Word learning and Statistical Inference in Children with Developmental Language Disorder

Word learning is a language component that usually appears to be impaired in children with developmental language disorders (DLD) (Kan & Windsor, 2010). Despite it has long been studied, the underlying mechanisms of these difficulties remain unclear. For several years, however, we assist to the emergence of Bayesian learning models, which have been applied to account for word learning process (Perfors, Tenenbaum, Griffiths & Xu, 2011; Xu & Tenenbaum, 2007). According to these models, human beings are equipped with a strong mechanism of probabilistic, inductive inference (as opposed to frequentist approaches) which allows the learning processes to take place. Consider this, our study aims at exploring the hypothesis of a deficit of inductive inference in children with DLD as explaining the lexical difficulties they exhibit. We decided to create a novel word learning task inspired by Xu and Tenenbaum (2007). In this task, sets of one or three objects, which correspond to exemplars of novel words (e.g. dax), are presented to children with DLD. Once exposed to the exemplars, children have to pick out the other objects of the same type (other dax) from a larger set of new objects. Objects from several semantic categories and different levels of taxonomy are proposed. We can then explore the ability of children to learn and generalize through inductive inference by exploring the distributional pattern of responses across categories and levels of taxonomy. The ability of adjustment of the child’s responses given the condition (one or three examples) will also be taken into account in our analyses.