The History of Language: 135,000 Years of Talking and 5,000 Years of Writing It Down

The 130,000-Year Silence

Here is the central problem with studying the history of language: spoken words do not fossilize. Stone tools survive for millions of years. Cave paintings last tens of thousands. But the first sentence ever uttered left no trace whatsoever. Everything we know about the origin of speech is inferred from bones, genes, stone tools, and the behaviour of living humans, which is a bit like reconstructing a symphony from the shape of the concert hall.

What we do know is this: Homo sapiens emerged roughly 230,000 years ago. By about 135,000 years ago, human populations had begun splitting geographically, migrating out of Africa in waves that would eventually populate every continent except Antarctica. Every single one of those populations developed language. Not some of them. All of them. There is no known human group, past or present, that lacks a fully developed language with grammar, syntax, and the ability to express abstract concepts.

This is the argument that MIT linguist Shigeru Miyagawa and colleagues laid out in a 2025 analysis published in Frontiers in Psychology. They examined 15 genetic studies spanning 18 years, covering Y chromosome, mitochondrial DNA, and whole-genome data, and concluded that language capacity must have been present before the first major population split. “Every population branching across the globe has human language, and all languages are related,” Miyagawa wrote. The first split occurred about 135,000 years ago, “so human language capacity must have been present by then, or before.”

The archaeological record seems to agree, cautiously. Around 100,000 years ago, humans began leaving evidence of symbolic thinking: ochre pigments used for decoration, meaningful marks on objects, shell beads that served no practical purpose. These are not proof of language, but they suggest the kind of abstract, referential thinking that language requires. Something was happening in human cognition that had not happened before.

What Makes Human Language Different

This is worth pausing on, because it is genuinely strange. Many animals communicate. Vervet monkeys have distinct alarm calls for eagles, leopards, and snakes. Honeybees perform dances that convey the direction and distance of food sources. Dolphins appear to use signature whistles as something resembling names. But none of these systems do what human language does.

The key difference is compositionalityThe ability to combine a finite set of elements (words, sounds, or symbols) into an infinite number of meaningful expressions through systematic rules. The defining feature of human language that allows a small vocabulary to generate unlimited sentences.: the ability to combine a finite set of elements (words, morphemes, phonemes) into an infinite number of meaningful expressions. With just 25 words for subjects, verbs, and objects, you can generate over 15,000 distinct sentences. Add tense, mood, negation, and subordinate clauses, and the number becomes effectively infinite. As the evolutionary biologist Mark Pagel put it in a 2017 paper in BMC Biology, human language is “qualitatively different” from any other animal communication system.

The trained chimpanzee Nim Chimpsky (named, with pointed humour, after Noam Chomsky) illustrated the gap neatly. His longest recorded utterance was: “give orange me give eat orange me eat orange give me eat orange give me you.” This is many words. It is not a sentence. It has no grammar. It communicates desire but cannot express time, causation, or hypotheticals. A three-year-old human can do all of these things effortlessly.

The History of Language in Writing: When Talking Was Not Enough

For at least 130,000 years, language existed only as speech. Then, around 3400 BCE, something changed in southern Mesopotamia. The Sumerians, who lived in what is now southern Iraq, began pressing wedge-shaped marks into wet clay tablets. They were not writing poetry. They were counting sheep.

The earliest cuneiform tablets are accounting records: inventories of grain, livestock, and trade goods. Writing was invented not to express the human soul but to track who owed whom how many goats. This is, if you think about it, deeply characteristic of our species. We spent 130,000 years telling stories, singing songs, arguing about the nature of the divine, and when we finally figured out how to make language permanent, we used it for bookkeeping.

Egyptian hieroglyphics emerged around the same period, roughly 3200 BCE, though the question of whether Egypt invented writing independently or borrowed the concept from Mesopotamia remains debated. Chinese script appeared around 1200 BCE in the form of oracle bone inscriptions, and Mesoamerican writing systems developed independently around 900 BCE. Writing, it turns out, was only invented from scratch a handful of times in human history. Almost every other writing system is borrowed, adapted, or inspired by one of these originals.

