Response times and the boundary definition

With the constrained option of 12 color terms, the perceptual regions corresponding to the main color categories on the CIE x-y diagram were carefully mapped out, as shown in all previous figures. However, the regions defined in those figures, such as the distinct color zones seen in Figure 5-3, were based on one single statistic; namely, the quantity of votes for a certain color term. Another way to help define the boundary between color territories is to take into account the task difficulty measure. In the study, the RTs in each sorting trial are rendered as dependent factors relative to the ease in making a color category judgment. It is assumed that the more ambiguous the color, the longer it takes to discriminate and sort the color into one of the given categories. The RTs were also considered important in the related studies.(Guest & Van Laar, 2000, 2002; Sturges & Whitfield, 1997) While the size of the mode is an index of the commonness of the stimulus, the RTs is an index of the distinctiveness of the stimulus. A stimulus that results in a rapid response plus a larger mode to the same color term signifies that it is well located in the center zone of a color category (i.e., it is a typical example of that category). The reverse situation, with a long RTs and fewer votes, indicates a stimulus located in the periphery of a category or the boundary between categories.

Figures 5-8 and 5-9 visualize two factors: the 50% and 75% vote threshold and the contour map of RTs, respectively. Both figures contain six luminance levels in the x-y diagram of the same scale. Figure 5-8 uses a unitary criterion to demarcate the boundary of color zones; namely, the vote frequency counts of 75% level (color fills) and 50% level (color lines). Figure 5-9 presents RT in terms of contour lines on the color space. The black area corresponds to RT below 1.5 seconds, while the white area corresponds to time beyond 2.25 seconds.

It is interesting to examine the connection between the spatial constitutions in these two figures. In Figure 5-9, there are several prominent hot spots (black areas) embedded in the

66

inert ground (white areas). The white and lightest grey areas, representing short RTs, are generally consistent with the areas of achromatic center and boundaries between categories in Figure 5-7. The Pearson correlation coefficient of the size of mode and the RT of each stimulus in low to high L conditions is -0.846 (p < 0.01), -0.874 (p < 0.01), -0.816 (p < 0.01), -0.81 (p < 0.01), -0.763 (p < 0.01) and -0.103(p = 0.67), respectively. High and stable negative correlation can be observed in most conditions, but this effect disappears in the L=170 condition. Generally, the center tendency index can demarcate the core zone of the category, as shown in Figure 5-7, while the RTs information gives robust weight to the boundary.

The RTs measure also reveals the distribution of perceptual distinctiveness (saliency) on the color space. Figure 5-10 presents the luminance-against-RTs line plot that connects the mean RTs of the color categories in certain L conditions. Note that the figure does not contain every category in every condition. To prevent the interference of the RTs of non-typical judgments, each line of color category only presents the L conditions in which obtained votes surpass 10% of all votes within the category. The white category is not included because its votes ratio reaches 10% only in L=170 condition. The line plot shows that the RTs is both category- and luminance-relevant. For the categories of green, blue and purple, the mean RTs are generally shorter across all L conditions. This suggests that observers can easily and rapidly decide whether a given color belongs to the green, blue or purple categories, even though these three are next-door neighbors on color category maps. However, the lengths of RTs in the other categories are relative to luminance variation. The RTs of Gray drop drastically, indicating that gray is easier to determine in higher L levels, whereas RTs of brown show the reverse trend. The rest of the color categories also have shorter RTs in their corresponding dominant L levels. The mean RTs of the orange category, for instance, drops at L=50, which is the luminance level at which the color is most frequently recognized.

Rapid RTs can be found in the highest L condition of L=170, as shown in Figures 5-9

67

and 5-10. This appears to be somewhat in conflict with the previous point that RT serves as an index of the ease level of the task. A color displayed in very high luminance should be diluted in hue and saturation, and it should thus become more difficult to determine its appropriate category. However, there are two possible reasons for the actual result. First, the limitation of the display gamut makes the colors of high L conditions vary in restricted numbers of categories. The second reason is that under such high luminance conditions, the observers actually make a color-or-white distinction; that is, they simply sort the stimulus into one of two main categories. The psychological distance between these two categories should be larger than that between many other color categories, such as green and yellow. With the limitation of the gamut display, the number of sub-categories under the broader ‘color’

category is even fewer, as designated by the yellow, green, blue, and pink zones in Figure 5-8.

These factors could reduce the task difficulty in L=170 condition and contribute to the quick response.

Figure 5.8. Zones of color categories in six luminance conditions. The boundaries are demarcated by 75% and 50% votes ratio, which are marked with color fills and color lines, respectively. These two threshold levels partition the x-y surface into distinct zones without overlapping.

68

Figure 5.9. Contour maps presenting RTs in six luminance conditions. The light areas indicate longer RTs, or the more difficult sorting decisions, while black areas indicate the faster RTs and easier response zones. The darker areas roughly correspond to the color zones in Figure 8, except in L=170 condition.

