Abstract
Aim: A hallmark of human intelligence is our ability to map any stimulus onto a response on the basis of an arbitrary rule. Extensive cognitive neuroscience research has focused on humans' ability to follow a pre-specified rule. But how do we learn arbitrary visual associations when the rule is not explicitly given, but must be discovered by trial-and-error? In this study we investigated whether arbitrary associative learning can be characterized as a single abstract ability or whether it is better described as a heterogeneous function. Methods: Human subjects learned, by trial-and-error, the one-to-one mapping of N visual images and N horizontally arrayed locations, a paradigm modified from WA Suzuki's studies. We varied the set size from 2 to 8, tested various types of visual images (natural scene, scrambled scenes, visual objects, abstract art, and words), and strategically employed articulatory suppression. Results: When verbal working memory was not suppressed, humans were best at mapping words onto locations, but long-term memory for the associations was poor. Articulatory suppression impaired word-location association, but did not impair learning of non-verbal images. Of the non-verbal images, learning was better for highly familiar scenes than for unfamiliar scenes, better for scenes from different categories than for scenes within a category, but equivalent for scenes, objects, and abstract art. Conclusion: Arbitrary associative learning can rely on at least two routes: verbal and visual, with verbal learning leading to fast mapping but quick forgetting, and visual learning leading to slower mapping but more robust long-term memory.
This study was supported by NIH MH071788, ARO 46926LS, and ONR YIP05.