Origin of Human Language
Language leaves no fossils, so when and how it arose is unknown; estimates run from about 2.5 million years ago to under 200,000.
Open in the interactive tree →Speech leaves no direct trace, and the Linguistic Society of Paris banned the topic in 1866. Views range from a gradual build-up from primate abilities to a sudden step such as Chomsky's Merge. Many researchers now treat language as several capacities, such as vocal learning, structure and social cooperation, each with its own history.
As of October 2026
A 2025 Science paper by Arnon, Fitch, Kirby, Fisher and others proposes a biocultural framework that combines biology with cultural transmission. In April 2025 Science reported that wild bonobo calls combine with non-trivial compositionality, which the authors take to suggest the trait is at least 7 million years old (an inference). Project CETI's 2024 paper on nearly 9,000 sperm whale codas found rhythm, tempo, rubato and ornamentation combining freely, but what the codas mean is unknown. No dating of language origin has consensus.
What is missing
- Dated evidence for when speech and grammar appeared
- Tests that separate human language from animal call combination
- Models of how shared signals turn into grammar, checked against experiments
- Decoded meaning in other species' communication
- Agreement on what counts as language
Becomes possible once solved
- Knowing what makes human language special
- A principled test for language-like systems in animals and machines
- Better theories of how children and communities build grammar
Open steps
- Indirect evidence for dating Low AI leverageSpeech leaves no fossils; dates range from under 200,000 years (Homo sapiens) to 2.5 million (proto-language in Homo habilis). Find testable proxies.
- Which capacities are uniquely human? Medium AI leverageCompare vocal learning, signal combination and cooperation across species; bonobo calls show non-trivial compositionality (Science, 2025), so combining alone may not be unique.
- Decode animal communication Medium AI leverageTest whether machine learning can find structure and meaning in sperm whale codas (nearly 9,000 analysed) and other species; the meaning of the codas is unknown.
- Simulate how grammar emerges Medium AI leverageModel populations in which signals gain structure through repeated learning and use, then compare with experiments and with new sign languages such as Nicaraguan Sign Language.
- Agree what counts as language Low AI leverageDisputes over gradual versus sudden origin rest on definitions; a framework treating language as several capacities (Science 2025) gives testable parts.
Where AI could help
Medium AI leverage. AI finds patterns in animal sound corpora and runs evolution models, but it cannot show what early humans spoke.
- Find structure in large collections of animal calls
- Run models of how communities turn signals into grammar
- Compare call combinations across species with shared methods
Shown so far
- Project CETI's May 2024 Nature Communications paper, led by members of its machine-learning team, analysed nearly 9,000 sperm whale codas and found freely combining rhythm, tempo, rubato and ornamentation. source
Prerequisites
- Spoken Languageat least ~135,000 years ago?
- Evolution by natural selection1859
- Formal Grammars (Chomsky)1956