Abstract
Many discrimination tasks (e.g. contrast discrimination) that normally exhibit perceptual learning when presented in blocked two-interval forced choice (2IFC) runs, cannot be learned when presented in a roving design in which one of four randomly chosen reference standards are presented on each trial. We recently reported (PLOS Biology, 2008) that learning can be restored if the levels are presented in a repeating pattern or if each 2IFC pair is preceded by a verbal pre-cue indicating the forthcoming standard, but not by a visual pre-cue of the standard. One hypothesis that could explain these findings is a floor effect whereby cases with less learning also have lower pre-training thresholds. We tested this hypothesis by measuring pre-training thresholds using a wide variety of pre-cues in the same observers. However, only by replacing 2IFC with a standard-always-first method (and compensating for the sqrt(2) factor) did thresholds change, giving us an important clue to a new hypothesis. Nachmias (2006) and Lapid et al. (2008) found that when the standard was in the first 2IFC interval, thresholds were [[gt]]30% lower than with the standard in the second interval. We believe their surprising finding goes well beyond an interval bias and well beyond Nachmias' article's modest title (the last 8 words of this abstract). We propose that when the levels are closely spaced in a roved 2IFC method, the standard becomes hard to pin down. Our results indicate that perceptual learning is not possible when roving makes it difficult to learn a virtual standard, forcing the observer to compare the two 2IFC intervals (an inefficient process). Perceptual learning may well involve learning a virtual standard, usable across trials. This finding not only can explain why learning is difficult or impossible under conditions of 2IFC with roving, it could also impact the general understanding of 2IFC.
Natural Science Foundation of China and NIH.