Skip to content
NatureHQ

Behaviour and cognitionability

Do animals have language?

Language is not one thing. Take it apart into its properties, and animals have several of them — just never all at once.

Not as a whole system, no — but that answer hides the interesting part. Language is a bundle of separate properties, and animals have been shown to have several of them individually. Dolphins have learned arbitrary labels. Sperm whale codas have combinatorial structure. Humpback song is culturally transmitted across an ocean. What no animal has been shown to have is the whole bundle at once, and in particular the ability to refer to something that is not there.

The question is usually posed as though there were a line, with animals on one side and people on the other, and the research consisted of working out where the line falls. That is not what the research looks like. Language is not one thing: it is a set of properties that happen to occur together in humans — learned arbitrary symbols, combinatorial structure that changes meaning, the capacity to refer to what is absent, open-ended productivity, and transmission by culture rather than by genes. Take them apart and each can be looked for separately, and several of them turn up. A bottlenose dolphin invents a whistle for itself, other dolphins learn it, and playing it back to the owner makes it answer — that is a learned arbitrary label with a receiver, which is a real linguistic property. Humpback whale song sweeps across the Pacific and is replaced wholesale, which is cultural transmission of form. Sperm whale codas vary along more dimensions than anyone catalogued. None of that adds up to language, and the reason is worth stating plainly: nothing in the list requires an animal to say anything about what is not in front of it. Displacement — talking about the elephant who left, or the fruit that will be ripe next week — is the property that most obviously separates human language from everything else, and it is also the hardest to test in an animal that cannot be asked. So the honest position is neither the deflationary one nor the romantic one. Several animals have pieces of what language is made of. Nobody has found an animal with all of them, and it is genuinely unclear whether that is because no such animal exists or because the pieces that are missing are the ones our experiments are worst at detecting.

Developed coverage · 68% complete · reviewed 2026-08-31

What this page covers

The question is asked of every animal and answered properly for very few. What follows draws on the systems that have actually been tested against linguistic criteria: cetaceans, elephants, a handful of primates, songbirds, bees and bats.

Often confused with: Animal communication, which is a much larger and much better answered question

Quick facts

Learned arbitrary labels
Yes — dolphins, and probably elephants
Cultural transmission of form
Yes — humpback song crosses an ocean and is replaced wholesale
Talking about what is absent
Not demonstrated in any species
Grammar carrying meaning
Not demonstrated — the one direct test failed on the videotape

The short answer

No — and the reason that is an unsatisfying answer is the most interesting thing here.

The short answer

Do animals have language?

No animal has been shown to have the whole of it. Several have been shown to have parts of it — learned names, structure that combines, traditions passed down by learning — and the parts are real findings rather than consolation prizes.

Treating language as a single possession an animal either has or lacks is what makes "no" unsatisfying. Linguists stopped treating it that way in the 1960s. List the properties separately and ask, for each, whether anything has it. Learned arbitrary signals: yes, in dolphins, and apparently in elephants, who address each other without imitating the animal addressed. Signals about specific external things: yes, in vervet monkeys, whose alarm calls send listeners up a tree or into cover with no predator anywhere. Combinatorial structure: probably, in sperm whales. Cultural transmission of form: certainly, in humpback whales. Then the ones nobody has found. Reference to what is absent has never been demonstrated in a natural animal signal. Syntax carrying meaning has been directly tested once, in a signing chimpanzee, and it collapsed when the videotape was slowed down and most of his signs turned out to follow a teacher making the same sign a second earlier. That does not establish that apes cannot do it. It establishes that watching in real time cannot tell.

Check it for yourself

What language is made of

Five properties, tested separately. This is where the interesting answers are.

Every one of these is a property human language has. None of them is language by itself, and an animal having one says nothing about whether it has the others — which is exactly why the question has to be asked property by property rather than species by species.

Diagram

The properties of language, and what has been found

Each row is a property human language has. Green means demonstrated in at least one non-human species with a receiver test; amber means structure has been found but no animal has been shown to use it; grey means not demonstrated anywhere.

