Seminar Series

Human Conceptual Processing

Rather than focusing on past research, I address issues that confront current understanding of how the human brain implements conceptual processing. In the theoretical framework adopted, a concept is a distributed neural network that integrates feature areas relevant for processing a concept’s instances. The concept of a hammer, for example, integrates feature areas that process the shape of a hammer (ventral stream), the actions in using a hammer (motor areas), the trajectory through space of manipulating a hammer in a goal-directed manner (supramarginal gyrus), and the visual motion that results (posterior temporal gyrus). The cumulative influence of processing a concept’s instances across many occasions causes its distributed network to become increasingly entrenched in memory, reflecting the aggregate experience. Once a network becomes entrenched, it produces an infinite number of temporary conceptual representations dynamically in a Bayesian manner that support prediction, evaluation, and situated action. On a given occasion, a concept’s distributed network activates a small subset of its content that includes: (1) information most likely to have occurred for the concept previously, together with (2) information that is most likely to be relevant in the current situation. Because concepts are highly associated with a variety of specific situations, this account of concepts is expanded to incorporate situational information. After establishing this basic account, I turn to possible directions for future research aimed at developing it further. One set of issues concerns how specific situated representations of concepts become established online as their instances are experienced. Of central interest are establishing the networks that process a situation’s elements (settings, agents, objects, self-relevance, actions, events, mental states), and how these networks interact to establish a coherent situational representation of the concept on that occasion. Of related interest is developing a computational account of how these situational representations become superimposed in memory to develop robust situational patterns of experience that underlie knowledge structures such as Bayesian models, schemata, etc. Finally, I turn to the implications of significantly expanded association areas in humans (relative to other primates). Based of a variety of past and current findings, augmented association areas appear to increase the power of human conceptual processing via three mechanisms: (1) the representation of abstract features (e.g., for aspects of self, mental states), (2) multi-modal data compression of information in modality-specific processing streams, and (3) the implementation of symbolic operations that integrate a situation’s elements (with the assistance of language). Finally, I suggest that these three mechanisms offer leverage in understanding abstract concepts, perhaps the greatest challenge to understanding conceptual processing.