Abstract: Relational learning and reasoning is central to higher-order cognition. It is essential in mathematics and science, and equally essential in social and ethical reasoning. Yet learning relational concepts is challenging, because, unlike concrete objects, relations are rarely obvious in perceptual experience. I suggest that relational knowledge is acquired largely through analogical comparison processes. Analogical comparison engages a process of structure-mapping that highlights common relational systems. The common relational structure then becomes more salient and more available for transfer—in short, a portable abstraction may be formed. The power of analogical comparison is amplified by language learning. For example, hearing a common label for two situations invites comparison between the situations, engaging a structure-mapping process that yields insight into the meaning of the term.
I will discuss work on how comparison fosters relational learning, on factors that impede or support this learning, and on how structure-mapping processes interact with relational language to accelerate relational learning. Finally, although analogical processing is thought of as a sophisticated process, there is evidence that the same kinds of structure-mapping processes occur in adults and children, and even in infants—implying that powerful learning processes are at work even in the first years of life.
Abstract: Basic (curiosity-driven) research typically relies on artificial tasks to answer fundamental questions about memory, attention, perception and decision-making. Such research often yields simple and useful models that serve to protect us from naïve intuitions, but no matter how useful those models are in the laboratory, they are often considered to be about as relevant to the real world as a fire-breathing dragon. That might be true of some models, but almost everything I know about eyewitness memory in the real world I learned from testing simple models of list-memory in the artificial world of the psychology laboratory. The assumptions of one such model – signal detection theory – happen to be at odds with some of the most influential ideas about eyewitness memory that have emerged from applied (problem-driven) research over the last 30 years. For example, surprisingly, simultaneous lineups are superior to sequential lineups, initial eyewitness confidence is strongly related to accuracy, and various conditions that impair overall accuracy (e.g., high stress, cross-race, short exposure duration, etc.) have little to no effect on the impressive accuracy of identifications made with high confidence. We obviously need applied research to address real-world problems, but, less obviously, models derived from basic research provide a necessary foundation for that endeavor.