directions.
  • State a rule, or a prohibition.
  • Tell a joke that turns on an ambiguity.
  • Say “I don’t know, but I know that you know.”
  • Describe something that has no name yet.
  • Compress a complicated story into three sentences.

Where it stands

It’s early. Right now Information Machines is a design and a method more than a finished language: there’s no fixed lexicon, phonology, or writing system, and there’s nothing here to report as a result. The first real move is small — pick one operator, write three rules and ten example sentences, and test them against a single real conversation. The point was never to finish. It’s to keep an object about language that can grow, fail, and be revised on the record.

Rules that turn out to be fun to discover, rather than just to describe, get spun off into RELAY — a puzzle game built on the same idea, where you reverse-engineer a machine’s language instead of designing your own.

← Back to the Playground