Cetacean Communication
Glossary

The scientific study of sound production, propagation, and reception in living organisms. In the context of cetacean research, bioacoustics covers the full range of sounds dolphins and whales produce — from echolocation clicks to social whistles to burst-pulse vocalizations — and the methods used to record, analyze, and interpret them.

The field has been transformed in the last decade by machine learning. Where analysts once had to manually review spectrograms, models can now process thousands of hours of recordings to detect patterns, classify call types, and track individual animals by their acoustic signatures. This is the technical foundation that makes projects like DolphinGemma possible.

Where the popular coverage goes wrong: Bioacoustics is a measurement science. It tells you what sounds are being made and in what contexts. It does not, by itself, tell you what those sounds mean. The gap between "we can classify dolphin calls with 90% accuracy" and "we understand dolphin communication" is large, and most journalism collapses it.

See also: Echolocation, Signature Whistle, DolphinGemma

A category of dolphin vocalizations distinct from whistles and echolocation clicks. Burst-pulse sounds are rapid sequences of clicks produced at such high frequency that they blur together and sound like buzzes, squawks, or creaks to human ears. They're used primarily in social contexts — aggression, excitement, courtship, mother-calf bonding — rather than for navigation.

Burst-pulse sounds are among the least studied aspects of dolphin communication because they're difficult to analyze with traditional methods. Machine learning approaches are beginning to find structure in them that wasn't apparent before. Some researchers believe burst-pulse communication may be more semantically rich than whistles, but the evidence is still preliminary.

See also: Signature Whistle, Echolocation

The order of marine mammals that includes dolphins, porpoises, and whales. Cetaceans evolved from land mammals roughly 50 million years ago and are divided into two main groups: toothed whales (Odontoceti), which includes all dolphins and porpoises and produces echolocation clicks, and baleen whales (Mysticeti), which filter-feeds and produces the low-frequency songs associated with humpbacks.

Most cetacean communication research focuses on odontocetes — specifically bottlenose dolphins (Tursiops truncatus), which are the most studied species due to their adaptability to captive research conditions and their large coastal populations. Results from bottlenose dolphin research don't necessarily generalize to other cetacean species, a caveat that gets dropped more often than it should.

See also: Signature Whistle, Echolocation

The study of animal cognition — mental processes including perception, memory, learning, and decision-making — in the context of natural behavior. The term was coined by Donald Griffin in the 1970s when he argued that it was scientifically legitimate to ask whether animals had mental states, against the then-dominant behaviorist position that mental states were either absent or unresearchable in non-human animals.

Cognitive ethology is the intellectual framework within which dolphin communication research sits. Whether dolphins have theory of mind, whether they experience something like grief, whether their vocalizations carry propositional content — these are cognitive ethology questions, and they require both behavioral evidence and careful reasoning about what that evidence actually supports.

The methodological challenge: It's hard to test cognitive claims in wild animals without contaminating the result with your own assumptions about what you're looking for. Well-designed cognitive ethology experiments control for simpler explanations before invoking complex cognitive processes — a principle frequently violated in popular accounts of animal intelligence.

See also: Theory of Mind, Umwelt

The frequency trajectory of a whistle over time — essentially the "shape" of the sound when plotted on a spectrogram. Dolphin signature whistles are defined primarily by their contour: the specific pattern of frequency rises, falls, and inflections that makes each individual's whistle distinctive.

Contour analysis is the foundational method for studying dolphin vocal identity and recognition. DolphinGemma and similar AI systems are largely trained on contour data. The precision of modern contour extraction algorithms — which can automatically identify and isolate whistle contours from hours of ambient ocean noise — is what makes large-scale bioacoustics research feasible today.

See also: Signature Whistle, Spectrogram, DolphinGemma

An AI model developed by Google in collaboration with the Georgia Tech Research Institute (GTRI) and the Wild Dolphin Project, designed to analyze bottlenose dolphin vocalizations. Released in 2025, it uses a transformer architecture fine-tuned on a dataset of wild dolphin recordings and can classify whistle contours, identify call types, and model temporal patterns in vocal sequences.

DolphinGemma is significant for two reasons. First, it demonstrates that large-scale AI resources can be applied productively to bioacoustics problems that were previously limited by computing constraints. Second, it's released as an open model, which means researchers can build on it without access to Google's infrastructure.

What it does and doesn't do: DolphinGemma is a classification and pattern-recognition system. It can identify that a particular sound belongs to a category with high accuracy. It cannot, at this stage, determine what those sounds mean in context, whether they have compositional structure resembling grammar, or how dolphins use them referentially. Coverage of DolphinGemma that describes it as a "translation" tool is incorrect.

