The short answer: nobody knows yet. Spoken words leave no bones, no footprints and no fossils. Every answer is built from indirect clues: throats and ear bones, genes we share with birds and bats, ancient DNA, experiments with living people and animals, and studies of how today’s brains handle words. A 2024 review in Nature puts estimates for when language emerged between about 100,000 and 1,000,000 years ago (Fedorenko, Piantadosi & Gibson, 2024, p. 575), and one well-known team argues for as recently as 70,000 years ago (Bolhuis et al., 2014, Abstract). A range that wide tells you how open this question still is. [Unknown]
Why start a history of the mind here? Almost everything later in this collection — naming a feeling, describing distress, giving testimony, writing a diagnosis — runs on language. And the story of how people first got words has to be told with most of the lights off.
What makes human language different from other animal signals?
Some animals communicate about things beyond their immediate surroundings. Honeybees encode the distance and direction of a food source or potential nest site in waggle dances (Menzel & Galizia 2024, Introduction and experiments). So pointing beyond the here-and-now is not a humans-only trick.
What does look distinctive is how open-ended human language is. As far as researchers know, every human society has language (Pinker & Bloom, 1990, §1) — and “language” includes signed languages, not just speech. The same left-side brain network responds to language whether it arrives spoken, written or signed (Fedorenko et al., 2024, p. 576).
Analogy, not evidence: a bee dance is like a GPS readout — precise about one location. A human language is more like a keyboard: a limited set of keys, and no limit on the messages.
Why can’t scientists just dig up the first word?
Because brains do not fossilize, as one research team points out (Fedorenko et al., 2024, p. 575), and spoken sound leaves no trace. So researchers climb a ladder of inference, the same kind this collection uses for prehistoric injuries. Each rung needs more evidence than the one below:
- A bone or gene is there. That can be documented.
- It may show a capacity — to hear or shape certain sounds.
- A capacity is not a behavior. Being able to make speech sounds doesn’t prove anyone spoke.
- A behavior is not a grammar. Calls or gestures aren’t automatically language.
- A grammar is not an experience. We can’t know what anyone said, or how it felt to say it.
Ask “which rung does this reach?” as you read the clues.
Clue 1: Does it all come down to a low voice box?
The human larynx (voice box) sits low in the throat. For decades, textbooks often said other primates can’t talk because their vocal tracts are the wrong shape. According to a 2016 study, that view rests largely on a 1969 analysis that modeled a macaque’s vocal tract from a plaster cast of a dead monkey (Fitch, de Boer, Mathur & Ghazanfar, 2016, Introduction).
The 2016 team X-rayed living macaques as they called, made faces and ate, and found their vocal tracts could produce clearly distinguishable vowels. Their conclusion: macaques have a “speech-ready vocal tract” but lack the brain circuitry to control it (Discussion). In 2017, Philip Lieberman of Brown University, lead author of that 1969 study, replied that monkeys still cannot make the full range of human speech sounds, including the vowels [i], [u] and [a], and that human anatomy mattered alongside the brain (Lieberman, 2017). [Disputed] So a low larynx can no longer be treated as the settled reason only humans talk.
Clue 2: FOXP2 — a famous gene and a famous misunderstanding
In the 1990s, geneticists studied a large British family, known in the scientific literature only by the initials “KE.” About half of its members, across three generations, had a severe speech and language disorder (Fisher, 2025). In 1998 the team traced it to a stretch of chromosome 7. In 2001 they named the gene: FOXP2, changed in affected family members and disrupted in an unrelated person known as “CS” (Fisher, 2025, historical perspective).
The family should not disappear behind the gene’s name. Fisher’s account describes research involving relatives across three generations, alongside the unrelated participant known as CS. These sources preserve the researchers’ account; they do not give us the participants’ own views of the work (Fisher, 2025, historical perspective).
Some press coverage called FOXP2 “the gene for language.” Fisher calls that a mischaracterization. FOXP2 makes a transcription factor — a protein that turns other genes up or down. It is active in the lung, gut and heart as well as the brain, and versions of it exist in mammals, birds, reptiles, amphibians and fish (Fisher, 2025, “Insights From Investigating FOXP2”). The U.S. National Library of Medicine puts it carefully: the protein “appears to be essential” for normal speech and language development (MedlinePlus Genetics).
Animal studies add clues:
- Songbirds. When researchers reduced FoxP2 in a song-learning brain region of young zebra finches, the birds copied their tutor’s song incompletely and inaccurately (Haesler et al., 2007).
- Bats. FoxP2 is unusually variable in echolocating bats. The authors suggest a link to echolocation but say the reason for the change “is not clear” (Li et al., 2007). [Plausible]
Humans carry two protein changes in FOXP2 that earlier work linked to a recent burst of natural selection. A 2018 reanalysis of genomes from around the world found no evidence of recent selection and noted that archaic humans shared the same changes (Atkinson et al., 2018). [Disputed] These sources don’t dispute FOXP2’s role in speech development; what’s contested is the idea that it marks language’s arrival.
FOXP2 became an important entry point into the genetics of speech and language, but Fisher’s review describes many other genes and methods involved in that research (Fisher, 2025).
