Abstract :
[en] Developmental Language Disorders (DLD) are severe and pervasive language disorders characterized among other things by word learning impairment (Kan & Windsor, 2010). Despite a lot of studies have been conducted, we still lack for evidence regarding the mechanisms which underlie these difficulties. Emerging for several years, Bayesian theories of cognition offer an interesting approach to study this phenomenon (Xu & Tenenbaum, 2007). These theories establish inductive inference as a chore component of the learning processes (Perfors, Tenenbaum, Griffiths, & Xu, 2011). Inductive inference is based on the interaction of prior knowledge, which can be defined as knowledge and biases a learner is equipped with, and environmental data. According with those theories, we formulate the hypothesis of a deficit of inductive inference in children with DLD as explaining their word learning difficulties. More specifically, our aim is to determine if children with DLD can use prior knowledge in a similar way than their typically developing peers in order to make inferences. Furthermore, Bayesian learning theories can account for abstract acquisitions (Tenenbaum, Griffiths, & Kemp, 2006). A secondary goal thus consists to see if children with DLD can make inferences at two levels of abstraction when we control prior knowledge, namely first-order inferences and second-order inferences (Perry, Samuelson, Malloy, & Schiffer, 2010; Smith, Jones, Landau, Gershkoff-Stowe, & Samuelson, 2002). We decided to create a design in which twenty school-aged children with DLD are exposed to a learning association task, followed by a novel word learning task and a generalization task. In the learning association task, children are taught associations of physical properties with housing environment, in order to create a baseline for prior knowledge. In the novel word learning task, children are taught new categories of insects associated with non-words. Each category is determined by particular physical characteristics (e.g. characteristics/shape of hands) that children have to infer, and each kind of insect has to be linked to his particular housing according to his physical properties. This phase represents the first-order inference step. To evaluate second-order inference step children are administered a generalization task. In this task, they are presented with insects from new categories which have not been learned. We can explore by this way the biases children use to make inferences as well as their learning ability at two levels of abstraction. Performances of children with DLD will be contrasted with those of their typically developing peers, matched for age and non-verbal IQ (age-matched) or lexical development (language-matched).