See also: Bioacoustics, Contour, Signature Whistle

Consistent, group-specific differences in vocalization patterns that are learned rather than genetically determined. Orca dialects are the best-documented case: different pods have distinct repertoires of calls, and calves acquire their pod's dialect by learning from family members. The dialects diverge over generations, such that distantly related pods have more different call repertoires than closely related ones.

Bottlenose dolphin dialects are less well-characterized but there is evidence for group-level variation in whistle repertoires in some wild populations. The existence of dialects has significant implications: it demonstrates cultural transmission of vocal behavior — a capacity previously thought to be rare or absent outside of humans and some birds.

See also: Signature Whistle, Vocal Learning

A biological sonar system in which an animal emits sound pulses and interprets the returning echoes to navigate and locate objects. In dolphins, echolocation clicks are produced in the nasal passages and projected forward through the melon — a fatty organ in the forehead that acts as an acoustic lens. The returning echoes are received through the lower jaw and transmitted to the auditory system.

Dolphin echolocation is extraordinarily precise: dolphins can detect objects a few centimeters in diameter at distances of 100 meters or more, distinguish between different materials, and form detailed three-dimensional representations of their environment. The system operates at frequencies up to 150 kHz — far above the range of human hearing.

Echolocation and communication: Echolocation clicks and social vocalizations use the same acoustic apparatus, and there is ongoing debate about whether echolocation is ever used communicatively — whether dolphins can share echolocation-derived information with each other. Some evidence suggests they can, but the mechanism is not well understood.

See also: Bioacoustics, Melon, Spectrogram

Changes in the frequency (pitch) of a sound over time. Dolphin signature whistles are defined almost entirely by their frequency modulation pattern — the specific trajectory of pitch changes that makes each whistle unique. Frequency modulation in dolphin calls carries social information: research suggests that the rate and degree of frequency modulation in non-signature calls can indicate emotional arousal, with more rapid modulation associated with heightened states.

See also: Contour, Signature Whistle

The rounded forehead structure of toothed whales, composed of fatty tissue with precise acoustic properties. The melon functions as an acoustic lens that focuses outgoing echolocation clicks into a directional beam. Dolphins can control the shape of the melon to some degree, allowing them to adjust the focus and direction of their echolocation signals.

The melon's role in dolphin anatomy is part of why dolphin heads look the way they do — the forehead is not merely cosmetic but a sophisticated acoustic instrument. Understanding the melon is relevant to any attempt to design artificial devices for two-way underwater acoustic communication with dolphins.

See also: Echolocation

A behavioral experiment designed to test whether an animal recognizes its own reflection. An animal is marked with a visible but odorless mark while anesthetized, then placed in front of a mirror. If the animal investigates the mark on its own body (rather than treating the reflection as another individual), this is taken as evidence of self-recognition.

Bottlenose dolphins passed the mirror test in research by Diana Reiss and Lori Marino published in 2001 — one of the first non-primate species to do so, and the first marine mammal. The test has methodological critics who argue that passing it doesn't necessarily indicate human-like self-awareness, and that other indicators of self-awareness should be considered alongside it. The debate over what the mirror test actually measures is ongoing and productive.

See also: Theory of Mind, Cognitive Ethology

The use of underwater hydrophones or recording arrays to detect and study animals without emitting any sound of your own. Passive means listening only. PAM is the standard ethical approach in cetacean communication research — it allows long-term, large-scale data collection without disturbing the animals being studied.

Modern PAM systems can record continuously for months at a time, producing enormous datasets. The limiting factor in cetacean communication research has historically been not recording capacity but analysis capacity — the ability to extract meaningful information from thousands of hours of recordings. This is where machine learning has had the greatest impact.

See also: Bioacoustics, DolphinGemma

Cetacean Translation Initiative — a large-scale research effort using machine learning to analyze sperm whale communication. Founded in 2020 with backing from major technology and academic institutions, CETI aims to record and decode sperm whale clicks (called codas) using AI models trained on the largest dataset of cetacean vocalizations ever assembled.

CETI is the most ambitious interspecies communication project currently active, and it's explicit about the analogy to AI language models: the hypothesis is that if you can collect enough data on a communication system, machine learning can find structure in it. Whether sperm whale codas have the kind of structure that would count as "language" is exactly the question the project is trying to answer. Progress has been significant at the data collection and classification layer; the semantic layer remains unsolved.

See also: DolphinGemma, Bioacoustics

A stereotyped, individually distinctive whistle that bottlenose dolphins develop in the first year of life and use to identify themselves throughout their lives. Each dolphin's signature whistle has a unique frequency contour — a specific pattern of pitch changes — that functions analogously to a name. Dolphins produce their signature whistle when separated from companions, when approaching others, and in contexts requiring individual identification.

