Baniwa classifiers as predictors of upcoming nouns

Abstract Theories of predictive processing postulate that the brain continuously generates predictions about future linguistic input. Many different linguistic elements have been found to function as predictive cues. One such element is classifiers, i.e., morphemes that classify noun referents according to some semantic property. However, very few studies have so far investigated the predictive function of classifiers, and all existing studies focus on major East Asian languages. The present study investigates whether listeners use classifiers as cues to following nouns in Baniwa (Arawakan, Northwestern Amazonia). Baniwa has a system of 53 classifiers that primarily encode physical shape. In a response time experiment, participants heard numeral-classifier-noun phrases and had to choose which one of two images depicted the noun. We tested two variables: informativity and degree of constraint. With respect to informativity, we found that response times were faster when the classifier could be used to identify the target image. With regard to degree of constraint, classifiers denoting shape yielded faster response times compared to the generic classifier. The results show that classifiers have a predictive function in Baniwa, and, by contributing data from a previously unexplored language family and geographical region, suggest that prediction may be a function of classifiers cross-linguistically.