Dogs seem to agree with humans that 'bouba' is round and 'kiki' is spiky
sciencealert.com
October 18, 2026
3 min read
51/100
Summary
A study in Proceedings of the Royal Society B: Biological Sciences found that domestic dogs appeared to associate the invented sound “Bouba” with rounded shapes and “Kiki” with spiky shapes, mirroring a well-known human tendency. The bouba–kiki effect, first proposed in the 1920s, has been observed across human languages and writing systems. Animal behaviorist Eleonora Biffi of the University of Parma and colleagues tested 70 dogs in two experiments. In the first, 51 dogs watched rounded and spiky shapes on screens while speakers played recordings of a human saying “Bouba” or “Kiki.” Video from a GoPro was analyzed with AI head-tracking software to measure the direction of each dog’s head as an indicator of attention. A second experiment tested 19 dogs under the same conditions, except their owners wore headphones playing rock music to prevent them from hearing the sounds and unintentionally cueing the animals. Dogs looked significantly more at rounded shapes during “Bouba” and looked at and explored spiky shapes during “Kiki,” the researchers reported. Dogs in the first experiment also showed an overall preference for looking at spiky shapes, but that bias disappeared when owners wore headphones. The researchers said more work is needed before concluding that dogs reliably make these sound–shape associations.
Key Takeaways
- Researchers tested 70 domestic dogs for a canine version of the bouba–kiki effect, in which “Bouba” is associated with rounded shapes and “Kiki” with spiky shapes.
- Dogs looked more at rounded shapes when “Bouba” played and looked at or explored spiky shapes when “Kiki” played, according to the research team.
- A second experiment used owners wearing headphones with rock music to limit potential owner cues; the reported Bouba–Kiki pattern remained, while an overall spiky-shape viewing bias disappeared.
- The study used AI analysis of GoPro video to track dogs’ head direction as a proxy for where they were attending.
What the discussion said
AI was barely the subject of this thread. Most commenters treated the dog study as a question about cross-modal perception: whether hard consonants, abrupt acoustic changes, and real-world sounds naturally map onto angular shapes, while softer voiced sounds map onto curves. Several readers argued that this shared intuition is precisely the puzzle, not an explanation; saying a sound feels sharp merely restates the association that needs explaining. Others suggested that physical experience may supply the bridge, since pointed objects often produce more percussive sounds and rounded ones gentler ones. The only direct AI angle came from a comparison between a human judgment of invented words and a GPT system’s classification. A commenter found that the model labeled one nonsense string as sharper while they judged the other sharper, using the mismatch to underline how fuzzy and personal these associations can become outside the classic examples. No one treated this as a meaningful benchmark failure or a broad claim about model capability; it was a curiosity appended to an otherwise animal-cognition and linguistics discussion. Consequently, the thread offers little evidence of either enthusiasm or alarm about AI, beyond a small note of skepticism that a model’s answer can diverge from an individual’s intuitive sensory mapping.
Where opinion split
The narrow AI disagreement is whether a GPT system’s choice of the spikier of two invented strings should be trusted over a person’s intuition. The human commenter saw the model’s opposite answer as evidence that borderline sound-shape judgments are ambiguous, while no participant made a case that the model had a uniquely valid interpretation.
Community Sentiment
Concerns
- A GPT system picked the opposite nonsense word as spikier from one reader, a small but telling reminder that subjective sound-shape intuitions do not automatically transfer cleanly into model judgments.
