Skip to content
NatureHQ

Claim check

Do sperm whales have a language?

Early evidence

Real findings exist, but too few or too new to settle the question.

What was found is structure — the clicks vary along more dimensions than anyone had catalogued, and the variation tracks conversational context. No meaning has been demonstrated, and the paper does not claim any.

The claim as it circulates

“Scientists have decoded a sperm whale alphabet and are close to translating their language.”

Where you may have met it: Coverage of the 2024 machine-learning analysis of sperm whale codas; Google autocomplete for "what do sperm whales talk about"; Artificial-intelligence and animal-communication reporting generally

Taking it apart

What was claimed
That sperm whales possess a language, that its alphabet has been identified, and that translation is within reach.
What was actually observed
Machine-learning analysis of roughly 9,000 codas from one identified Eastern Caribbean clan, examining each coda in the context of those around it, identified two previously uncatalogued dimensions of variation — smooth variation in overall duration, and an extra click added at the end. These vary with conversational context rather than randomly, and combine with known rhythm and tempo categories to yield a far larger inventory of distinguishable coda types. Separately, coda repertoires are known to define socially learned clans that cross-cut geography.
What the evidence supports
That sperm whale vocalisation carries combinatorial, context-dependent structure across a much larger signal space than the traditional catalogue allowed; and, from earlier work, that which repertoire a whale uses is learned from the whales it grows up among rather than inherited or determined by location.
What it does not support
Semantics. No coda type has been linked to any referent, and none has been shown to produce any specific response in a listening whale. Combinatorial structure occurs in birdsong without meaning. The analysis also covers one clan in one region, so even the structural finding is not yet known to generalise.

Where researchers disagree

Researchers agree on what was measured and differ on what it portends. One reading is that a combinatorial signal space of this size in a long-lived social animal is unlikely to be meaningless, and that the tools to test for semantics now exist. The other is that birdsong and humpback song both have rich combinatorial structure with no semantic content, so structure alone predicts nothing — and that the history of animal-communication research is largely a history of structure being over-read. Both positions accept that no coda has been tied to a referent or to a behavioural consequence.

The gap between the paper and the coverage is unusually instructive. The researchers mapped a search space and were careful to say so. What got reported was that the search had succeeded.

It is worth separating the two sperm whale sounds, because coverage merges them constantly. Echolocation clicks are foraging sonar — enormously loud, beamed forward, aimed at squid. Codas are short rhythmic patterns, far quieter, exchanged while socialising. The decibel figures belong to the first; the conversations belong to the second.

The genuinely established result here is the one that gets less attention: repertoire is socially learned. Two units sharing the same water can share nothing, while units thousands of kilometres apart share a repertoire — ruling out place and inheritance and leaving learning. Whales learn what to say from other whales. That is a large claim, and it is a different claim from having something to say.

The claims underneath

Each one carries its own evidence, scope and caveats. Expand any of them to reach the studies.

The clicks have far more structure than anyone knew, and no demonstrated meaning

Emerging evidence

Real findings exist, but too few or too recent to be settled.

Machine-learning analysis of roughly 9,000 codas from an identified Eastern Caribbean clan identified two previously uncatalogued dimensions of variation — smooth variation in overall duration, and the addition of an extra click at the end of a coda — which vary with conversational context rather than randomly and combine with existing rhythm and tempo categories to yield a substantially larger inventory of distinguishable coda types. No semantic content has been demonstrated: no coda type has been linked to any referent, and no behavioural consequence of a given type has been established.

Who this applies to
one identified clan in the Eastern Caribbean
Studied in
Physeter macrocephalus

You may have heard

“Scientists have discovered a sperm whale alphabet or language”

What was found is that the signal varies along more dimensions than anyone had catalogued, and that the variation tracks the conversational context rather than being noise. That is a real and substantial result, and it is a result about *structure*. Nothing was linked to a meaning, no coda was shown to make a whale do anything, and the paper does not claim otherwise. Birdsong has rich combinatorial structure and no semantics. The honest position is that the search space has been mapped and the search has not yet been run.

Why we rate it this way, and what the caveats are
Emerging evidenceModerate confidence

The structural analysis is rigorous and the dataset large for this species, but it covers one clan, and the finding is about the shape of the signal rather than about what it does.

Caveats

  • One clan in one region; whether the same structure exists elsewhere is untested.
  • Context is coded from the surrounding exchange, not from any independently known referent.
  • Combinatorial structure occurs in birdsong and whale song without semantics, so structure alone does not indicate a language.

Still unanswered

  • Does any coda type reliably predict a behavioural response in the receiving whale?
  • Is the additional structure information-bearing, or a by-product of how codas are produced?

Last reviewed 2026-08-11

The evidence (2 studies)

Sperm whales learn what to say from the whales they grow up with

Well supported

Good evidence backs this, though some details remain open.

Sperm whale coda repertoires cluster into a small number of distinct types — vocal clans — whose distribution does not correspond to geography: units from different clans overlap in the same waters without sharing repertoire, while units thousands of kilometres apart share one. Clan membership is not predicted by maternal genetic lineage alone, and units associate preferentially with others of the same clan. This satisfies the standard definition of culture as group-typical behaviour transmitted socially rather than genetically.

Who this applies to
sperm whales, established chiefly in the South Pacific and Caribbean
Studied in
Physeter macrocephalus
Why we rate it this way, and what the caveats are
Well supportedHigh confidence

The design excludes the two obvious alternatives — geography and maternal inheritance — by showing repertoire varying independently of both, and the pattern has been recovered repeatedly.

How far it can be extended

Clan structure has been recovered in multiple ocean basins, though recording effort is uneven and some regions are unsampled.

Caveats

  • Repertoire clustering depends on the similarity measure chosen.
  • Social transmission is inferred by excluding alternatives rather than observed directly.
  • Recording effort is uneven, so clan counts are lower bounds.

Still unanswered

  • What function does clan membership serve — does sharing a repertoire bring a measurable benefit?

Last reviewed 2026-08-11

The evidence (3 studies)

Go deeper

Sperm whale

A third of it is a nose rebuilt into a sound cannon, and it uses that to find squid a kilometre down in the dark.

Last reviewed 2026-08-11