Knowledge Representation in Human Reasoning and Memory
Speaker: Nick Ichien
A longstanding controversy in cognitive science concerns whether or not representations of relations between entities are distinct from the features of individual entities. Although some experimental evidence has been interpreted as supporting such a dissociation, recent advances in machine learning raise the possibility that relations may be represented in the same representational space as features, using some form of conjunctive coding. I will discuss a series of studies in which I attempt to distinguish relational from featural representations, using both behavioral tests and applications of alternative computational models. These findings show that for word pairs that express semantic relations, separate relational representations are required to explain both explicit judgments of similarity and implicit measures of recognition memory. Furthermore, due to the greater capacity demands imposed by relations, similarity judgments are more sensitive to relations than are difference judgments. I conclude that in human knowledge representations, relations are distinct from features of individual entities.