Your brain may store language as flat chunks, not just grammar trees

Linguists have long said that sentences live in your head as hierarchical trees: nested phrases built by abstract syntax rules. That idea has dominated since Chomsky's work in the 1950s. A study in Nature Human Behaviour by Morten Christiansen (Cornell) and Yngwie Nielsen (Aarhus) argues the trees aren't the whole story.

The pair ran experiments, including eye-tracking, and analysed real phone conversations. They found that flat, linear sequences of word classes get primed: people process them faster the second time. That includes strings no grammar tree treats as a unit, such as "can I have a" or "in the middle of the". Priming is a classic sign that the brain has stored something as a unit. So these "beads on a string" chunks seem to be part of how language is represented, working alongside grammatical structure.

The headline claims going around are louder than the paper. It doesn't say Chomsky was "proven wrong" or that hierarchy doesn't exist. Its claim is that tree-based grammar "cannot capture all" of our mental language structure. The study only covers English. Christiansen adds that if language needs less hierarchy than assumed, the gap between human language and animal communication may be smaller than we thought.

Why it matters: chunk-and-sequence processing looks a lot like how LLMs model text, which reopens the debate over whether next-token prediction is really so unlike human language.