AI helps researchers hear patterns in dolphin speech — but a true translation is still far away

AI is helping Denise Herzing’s Wild Dolphin Project identify patterns in dolphin vocalizations and test early two-way signaling with Atlantic spotted dolphins, but the researchers say a true translation of dolphin language remains far off. The work, reported by WIRED, shows how machine learning is speeding up animal-communication research without pretending to have cracked interspecies speech.

What the Wild Dolphin Project is trying to do

For decades, the Wild Dolphin Project has built up a large archive of underwater video and audio, giving the team a rare long-term record of how dolphins behave and sound in the wild. That data has now become the raw material for a new phase of research: using AI to sort through vocalizations and search for structure that humans could miss.

Herzing, who founded and directs the project, told WIRED that the goal is not to leap straight to a machine-made dolphin-to-English dictionary. Instead, the team is using AI to organize sound patterns, relate them to behavior, and explore whether a limited form of back-and-forth signaling can be tested in the ocean.

The project focuses on Atlantic spotted dolphins, a species Herzing has followed since a 1985 trip to the Bahamas first drew her attention to dolphin communication. That long timeline matters because the team’s work is not based on a single field season or a short recording set. It rests on years of observation, repeated encounters, and a growing record of how these animals use clicks, whistles and other sounds in social settings.

How is AI being used in dolphin research?

AI is being used as a pattern-finding tool. Herzing said the technology is helping the team look through its archives of dolphin vocalizations and identify relationships that would be difficult to spot manually, especially when audio and video must be matched across many encounters and behavioral contexts.

WIRED reports that the Wild Dolphin Project is working with scientists at Georgia Tech and Google DeepMind. That collaboration suggests the study is not only an animal-behavior project but also a computational one, where methods built for large datasets are being adapted to the messy reality of field research.

The point is not full translation. The team is trying to add structure to dolphin language, not declare that it has decoded it. That distinction is central to how the researchers describe the work: AI may help classify sounds, identify recurring patterns and connect them to context, but meaning is still another step entirely.

What the machines can do now

According to WIRED, the system is helping the researchers find patterns in vocalizations and test early two-way signaling. In practice, that means the team can use computation to work through large audio and video collections faster than a human listener could and then check whether certain sounds appear again under similar conditions.

This kind of analysis is especially useful in the wild, where animal communication is noisy, intermittent and tied to movement, group behavior and environment. Dolphins do not speak in a lab-like setting, and their signals are intertwined with everything from travel and play to hunting and social interaction.

The challenge, Herzing says, is not just identifying repeated sounds. It is figuring out what those sounds mean in dolphin culture.

Why the project matters

The Wild Dolphin Project’s work matters because it sits at the edge of a broader scientific question: how far can AI go in decoding communication outside the human species? If the technology can help researchers move from raw recordings to meaningful patterns, it could change how biologists study animals that have always been difficult to observe and interpret.

For readers, the implications go beyond dolphins. The same basic problem appears across animal-behavior research: large datasets are now common, but understanding the signals inside them takes time, context and restraint. AI can accelerate the first part of that job. It cannot, by itself, prove semantic understanding.

That caution is especially important in a field where claims of “talking to animals” can outrun the evidence. Herzing’s comments, as reported by WIRED, push the conversation back toward what can actually be shown: pattern detection, sound-behavior pairing and controlled experiments that test whether dolphins respond in predictable ways.

Why readers should care

• AI can help researchers process decades of underwater audio and video faster than manual review.
• The work may improve understanding of dolphin social structure and communication.
• It highlights the limits of machine learning in translating animal sounds into human language.
• It shows how long-running field projects can become more powerful when paired with modern computation.

Background: how Herzing got here

Herzing’s interest in dolphin communication began in 1985, after a trip to the Bahamas. That early focus eventually grew into the Wild Dolphin Project, which has spent years observing Atlantic spotted dolphins and documenting their sounds, behavior and group interactions.

Herzing says she waited roughly 40 years for AI to become useful enough for this work. That long wait reflects a practical reality in animal communication research: the question has always been bigger than the available tools. A single field note or a handful of recordings cannot show the structure of a complex communication system. A large, carefully labeled archive can.

The broader scientific background is straightforward. Dolphins use clicks, whistles and other sounds. Researchers study those vocalizations alongside behavior and social structure in an effort to infer meaning. Machine learning is now useful because it can sift through enormous datasets and surface recurring forms, sequences and correlations that might otherwise remain hidden.

