In this paper we shall point to some principles of neural computation as they have been derived from experimental and theoretical studies primarily on vision. We argue that these principles are well suited to explain some characteristics of the linguistic function of semantic concept recognition. Computational models built on these principles have been applied to morphological-grammatical categories (aspect), function words (determiners) and discourse particles in spoken language. We suggest a few ways in which these studies may be extended to include more detail on neural functions into the computational model.