Don’t animals talk too?

“Isn’t language just more sophisticated animal communication?”

Terrence Deacon — a Berkeley neuroscientist and anthropologist, and no partisan of any origin scenario — spent The Symbolic Species (1997) dismantling exactly this assumption. His distinction, in Peirce’s terms: animal signals are indexical — learned correlations, the way Pavlov’s dog learned the bell. Words look like the same thing, a sound paired with an object. They are not, and Deacon’s test is operational. Break the correlation and watch what happens. Stop feeding the dog after the bell, and the bell loses its meaning — indexical associations decay when the world stops cooperating. Now notice that the word “unicorn” works fine, that “dodo” did not die with the bird, and that no amount of dinner arriving without anyone saying “dinner” weakens the word. Word reference survives broken correlations because it is not held in place by the world at all — it is held in place by other words, a system of mutual definition. That is why he concludes:

Two consequences matter for everything below. Because a symbol’s reference lives in its relations to other symbols, symbols cannot be acquired the way associations are — one at a time. And on Deacon’s reading of the neuroscience, the two systems are housed differently in the brain: human symbolic language is not an upgraded call system but a different faculty. Hold both points; they return with force at the end.

“But vervet monkeys have words for eagle and leopard.”

The vervet studies are the strongest card the continuity view holds, so play it at full strength. In 1980, Seyfarth, Cheney and Marler showed in Science that vervets give acoustically distinct alarm calls for leopards, eagles, and snakes — and that a recorded call played back with no predator present sends the troop into the correct evasive action. The calls are, as the field came to say, functionally referential. Watch:

Vervet monkey alarm calls — BBC, Talk to the Animals

Now stay with the researchers, because they did the damage themselves. Vervets go on giving leopard alarms after every member of the group has seen the leopard (Cheney & Seyfarth, 1990). Sit with that: a speaker who keeps shouting “leopard!” at people watching the leopard is not informing anyone — the call is not for anyone; it fires. The follow-up experiments located the reason. Monkeys do not track what their audience knows:

A human utterance is aimed at a gap in someone else’s knowledge; the vervet call is a triggered broadcast, indifferent to whether anyone needs it. That is the indexical/symbolic line from the last section, observed in the field. And the field itself eventually said so: Wheeler and Fischer’s 2012 review concluded that “functional reference” had become “a red herring” in the search for language’s precursors. The best case for animal words, pressed hard, turns into evidence for the difference.

“What about the apes who learned sign language?”

Herbert Terrace is the crown witness here, and what makes his testimony unanswerable is its direction: he ran Project Nim in order to refute Chomsky — the chimp’s name, Nim Chimpsky, was the boast — and every incentive he had pointed toward finding language. Nim learned 125 signs and produced thousands of multi-sign sequences that looked, in the daily logs, like sentences. Then Terrace did the one thing the other projects never did: he re-watched the videotapes frame by frame. What the tape showed ends the argument, which is why it deserves more than a clause: Nim’s teachers were signing first. A fraction of a second before each of Nim’s “spontaneous” utterances, his teachers — unconsciously, in expectation — were producing the very signs he then mirrored back. The “sentences” were reflections. At normal speed, the humans in the room could not see themselves prompting; the experiment was measuring its own experimenters. Terrace published the reversal in Science (“Can an Ape Create a Sentence?”, 1979), and his data made the point brutally on its own — here is Nim’s longest recorded utterance, sixteen signs:

Sixteen signs, zero syntax, one message: give. More signs never added up to a sentence — length grew, structure didn’t. Koko the gorilla, the most famous case, is scientifically the weakest form of the same story: across decades, no controlled data was ever published — Stanford primatologist Robert Sapolsky’s verdict was “no data,” just “several heartwarming films” — and Koko’s celebrated utterances reached the public only through her handler’s interpretive glosses. An animal exquisitely tuned to an expectant handler’s cues, and no way from outside to tell performance from language: the same structure the Nim tapes exposed, minus the tapes. Terrace’s summary of the whole generation — Washoe and Koko included — was that the apes “only learned to use imperatives that were involuntary demands for primary rewards.” In his own blunter phrase: Nim “learned how to beg.”

Project Nim (2011) — official trailer

“Kanzi, though. Kanzi understood spoken English.”