Still, the leap from pattern to meaning is large. Researchers in animal communication often separate sound mimicry from evidence of semantic understanding. That distinction runs through Herzing’s work as well. A sound may be repeatable, and a dolphin may copy it, but that does not automatically show the animal understands the function behind it.

What the team has tested in the water

One of the most intriguing parts of the project is its underwater wearable computer, which can play dolphin sounds. WIRED reports that the team has used the device in paired behavior trials to see how dolphins respond when particular sounds are played back in the water.

The sounds have been linked to objects such as a scarf and to environments such as sargassum. That matters because it gives researchers a way to test whether a sound is associated with a specific context, rather than being a random vocal pattern.

These experiments do not solve the translation problem, but they do move the research beyond passive observation. If a sound reliably triggers a response, or if dolphins react differently to different cues, that gives scientists a sharper way to ask what the signals may be doing.

Herzing also says decoding mimicry is a major obstacle in the wild. A dolphin may echo or imitate a sound for reasons that do not amount to communication in the human sense. That is one reason she believes a sanctuary or some other more controlled setting could be helpful: it may allow researchers to separate mimicry from meaning more cleanly than in open-ocean conditions.

By the numbers

The timeline and scale of the project help explain why the current AI phase is getting attention.

What did Herzing say about the AI tools?

Herzing framed the technology as a practical accelerator rather than a magic decoder. In comments to WIRED, she said, “With AI, we can find patterns.”

“With AI, we can find patterns.”

She also described the technology as something she had been waiting for during decades of fieldwork. “This is a powerful new tool. I’ve personally been waiting for it.”

“This is a powerful new tool. I’ve personally been waiting for it.”

Even so, Herzing cautioned that the work is still far from proving dolphins and humans can converse. She said researchers need to understand not just the sounds, but the patterns and what they mean inside dolphin culture.

Why this research is harder than it sounds

Animal communication studies often run into the same obstacle: a sound may appear meaningful in one setting and irrelevant in another. The wild adds another layer of complexity because animals do not produce signals on demand, and researchers cannot control every variable in the environment.

That is why the Wild Dolphin Project’s archive matters. The more video, audio and behavioral context the team has, the better chance it has of linking a sound to a specific interaction, object or response. But even a large archive does not automatically produce interpretation. It produces better questions.

The Georgia Tech and Google DeepMind collaboration suggests AI may help answer some of those questions at scale. A model can sort signals, group similar events and flag patterns in a way that is hard for a person to do by ear alone. But the final step still depends on human interpretation, careful experiments and a willingness to reject premature conclusions.

What happens when a dolphin sound is played back?

Researchers use playback tests to see how dolphins respond when a sound is returned to them. In the Wild Dolphin Project’s case, that has included sounds associated with a scarf and with sargassum environments, according to WIRED.

These tests are important because they move the research from observation to interaction. If dolphins react in consistent ways, the team can start building stronger hypotheses about what a sound might signal. If responses vary, that may show the signal depends on context, audience or social setting.

What happens next

The next phase of the work is likely to be defined by more testing, more comparison and more restraint in how results are described. The core open questions remain unresolved: what specific patterns the AI is finding, whether those patterns can be tied to reliable meanings, and whether more controlled settings can separate mimicry from intentional signaling.

Herzing has been clear that a true dolphin conversation is still a distant goal. The immediate priority is more modest but still significant: determine whether the tools can help the team understand the structure of dolphin communication well enough to ask better questions about culture and meaning.

WIRED noted that Herzing discussed the research at WIRED World Fair on 2026-11-05. That appearance suggests the project is entering a more visible phase, but visibility is not the same as certainty. The science still has to be tested in the water, against real animal behavior, and under conditions that do not distort the signals the researchers are trying to study.

For now, the main takeaway is simple: AI is making it possible to look at dolphin communication with sharper tools than before, but it is not yet translating dolphin speech. The project’s value lies in how it narrows the gap between data collection and interpretation — and in how honestly it acknowledges how much remains unknown.

Open questions to watch

• What specific repeated patterns will AI surface in Atlantic spotted dolphin vocalizations?
• Can the team connect those patterns to behavior strongly enough to support interpretation?
• Will controlled or sanctuary settings reveal clearer distinctions between mimicry and meaning?
• Can playback experiments show whether dolphins treat certain sounds as signals about objects or environments?
• How much of dolphin communication can be understood without overstating the case?