Do red things differ in visual appearance from non-red things?
Recall that our question is,
Do red things differ in visual appearance from non-red things?
An immediate problem arises from the fact that pairs of things just one of which is \emph{red} (i.e. has the property denoted by ‘red’) differ not only in this way but also in which particular shade of colour they have.
How can we tell whether differences in visual appearance are due in part to differences in being \emph{red} rather than entirely due to differences in shade?
The problem can be overcome by introducing a third thing. Consider constructing a sequence of three things which are indiscriminable except by colour. Let the middle thing be \emph{red}, and let the first thing not be \emph{red}. Now consider the difference between the particular hues of these two things. Ensure the difference is small, but large enough that any two things which differ in hue by this amount are readily discriminable.
Finally, let the third thing be \emph{red}, and let the difference with respect to hue between the third and second thing be the same as the difference between the first and second thing. (Research on colour perception shows that such differences can be equated, and that sequences of this type exist; see
\citealp{kuehni:2001_color} on perceptual uniformity and \citealp{witzel:2013_categorical} on categories.)
Consider the visual appearances of these three things.
The difference in differences ...
Here is the argument. Consider two sequences of sensory encounters: (a) a sequence of sensory
encounters with two phonetic events that do not differ with respect to category (both are
realisations of /d/, say), and (b) a sequence of sensory encounters with two phonetic events that do
so differ (one is a realisation of /d/ the other of /g/, say).11 Let the events encountered in the
first sequence differ from each other acoustically in the same way and by the same amount as the
events encountered in the second sequence differ from each other. (That it is possible to find two
such pairs of events follows from the fact that we enjoy categorical perception of speech.) The two
sequences are depicted in Fig. 3. Now:
\begin{center}
\includegraphics[scale=0.25]{img/categorical_colour_difference3.jpg}
\end{center}
Hypothesis: Red things differ in visual appearance from non-red things.
Let me show you how speed an accuracy of discimination are usually measured.
This is a 2-alternative forced-choice task (2AFC).
Witzel & Gegenfurtner 2018, figure 5
\begin{center}
\includegraphics[scale=0.25]{img/categorical_colour_difference3.jpg}
\end{center}
Hypothesis: Red things differ in visual appearance from non-red things.
Predictions:
1. Redness will influence discrimination.
2. Redness will influence similarity judgements.
3. Redness will influence pop-out effects.
4. Redness will influence perceptual grouping.
‘Category effects on color perception. (a) Categorical sensitivity. The diagram shows how the sensitivity to color differences changes across hue. The x-axis corresponds to hue in DKL color space, and the y-axis to discrimination thresholds. Adapted from Witzel & Gegenfurtner (2013). Note that the green-blue boundary is unlike the others in that it coincides with a local minimum of discrimination thresholds. (b) Categorical facilitation. The bars along the x-axis of the main graphic correspond to color pairs that are composed of the colors illustrated by the inset. The y-axis of the main graphic represents response times in speeded discrimination. Note that discriminating the color pair BC, in which B and C belong to different categories (i.e., red and brown), is fastest even though the distances between colors control for sensitivity to color differences. Adapted from Witzel & Gegenfurtner (2016). Abbreviation: DKL, Derrington-Krausfopf- Lennie. ∗∗p < 0.01; ∗∗∗p < 0.001.’
Kay & Kempton 1984, figure 3
In brief: Kay and Kempton contrasted the responses of native English
speakers (who have words for green and blue) with native Tarahumara (a Uto-Aztecan language of
northern Mexico) speakers, whose basic colour words mark no such distinction.
Here you see a triad of colours, A, B and C.
Kay and Kempton first measured ‘discrimination distance’.
That is, how far apart are each of these in terms of JNDs?
As it turns out, JNDs are not affected by which categorical colour properties you can name
\citep{witzel:2013_categorical}.
So we would expect discrimination distance to be approximately the same for all
participants. (Some studies do measure discrimination distance for
each subject individually and find indiviudal differences; e.g. \citep{witzel:2014_categorical}).
In this case, you can see that A is further from B in JNDs than B is from C.
‘discrimination distance’:
‘The scale of psychological distance between colors we take as the "real" scale
for present purposes is called discrimination distance. The unit of this
scale is the just noticeable difference (jnd), that is, the smallest physical
difference in wavelength that can be detected by the human eye.’