The Alphabet: History’s Greatest Hack

Cuneiform had hundreds of signs. Egyptian hieroglyphics had over 700. Chinese characters number in the tens of thousands. Learning to write in any of these systems was a career, not a skill, which meant that literacy was restricted to a small priestly or bureaucratic class.

Then, around 1800 BCE, Semitic-speaking workers in the Sinai Peninsula did something revolutionary. They looked at Egyptian hieroglyphics and thought: what if each symbol represented just one sound? They created the Proto-Sinaitic script, a set of 22 letters based on the acrophonic principleA method of writing where each character represents the initial sound of the word it depicts. For example, in Proto-Sinaitic script, a symbol of an ox's head (aleph) represents the sound 'a'. This principle enabled the creation of alphabetic writing systems., where the sign for a word stands for the first sound of that word. An ox head (aleph) became the sound “a.” A house (beth) became “b.”

This was, by any measure, one of the most consequential inventions in human history. A writing system with 22 symbols could be learned in weeks rather than years. The Phoenicians refined it around 1050 BCE and spread it across the Mediterranean through their trade networks. The Greeks borrowed it, added vowels (Phoenician, like most Semitic scripts, wrote only consonants), and produced the ancestor of every European alphabet. The Aramaic branch evolved into Hebrew, Arabic, and eventually the scripts of South and Central Asia. Nearly every alphabet on Earth descends from those 22 Proto-Sinaitic characters.

This matters because literacy is power. When writing required years of specialized training, information was controlled by whoever could afford scribes. The alphabet did not eliminate that power imbalance, but it cracked the door open. The Protestant Reformation, the Enlightenment, the spread of democratic ideas: none of these were possible without mass literacy, and mass literacy was not possible without a writing system simple enough for ordinary people to learn.

Languages Die. The Rate Is Accelerating.

Of the roughly 7,000 languages spoken today, linguists estimate that between 40 and 50 percent are endangered, meaning they have fewer speakers than needed to sustain transmission to the next generation. UNESCO’s Atlas of the World’s Languages in Danger catalogues hundreds of languages with only a handful of elderly speakers. When they die, their languages die with them.

This is not new. Languages have always gone extinct. Latin “died” (or rather, evolved into the Romance languages). Sumerian vanished as a spoken language around 2000 BCE, surviving only as a literary and liturgical tongue. But the current rate of language death is historically unprecedented. Globalisation, urbanisation, and deliberate government policies have accelerated the process enormously. Some linguists project that 50 to 90 percent of the world’s languages could disappear by 2100.

Each loss is irreversible and carries consequences beyond sentimentality. Languages encode knowledge: botanical terminology in indigenous Amazonian languages that has no equivalent in Portuguese, navigational concepts in Pacific Islander languages that Western linguistics is still working to understand, grammatical structures that reveal things about human cognition that would be invisible in a world with fewer linguistic options.

The History of Language Is Still Being Written

The history of language does not end with the present. Language is evolving faster now than at any point in recorded history. The internet has created written dialects that would have been unrecognisable a generation ago. Emoji constitute a new quasi-pictographic system layered on top of alphabetic text. Machine translation is making cross-linguistic communication possible at a scale that would have seemed miraculous to the Phoenician traders who spread the alphabet.

And somewhere in a lab, researchers are still trying to answer the question that Paris banned in 1866: how did language begin? The honest answer, after 160 years of renewed inquiry, is that we still do not know for certain. We know roughly when (at least 135,000 years ago). We know roughly where (sub-Saharan Africa). We know it happened only once, in the sense that all human languages appear to share fundamental structural properties. But the mechanism, the moment when a hominid brain first assembled a thought that required grammar to express, remains the hardest problem in linguistics.

The Linguistic Society of Paris was right about one thing: this question generates a lot of heat. They were wrong to ban it. The heat, it turns out, was worth enduring.

The Dating Problem: When Did Language Emerge?

Understanding the history of language requires answering when human language first emerged, which is complicated by a fundamental methodological obstacle: speech leaves no direct archaeological trace. Unlike stone tools or cave paintings, vocalisations do not fossilize. Researchers must rely on indirect evidence from genetics, anatomy, archaeology, and comparative linguistics.

