From robotic toil to symbolic theft: Grounding transfer from entry-level to higher-level categories1

摘要

Neural net models of categorical perception (compression of within-category similarities and separation of between-category differences) are applied to the symbol grounding problem (of how to connect symbols with meanings) by connecting analog sensory projections to arbitrary symbolic representations via learned category invariance detectors in a hybrid symbolic/nonsymbolic system. Our nets are trained to categorise and name 50x5O pixel images (circles, ellipses, squares and rectangles) projected onto the receptive field of a 7x7 retina. They first learn to do prototype matching and then entry-level naming for the four kinds of stimuli, grounding their names directly in the input patterns via hidden-unit representations ("sensorimotor toil"). We show that higher-order categorisation (e.g., "symmetric" vs. "asymmetric") can be learned in two different ways: either (1) directly from the input, just as with the entry-level categories (i.e., by toil), or (2) indirectly, from boolean combinations of the grounded category names (symbols) in the form of propositions DESCRIBING the higher-order category ("symbolic theft"). We analyse the architectures and input conditions that allow grounding (in the form of compression/separation in internal similarity space to be "transfered" in this second way from directly grounded entry-level category names to higher-order category names. Such hybrid models have implications for models of the evolution and learning of language.