\citep[p.~68]{kay_what_1984}
Next Kay and Kempton measured how visually similar their subjects judged these
samples to be.
To do this, they showed them different triads of colours and asked them,
Which is the most different from the other two?
For each pair they then computed the proportion of times that pair was split.
(As they write: ‘The psychological distance between A and B relative to other
stimulus pairs in the set is given by the proportion of times A and B are split
by the subject's selection of one of them as the most different item in the
triad.’ p. 70.)
This is what you see from the Tarahumara speakers and the English speakers
under the circles.
Looking at the numbers, you can see that
the English speakers tended to split B and C more often than A and B,
whereas the Tarahumara speakers did the opposite.
And note that the B-C pair crosses the blue-green boundary.
What does this mean?
‘The presence of the blue-green lexical category boundary appears to cause
speakers of English to exaggerate the subjective distances of colors close to
this boundary. Tarahumara, which does not lexicalize the blue-green contrast,
does not show this distorting effect.’
\citep[p.~77]{kay_what_1984}:
‘the English speaker judges chip B to be more similar to A than to C because
the blue-green boundary passes between B and C, even though B is perceptually
closer to C than to A.’
This is exactly the sort of evidence that should persuade us
that red things differ in visual appearance from non-red things.
Kay & Kempton 1984, figure 4
Here are the results from experiment 2.
In this case, there are only English speaking subjects.
And the numbers for the subjects are
‘simply the number out of 21 subjects who chose the indicated pairwise subjective distance as larger.’
The results now are quite different.
Top left: subjects appear to go with discriminability (JND distance)
rather than colour category.
Bottom left: when discriminability (JND distance) is equated,
subjects show no significant effect of colour category (although there is a trend).
‘Subjective similarity judgments follow
discrimination distance and reflect no influence from lexical category
boundaries.’ \citep[p.~73]{kay_what_1984}
\begin{center}
\includegraphics[scale=0.25]{img/categorical_colour_difference3.jpg}
\end{center}
Hypothesis: Red things differ in visual appearance from non-red things.
Predictions:
1. Redness will influence discrimination.
2. Redness will influence similarity judgements.
3. Redness will influence pop-out effects.
4. Redness will influence perceptual grouping.
‘Category effects on color perception. (a) Categorical sensitivity. The diagram shows how the sensitivity to color differences changes across hue. The x-axis corresponds to hue in DKL color space, and the y-axis to discrimination thresholds. Adapted from Witzel & Gegenfurtner (2013). Note that the green-blue boundary is unlike the others in that it coincides with a local minimum of discrimination thresholds. (b) Categorical facilitation. The bars along the x-axis of the main graphic correspond to color pairs that are composed of the colors illustrated by the inset. The y-axis of the main graphic represents response times in speeded discrimination. Note that discriminating the color pair BC, in which B and C belong to different categories (i.e., red and brown), is fastest even though the distances between colors control for sensitivity to color differences. Adapted from Witzel & Gegenfurtner (2016). Abbreviation: DKL, Derrington-Krausfopf- Lennie. ∗∗p < 0.01; ∗∗∗p < 0.001.’
\begin{center}
\includegraphics[scale=0.25]{img/categorical_colour_difference3.jpg}
\end{center}
Hypothesis: Red things differ in visual appearance from non-red things.
Predictions:
1. Redness will influence discrimination.
2. Redness will influence similarity judgements.
3. Redness will influence pop-out effects.
4. Redness will influence perceptual grouping.
‘Category effects on color perception. (a) Categorical sensitivity. The diagram shows how the sensitivity to color differences changes across hue. The x-axis corresponds to hue in DKL color space, and the y-axis to discrimination thresholds. Adapted from Witzel & Gegenfurtner (2013). Note that the green-blue boundary is unlike the others in that it coincides with a local minimum of discrimination thresholds. (b) Categorical facilitation. The bars along the x-axis of the main graphic correspond to color pairs that are composed of the colors illustrated by the inset. The y-axis of the main graphic represents response times in speeded discrimination. Note that discriminating the color pair BC, in which B and C belong to different categories (i.e., red and brown), is fastest even though the distances between colors control for sensitivity to color differences. Adapted from Witzel & Gegenfurtner (2016). Abbreviation: DKL, Derrington-Krausfopf- Lennie. ∗∗p < 0.01; ∗∗∗p < 0.001.’
