Prototypical Relations for Cortex-Inspired Semantic Representations

Cognitive systems for the representation of declarative knowledge like semantic networks and other graph-based systems are widely unrelated to characteristic neurobiological mechanisms in the brain. In this contribution we report on our efforts in bridging the gap between typical semantic relations like “is part of”, “has property” etc. and the laminar wiring pattern of the neocortex. Central to our approach is the identification of the cortical column as a basic building block within the relational network. These columns are typically sectioned into subsystems which comprise different horizontal layers and thereby provide different links for forward, backward and lateral processing. We show how these intercolumnar connections can be related to semantic links, which reflect hierarchical knowledge, temporal ordering and ontological relationship. These dimensions are of outstanding interest for most cognitive tasks. But also arbitrary n-ary relationships can be build by representing the relations as nodes and using only the proposed basic link types. As inference mechanism, a simple locally controlled activation spread was applied. It results directly from the intra-columnar connectivity which is uniform for all nodes. We tested the system with large commonsense databases and obtained promising results including predictions, context influences and feature inheritance.