Key established findings on signature whistles: dolphins respond selectively to their own signature whistle when it's played back to them; they answer to their signature when called by others; they can recognize the signatures of individuals they haven't encountered in over 20 years; and they copy each other's signatures as a form of address (essentially calling someone by name before engaging with them).

What's still unclear: Whether the signature copying behavior implies a referential understanding of the other as an individual (the strongest interpretation) or is better explained as a simpler learned association. Whether the modifications dolphins make to their signatures in different social contexts carry semantic content. Whether species other than bottlenose dolphins have comparable systems.

Why it matters for AI research: Signature whistles are the most tractable target for machine learning in cetacean communication because they're stable, individually distinctive, and well-documented. Most large-scale bioacoustic AI work — including DolphinGemma — begins with signature whistle identification and classification. They're the entry point, not the endpoint.

See also: Contour, DolphinGemma, Vocal Learning

A visual representation of the frequency content of a sound over time, with time on the horizontal axis, frequency on the vertical axis, and amplitude represented by color or brightness. Spectrograms are the primary tool for analyzing and displaying cetacean vocalizations — a signature whistle appears as a distinctive curved line on a spectrogram, and its unique contour shape is immediately visible.

Machine learning models for cetacean communication are often trained on spectrogram images rather than raw audio, treating the acoustic classification problem as an image recognition problem. This approach has proven effective and is the basis for how DolphinGemma processes whistle data.

See also: Bioacoustics, Contour

The capacity to attribute mental states — beliefs, desires, intentions, knowledge — to other individuals, and to understand that those mental states can differ from your own. "Theory of mind" in its full form means understanding that other agents have minds, that those minds can hold false beliefs, and that behavior is driven by mental states rather than just by external stimuli.

Theory of mind in dolphins has been studied through behavioral experiments. The clearest evidence comes from research showing that dolphins in competitive contexts strategically withhold information from competitors — behavior that implies some model of what another individual knows. The evidence is suggestive but not definitive; controlled experimental demonstrations of false-belief understanding (the standard test in comparative psychology) are harder to design for dolphins than for primates.

Why it matters for communication: If dolphins have theory of mind, then their signature-copying behavior (addressing another by name, anticipating that the other will recognize and respond to the name) implies a model of the other individual's mental states. That would make dolphin "naming" genuinely similar to human naming in its cognitive underpinnings, not just superficially. This is a contested interpretation, but it's the direction the evidence points.

See also: Mirror Self-Recognition Test, Cognitive Ethology

A concept introduced by the biologist Jakob von Uexküll in 1909 to describe the subjective sensory and perceptual world of an organism — the totality of signals from the environment that are meaningful to it, filtered through its sensory and cognitive apparatus. A tick's umwelt consists primarily of temperature, light/dark, and the smell of butyric acid; a bat's includes a rich three-dimensional acoustic map; a dog's is dominated by smell in ways humans can barely conceptualize.

The umwelt concept is useful and underused in cetacean communication research because it's a check against projecting human perceptual frameworks onto dolphin experience. A dolphin's umwelt is organized around acoustic information at frequencies and resolutions humans don't experience. When we study dolphin communication, we are studying a system designed for and operating within an umwelt that is largely inaccessible to us.

The practical implication: Communication research that focuses only on what dolphins do with their voices may miss significant communicative information that operates through other channels — physical touch, hydrodynamic signals, perhaps shared echolocation imagery. Ed Yong's An Immense World (2022) is the most accessible recent treatment of the umwelt concept across species.

See also: Cognitive Ethology, Echolocation

The ability to acquire new vocalizations through imitation, as opposed to producing only genetically pre-programmed calls. Vocal learning is rare in the animal kingdom — it has been confirmed in humans, some songbirds, parrots, hummingbirds, bats, elephants, seals, and cetaceans, but is absent or severely limited in most other mammals including other primates.

Dolphins are confirmed vocal learners. Their signature whistles are not genetically determined but are acquired early in life through a process that involves both imitation and innovation. They can also imitate novel sounds produced by humans or other species — a capacity that has been demonstrated experimentally. Vocal learning is one of the prerequisites for anything resembling language, and its presence in dolphins is part of what makes cetacean communication research particularly interesting.

The comparative relevance: The convergent evolution of vocal learning in such disparate groups — dolphins and songbirds share no recent common ancestor with vocal learning — raises deep questions about what cognitive and ecological conditions select for it. Understanding why vocal learning evolves is part of understanding what communication systems can do.

See also: Signature Whistle, Dialect