Children learn words astonishingly skilfully. Even infants can reliably “fast map” novel category labels to their referents without feedback or supervision (Carey & Bartlett, 1978; Houston-Price, Plunkett, & Harris, 2005). Using both empirical and neural network modelling methods this thesis presents an examination of both the fast and slow mapping phases of children's early word learning in the context of object and action categorisation. A series of empirical experiments investigates the relationship between within-category perceptual variability on two-year-old children's ability to learn labels for novel categories of objects and actions. Results demonstrate that variability profoundly affects both noun and verb learning. A review paper situates empirical word learning research in the context of recent advances in the application of computational models to developmental research. Data from the noun experiments are then simulated using a Dynamic Neural Field (DNF) model (see Spencer & Schöner, 2009), suggesting that children's early object categories can emerge dynamically from simple label-referent associations strengthened over time. Novel predictions generated by the model are replicated empirically, providing proofof- concept for the use of DNF models in simulations of word learning, as well emphasising the strong featural basis of early categorisation. The noun data are further explored using a connectionist architecture (Morse, de Greef, Belpaeme & Cangelosi, 2010) in a robotic system, providing the groundwork for future research in cognitive robotics. The implications of these different approaches to cognitive modelling are discussed, situating the current work firmly in the dynamic systems tradition whilst emphasising the value of interdisciplinary research in motivating novel research paradigms.
| Identifer | oai:union.ndltd.org:bl.uk/oai:ethos.bl.uk:582025 |
| Date | January 2013 |
| Creators | Twomey, Katherine Elizabeth |
| Publisher | University of Sussex |
| Source Sets | Ethos UK |
| Detected Language | English |
| Type | Electronic Thesis or Dissertation |
| Source | http://sro.sussex.ac.uk/id/eprint/45963/ |
Page generated in 0.0053 seconds