Clue 3: Did Neanderthals talk?
In 2013, Dan Dediu and Stephen Levinson argued that the case for Neanderthal language is stronger than many assumed. They pointed to an essentially modern-looking Neanderthal hyoid (a small throat bone) from Kebara, Israel; ear anatomy suggesting modern-like hearing in roughly 500,000-year-old fossils from Sima de los Huesos, Spain, which they assign to an earlier species, Homo heidelbergensis; and the shared FOXP2 changes. Their conclusion: recognizably modern language likely goes back at least to the common ancestor of Neanderthals and modern humans, about half a million years ago (Dediu & Levinson, 2013). They also warn that reasoning from gene sequences to speech is “necessarily tentative.”
Johan Bolhuis, Ian Tattersall, Noam Chomsky and Robert Berwick argued the opposite in 2014: language emerged only some 70,000 to 100,000 years ago. They noted symbolic objects appear around 100,000 years ago, well after anatomically recognizable Homo sapiens (Bolhuis et al., 2014, Abstract and “Paleoanthropology”). The Smithsonian now dates Homo sapiens to about 300,000 years ago (Smithsonian Human Origins). [Disputed] On the ladder, a modern-looking hyoid or ear reaches rung 2, a capacity. It cannot tell us what anyone said.
Why did language evolve? Four big ideas, none proven
- Gossip instead of grooming. Robin Dunbar argues that monkeys and apes bond by grooming, which takes time and works one-on-one. As human groups grew, conversation let people bond with several others at once and keep track of who is doing what (Dunbar, 2004, pp. 102–103). [Plausible]
- Gesture first. Some researchers propose that language began with the hands. In a 2014 lab study, pairs of university students inventing signals from scratch did better with gestures than with non-word sounds. The authors caution that their volunteers already had modern brains and already spoke a language (Fay, Lister, Ellison & Goldin-Meadow, 2014). [Plausible]
- Ritual builds trust. Chris Knight argues that word-like signals only work when listeners can trust speakers, since cheap signals are easy to fake. Costly, shared ritual may have built the trust that let speech take hold (Knight, 1998, pp. 72, 85–86). [Plausible]
- One small step — or many? Bolhuis, Chomsky and colleagues argue that the heart of language is a single combining operation they call “Merge,” which could have appeared quickly and recently (Bolhuis et al., 2014). Steven Pinker and Paul Bloom argued in 1990 that grammar is a complex adaptation built by ordinary, gradual natural selection (Pinker & Bloom, 1990, §6, author preprint). [Disputed]
These ideas aren’t necessarily rivals; several could be partly right. The evidence can’t yet rank them. [Unknown]
The question that was once off the table
On March 8, 1866, the French government approved the statutes of the Société de Linguistique de Paris. Article 2 said the society would accept no communications on the origin of language or on creating a universal language (Société de Linguistique de Paris, Statuts de 1866). Writing in an encyclopedia of the history of anthropology, Claudine Gauthier reads the rule as the linguists distancing themselves from the language theories of the Société d’anthropologie de Paris (Gauthier, 2008). You may read that this rule chilled the whole topic until the late 20th century. The sources here don’t support that larger claim. [Unknown]
The rule did not end the question. By 1998, Knight could describe Darwinian evolution as setting a “new research agenda,” citing work from 1989 onward, including Pinker and Bloom’s 1990 paper (Knight, 1998, p. 68).
How we know
What evidence exists: a few fossil throat and ear bones; ancient and modern DNA; family genetic studies; experiments with monkeys, songbirds, bats and modern volunteers; symbolic objects; brain studies of people who lose language.
Whose voices are missing: every early speaker. No first word, first sign or first conversation survives. The KE family appears in the sources we opened only through researchers’ descriptions, not their own accounts. Much origin-of-language writing assumes speech, which can sideline Deaf and signing communities. Indigenous and other communities’ own accounts of language’s origins aren’t covered here; they need their own sources and community permission, not a sidebar.
How power shaped the record: in 1866 a learned society decided which questions it would hear. The research we opened comes mostly from universities in Europe and North America and is published mostly in English. The fossil sample is tiny and shaped by where bones survive and where excavators dig. Some press coverage turned a careful genetic finding into a “gene for language.”
Then vs. now
Then. What the first speakers believed about their own words is [Unknown], because no record exists. In 1866, one Paris society refused to hear papers on the origin question.
Now:
-
Is language the same thing as thinking? [Disputed] Bolhuis and colleagues describe language as a “computational cognitive system” and call its use for communication “largely irrelevant” to explaining its origin (2014, “Conceptualizations of Language”). Fedorenko and colleagues argue it is mainly a tool for communication. Their review distinguishes language from other forms of thought and asks how the two interact (Fedorenko et al., 2024, pp. 575–582). That disagreement concerns the role of language; it does not establish a date for its beginning.
-
And the rest? There is no single “language gene”; FOXP2 is one gene among many. The low-larynx story is [Disputed]. The start date is [Unknown]: the published estimates we opened run from about 70,000 to 1,000,000 years ago.