Stephen Grossberg

Steels & Belpaeme (S&B) ask how autonomous agents can derive perceptually grounded categories for successful communication, using color categorization as an example. Their comparison of nativism, empiricism, and culturalism, although interesting, does not include key biological and technological constraints for seeing color or learning color categories in realistic environments. Other neural models have successfully included these constraints.

To understand schizophrenia, a linking hypothesis is needed that shows how brain mechanisms lead to behavioral functions in normals, and also how breakdowns in these mechanisms lead to behavioral symptoms of schizophrenia. Such a linking hypothesis is now available that complements the discussion offered by Phillips & Silverstein (P&S).

Boundary completion and surface filling-in are computationally complementary processes whose multiple processing stages form processing streams that realize a hierarchical resolution of uncertainty. Such complementarity and uncertainty principles provide a new foundation for philosophical discussions about visual perception, and lead to neural explanations of difficult perceptual data.

Recent neural models clarify many properties of mental imagery as part of the process whereby bottom-up visual information is influenced by top-down expectations, and how these expectations control visual attention. Volitional signals can transform modulatory top-down signals into supra-threshold imagery. Visual hallucinations can occur when the normal control of these volitional signals is lost.

ACT is compared with a particular type of connectionist model that cannot handle symbols and use nonbiological operations which do not learn in real time. This focus continues an unfortunate trend of straw man debates in cognitive science. Adaptive Resonance Theory, or ART-neural models of cognition can handle both symbols and subsymbolic representations, and meet the Newell criteria at least as well as connectionist models.

“Chorus embodies an attempt to find out how far a mostly bottom-up approach to representation can be taken.” Models that embody both bottom-up and top-down learning have stronger computational properties and explain more data about representation than feedforward models do.

Neural models have proposed how short-term memory (STM) storage in working memory and long-term memory (LTM) storage and recall are linked and interact, but are realized by different mechanisms that obey different laws. The authors' data can be understood in the light of these models, which suggest that the authors may have gone too far in obscuring the differences between these processes.

Functional roles for cortical synchronization in self-organizing neural networks are described. These properties are best understood by models that link brain to behavior. Synchrony can express itself differently in cortical circuits that perform different behavioral tasks. Cortical temporal properties that seem inexplicable by synchrony are also mentioned.

The brain contains ubiquitous reciprocal bottom-up and top-down intercortical and thalamocortical pathways. These resonating feedback pathways may be essential for stable learning of speech and language codes and for context-sensitive selection and completion of noisy speech sounds and word groupings. Context-sensitive speech data, notably interword backward effects in time, have been quantitatively modeled using these concepts but not with purely feedforward models.