A Century of Bouba/Kiki Replication Never Produced Mechanism

Dogs match sounds to shapes the way humans do — show them a rounded blob called "bouba" and a jagged star called "kiki," and they'll pick the right pairing.

The 2022 finding arrived with appropriate fanfare, framed as a window into animal cognition and evidence of a capacity reaching across species. What nobody mentioned was that researchers had made this exact claim a century earlier.

Wolfgang Köhler ran the bouba/kiki experiment in 1929 with human subjects and found it solid and reproducible. Since then, researchers have verified it across infants who hadn't learned language, across people who'd never seen written letters, across synaesthetes and non-synaesthetes, across languages and cultures — one of psychology's most reliable findings and also one of its most mechanically mysterious.

Nobody has produced a unified explanation for why sound and shape bind together this way. The neural substrate remains unclear, the evolutionary pressure unidentified, the learning mechanism. If there is one, opaque. What the dog study reveals is not new data about animal minds but how science treats reproducibility as a substitute for understanding.

When a phenomenon replicates across a hundred different conditions without yielding mechanism, that's not a mystery pending solution.

When something replicates consistently, institutions treat that consistency as evidence that a mechanism must exist. Researchers repeat the same test in a new species, hunting for the mechanism in the wrong direction. A century of human bouba/kiki work produced no mechanism. Adding dogs won't produce one either. What changes is only what researchers will claim they're hunting for, framed as cross-species cognition or evolutionary depth. When a phenomenon replicates across a hundred different conditions without yielding mechanism, that's not a mystery pending solution. That's a signal you're looking for the wrong kind of answer entirely.

Watch what happens next time you see a study claim to have "finally explained" something that's replicated consistently for decades. The explanation will fit the data. It won't explain why the data arrived unchanged across so many different tries.