The Pivot — Summary

Nasrin Zargar has interpreted in immigration court in Newark for nineteen years. She keeps a spiral notebook beside the microphone, and in the right margin she writes notes that never enter the record: he softened that, she is addressing the judge and not me, this is a form he would only use to a stranger. The transcript gets the sentences. The margin gets everything the sentences were doing.

On a Tuesday in March she interprets for Sohrab Nazari, who is describing what happened to his brother. He speaks in Dari, and twice he uses the form that marks a thing he did not see himself but was told. In the transcript this becomes my brother was taken. Nasrin writes in the margin: he is saying he did not witness it. Nobody reads the margin. The adjudicator will later note that the account is inconsistent, because in an earlier interview the applicant appeared to claim direct knowledge.

In 1968 Bernard Vauquois drew a triangle. The base is direct translation, word against word; the middle is transfer; the apex is interlingua, where the source is analyzed up into a representation belonging to no language and the target generated from it. Everyone understood the apex was the prize, because a hundred languages need a hundred pairs of components rather than nearly ten thousand. Decades went into specifying the middle, and every designed interlingua failed. The projects ran, demonstrated on constrained domains, and did not generalize, and by the late nineties the field had moved to statistical methods that ignored meaning entirely. Then a multilingual model was trained on many languages at once and something that behaves like an interlingua appeared in the middle of it. Sentences with the same content land near each other, translation between untrained pairs works, and the apex exists, in production, at scale. Nobody wrote it and nobody can read it, and it runs under a substantial fraction of the sentences that cross a linguistic boundary on any given day, including the ones in Nasrin’s courtroom when no interpreter was available.

To a philosopher this is the first empirical instance of something that had only been argued about. Quine held that translation is indeterminate, that multiple incompatible manuals fit all possible evidence. A working pivot is a wager against that, settled by engineering rather than by philosophy. Not answered. Settled, in the sense that the infrastructure behaves as though the question has an answer, which is a different thing and a worse one. Quine’s indeterminacy is taught with a linguist in a field and a rabbit. The working version is an adjudicator deciding whether two accounts are inconsistent, where the manual producing the second is one of several that would fit and nothing in the record indicates a manual was applied at all.

I wonder whether a discipline can be said to have lost an argument it was never told had been submitted for judgment.

A linguistic anthropologist asks a different question. Not whether the pivot is metaphysically sound, but what a deployed one does. An utterance says something and it also does something: it marks who is speaking, to whom, in what relation, with what claim to knowledge. Hymes made the case that competence is mostly competence in the second function. The pivot preserves the first very well and handles the second by dropping it, and this is not an implementation gap a better model closes. The middle is organized around content, and deference, hedging, evidential marking and register are not content. They are relations between speaker, addressee and claim, and a representation organized around what a sentence is about has nowhere to put them. Irvine and Gal give the name: erasure, by which an ideology renders distinctions invisible not by suppressing them but by having no place for them. The distinctions do not appear as losses. They do not appear at all.

The series does not correct systems. It equips them to report their own insufficiency, and the eighth operation is the indexical audit. Before a pivoted output is treated as equivalent to its source, the system enumerates the features of the source that mark relation rather than content, determines whether the target carries a corresponding feature, and where it does not, emits a flag. The flag repairs nothing and does not guess what the feature was doing. It would have put a line in Sohrab’s record where Nasrin’s margin note is, saying the English has no place for a form the Dari carried and that the inconsistency the adjudicator is about to note may be an artifact of the pivot. Run it on a procurement director in Stuttgart whose careful conditional arrives as a commitment and it flags the same class of loss. The method is universal. The urgency is differential.

Philosophy could once afford an open question, because nothing was standing on it. Something is now built on top of this one, and the disciplines that would adjudicate the middle are downstream of a deployment they did not authorize.

Nasrin has nineteen years of margins and has never been asked for them. The transcript is the record and the margin is not, and that was true before any of this existed. The pivot did not create the problem. It industrialized a loss the institution had already decided not to count. The notebook is full. Nobody has ever asked to see it.