On June 10, 1898, Tuone Udaina died in a mine explosion on the island of Krk, in the Adriatic. He was the last speaker of Dalmatian, more precisely of the Vegliot dialect. With him, a Romance language spoken for over a thousand years along the Dalmatian coast went extinct.
The linguist Matteo Bartoli had interviewed him in the years before his death and published Das Dalmatische in 1906, which remains the principal source on this language. But Bartoli was working on a ruin. Udaina had not spoken Dalmatian in some twenty years. He had learned it as a child by listening to his parents, who themselves no longer used it except in fragments. He was partially deaf and had lost his teeth, which distorted his pronunciation. He was reconstructing from memory a language he had never truly practiced as a living tongue.
The last witness to a system is already a ruin of that system.
The problem is not merely sentimental. It is logical. When Bartoli records a Dalmatian form from Udaina's mouth, he cannot know whether that form is a property of the language or a peculiarity of this single speaker. A lapse of memory, a borrowing from the Italian or Croatian that Udaina otherwise spoke, a distortion due to age or deafness: all of this is indistinguishable from an authentic feature. There is no one else to compare him against. A form attested by a single witness cannot be distinguished from the idiosyncrasy of that witness.
For a feature to be a property of the language, it must be shared. A language is a social system: its rules are what several speakers jointly validate through usage. A single speaker has no one to validate his own. His language is no longer a social system. It has become his private memory of a vanished social system.
The language was therefore dead as a system before the death of its last speaker. Udaina was not the last bearer of a living language. He was the survivor of an already-dead language, of which he alone carried a trace that no one could any longer correct or confirm.
The same structure recurs in conservation biology. Celia, the last Pyrenean ibex — the bucardo — died in January 2000, crushed by a tree. Cells had been collected. In 2003, an attempt at cloning produced an individual that died a few minutes after birth, of a pulmonary malformation. Cloning could reproduce an individual genome. It could not reproduce the species.
For a species is not an individual. It is a population: a genetic diversity, an age structure, learned behaviors, ecological relationships. A single individual carries only a sample of a genome, not the variability of the species. Long before the death of the last bucardo, the species had crossed the threshold of minimum viable population; inbreeding and the loss of diversity had already extinguished it as a population. The last individual was a survivor, not a conservatory.
The Allee effect describes this threshold: below a certain density, a population declines even if each individual is healthy, because reproduction, defense, and foraging require a minimum number of cooperating individuals. The species goes functionally extinct before the last individual dies. The death of the last specimen is not the cause of extinction. It is its belated confirmation.
Philology knows a kindred form of the problem under the name codex unicus. Certain ancient texts survive only through a single manuscript. Beowulf hangs on a single codex, the Nowell Codex, partly damaged in the 1731 fire at the Cotton library. When a text survives in a single copy, the philologist can no longer compare variants. Textual criticism, which establishes the original by confronting the divergences between independent copies, becomes inapplicable. There is nothing to confront. Every error of the last copyist has become indistinguishable from the original text, because no other lineage allows it to be detected. The unique text cannot be corrected. It can only be transcribed, with its errors now made definitive.
These three cases share a structure. The last specimen does not preserve the type: it marks its end. The language, the species, the text were each defined by a multiplicity of instances that validated one another. Reduced to a single instance, they cease to be types. They become individuals from which the norm can no longer be inferred.
Information theory gives this its most abstract formulation. A message without redundancy cannot be corrected. If a piece of information exists in only a single copy, with no duplicate or error-correcting code, the slightest corruption is undetectable, for want of another version against which to check it. Redundancy is not waste. It is the condition for detecting error. A system without redundancy is a system in which every error becomes invisible, and therefore definitive.
Uniqueness is not maximal conservation. It is the loss of the possibility of generalizing.
Doctrine
A type holds only through the multiplicity of instances that validate one another: redundancy is not a surplus, it is the condition for distinguishing the rule from the accident. Reduced to a single case, a type loses the very thing that defined it — idiosyncrasy and norm become indistinguishable. The last specimen does not save the class; it marks the moment when the class ceases to be one. To preserve the last witness is not to preserve the system: it is to keep an individual who can no longer attest to anything general.
Open vector
The threshold is not at one. It is crossed well before. A language dies when its community of speakers falls below the number that allows daily use, not when the last person who remembers it dies. A species goes extinct when its population drops below the threshold of viability, not at the death of the last individual. A text becomes incorrigible when only a single lineage of copies remains, not when the last copy burns.
The last specimen is an effect, not a cause. It makes visible an extinction that has already occurred. What it preserves is not the system: it is the interval between the functional death of a type and the disappearance of its last trace.
The question, then, is not: how do we save the last specimen? The question is: at how many specimens does a type cease to be a type?
