Learning and generalization of lexical categories in Developmental Language Disorder

Background : Developmental Language Disorders (DLD) are characterized by word learning difficulties (Sheng & McGregor, 2010). These difficulties are well described but the underlying mechanisms remain unknown. Bayesian theories of cognition offer an interesting approach to study this phenomenon (Xu & Tenebaum, 2007). They define inductive inferences as a core component of the learning processes resulting from the interaction between prior knowledge and environmental data (Perfors, Tenenbaum, Griffiths, & Xu, 2011). Furthermore, they account for abstract and hierarchized acquisitions (Tenenbaum, Griffiths, & Kemp, 2006) in line with the idea of a multi-level generalization (e.g. Perry, Samuelson, Malloy, & Schiffer, 2010). Methods: Our research aims at determining if children with DLD can learn and generalize classification rules, making inductive inferences at two levels of abstraction. We further assume that, along with a limitation of processing resources (Im-Bolter, Johnson, & Pascual-Leone, 2006), children with DLD will encounter difficulties specifically when the tasks become more complex. We exposed twenty children with DLD (6 to 12 year-old) and twenty age-matched typically developing (TD) children to a word-learning task (1st order generalization) in which they learn categories of insects. These categories are either perceptually or relationally defined. The learning is progressive, such as children learn one classification criteria at a time until they reach a maximum of three. In a generalization task (2nd order generalization), they learn unfamiliar categories using the previously acquired rule. Results and conclusion: Compared to TD children, children with DLD learned at a slower rate when learning one and two categorisation rules. They also perform less well in the generalization task. This study is in line with our previous work using similar methods, showing that older children with DLD (9 to 12 year-old) could acquire and generalize a classification rule based on one perceptual feature, while younger ones (6 to 8 year-old) needed more exemplars and could not generalize as well as their TD peers