Cornell study of ~500 people: brain stores linear word sequences alongside grammar trees

Yngwie Nielsen and Morten Christiansen of Cornell University published a study in Nature Human Behaviour in January 2026 on how language is stored in memory, according to Perfil. They ran four experiments with almost 500 people and analysed two large datasets: a reading corpus with eye tracking and a corpus of real phone conversations perfil.comPerfilperfil.comOpen source ↗.

What the experiments showed

Linguistics has held for about seventy years that language is organised in nested units, which are drawn as a tree. In the Spanish sentence "el perro de mi vecino ladra", the phrase "de mi vecino" sits inside "el perro de mi vecino", which in turn sits inside the whole sentence perfil.comPerfilperfil.comOpen source ↗.

The authors found that, in addition to such units, people keep in memory linear chunks that do not form any grammatical unit. Perfil cites the sequence verb, preposition, article, shared by phrases such as "llegó a la" and "habló con él", and formulas such as "en el medio de la" perfil.comPerfilperfil.comOpen source ↗. When such a sequence appears a second time, it is processed faster. This effect is called priming, and the authors treat it as a trace of storage perfil.comPerfilperfil.comOpen source ↗.

What the paper does not claim

The study does not state that hierarchy is absent. One of the authors said in an interview that they cannot claim there are no trees, but that there is something more than trees perfil.comPerfilperfil.comOpen source ↗. The work measures recognition of previously seen sequences, not prediction of the next word. The authors say the result challenges both followers of Noam Chomsky and the construction-based linguists who oppose them perfil.comPerfilperfil.comOpen source ↗.

In Chomsky's model, internal grammar is hierarchical and the linear order appears only in speech, because the mouth produces one word at a time. One of the analyses, however, covers reading, where there is nothing to flatten for pronunciation, and the question concerns what is held in memory rather than how speech comes out perfil.comPerfilperfil.comOpen source ↗.

The debate remains open. A supporter of Chomsky could argue that these chunks are usage shortcuts and not part of grammatical knowledge. The authors present their interpretation as a hypothesis, and Christiansen acknowledged that they are not there yet. All the experiments were conducted in English, so the result still has to be tested in other languages perfil.comPerfilperfil.comOpen source ↗.