Dr Elizabeth Wonnacott, University College London
The role of input variability in language learning and generalization: Evidence from language training studies with child learners
Language learning involves abstracting generalizations which operate across linguistic items. According to statistical learning approaches, these emerge on the basis of exposure to the input. For example, if a set of words occurs in the same linguistic environments, learners may use this as evidence that these words at as syntactic category within that language. Much of the evidence for this work comes from artificial language experiments which are generally conducted with adult learners (Reeder & Newport 2014), although the authors make inferences from these data as to the processes involved in child language acquisition. In contrast, there has been relatively little experimental work directly exploring children’s ability to use distributional statistics. I will present data from three different experiments in which primary school aged children (5-8years) are exposed to, and tested on, some new linguistic construction under experimental conditions. Specifically, we have explored: (1) learning a novel word order within the L1; (2) learning gender classes in an L2 (English children learning Italian); (3) learning preposition constructions in a L2 (English children learning Japanese). We find evidence that, in some circumstances, children’s generalization depends upon witnessing sufficiently varied exemplars in their input, presumably because this allows them to dissociate the structures from the particular trained instances (Bybee 1995; Ramscar et al. 2010). However there are also important interactions with input complexity, so that input variability may actually decrease performance when learning from the relatively small amount of input which can be provided in an experiment (or, perhaps, a classroom).