Study Finds Sperm Whales Communicate Using Complex Alphabet That Includes Vowels

Sperm whales produce rapid-fire sequences of clicks that echo through the deep ocean. For decades, scientists have known these clicks carry social information. Now a team of linguists and marine biologists claims to have found something far more specific: vowel-like sounds embedded in those clicks, patterned in ways that parallel the phonological rules of human languages.

The study, published April 16, 2026 in Proceedings of the Royal Society B, identifies two distinct vowel types — dubbed "a-codas" and "i-codas" — in recordings of sperm whales off the coast of Dominica [1]. The research represents the latest and most linguistically detailed output of Project CETI (Cetacean Translation Initiative), a multi-institution effort that has been systematically expanding what we know about sperm whale communication since its founding in 2020 [2].

But the central question remains unresolved: are these patterns evidence of something language-like, or do they reflect acoustic and social dynamics that have nothing to do with symbolic communication?

What Exactly Was Found

The paper, led by UC Berkeley linguist Gašper Beguš alongside co-authors Maksymilian Dąbkowski, Ronald L. Sprouse, David F. Gruber, and Shane Gero, analyzed the internal acoustic structure of sperm whale codas — the short bursts of clicks that whales exchange during social interactions near the ocean surface [1].

Previous research, including Project CETI's landmark 2024 paper in Nature Communications, had already expanded the known coda inventory from 21 types to 156 distinct codas, each defined by combinations of tempo, rhythm, rubato (tempo variation within a sequence), and ornamentation (extra clicks added to a pattern) [3]. That study established the combinatorial structure of whale communication — the idea that a finite set of building blocks could be mixed and matched to produce a large number of distinct signals.

The new study goes further by examining what happens when the silent gaps between clicks are removed and the remaining acoustic pulses are analyzed as continuous sound. Using spectrograms — visual representations of sound frequency over time — the researchers identified two categories of formant structure within the click pulses. Formants are resonance frequencies that define vowel quality in human speech; "a-codas" showed a single formant, while "i-codas" displayed two formants, analogous to the difference between the vowels "a" and "i" in human languages [1][4].

Source: Project CETI / Nature Communications (2024), Proceedings B (2026)

Data as of Apr 16, 2026CSV

Structural Parallels to Human Language

The researchers did not stop at identifying two vowel types. They found that these vowels behave according to distributional rules that mirror patterns documented across human languages [1].

A-codas tend to be longer in duration than i-codas — a property the authors compare to intrinsic vowel length differences seen in many human languages [4]. I-codas, meanwhile, show a bimodal distribution, with evidence of both short and long variants, paralleling the short/long vowel contrasts found in Latin, Arabic, and other languages with phonemic length distinctions [1].

The team also documented what they describe as coarticulation — the phenomenon where the quality of one sound is influenced by the sounds immediately adjacent to it. In human speech, coarticulation is pervasive: the "n" in "tenth" is produced differently from the "n" in "noon" because of the surrounding sounds. The researchers report that the first click in a sperm whale coda can be shaped by the quality of the immediately preceding coda, suggesting a similar process [1][4].

Some coda sequences also showed what the researchers term diphthong-like transitions — gradual shifts from one vowel quality to another within a single coda, comparable to the vowel glide in the English word "coin" [5].

"On the surface, they sound like this alien, ocean intelligence that has nothing to do with us," Beguš said. "But when you actually look at it closely, you realize, 'Oh, we're way more similar'" [4].

The Dataset: 15 Whales, One Ocean Basin

The study draws on recordings from the Dominica Sperm Whale Project (DSWP), a long-running field effort that has been tracking a community of sperm whales in the Eastern Caribbean since the mid-2000s. The analyzed dataset comprises over 1,100 codas from a population of 15 females and calves, with four particularly vocal individuals — nicknamed Atwood, Fork, TBB, and Pinchy — contributing a large share of the data [6]. The recordings were made between 2014 and 2018 using hydrophones and animal-worn acoustic biologging tags [3].