The most recent systematic attempt to date language emergence comes from Miyagawa, DeSalle, Nóbrega, Nitschke, Okumura, and Tattersall, whose 2025 meta-analysis in Frontiers in Psychology examined 15 genetic studies spanning 18 years. Their dataset included three Y chromosome studies, three mitochondrial DNA studies, and nine whole-genome studies. The core argument is phylogenetic: since every known human population possesses fully developed language, and since the first major population split occurred approximately 135,000 years ago, language capacity must predate that divergence.

This estimate is conservative. Homo sapiens emerged roughly 230,000 years ago, and some researchers argue that the cognitive architecture for language could have been present from the species’ origin. The archaeological record shows increased symbolic activity (ochre use, shell beads, deliberate markings) beginning around 100,000 years ago, which may indicate language use but does not constitute proof of it.

A competing view, associated with archaeologist Richard Klein, places the emergence of behavioural modernity (and by implication, language) at roughly 50,000 years ago, coinciding with the “Upper Paleolithic revolution” in Europe. This hypothesis has weakened as earlier evidence of symbolic behaviour has emerged from African sites, but the debate illustrates how heavily conclusions depend on which proxy evidence one privileges.

The Genetics of Language: FOXP2 and Beyond

The discovery of the FOXP2 gene in the late 1990s by Simon Fisher’s group at Oxford initially generated enormous excitement. The gene was identified through the study of the KE family, a British family in which a dominant mutation caused severe verbal dyspraxiaA neurological speech disorder where the brain struggles to coordinate the precise mouth and jaw movements needed to produce speech sounds, despite normal muscle strength and language comprehension. (difficulty with the coordinated movements required for speech) across three generations. Media coverage quickly labelled FOXP2 “the language gene.”

This label was premature. FOXP2 is a transcription factor that regulates other genes during embryonic development, and it is not unique to humans. Homologues exist in mice, birds, and other vertebrates. A 2008 finding that Neanderthals shared the same derived FOXP2 variant as modern humans further undermined the idea that this gene alone explains human language capacity, since Neanderthals show “almost no evidence of the symbolic thinking” characteristic of contemporaneous Homo sapiens, as Pagel noted in BMC Biology.

Current research treats FOXP2 as one component of a complex polygenicDescribes a trait or disease influenced by many genes, each contributing a small effect. Most common diseases like diabetes and heart disease are polygenic. trait. The gene influences orofacial motor control and certain aspects of procedural learning, but individuals with FOXP2 mutations can still comprehend language. The lesson is instructive: language is not a single adaptation produced by a single gene. It is a suite of capacities (articulatory control, working memory, hierarchical processing, social cognition) that likely evolved incrementally over hundreds of thousands of years.

CompositionalityThe ability to combine a finite set of elements (words, sounds, or symbols) into an infinite number of meaningful expressions through systematic rules. The defining feature of human language that allows a small vocabulary to generate unlimited sentences.: The Feature That Defines Human Language

The central distinction between human language and all known animal communication systems is compositionality, the ability to combine discrete units into structured, hierarchically organised expressions with compositional semantics. This is the property that Noam Chomsky calls “discrete infinityA core property of human language in which a finite set of elements can be combined recursively to generate an unbounded number of distinct meaningful expressions. A key distinction between human language and animal communication.”: a finite inventory of elements generates an unbounded set of meaningful expressions through recursive combination.

Mark Pagel’s 2017 analysis in BMC Biology quantifies this: with 25 words each for subjects, verbs, and objects, the combinatorial space exceeds 15,000 sentences before accounting for tense, aspect, mood, or embedding. No documented animal communication system approaches this generative capacity.

The theoretical debate over compositionality divides broadly into two camps. The nativist position (Chomsky and successors) holds that the capacity for hierarchical syntactic structure is innate, species-specific, and domain-specific, encoded as a “Universal Grammar” in the human genome. The usage-based position (Tomasello, Bybee, Goldberg, among others) argues that linguistic structure emerges from general cognitive capacities (pattern recognition, analogy, social learningIn criminology, the process by which individuals acquire criminal behavior through observation and imitation of others, particularly authority figures. Children who witness violence learn to normalize it as a method of control., joint attention) applied to communicative interaction over developmental time. This debate remains unresolved, though the strong nativist position has lost ground as typological and acquisition data have accumulated.

