The paper is real, but the post overstates it. Morten H. Christiansen (Cornell) and Yngwie A. Nielsen (Aarhus) published "Evidence for the representation of non-hierarchical structures in language" in Nature Human Behaviour on January 21, 2026, nine months before this post (DOI 10.1038/s41562-025-02387-z; the article).
What the study did: four preregistered experiments (total N = 497) found that sequences of part-of-speech categories, such as verb + preposition + determiner, can be primed even when they are not grammatical constituents. Two corpus checks, eye-tracked reading (N = 68) and phone conversations (N = 358), suggested the effect shows up outside the lab. The authors argue that some linear chunks are stored alongside hierarchical structure, which challenges the idea that hierarchy is the only mental representation.
What the post gets wrong: the paper does not say hierarchy is 'completely wrong' or that humans only predict the next word. It does not study large language models, and it does not claim the brain runs a 'simpler algorithm.' The authors present it as a partial revision of the tree-based view, and linguists still debate how much weight it deserves. The post's 'since 1957' framing is also loose: Chomsky's 1957 Syntactic Structures did not claim a single textbook consensus.
If you want to judge it yourself, read the abstract and the Cornell summary, which quotes the authors directly.
