People learn by both decomposing and combining concepts; most accounts of combination are either compositional or conjunctive. We augment the DORA model of representation learning to build new predicate representation by combining (or compressing) existing predicate representations (e.g., building a predicate a_b by combining predicates a and b). The resulting model learns structured relational representations from experience and then combines these relational concepts to form more complex, compressed concepts. We show that the resulting model provides an account of a category learning experiment in which categories are defined as novel combinations of relational concepts.

Many tasks that are easy for humans are difficult for machines. Particularly, while humans excel at tasks that require generalising across problems, machine systems notably struggle. One such task is the Synthetic Visual Reasoning Test (SVRT). The SVRT consists of a range of problems where simple visual stimuli must be categorised into one of two categories based on an unknown rule that must be induced. Conventional machine learning approaches perform well only when trained to categorise based on a single rule and are unable to generalise without extensive additional training to tasks with any additional rules. Multiple theories of higher-level cognition posit that humans solve such tasks using structured relational representations. Specifically, people learn rules based on structured representations that generalise to novel instances quickly and easily. We believe it is possible to model this approach in a single system which learns all the required relational representations from scratch and performs tasks such as SVRT in a single run. Here, we present a system which expands the DORA/LISA architecture and augments the existing model with principally novel components, namely a) visual reasoning based on the established theories of recognition by components; b) the process of learning complex relational representations by synthesis (in addition to learning by analysis). The proposed augmented model matches human behaviour on SVRT problems. Moreover, the proposed system stands as a more realistic account of human cognition, wherein rather than using tools that have been shown successful in the machine learning field to inform psychological theorising, we use established psychological theories to inform developing a machine system.

The debate about semantic reference between Frege’s (1948) descriptivism and Kripke’s (1972) causal theory of reference has recently been approached through experimental psychology. However, no consensus has been reached on the direction of the results. While some studies face clear methodological charges, even those that are currently uncontested do not reach a mutual conclusion. We propose a novel experimental paradigm with methodology designed to evade the problems of previous studies. Contrary to the past literature, we find a prevalence of descriptivists under lenient criteria for consistency across trials, while under strict criteria we find an equal amount of descriptivists and hybrids, with low numbers of referentialists (causal theory of reference) under both criteria. We suggest an interpretation of this result, and where future research might head.

Two studies, an experimental category learning task and a computational simulation, examined how sequencing training instances to maximize comparison and memory affects category learning. In Study 1, 2-year-old children learned color categories with three training conditions that varied in how categories were distributed throughout training and how similarity between exemplars progressed across instances. The results indicate that beginning learning by interacting with a limited set of highly similar exemplars leads to more learning than when the instances are distributed and dissimilar. In Study 2, order effects were examined with a symbolic connectionist model of general learning and representation discovery (DORA). The results of the studies suggest that when the presentation of instances is ordered in such a way that discrete instances of a category can be more readily connected in memory, category learning and discovery are more likely to occur.