Before 1977, most deaf children in Nicaragua stayed home. Schools taught no sign language, no deaf community had formed, and many children never met another deaf person in their lives. They communicated with their families through gestures they made up themselves — a system with a name: home sign.
That year a school for the deaf opened in the capital, Managua; enrolment reached a hundred by 1979. In 1980 a vocational school opened in Villa Libertad on the edge of the city, and by 1983 the two together held over four hundred students. The curriculum was lipreading and spoken Spanish, and that part went badly. The part nobody planned worked: on the buses and in the schoolyard, dozens of unrelated home-sign systems landed in the same pool for the first time.
In 1986 the linguist Judy Kegl was invited to look at what these children were doing with their hands. What she found was not a heap of gestures. It was a language.
Three things this time: which group of children the grammar actually appeared in (not the ones who built it), how you separate "practised longer" from "received it earlier", and how the same thing happened again in a village in the Negev — and swerved into a word order nobody around it uses.
Start with the starting point. A deaf child who has never met a sign language, growing up with hearing parents who do not sign, does not simply do nothing. Goldin-Meadow and colleagues followed such children for years and found that the gesture systems they invent carry consistent ordering: for who-did-what-to-whom, which gesture comes first and which comes second stays stable within a child.
Here is the crucial comparison: their mothers' gestures do not have these properties. A mother waves her hands while talking — loose, and different from situation to situation. So the structure in the child's hands was not copied. The harder result is cross-cultural: deaf children in Turkey and in the United States converge on the same way of packaging an event into a single action phrase, even though adult gesture habits in the two societies differ a great deal.
But home sign has a ceiling: it is one person's system, used in one household, with no second user treating it as something to be learned — it never grows the layered machinery a community language has.
So what happened in that school in 1977 was dozens of such systems meeting each other for the first time. What comes out of that? Senghas, Kita and Özyürek, in 2004, played a cartoon: a ball spinning as it rolls down a slope. Describe it.
Hearing Spanish speakers gesture with the hand swirling while it travels downward — manner and path fused into one movement. First-cohort deaf signers largely did the same. The second and third cohorts did not: they split it in two, first a "spin" held in place, then a "downward", one after the other.
Now a finer piece of machinery. A sign language can aim a verb at a location in space that temporarily stands for a person: push the action "pay" out toward the empty spot held by "the one being paid", and point back at that spot later when the same person comes up. One location, one referent, and who-is-who never has to be spelled out separately.
Senghas and Coppola went looking, in 2001, for when this device appeared. Every cohort aims signs at spots in space, it turns out, but only the second cohort had turned it into a system — the same location bound stably to the same person. For the first cohort the space carried direction, not coreference.
The methodological move is the real point. "The second cohort is better" has a boring explanation: they were younger, so they learned faster. So the researchers separated "how old you were when you entered this community" from "how many years you have been signing" — they grouped each cohort by age at entry and compared across cohorts only within matching age-at-entry groups. The result: early-exposed first-cohort signers had more years of signing behind them, yet early-exposed second-cohort signers used the device more, and more systematically. Those extra years did not carry the first cohort past them.
What drives it is not fluency, but what you received while you were still young. Senghas later followed the same people across years and found individuals mostly settle after childhood, each carrying their own version forward — the change lives in the children, not in the adult users.
Evidence strength: this is a cross-sectional cohort comparison within one community, with no way to randomise and no way to run it again. The cohorts differ in more than age — by the time the second cohort arrived the community was several times larger, with far more people to talk to, and that is tangled up with age in a way nobody can untangle. This is the standing limitation of work like this.
It runs on the same line as the issue on what children's mistakes reveal: there, one child spread the "add -ed" pattern onto a well-memorised went, output more systematic than input. Here a whole cohort spread the previous cohort's scraps into a system. Same direction, scaled up to a community.
In the Negev, in southern Israel, there is a Bedouin village of about four thousand people, more than a hundred and thirty of them deaf — a recessively inherited congenital deafness traceable to a few early families. Some seventy to eighty years ago a sign language grew up in the village on its own, and a good number of the hearing villagers use it too.
Sandler, Meir, Padden and Aronoff reported in 2005 that by the second generation of signers the word order had settled: subject–object–verb. To say that a woman feeds a child, the order signed is "woman child feed", with the verb resting at the end.
Why that matters: no language around it has that order. The Arabic dialect the hearing villagers speak is subject–verb–object, and so is Hebrew; Classical Arabic is verb–subject–object; the Israeli Sign Language already present in the region is not it either. So the order was not borrowed — it was fixed inside this language, and the point at which it was fixed was, again, the second generation.
Pull in another line: creole languages. Plantation economies threw together adults with different mother tongues, and among themselves they improvised a working means of exchange — a pidgin, with words but little settled grammar. Bickerton's language bioprogram hypothesis says the children of those adults turned the pidgin into a creole with a full grammar, and that the suspiciously similar parts across distant creoles come from a program people carry. His pivotal case was Hawai'i.
The other side struck back at exactly that case. Siegel, in 2007, re-examined the history and pointed out three things: formation spanned more than one generation; what the children received was not a rudimentary pidgin but a fairly stabilised one; and a good many features can be traced to the languages immigrants brought, Cantonese and Portuguese among them. DeGraff goes further: treating creoles as a class of languages exceptional in how they arise does not hold up as a premise at all.
To be clear: these are all full languages, merely young. A creole is not a simplified version of anyone, and the Negev sign language is not a local variant of Israeli Sign Language.
So "one generation is enough" has to be narrowed. What holds is this: a single pass through learners makes a system more systematic — seen once each in two independent places, landing both times on the second cohort or generation. What does not hold is that one generation of children builds grammar out of nothing. One is not enough. What is needed is somebody to take the previous generation's pile as the thing to be learned.
What they contributed cannot be measured by asking whether it had grammar: a public system, in use, worth learning. That is exactly what home sign lacks — not structure, a second person. So the question becomes how many users are enough. The first few dozen students in Managua were; one child in one household is not; where the line between them falls, neither case can say.
Its word order differs from every language around it, which looks solid. But how far does "around" extend — the next village, television, people who worked in the city? An argument by elimination is only as strong as the candidate list, and the list is drawn up by people. Turn it around: had the order matched the local Arabic dialect, would that have settled it as borrowing? Then the evidence only ever falsifies borrowing; it never establishes spontaneity.
It has been seen in two places: one child's overregularization, one cohort's systematisation. But the two differ by two orders of magnitude in timescale, and the layers in between — a household, a classroom, a village — have barely been measured. A shape that recurs across scales may be one mechanism, or several leaving similar traces. To tell them apart, you first need a prediction that holds at only one of the scales.