Writing Systems: Independent Invention and Diffusion

Writing was independently invented at most four times in human history: in Mesopotamia (cuneiform, c. 3400 BCE), Egypt (hieroglyphics, c. 3200 BCE, possibly influenced by Mesopotamia), China (oracle bone script, c. 1200 BCE), and Mesoamerica (Zapotec/Maya, c. 900 BCE). Every other writing system in use today derives from one of these through borrowing, adaptation, or stimulus diffusion.

The development of alphabetic writing represents a critical phase transition. Cuneiform and hieroglyphics used logographic and syllabic principles requiring hundreds of signs. The Proto-Sinaitic script (c. 1800 BCE), developed by Semitic-speaking workers in the Sinai Peninsula, reduced writing to approximately 22 consonantal signs using the acrophonic principleA method of writing where each character represents the initial sound of the word it depicts. For example, in Proto-Sinaitic script, a symbol of an ox's head (aleph) represents the sound 'a'. This principle enabled the creation of alphabetic writing systems.: each sign represented the initial phoneme of the depicted object. The Phoenician alphabet (c. 1050 BCE) refined this system and, through Greek adaptation (which added vowel signs), became the ancestor of virtually every alphabetic script in use today, including Latin, Cyrillic, and (via Aramaic) Hebrew, Arabic, and Brahmic scripts.

The cognitive implications are significant. Alphabetic literacy can be acquired in months rather than years, dramatically lowering the barrier to textual participation. This shift correlates historically with expanded literacy rates, though the causal relationship is complex (political, economic, and religious factors mediate the connection heavily).

Language Endangerment: Current Data

Of approximately 7,000 languages currently spoken, UNESCO and the Endangered Languages Project classify between 40 and 50 percent as endangered. The distribution is highly unequal: roughly 23 languages account for more than half the world’s speakers, while thousands of languages have communities numbering in the hundreds or fewer.

Language death is not new (Sumerian ceased as a spoken language c. 2000 BCE; Etruscan disappeared under Roman assimilationThe process by which a minority group's cultural, linguistic, or ethnic identity is gradually absorbed into a dominant culture, often through institutional pressure such as education policy.), but the current rate is unprecedented. Estimates vary, but projections suggest that 50 to 90 percent of extant languages may cease to be spoken by 2100. The primary drivers are urbanisation, economic integration, education policies mandating dominant languages, and deliberate state-level language policies.

The scientific cost of language loss extends beyond cultural heritage. Languages encode categorisation systems, spatial reasoning frameworks, and ecological knowledge that may not be recoverable through translation. The Pirahã language’s disputed lack of recursion, Guugu Yimithirr’s absolute spatial reference system, and the elaborate botanical taxonomies of Amazonian languages all represent data points for understanding the boundaries and flexibility of human cognition. Each language lost narrows the empirical base for testing theories about what human minds can and cannot do with language.

Open Questions

The history of language contains more unknowns than knowns, which is itself a useful piece of information about the state of the science. Among the major unresolved questions:

• Monogenesis vs. polygenesis: Did language emerge once, with all languages descending from a single proto-language? Or did it emerge independently in multiple populations? The genetic evidence (language capacity predating population splits) favours monogenesis, but this remains debatable.
• Gradual vs. saltational emergence: Did language evolve incrementally over hundreds of thousands of years, or did a single mutation (Chomsky’s “great leap forward” hypothesis) enable the full capacity relatively suddenly?
• The Neanderthal question: Did Neanderthals have language? They possessed the hyoid bone, FOXP2, and large brains, but left minimal evidence of symbolic behaviour. The question is unresolved.
• The role of gesture: Some researchers (notably Michael Corballis) argue that language began as manual gesture and only later transitioned to speech. Sign languages’ full grammatical complexity supports the plausibility of this pathway, though direct evidence is unavailable.

The Linguistic Society of Paris lifted its 1866 ban in spirit, if not formally, during the late 20th century. The question of language origins is now considered a legitimate field of inquiry. The honest assessment of 160 years of renewed investigation: we know considerably more about when and where, but the how remains the hardest problem in the cognitive sciences.