Figure 5.10. Line plot of the mean RTs of 11 color categories (white is excluded) in their frequently identified luminance conditions. The y-axis shows milliseconds and the x-axis shows luminance.

5.3.3. Summary

The experiment presents the formation of color categories through a 12-color-terms sorting experiment that employs native Mandarin speakers as participants. The adopted categorical color terms were determined to be universal among human cultures (e.g. Berlin &

69

Kay, 1969; Guest & Van Laar, 2000; Lin, et al., 2001a, 2001b, 2001c; Lindsey & Brown, 2006; Lu, 1997; Shinoda, et al., 1993), and were confirmed to be frequently used by a free-recall pretest. The range of each term-related color category among observers was carefully plotted on the CIE 1931 chromaticity diagram on six luminance surfaces. Unlike many studies adopting reflective materials, limited saturation or luminance setting, or irregular sampling in designing stimuli, this study’s illuminant stimuli vary regularly in terms of hue, lightness and saturation and can systematically capture the spatial structure of color categories in different perceptual dimensions. In general, this experimental design leads to an intriguing finding in the results; namely, the changing shape of the color zone depending on purity and luminance. These two colorimetrical parameters correspond roughly to saturation and lightness. In the seminal Color Categories in Thought and Languag (Jameson &

D'Andrade, 1997), Jameson and D’Andrade argue that within the internal perceptual color space, hue interacts with saturation and lightness to produce ‘bumps.’ Bumps are defined as the salient representation of color categories, or the foci colors. The formation of focal color zones located at different luminance levels and eccentricities apparently support, and

‘visualized’ this theory.

The formation of color categories shows the various degrees of the luminance effect.

The most luminance-irrelevant cluster includes green, blue, purple, and gray. These four colors, particularly green and blue, are identified across all luminance levels. Additionally, the shape of the corresponding contour map remains stable, and the location of the foci of these categories is consistent across all conditions. Moreover, the RTs of the green, blue, and purple categories are the shortest among all colors, and are unrelated to variances in luminance. All measures indicate that these three color concepts, particularly green, are more psychologically distinctive, salient and robust than others. Green gained the most votes in the experiment with the lowest mean RTs, and its zones are encircled by sharp contour edges. Interestingly, the locations of these three categories on the color space are close. Blue is adjacent to green and

70

purple is adjacent to blue. These color categories are similar in terms of chromaticity, but distinct in category distinguishing. Nevertheless, many categories can be frequently identified and appear typical only in certain restricted luminance ranges. Red is typical in L = 10-25, deep pink in L = 50, orange in L = 50-100, pink in L = 100, and yellow in L = 100-170.

Conceptually, these color categories are different shades of the ‘warm’ color cluster, and are bound tightly by luminance conditions. In low luminance levels, the same chromaticity location of warm colors could easily be identified as brown. Additionally, the red, deep pink and pink categories, which belong to the ‘Hong’ (red) cluster in Mandarin, appear to be typical in three distinct ascending luminance levels. Their foci locations do not overlap. These factors indicate that Hong, Fen-Hong and Tao-Hong could be independent categories. Also, the claims of earlier studies of Mandarin, which accounted for only six color categories(Berlin

& Kay, 1969), could be inappropriate to apply to the contemporary Mandarin environment. In Berlin and Kay’s survey on the development of color terms in worldwide languages, Mandarin has only four chromatic color terms: red, green, yellow and blue. Some researchers argue that these limitations are refutable and have tried to propose new evidence (Lu, 1997).

Furthermore, it is important to note that the foci of brown and gray are located symmetrical to the reference white. Traditionally, gray should serve as a representation of achromatic stimuli, but the results show that it actually stands in for ‘cold’ colors in low saturation conditions, while brown stands in for warm colors in similar conditions. The exact neutral gray may only exist in perfectly controlled viewing conditions, which are seldom found in the real world. Supposedly, these two wild-card color concepts (Greenfeld, 1986) are capable of conveying near achromatic shades of cold– and warm–tinted colors.

A previous study uses similar viewing conditions and color space to examine Japanese speakers as observers (Shinoda, et al., 1993). In the Japanese study, the location of boundaries between blue and green are different than the location observed in this study. The green areas in this study’s color zone maps extended further than the blue areas, while the reverse was

71

found in the comparative study. The red area in this study is narrower than that in the Japanese study. Other than the differences in area size of red and the boundary location of blue and green, the remaining color categories were similarly spaced in both studies. Interestingly, blue and green can be loosely represented by a term in a literary language used by ancient Chinese, and this ancient Chinese written language influenced both modern Mandarin and Japanese.

Perhaps the conventional definitions of blue and green in modern Mandarin and Japanese developed differently. In the fields of color categorization and naming, a conventional view is still developing. A greater quantity of substantial empirical data would undoubtedly improve the overall understanding of the fields.

72