Cross-linguistic regularities in pronominal number neutralisation are not driven by learning biases: Limits of the Typological Prevalence Hypothesis for morphosyntactic categories

Number neutralisation—the loss of singular-plural distinctions—is one of the most common homophony patterns in pronominal paradigms across languages but follows systematic asymmetries across person values. Specifically, number distinctions are claimed to be most frequently neutralised in third person, less so in second person, and least often in first person. While this generalisation has been widely noted, its empirical foundation to date remains relatively weak, and no clear mechanistic explanation has been established for why this typological asymmetry might arise. To address this gap, we test the Typological Prevalence Hypothesis (TPH)—which predicts that more frequent linguistic patterns should be easier to learn—as a potential explanation for these cross-linguistic patterns. We first conduct a large-scale synchronic analysis of pronominal paradigms (1,250 languages), confirming a frequency-based hierarchy (3rd > 2nd > 1st) but challenging its previously assumed implicational nature (i.e., if 1st, then 2nd, and if 2nd, then 3rd). A diachronic phylogenetic analysis finds weaker evidence for this frequency effect but stronger for the implicational aspect. Finally, an artificial language learning experiment testing the learnability of different neutralisation patterns reveals that learners do not mirror typological tendencies: contrary to predictions—and in line with a speaker-based bias— first-person neutralisation is easiest to acquire. These findings suggest that the typological evidence for the homophony hierarchy might not be as clear as previously assumed, and that learning biases alone might not straightforwardly explain typological regularities. This challenges the TPH, suggesting that pronominal paradigms evolve in a manner that is less constrained by learning pressures, and/or that alternative processing and communicative pressures may better account for cross-linguistic distributions of pronominal number neutralisation.