“Biased vocabulary + visual attention and memory processes = a shape bias: A Dynamic Neural Field model”
From roughly 24-months-of-age, children generalize novel names for novel solid objects to new instances based on similarity in shape (Landau, Smith & Jones, 1988). In this talk, I examine the mechanisms behind both the development of the “shape bias” and its manifestation in real-time via simulations in a Dynamic Neural Field model. In particular, I ask whether the same model that Samuelson et al. (2011) used to demonstrate the role of spatial memory in early word learning also produces a shape bias when taught a vocabulary with the same statistics as the early noun vocabulary. Four sets of simulations capture the emergence of the shape bias from the growing noun vocabulary; differences in the bias depending on specifics of the novel noun generalization task; acceleration of vocabulary development following the training of a precocious bias; and differences in the bias individual differences in vocabulary (Perry & Samuelson, 2011). New data from a looking-while-listening version of the novel noun generalization task further reveal the relationship between visual exploration and biased noun generalization at a moment-by-moment timescale. Together the model and empirical work support established links between the bias and the developing noun vocabulary and provide insight on connections between visual cognition and word learning biases. I argue the model compares favourably to existing models of word learning and the shape bias on several model comparison metrics (Sims & Colunga, 2013; Christiansen & Chater, 2001).