ABSTRACT
In SG models, we find the following forms of classical inferences, hidden in connectionist approaches to the problem.3 Although it is not always very clear where to put one or the other technique, the following system is suggested. In the case of the indexical aspects of SG models, it seems natural to identify them with the phase of unsupervised categorization. The actual formation
--C:=5=1F"’ t t
As a consequence of this analysis, I conclude that transcendental logic and symbol grounding are based on the very same principles of logic. Since SG systems try to form concepts and label these concepts, they try to implement this view of transcendental logic. This implementation comes together with a search for principles that will make these systems work. In SG papers these principles are sometimes called "invariants" of sensory experience. SG model builders are therefore in search of rules which will allow the system to find the object in its visual field when given a name for it. In Kantian terms this means to search for the transcendental basis of cognition.
