Since the 1950s, the dominant framework in linguistics has treated hierarchical structure as the defining feature of human language. Words combine into phrases, phrases into clauses, clauses into sentences — nested trees of constituents, each node containing its children. This hierarchy is what separates human language from animal communication. It's what formal grammars formalize. It's what generative linguistics was built to explain.
Nielsen and Christiansen (Nature Human Behaviour, January 2026) found that the most common multi-word sequences in actual speech are not constituents. “Can I have a,” “it was in the,” “in the middle of” — the three- and four-word sequences people produce most frequently don't form grammatical units in hierarchical syntax. They don't parse into subtrees. They are invisible to the theory.
The evidence comes from priming: when speakers encounter a nonconstituent sequence, they process it faster the next time. Eye-tracking confirms this. The sequences are mentally represented — stored, retrieved, and deployed as units. They function as chunks, linearly assembled, more like snapping pre-assembled LEGO pieces together than like building tree structures from atomic rules.
The framework was built to describe the exceptional case — the well-formed, fully parsed sentence — and treated it as the typical one. But the typical case is the chunk. The most frequent structures in natural language are the ones the dominant theory cannot describe. Not because they're noise, not because they're performance errors, not because they're degenerate. Because they operate by a different mechanism — linear sequence, not hierarchical composition.
This is subtly different from discovering that a model described a special case of a more general mechanism. In that pattern, the special case is what the general mechanism looks like under simple conditions. Here, the dominant cases aren't a special case of hierarchy at all. They're not simplified trees. They're a different kind of structure entirely — and they're more common than the structures the theory was designed for. The theory didn't describe the easy version of the phenomenon. It described the rare version and didn't notice.
The implication for language evolution: if the common mechanism is linear chunking rather than hierarchical composition, the gap between human language and animal communication narrows. Sequential chunking is something many species do. The hierarchy was the supposed uniqueness claim. If the hierarchy describes only the formal, infrequent case, and the workhorse mechanism is shared with other species, then the discontinuity between human and animal communication is an artifact of the framework — visible only because the theory was built around the wrong typical case.