Learnability as a window into universal constraints on person systems

Abstract

Zwicky (1977) made the following observation regarding the cross-linguistic distribution of person systems: Languages that do not have a dedicated form for the inclusive meaning (speaker+addressee) always assimilate it into the first person and never into the second or third. Different theories of person have been put forward to explain this asymmetry, positing different kinds of semantic constraints on the human capacity to categorize the person space. Here we use an Artificial Language Learning methodology to investigate whether learners are sensitive to these typological asymmetries. Our results reveal that (1) learners prefer person systems where there is homophony between inclusive and first person meanings; and (2) they find it easier to learn that the inclusive is a form of ‘you’ than a form of ‘them’. Given that second-inclusive and third-inclusive homophony patterns are both absent in the typology, our findings suggest that not all typological tendencies are built equal: while some might be seen as the result of strong, universal cognitive constraints on grammar, others should be modeled as weaker learning biases.