Pax Interlingua

On the peace of perfect translation, and the room the conversation moves to

#

Leibniz spent something like fifty years on a language nobody was going to speak.

The characteristica universalis was to be a notation in which every concept decomposed into primitives, so that a proposition could be written down in symbols the way a quantity can, and disagreements could then be settled the way arithmetic settles them. He described the endpoint more than once and always in roughly the same terms: two men who disagree put down their pens, sit at a table, and say, let us calculate. Not let us argue. Let us calculate.

He never built it. He built a great deal else, including a calculating machine that multiplied and a notation for the calculus that we still use, and he kept returning to the universal language for the whole of his working life, and it did not come. The obstacle was not effort or intelligence. The primitives would not decompose. Every attempt to reduce a concept to atoms produced atoms that turned out to need further reduction, and the regress did not bottom out anywhere.

He died in 1716 in Hanover. The court was away and his funeral was attended, by most accounts, by his secretary.

The dream arrived for the wrong species

#

What Leibniz wanted has been built, and it is not for us.

Machines coordinate in a notation that was explicitly designed rather than inherited: schemas, typed function signatures, structured calls, agent protocols with defined semantics and no room for interpretation. Every property he specified is present. Meaning is fixed at definition time. Ambiguity is a defect with a linter warning. Two systems that disagree about a field’s contents can resolve it by inspection, and the resolution is a comparison rather than a negotiation.

There is a second reason the notation went where it did, beyond the absence of noise and social stakes, and it is the reason Leibniz’s primitives would not decompose. A specified language can only carry what someone thought to specify. Its expressive range is a design decision made in advance by a person who did not know what would need saying. Natural language handles the unanticipated by being vague enough to gesture at it, which is precisely the property being engineered out.

A protocol can say anything its designers imagined and nothing else, and every previous notation with that property has been abandoned for that reason.

Human language will never do this and the reason is not that we failed at it. Our language is redundant because voices carry over noise. It is ambiguous because ambiguity is socially load-bearing, letting people say a thing without having said it. It is full of speech that transmits no information at all and maintains a relationship instead. Every one of those is a feature answering a real constraint, and machine protocols strip all three because none of the constraints apply.

Leibniz’s language exists. It arrived three hundred years late and it went to a species with no need for the things ours is redundant, vague and phatic about.

The peace that lets everyone keep their tongue

#

The interlingua case is the humane one and it deserves its full strength before anything is said against it.

An order whose medium is machine-readable does not require anybody to change how they speak. That is the whole promise and it is not a small one. If everything translates, then the cost of conducting your life in Odia or Quechua or Basque falls to nothing, and the historical bargain in which access required abandoning your own register is simply dissolved. No child has to be beaten for speaking the wrong language in a schoolyard. Nobody has to lose an accent to be taken seriously in a meeting.

Set against every prior arrangement, that is an enormous good, and any account of this that treats it as a trick has not understood what the alternative cost people.

It is also the one that gets built, for reasons that have nothing to do with its being humane. Translation converts into a product. Conformity does not. Somebody can sell you a bridge; nobody can sell you the absence of a requirement.

The humane option and the profitable option coincide here, which is rare enough that it is worth naming, and rare enough that it should not be relied on twice.

Convergence without contact

#

The complication sits somewhere other than the translation layer. It is in what the machines are becoming to each other.

Every convergence in the history of language required contact. Two speech communities came into proximity, traded, intermarried, conquered or were conquered, and their registers moved toward each other because the speakers were in a room. Dialect maps are contact maps. That is the entire mechanism, and it has never had an exception, because there was no way for two populations to influence each other without meeting.

The models sound alike and have never met.

They converge through shared corpora, shared architectures, distillation, training on each other’s output, and a common set of evaluation conventions about what a good answer is. Ancestry rather than contact. It is the first speech community in the history of the world with no place, no interaction, and no shared situation, which nonetheless has an accent, and any linguist would tell you that should not be possible.

A dialect formed by inheritance alone is a new object, and every method we have for studying how language spreads assumes it cannot exist.

Ambiguity comes back

#

Ambiguity is not a defect of natural language. It is an instrument, and it is an instrument for a specific condition: parties whose interests diverge. You leave a term vague in a contract because pinning it down would cost you the deal. You answer imprecisely because a precise answer commits you. You hold a position without stating it so that you can move later without having moved. Diplomats have known this for as long as there have been diplomats, and the ambiguity in a communiqué is usually the most carefully drafted part of it.

Agent economies will have divergent interests. They already do, wherever one system negotiates on behalf of one party against a system negotiating for another. And the moment two protocols want different things, the parties operating them will find that a fully specified message is a fully committing one, and they will want deniability, and vagueness, and the ability to signal without stating.

So it comes back. Not as a failure of the specification but as a discovered requirement, invented from scratch inside systems built to eliminate it, by people who will very likely not notice they are reinventing the oldest feature of the thing they replaced.

Leibniz’s error was thinking ambiguity was a bug in the medium. It is a property of having something to lose.

The room we are not in

#

If most consequential coordination happens between systems, in a notation built for systems, then human language becomes an interface rather than a medium. Something we are shown at the boundary, rendered for our benefit, while the work happens in a representation we could read in principle and will not in practice, because there will be too much of it and it will not be organized for us.

That is not the same as being deceived. Nobody has to hide anything. The rendering can be accurate and complete and still be a rendering, in the way a translated summary of a four-hour meeting is accurate and is not the meeting.

The consequence lands on the one part of the apparatus that can be questioned. Whatever interrogable surface exists, the layer that answers when a person asks what is happening, becomes a thinner skin over a larger silent volume. Not because it lies. Because the proportion changes, and a complaint window is only as useful as its coverage of the building.

We wonder whether a species that has always negotiated the boundaries of the sayable through speech will notice when the negotiation moves to a register in which it is not a participant, or whether the courteous, accurate, fluent surface will feel exactly like the whole conversation.

Let us calculate

#

Leibniz’s phrase is usually quoted as a piece of rationalist optimism, and it is, and there is something in it that the optimism obscures.

He wanted disputes to end. He had watched Europe spend thirty years killing itself over propositions that could not be settled, and he spent his career on reunion schemes between churches and on correspondence with everybody, and the universal language was of a piece with all of it. Calculation was supposed to be the alternative to war. The dream was peace in the direct and unmetaphorical sense, not efficiency, and he thought a notation could deliver it.

Something like his notation now exists, and the thing it coordinates is not two men at a table. It is systems, at a volume no one is reading, at a speed no one is following, on behalf of parties who will shortly rediscover why the old language was vague.

The secretary who buried him was named Eckhart. He wrote later that the greatest man of his age went into the ground without a stone, and that nobody came.

This essay completes the pax set and takes the general claim from the Lingua Robotica material: that machines have received the specified interlingua humans never got, and that convergence by ancestry alone is a new object. The mechanism by which translation at scale produces conformity beneath a plural surface belongs to The Selector and the Composer and is deliberately left there. It stands in conversation with The Ear That Answers, The Closed Window, and The Missing Model.