Kanzi the bonobo is the honest hard case, and he survives the Nim critique: in the landmark 1993 study he carried out novel spoken requests (“put the pine needles in the refrigerator”) about as well as a two-year-old child — many delivered from behind a mask, so no cueing. That is real comprehension of arbitrary vocabulary, the best any nonhuman has shown. Watch it before reading on:

Kanzi and novel sentences — Great Ape Trust footage

Then look at where it stops, because the boundary is precise. Reanalyzing the same 1993 data, the linguist Robert Truswell found Kanzi at chance exactly where sentence structure has to do the work: asked to “fetch the toy and the water,” he brings the toy, the water, or both — at random. He hears words; the grammar connecting them never arrives. Production tells the same story: across the systematic studies of language-trained apes, 96–98% of everything they produce is a demand (Tomasello & Call, 2019). And after sixty years of these projects, linguist Geoffrey Pullum’s standing challenge has no counterexample: “I do not believe that there has ever been an example anywhere of a nonhuman expressing an opinion, or asking a question. Not ever.” A creature that can fetch the pine needles but cannot ask, tell, or wonder aloud is not partway up a slope to language. It is at the top of a different hill.

“So what exactly is missing?”

There is a precise answer, and a child demonstrates it every day. Attention comes in orders. First order: I see X — plain directed attention; nearly everything with eyes has it. Second order: I see you see X — tracking another’s attention; social animals have it, and it is enough for hunting, stalking, and pecking orders. Third order: I see you see me see X — attention to shared attention, mutually acknowledged. The entire difference documented above lives on that last step.

The decisive experiments are Michael Tomasello’s, from three decades of side-by-side child and ape studies at the Max Planck Institute. At around nine months — his phrase is “the nine-month revolution” — human infants begin following gaze, checking back, and, around the first birthday, before language, pointing. Not only to get things: to show them. In the classic experiment, a twelve-month-old points; if the adult just looks at the object, or just smiles at the child, the infant persists, unsatisfied — satisfied only when the adult looks and shares the moment back. The infant is not requesting; it is checking that the two of them are attending together. Apes, in thirty years of gesture research, never take this step:

Captive apes do point — imperatively, at food, for humans. Never for each other, and never to share. The published version of Tomasello’s much-quoted remark about chimpanzees carrying a log together puts it at the species level: collaboration on a genuinely joint intention is “almost unimaginable” between chimps. The same line explains Hockett’s classic finding that displacement — talking about the absent, the past, the hypothetical — is essentially unique to human language: you can only talk about what isn’t there with someone you can jointly attend with about what isn’t there. (The honeybee waggle dance, the textbook near-exception, encodes exactly one displaced fact — the last food source — in an innate, continuous code that cannot negate, ask, or say anything else.) A fourteen-month-old sails past all of it:

Warneken & Tomasello — spontaneous helping in toddlers (and chimps)

“Fine — a real difference. Why would it need an origin event?”

Everything above is other people’s science. Here is the step that leads to this archive. A sign is a convention: it means something only because everyone party to it takes it to mean that. So ask the gradualist question honestly — what would half a convention be? A signal that only some understand as a sign is not a weak sign; it is a strange gesture. Claude Lévi-Strauss — no friend of origin stories — saw the consequence in 1950:

And here is the convergence worth the whole page: Deacon reaches the same structure from the learning side, with no stake in the conclusion. Because a symbol’s reference depends on the system of other symbols, he argues, “symbols cannot be acquired one at a time, the way other learned associations can” — crossing the symbolic threshold is “a restructuring event” that “essentially takes no time.” The structuralist and the neuroscientist, from opposite ends, both find that signifying has no gradual on-ramp. The archive’s texts draw the conclusion:

Two honest fine-prints. This is not a denial of development — “there is gradualism and gradualism,” Gans writes; what is excluded is not stages but a phase-in of meaning itself. And it is not a one-place dogma: Gans allows the sign may have been invented “in more than one place” — “the originary hypothesis refers to the causality of the event, not to its uniqueness.” What cannot be given up is the event-form: somewhere a first sign was issued and taken up together, because a sign nobody else takes up is not a sign.

“And this is where your originary scene comes in.”

Now run the chain: the difference is qualitative (the science), so meaning cannot have phased in (the logic), so its origin was an event — and an event with specific required contents: more than one participant, a shared object of attention, and a gesture whose whole force is that everyone takes it up together. The originary hypothesis is the minimal reconstruction of an event with exactly those contents. It is not an ornament added to the human/animal difference; it is the bill the difference presents:

And if you still suspect the difference is one of degree, there is a test you can run on yourself, from Katz: tell someone you see the same thing they do — then say what exactly makes it the same, and for whom. “Can any of this pertain to anything any other species does?”

Further viewing

Michael Tomasello — The Origins of Human Collaboration

Terrence Deacon — Language and Complexity: Evolution Inside Out