This geographic and demographic concentration is one of the study's acknowledged limitations. The whales studied belong to a single social clan in one ocean basin. Sperm whale society is organized into clans that span thousands of kilometers and contain thousands of individuals, with each clan distinguished primarily by its vocal dialect [7]. Two or more clans may occupy the same waters but socialize only within their own group, and dialect differences are learned socially within matrilines — groups of mothers and their offspring [7].

Whether the vowel patterns hold across other clans and ocean basins is unknown. The Eastern Caribbean population represents one cultural group within a species that is distributed across every major ocean. Other clans may use different systems entirely, or the same vowel categories may prove universal. Without data from additional populations, neither possibility can be ruled out.

The Technology Behind the Claims

Project CETI has from the outset combined marine biology with computational tools. The 2024 Nature Communications study used machine learning — specifically generative adversarial networks (GANs) — to detect statistical patterns in the coda dataset [3][8]. The new vowel study relies more on classical phonological analysis, with linguists applying spectrographic methods standard in human speech research [1].

The project's scientific roadmap, published as a peer-reviewed paper in iScience, outlines a multi-phase approach: collect massive datasets, detect communication units and language-like structures computationally, and then validate models through interactive playback experiments in which researchers play synthesized codas back to whales and observe their responses [9].

Beguš has stated that his team hopes to understand roughly 20 sperm whale expressions — including verbs related to diving and sleep — by 2031 [1]. David Gruber, a co-author and Project CETI leader, has noted that scientists are "already starting to predict simplistic behaviors, like diving, based on their voices," but emphasized "we have a long way to go to truly understand their complex communications" [6].

Source: OpenAlex

Data as of Jan 1, 2026CSV

The Skeptics

Not everyone is persuaded that the linguistic framing is appropriate. Luke Rendell, a marine mammal scientist at the University of St. Andrews who has studied sperm whale communication for years, has been among the most vocal critics of the "alphabet" interpretation.

"The way the tempo variation is used is completely different to how we use elements of an alphabet to construct linguistic expression," Rendell told Science magazine. "There's no evidence of that, and it's not a super helpful interpretation because it forces everything into a restricted and somewhat over-sold perspective of 'is it like human language or not'" [10].

Rendell's alternative hypothesis is that coda exchanges may function primarily as social bonding mechanisms. Given the importance of clan cohesion in sperm whale society, he suggests that the experience of matching coda tempos could strengthen mutual connection — with the synchrony itself being the point, rather than any information encoded in the signal [10].

"Machine learning is great for finding patterns in large datasets," Rendell cautioned, "but it doesn't create meaning" [10].

Simon Townsend, a comparative communication researcher, has raised a related concern: the importance of determining "to what extent these sounds also play a role in meaning" [6]. Identifying combinatorial structure — patterns that can be described using the vocabulary of linguistics — is not the same as demonstrating that those patterns carry semantic content, the way words and sentences do in human language.

The strongest falsification test would involve playback experiments: if researchers can demonstrate that whales respond differently to codas with different "vowel" structures in controlled conditions, that would support the claim that the distinctions are functionally meaningful. Without such experiments, the patterns documented so far could reflect acoustic physics, individual variation, or social coordination rather than symbolic communication.

The Gap Between Structure and Meaning

The researchers themselves acknowledge this gap directly. "While in human languages, different tones can be associated with different meanings, the meanings conveyed by sperm whale codas have not been established," the study states [1].

What has been established is that codas are produced most actively during surface socializing — when whales are in close physical contact — and at the onset of deep foraging dives [9]. Codas are exchanged in duet-like sequences between two or more whales, with apparent turn-taking: individuals respond within two seconds of each other, frequently overlapping and matching identical call types [9]. Individual codas carry information about the caller's identity at three levels: individual, family, and clan [7].

But whether specific coda sequences correlate with specific behaviors — whether there is a "dive" coda distinct from a "danger" coda or a "food" coda — remains unproven. The study investigates structural elements, not semantics. This places sperm whale communication research at a stage roughly analogous to knowing that a foreign text uses an alphabet without being able to read any of the words.

