Abstract
Recent research in adult word learning has shown that associative learning performance depends on the temporal order of the learned pairs: associations between novel objects and labels are learned more successfully when objects precede labels than when labels precede objects (Ramscar, Yarlett, Dye, Denny, & Thorpe, in press). Here we investigated whether this effect may be driven by labels being less confusable than objects, and whether the effect can be replicated in a non-linguistic domain if this asymmetry in confusability among cues and associates is preserved. To test this hypothesis, we constructed two-tone shapes to serve as either labels or objects depending on their level of confusability. Specifically, two pairs of visually similar “F” shapes served as objects with confusable Features and two pairs of visually dissimilar shapes “L” shapes served as easily discriminable Labels. Each F shape was arbitrarily assigned an L shape. Thirty-two subjects passively observed sequential presentation of these cue-associate pairs in an unsupervised learning paradigm. We manipulated the temporal order across pairs so that each subject learned two pairs in the F-L order and two in the L-F order. After 8 minutes of unsupervised learning, we tested whether subjects had learned the associations in a 4-AFC task: On each trial, subjects were prompted with a target shape and asked to pick the appropriate associate from amongst the four possible complementary shapes. Subjects demonstrated better associative learning on pairs learned in the F-L order compared to the L-F order, regardless of the order in which they were tested, analogous to the result observed in the linguistic domain. This provides preliminary evidence that the ordered learning effect found in word-learning may be the result of a domain-general mechanism. We frame our finding in terms of error-driven learning and explore the theoretical implications for recent word-learning research.