I’m sure you’re familiar with the idea of a universal grammar — the long-standing claim that humans share an innate blueprint for language, shaping how all grammars are built. It’s been one of the most influential & contested ideas in linguistics. A new open-access paper in Nature Human Behaviour by Verkerk, Shcherbakova, Haynie, Skirgård, Rzymski, Atkinson, Greenhill & Gray puts this assumption to the test using an extraordinary dataset & some very sophisticated modelling.
The study
The researchers examined 191 proposed grammatical universals — long-standing claims like “if a language has X, it tends to have Y”. These are the classic implicational patterns that have shaped typology since Greenberg — the long-standing idea that languages often follow “if–then” rules (for example, if a language puts the verb at the end, it often uses postpositions too). Greenberg’s work in the 1960s popularised these patterns, so this study is essentially testing whether those assumptions really hold across the world’s languages.
To test them rigorously, the team used a database of 2,430 languages, coded for 195 grammatical features (word order, agreement, case, negation, possession etc). They then applied methods designed to avoid the biggest pitfalls in cross-linguistic research:
- Bayesian generalised linear mixed-effects models — essentially a way of checking whether two grammatical features really correlate without being misled by the fact that related languages resemble each other. Think of it as filtering out “family resemblance” so only genuine patterns remain.
- 100 phylogenetic trees — multiple possible “family trees” of the world’s languages. Using many trees ensures the results don’t depend on a single reconstruction of linguistic descent.
- BayesTraits — a tool that tests whether two features tend to evolve together over time. For example, do languages that develop postpositions also tend to develop OV word order?
In short: a careful, large-scale test of which grammatical universals genuinely hold once you strip away shared ancestry, geography & coincidence.
The findings
Support for universals dropped sharply once ancestry & geography were controlled for: from 91% to 47%. After adding evolutionary modelling, only 60 universals (31%) remained robust.
The strongest patterns were:
- hierarchical universals (80% supported) — e.g. if a language marks a rare category, it almost always marks the more common one
- narrow word order universals (37% supported) — especially those involving head–dependent harmony
- broad word order universals (11% supported) — mostly unsupported
- “other” universals (17% supported) — scattered & inconsistent
The study also shows that languages tend to evolve toward certain “attractor states”. Despite the vast design space of possible grammars, unrelated languages repeatedly converge on similar solutions — likely due to shared cognitive & communicative pressures.
To illustrate: imagine a language where adjectives follow nouns (“house big”). If numerals also follow nouns (“house three”), that’s a harmonic pattern. The study suggests languages are more likely to evolve towards such harmony than away from it.
Teacher takeaways?
- Although the study reshapes how linguists think about universals, it has no direct implications for classroom practice — its value for teachers lies more in broadening our understanding of how languages evolve than in changing what we do on Monday morning.
- It’s a useful reminder that learners’ L1 structures are not “wrong” versions of English but different grammatical solutions shaped by their own linguistic histories. This can help us frame learner language more constructively.
- When teaching grammar, it’s worth highlighting that English represents just one pathway among many. Using cross-linguistic examples can help learners see patterns, reduce anxiety about “breaking rules” & build metalinguistic awareness.
Which grammar patterns in English do your learners find most counterintuitive?
Leave a Reply