How can we separate changes to *perception itself* from changes to higher-level judgments and responses? Our latest work (preprint @ perception.jhu.edu/color) introduces a new approach to this deep & difficult problem, by exploring various "memory color" effects as a case study.
Memory color effects are said to occur when the known color of an object changes how it visually appears. For example, a gray banana might appear to be a bit yellow, or an orange heart might appear redder than it really is, because we know what color they're *supposed* to be. 
There are many ways to study memory color effects, one of which involves asking subjects to say when an object looks "gray". For a banana, subjects identify a slightly *bluish* looking banana as gray—as if the mind is adding some yellow, so that blue is needed to counteract it.
Does this truly reflect a change in perception? There are some reasons for doubt (see the paper for a few arguments). But one prediction, if these claims are true, is that they should show up not only in tasks requiring explicit color naming, but also in cases of *comparison*.
What do I mean? Consider this example. Here, two objectively teal ovals have been made to look purplish or greenish due to the Munker illusion. You *could* measure this by asking people "does this look purple? does this look green?". But you could also measure it by comparison:
Placing an objectively purple oval beside the two distorted ovals makes the green-looking look *different* from the other two. In panel B, if someone said "the left two look similar, but the right one looks different", that would be evidence that the illusion is working.
We suggest using this kind of "odd-one-out" judgment as a general strategy. It's one thing to discover that people call a bluish banana "gray"; but if it really *looks* gray, then it should also look more like an objectively gray thing than like an objectively bluish thing.
Our paper applies that test to memory color effects. Using stimuli from earlier studies, we show people triplets of objects: eg, bluish banana between a bluish disk and a gray disk (with the bluish-level chosen to match estimates from earlier work). (Here, A=bluish and C=gray.)
We find that memory color effects "fail" this test. Unlike in the Munker illusion, people can fairly easily tell that the bluish banana is blue, and not gray, as long as they have a truly blue and truly gray reference in front of them.
In other words, they pick C as the odd color out (which it is); but if bluish bananas really look gray, then A should have looked like the odd color out. And this happens not only in our studies, but also also in a fun little "replication" here on Twitter!
The theme is to exploit the "logic" of perception, which is good at reasoning transitively: eg if X looks like Y and Y looks like Z, X will look like Z. But sometimes these chains are too much for explicit judgments to follow.
So, transitive judgments like these, especially when they are the result of tortuous chains of reasoning, can help separate seeing from thinking! And this strategy can be used to study many kinds of perceptual effects, even beyond color or cognition-to-perception. /fin
I should add: "we" here includes lab undergrad JJ Valenti!
perception.jhu.edu/#jj
You may remember JJ from his appearance at #VSS2018, where he did a fantastic job with a 4-hour poster presentation!
perception.jhu.edu/#jj
You may remember JJ from his appearance at #VSS2018, where he did a fantastic job with a 4-hour poster presentation!
