At a recent forum hosted at the OpenAI headquarters in San Francisco, Linguistics Professor Gašper Beguš discussed how AI can act as a catalyst for biological discovery and a bridge between animal and human communication.
He highlighted his recent study with Project CETI, where he serves as the linguistics lead. The project used AI to decode the communication structures of sperm whales and revealed striking similarities to human speech, as a concrete example of this work.
“I think of AI as a tool for scientific discovery,” Beguš said during his talk. “It’s not yet an end product, but it’s a tool that can help push science forward.”
He outlined three main “case studies” illustrating how AI interpretability provides causal explanations for things that were previously unexplained. The first case study uses LLMs (large language models) for testing what is uniquely human — in this case, focusing particularly on human language. The second uses LLMs to simulate what is possible and test the limits of technological intelligence. The third centers on internal interpretability of AI models, or “peeking under the hood" of these models to uncover new scientific insights.
A breakthrough in his first case study revealed to Beguš that LLMs possess metalinguistic capabilities — or the ability to analyze language itself. He found that these models don’t just mimic speech; they analyze and generate complex sentences like humans would.
“This kind of experiment shows us that something we thought was uniquely human — or that you have to be a human to learn language — is not true anymore,” he said. “And that really changes our perspective about languages, and who can learn languages. If you don't have to be human to learn language, maybe other species have something complex and interesting as well.”
Beguš and his team at Project CETI put this idea into action by using Generative Adversarial Networks (GANs) — which learn by listening and imitating, much like human children — to study whale calls. They discovered that whale calls contain spectral patterns that resemble human vowels and diphthongs.
"Gašper’s research for Project CETI demonstrates that linguistics is one of the major keys to understanding intelligence across species. Discovering vowel and diphthong-like structures in sperm whale communication fundamentally reshapes how we think about non-human language,” said David Gruber, founder and president of Project CETI. “For Project CETI, this work shows how AI, guided by linguistic insight, can open entirely new windows into how other minds on this planet may be organizing and expressing information."
Beguš hopes that through work like this, AI can help to bridge the gap between animal and human communication, sparking deeper discussions about animal rights and conservation efforts.
“We’ve learned that humans are not as exclusive as we used to think,” Beguš said. “So in a sense, it's very humbling and exciting because for the first time, we're really studying in parallel these three mutually informative intelligences — the humans, the animals and the machines.”
Beyond his work with Project CETI, Beguš is also exploring AI’s ability to generate endless new languages. Together with his colleagues, he developed ConlangCrafter, an LLM-driven system that automatically generates coherent and typologically diverse artificial languages.
ConlangCrafter allows creators to generate “languages” that can be used in novels, games, worldbuilding and other creative outlets. It also allows researchers to examine how languages develop and evolve.
Despite the power of AI, Beguš emphasized that human scientists still have certain traits that make them irreplaceable. AI lacks the “boldness and originality” required to pursue "impossible" ideas that a model might instinctively dismiss.
Looking ahead, Beguš says he hopes to see the expansion of AI interpretability — or the ability to further analyze AI models and understand what the models are actually learning. He is also interested in seeing AI models learn to decipher what was previously undecipherable and to create something genuinely original — opening new ways for linguists like himself to contribute to the broader scientific conversation.
“My hope is that linguistics and these kinds of discoveries can feed into this loop of science, where linguistics informs science and science informs linguistics,” he said.
