Models of color working memory with color perception as a variable
23.548, Saturday, 17-May, 8:30 am - 12:30 pm, Pavilion
Gi-Yeul Bae1, Maria Olkkonen2, Sarah Allred3, Colin Wilson1, Jonathan Flombaum1; 1Johns Hopkins University, 2University of Pennsylvania, 3Rutgers-The State University of New Jersey
Surprisingly, the rapidly expanding literature on models of color working memory (CWM) has made little contact with research on color perception. With the aim of integrating these literatures, we applied standard color perception methodology to an influential CWM task. First, we scrutinized methodological practice for color rendering. Published CWM research typically specifies colors in CIELAB space, but does not verify that produced color coordinates match those requested. When we applied standard calibration procedures to our display, we discovered that typical CWM methods produce colors that vary considerably in luminance, and that many were out of gamut. Thus the space from which colors were sampled was not a ring, undermining the common practice of collapsing across trials with different colors when estimating model parameters. Next, we generated a CIELAB color ring with fixed luminance, and we verified that color rendering was accurate. With this color ring, we looked for color-specific variability in memory and perception. We found considerable color-dependent variability in response dispersion (precision) and central tendency (bias) in two delayed-estimation experiments. Color-specific variability in memory correlated significantly between observers and memory loads. Moreover, it correlated with the perceptual variability obtained in a third experiment without a delay. Finally, we investigated potential interactions between color memory and color categories defined by verbal naming. In two experiments, observers identified focal colors and category boundaries. We found that response precision and bias may depend on category representations associated with remembered colors. Using these empirical results, we developed several new models of CWM that include category-influenced effects, as well as binding computations relying on categorical similarity. These results suggest a reevaluation of previous CWM models. More broadly, they suggest that not all colors are structured identically in color working memory; instead, both perceptual and memory representations vary in complex ways throughout color space.