For context, other well-studied animal communication systems have clearer behavioral correlations. Vervet monkeys produce distinct alarm calls for eagles, leopards, and snakes, with each call triggering a specific escape behavior [11]. Honeybee waggle dances encode the distance and direction of food sources with measurable precision. Sperm whale codas, despite their structural complexity, have not yet been linked to specific referents with comparable confidence.

Population Loss and the Dialects We May Never Hear

The question of how many distinct communication systems exist among sperm whales is entangled with the species' population history. Commercial whaling devastated sperm whale numbers over two centuries. An estimated 184,000 to 236,000 sperm whales were killed in the 19th century, and at least 770,000 more were taken in the 20th century, mostly between 1946 and 1980 [12].

Source: Whitehead (2002), Pirotta et al. (2022) in Scientific Reports

Data as of Nov 14, 2022CSV

Pre-whaling population estimates range from 1.1 million [12] to nearly 2 million [13], depending on the model used. A 2022 study in Scientific Reports calculated a global estimate of roughly 736,000 sperm whales as of 1993, with a current population estimated between 300,000 and 450,000 [13]. The International Whaling Commission moratorium on commercial whaling, adopted in 1986, halted most direct killing, but the population has not recovered to pre-whaling levels [14].

Because sperm whale clans are culturally defined — with dialects learned rather than genetically inherited — population bottlenecks carry an additional cost beyond simple numbers. When a clan is reduced to a handful of individuals, its entire vocal tradition may be lost. There is no way to reconstruct a dialect that was never recorded. The cultural diversity of sperm whale communication before industrial whaling is permanently inaccessible, and it is plausible that distinct phonological systems — including vowel structures unlike anything observed in the Dominica population — have already gone silent.

Legal and Ethical Implications

The framing of whale communication as "language" or "alphabet" carries implications beyond marine biology. The 2010 Declaration of Rights for Cetaceans, drafted by a group of whale scientists and ethicists, asserted that all cetaceans should be recognized as persons with rights to life, liberty, and wellbeing [15]. India's 2013 declaration of dolphins as "non-human persons" and the 2024 He Whakaputanga Moana treaty — in which Pacific Indigenous leaders from the Cook Islands, French Polynesia, New Zealand, and Tonga recognized whales and dolphins as legal persons — show that this framing has real legal traction [16].

In the United States, the Marine Mammal Protection Act prohibits the "take" of marine mammals, including harassment, but does not confer personhood or language-based protections [17]. The Fordham Law Review has examined how advances in animal communication research could reshape legal arguments for expanded protections, noting that demonstrated communication of preferences by an animal "may necessitate that we seriously consider conferring legal rights to those animals" [18].

Opponents of expanded legal personhood for cetaceans argue that structural complexity in a communication system does not demonstrate the kind of intentional, referential language use that would justify the analogy to human personhood. From this perspective, the gap between "has vowels" and "has something to say" is precisely the gap that matters for legal and ethical reasoning.

What Comes Next

Project CETI's roadmap calls for playback experiments as the next major phase — testing whether whales respond meaningfully to synthetic codas constructed with specific vowel structures and combinatorial features. These experiments would begin to address whether the patterns identified statistically in the recordings carry functional significance for the animals themselves.

The project also plans to expand data collection beyond Dominica, which would test whether the vowel system generalizes across clans. Given that sperm whale clans are culturally distinct and geographically widespread, finding similar vowel structures in Pacific or Indian Ocean populations would substantially strengthen the case for a species-wide phonological system. Finding different systems would be equally significant, suggesting that sperm whale communication is even more culturally diverse than current models predict.

Beguš and his collaborators have been transparent that their work describes structure, not meaning. But the structure they describe is striking: a communication system with combinatorial elements, vowel-like categories that follow distributional rules, coarticulation effects, and length contrasts — features that, in human speech, are hallmarks of phonological organization. Whether sperm whales are "speaking" in any sense recognizable to linguists, or whether these patterns serve an entirely different function that human linguistic categories cannot capture, is the question that will define the next decade of cetacean communication research.

What is already clear is that sperm whales produce signals of a complexity that, until recently, few scientists believed possible outside of humans. Whatever these signals turn out to mean — or not mean — they demand a reckoning with assumptions about the boundaries of sophisticated vocal communication in the animal kingdom.