Webster & Kay 2012, figure 1
Converging evidence that things labelled with different basic colour terms
do not thereby differ in visual appearance involves a third method for
detecting visual appearances (\citealp{webster:2012_color}). The idea, as
illustrated in Figure \vref{fig:perceptual_grouping}, is that how things
appear with respect to colour should influence how likely it is that they
will be grouped together perceptually. The results are consistent with the
view that whether objects are labelled with the same basic colour term
makes no measurable difference to how they are grouped perceptually. This
further supports suspicion that properties denoted by colour terms like
‘red’ make no difference to visual appearances (see also
\citealp{davidoff:2012_perceptual}).
Let’s consider the experiment in more detail.
‘circles at opposite diagonal corners of the square had the same color,
while the two diagonals differed in color by a fixed angle of 30°’.
The center circle could either be the same colour as one of the diagonals
(as in the left and right panels),
or it could be an inbetween colour (as in the middle panel).
When the central colour is the same as one of the diagonals, you should perceptually
group the diagonal as a line, like \/ or \\.
Subjects’ task was indeed to judge which way the diagonal went.
\citet{webster:2012_color} varied the center colour to find at what point
a given subject was equally like to judge that the line was \/ or \\.
Call this the ‘grouping midpoint colour’.
(p. 378: ‘Observers made a two-alternative forced choice response to indicate
whether the perceived orientation was clockwise or counterclockwise. A
staircase varied the center color angle to estimate the an- gle at which
both orientations appeared equally likely, with the point of subjective
equality estimated from the mean of the final 10 of 13 reversals in the
staircase.’)
Webster & Kay 2002, figure 2
\citet{webster:2012_color} reasoned like this:
\begin{enumerate}
\item Where the two diagonals are from within the same colour category,
the grouping midpoint colour should be the midpoint in a colour space that
represents retinal colour discrimination.
\item Where the two diagonals are from different colour categories (one is
green, the other is blue), and one of the diagonals is just over a boundary,
then the grouping midpoint colour should be closer to the colour of the diagonal
that is just over the boundary in a colour space that
represents retinal colour discrimination.
\end{enumerate}
This prediction is illustrated in the figure.
The dotted line shows the prediction if there is no effect of colour category
on perceptual grouping; the solid lines show different strengths of effect.
They found that almost no indicators of an
effect of colour category on perceptual grouping
(there are a variety of measures and one was signifiant:
p. 381: ‘the participants’ settings thus trended toward a (very weak) CP effect,
with an average bias of 0.10 (which was nevertheless significantly differ-
ent from zero; t (7) = 3.73, p < .01).’).
They also suggested that these effects may be due to the use of
CIELAB as a colour space
(p. 382: ‘ the small biases we found in the observer’s settings may in part
include an artifact of the stimulus space, weakening further the evidence for a
clear CP effect in the grouping task.’)
\begin{center}
\includegraphics[scale=0.25]{img/categorical_colour_difference3.jpg}
\end{center}
Hypothesis: Red things differ in visual appearance from non-red things.
Predictions:
1. Redness will influence discrimination.
2. Redness will influence similarity judgements.
3. Redness will influence pop-out effects.
4. Redness will influence perceptual grouping.
‘Category effects on color perception. (a) Categorical sensitivity. The diagram shows how the sensitivity to color differences changes across hue. The x-axis corresponds to hue in DKL color space, and the y-axis to discrimination thresholds. Adapted from Witzel & Gegenfurtner (2013). Note that the green-blue boundary is unlike the others in that it coincides with a local minimum of discrimination thresholds. (b) Categorical facilitation. The bars along the x-axis of the main graphic correspond to color pairs that are composed of the colors illustrated by the inset. The y-axis of the main graphic represents response times in speeded discrimination. Note that discriminating the color pair BC, in which B and C belong to different categories (i.e., red and brown), is fastest even though the distances between colors control for sensitivity to color differences. Adapted from Witzel & Gegenfurtner (2016). Abbreviation: DKL, Derrington-Krausfopf- Lennie. ∗∗p < 0.01; ∗∗∗p < 0.001.’
Do red things differ in visual appearance from non-red things?
So do red things differ in visual appearance from non-red things?