Inconsistent generalisation of structured knowledge between 2-dimensional conceptual spaces

Recent experimental and theoretical work has supported geometric theories of conceptual representation, where concepts are defined as regions in an abstract n-dimensional space, with each dimension specifying a certain characteristic. A key question in cognitive neuroscience concerns the generalisation of information between such concepts, i.e., how existing domains of conceptual knowledge can accelerate learning of new ones. To understand this phenomenon, an experimental paradigm is needed that would allow careful and systematic manipulation of stimuli and task characteristics.We developed an online paradigm for capturing generalisation between concepts. Participants successively learned two new 2-dimensional concepts: birds defined by the size of their neck and legs and birds defined by the size of their beak and tail. The participants learned that specific exemplars of each concept were associated with arbitrary target stimuli, forming paired-associates which served as “landmarks” in the corresponding 2D conceptual spaces. Crucially, for one group of participants, the (congruent) geometric arrangement of paired-associates was identical between the two bird-spaces, whereas for another (incongruent), it was maximally different. We predicted that in the congruent group, the shared geometric arrangement of paired-associates across the conceptual spaces would accelerate learning for the second bird-space.We preregistered a Bayesian sequential analysis to examine relative support for the null or the alternative hypothesis. Across two experiments, only groups starting with certain arrangement of paired-associates showed behavior consistent with generalisation. This interaction is potentially explained by stimulus characteristics of one of the bird-exemplars leading to ceiling effects. We suggest the use of a different stimulus set and argue for the importance of developing a fast and a reliable paradigm for studying generalisation across conceptual spaces.