Karuza, E. . 1 , Farmer, T. . 2 , Fine, A. 3 , Smith, F. 2 & Jaeger, T. F. 1
1 Department of Brain and Cognitive Sciences, University of Rochester, Rochester, NY
2 Department of Psychology, University of Iowa, Iowa City, IA
3 Department of Psychology, The Hebrew University of Jerusalem, Israel
In the present work, we adopted a self-paced moving window display to examine the on-line predictive processes underlying non-adjacent dependency learning. Participants were visually presented with glyph triplets generated according to the artificial grammar A-X-B. In this grammar, A and B glyphs always co-occurred, while the intervening X glyphs were comparatively unpredictable (Gómez, 2002). As participants pressed a button to advance through the triplets, we measured the amount of viewing time they allocated to each glyph. Thus, motor responses offered an implicit index of changing predictions during learning; as participants' expectations converged with their input, they should process the predictable B glyphs increasingly faster than the predictive A glyphs. Indeed, we find that, over the course of exposure, processing of B was progressively facilitated relative to A. We also demonstrate through an individual differences analysis that participants with the greatest sensitivity to element predictability, those whose on-line expectations aligned most closely with the statistics of the grammar, performed better on an off-line measure of learning. Finally, we rule out alternative explanations for these results with two additional control conditions: one, a random concatenation of glyphs, and the other, a frequency-matched grammar with a weak relationship between A and B.