Language, taken apartEach row is a property human language has. Animals have several. Nothing has them all.Learned arbitrary signalsthe signal is invented, not inheritedDolphins · elephantsReference to external thingsthe call is about something out thereVervets · chickadeesCombinatorial structureparts recombine into more signalsSperm whalesCultural transmission of formthe signal itself is learned sociallyHumpbacks · orcasReference to what is absenttalking about what is not hereNothing, anywhereAmber: structure found in recordings, with no animal shown to use it.
The same explanation in words

A table of five properties. Learned arbitrary signals: demonstrated, in bottlenose dolphins and apparently African elephants. Reference to external things: demonstrated, in vervet monkeys and black-capped chickadees. Combinatorial structure: structure found in sperm whale codas, but no whale has been shown to treat the variants as distinct. Cultural transmission of form: demonstrated, in humpback whales and killer whales. Displacement, meaning reference to what is absent: not demonstrated in any species.

Several animals have one or two of the things that make human language what it is. Nothing has been shown to have the whole set.

Well supported

Good evidence backs this, though some details remain open.

Individual design features associated with human language — learned arbitrary labels, functionally referential signals, combinatorial structure, cultural transmission of form — are separately documented in non-human species. No non-human system has been demonstrated to combine open-ended productivity, syntactic structure carrying meaning, and displacement.

Who this applies to
Concerns natural communication systems. Claims about apes trained on human symbol systems are assessed separately and on their own evidence.
Studied in
Animalia

You may have heard

“Scientists have discovered that animals have language.”

Each discovery of this kind is a discovery of one property. Learned labels in dolphins, name-like calls in elephants, structure in sperm whale codas — all real, all genuinely surprising, and none of them the full set. The headline collapses "has a property of language" into "has language", and the two are separated by the properties nobody has found.

Why we rate it this way, and what the caveats are
Well supportedModerate confidence

Confidence is moderate rather than high because this is partly an argument from absence, and because the properties in question are hard to test in the wild. What supports it is that the best-studied systems have been examined specifically for these properties and have not shown them.

How far it can be extended

The absence is across everything examined so far; it is a statement about the current evidence rather than about what is possible.

Caveats

  • This is not a claim that animals are simple, or that their systems are worse. A signal that does exactly what the animal needs is not a failed language.
  • Absence of evidence is doing real work in this claim, and displacement in particular is close to untestable in a wild animal.
  • Every year or two a system is reported to have one more of these properties, and that trend may continue.

Still unanswered

  • Whether any animal can refer to something that is not present — the property that is both most diagnostic and hardest to test.
  • Whether the design-feature framework, built from human language, is the right way to describe systems that evolved for other purposes.

Last reviewed 2026-08-31

The evidence (4 studies)

  • Supports · primary

    Can an ape create a sentence?

    Terrace et al., 1979 · Science

    The most direct test of syntactic productivity in a non-human, and it did not survive frame-by-frame analysis of who signed first.

  • Context · supporting

    Contextual and combinatorial structure in sperm whale vocalisations

    Sharma et al., 2024 · Nature Communications

    The strongest current case for combinatorial structure in a natural system — and it identifies structure without identifying meaning or testing a receiver.

  • Qualifies · supporting

    Bottlenose dolphins can use learned vocal labels to address each other

    King and Janik, 2013 · Proceedings of the National Academy of Sciences

    Shows one language-like property genuinely present: learned, arbitrary, individually specific vocal labels that receivers respond to.

  • Qualifies · supporting

    Prairie dog alarm calls encode labels about predator colors

    Slobodchikoff et al., 2008 · Animal Cognition

    The origin of "prairie dogs have a language". The acoustic finding — that calls vary with an intruder’s size, shape and even shirt colour — is real and replicated. What is much thinner is the receiver work needed to show that other prairie dogs extract those distinctions, which is what would make the variation descriptive rather than merely present.

The result that changed how this is studied. In 1979 a team analysed some twenty thousand sign combinations from a chimpanzee who appeared to be producing sentences. Slowed down on videotape, most of his signs had been preceded within the previous second by a teacher making the same sign — too fast for anyone in the room to notice.

Based on Several animals have one or two of the things that make human language what it is. Nothing has been shown to have the whole set.

What would actually settle this

The experiments nobody has managed to run.

  • A demonstration that an animal can refer to something not present — the property that most clearly separates language from everything else, and the one no design has convincingly reached in a wild animal.
  • A playback showing that a change in signal *order* changes what a receiver does, which would be syntax carrying meaning rather than structure sitting in a recording.
  • For sperm whales: any test of whether the fine coda distinctions found by machine analysis are distinctions the whales themselves make.
  • For elephants: identification of what the label physically is, since the finding so far is that a model can predict the addressee, not what feature it is predicting from.

  • Is the design-feature framework the right instrument at all for systems that evolved to solve other problems?

