Sub sections:
Language and Robots: Computational Models of the Evolution of Language
Semester II 2005
A/Prof. Janet Wiles
wiles @ itee.uq.edu.au
(Education 120, 8 August 2005, 12.00 noon)
Abstract
Human languages are infant-friendly, in that they are learnable by infants at very young ages from interacting with others in their natural environments. The features of languages that make them infant-friendly and the abilities of infants that make them language-savvy are much debated. This seminar will discuss computational models of the evolution of language that focus on how languages can adapt themselves to become infant friendly and a novel approach to studying language evolution using robots that engage in language games. The aims of the project are to develop mobile robots that can evolve robot-friendly languages that are grounded in their own rich perceptual and spatial representations, constructed as they explore the environment around them. Using real robots provides both benefits and challenges. Robots, like humans, are fully embodied, and their internal representations are constructed solely from their perceptions and cognitive processes. In particular, we use robots with navigation systems based on the rodent hippocampus. Our robot language games are designed to communicate symbols that can be compositional and productive analogous to human languages. However, these symbols are grounded in the robots' representations of locations and objects in their own environments. The seminar will review research in the computational evolution of languages using language games and describe progress to date on the evolution of robot friendly languages.