    Why it matters: The properties were derived by describing human language and then looking for its parts elsewhere. A system that does something we have no word for would score zero on every criterion and still be remarkable.

  • Is reference to what is absent genuinely missing, or merely undetectable?

    Why it matters: Displacement is the single most diagnostic property, and no design has convincingly tested for it in a wild animal. Absence of evidence is carrying more weight here than anywhere else in this question.

  • Is any animal system open-endedly productive, rather than large but bounded?

    Why it matters: A repertoire of thousands of distinguishable signals is still a repertoire. Productivity means being able to construct one that has never been produced before and have it understood, and nothing has been shown to do that.

Follow the argument

The pieces of this question, each with its own evidence.

Follow the properties

  • Animal communication

    What a signal is, and how anybody establishes that one carries information

  • Sperm whale

    Codas, clans, and the analysis that became an “alphabet” in the reporting

  • Dolphin

    Signature whistles — the strongest case for a learned label

  • Animal culture

    Transmission by learning rather than by genes

  • Bioacoustics

    What machines can and cannot get out of a recording

The research behind this page

12 studies, newest first. Each one has a page explaining what it found and what it could not show.

2024Nature Communications

Contextual and combinatorial structure in sperm whale vocalisations

Coda structure varies systematically along dimensions the authors name rubato — smooth variation in overall duration — and ornamentation, an extra click added at the end of a coda.

2024Nature Ecology & Evolution

African elephants address one another with individually specific name-like calls

The model identified intended receivers better than chance, and elephants responded more strongly and more quickly to calls originally addressed to them.

2022PeerJ

Computational bioacoustics with deep learning: a review and roadmap

Deep learning has substantially improved detection and classification of animal sounds, particularly for large passive-acoustic datasets.

2019Science

Evolution of vocal learning and spoken language

Vocal production learning is documented in a small number of distantly related lineages, and the independently evolved brain circuits supporting it share a recurring feature: a direct projection from forebrain to the motor neurons controlling the vocal organ, which is absent in non-learners.

2016Royal Society Open Science

Individual, unit and vocal clan level identity cues in sperm whale codas

Different coda types carry information at different levels.

2016Scientific Reports

Everyday bat vocalizations contain information about emitter, addressee, context, and behavior

Classifiers recovered caller identity well above chance, and could identify the addressee and the behavioural context — food, mating, sleeping position, perch — from the call alone.

2013Proceedings of the National Academy of Sciences

Bottlenose dolphins can use learned vocal labels to address each other

Dolphins responded to their own synthetic signature whistle by calling back, and did not respond in the same way to the whistles of others.

2008Animal Cognition

Prairie dog alarm calls encode labels about predator colors

Calls differed consistently with the colour of the walker’s shirt, as well as with the size and shape of the intruder.

2001Science

Dynamics of the vocal imitation process: how a zebra finch learns its song

Song development proceeds from highly variable, unstructured subsong through progressive differentiation of individual syllables, with different syllables converging on their targets at different times and rates rather than the song improving as a whole.

1987Animal Learning & Behavior

Acquisition of the same/different concept by an African Grey parrot (Psittacus erithacus): learning with respect to categories of color, shape, and material

The parrot answered correctly well above chance with novel objects, and could report which category the objects differed in rather than only that they differed.

1979Science

Can an ape create a sentence?

The great majority of the chimpanzee's signs were prompted: they reproduced signs a teacher had made moments earlier, or were produced while the teacher was signing.

1965Zeitschrift für Tierpsychologie

The role of auditory feedback in the control of vocalization in the white-crowned sparrow

Birds deafened before singing produced highly abnormal, unstructured song even when they had already heard a tutor.

This page is a stop on a longer route

A guided journey reads several subjects in a deliberate order, with an argument for why one follows another. You can join in the middle.

Where to go from here

Each of these follows from something on this page — a relationship in the evidence, a claim people ask about, or the next mechanism along.

How complete this page is, and what it is still missing

NatureHQ publishes its own gaps. This page is at 68% completeness against what we would call a finished subject, and was last reviewed on 2026-08-31. It carries 5 claims and answers 4 mapped search questions.

  • no popular claim about this subject has been checked yet
  • Birdsong is a major part of the vocal-learning evidence and is treated only in passing here.
  • Trained-ape symbol research after 1979 reached different conclusions with different methods, and is summarised rather than assessed in depth.
  • The philosophical literature on what would count as animal language is not covered at all.