Banana Blues

Gray bananas look more yellow than equally gray non-bananas. Your conceptual categorization of a stimulus as a banana has something to do with the perceptual appearance of the stimulus. But what more can be said about this “having something to do with” stuff? More later.

Fig. 1. These bananas are bluer than they seem.

Karl Gegenfurtner and his colleagues did some pretty convincing experiments on this stuff.

We asked human observers to adjust the color of natural fruit objects until they appeared achromatic. The objects were generally perceived to be gray when their color was shifted away from the observers’ gray point in a direction opposite to the typical color of the fruit. These results show that color sensations are not determined by the incoming sensory data alone, but are significantly modulated by high-level visual memory.

I saw Gegenfurtner demonstrate some of this in a talk he gave at the Pasadena Neurophilosophy conference in June ‘05. Gray bananas did indeed look more yellow than their equally gray non-banana counterparts.

So, what’s going on here? One hypothesis, call it the “phenomenal hypothesis” is that the activation of “high-level visual memory” influences, but is in no way constitutive of, the appearance. Another, call it the “conceptual hypothesis” is that the activation of high-level visual memory partially constitutes the appearance. I favor the latter hypothesis. Why? More later. Which do you favor? Why?
Reference: Hansen, T., Olkkonen, M., Walter, S. & Gegenfurtner, K.R. (2006). Memory modulates colour appearance. Nature Neuroscience, DOI:10.1038/nn1794.

15 Responses to “Banana Blues”

  1. Ken Aizawa says:

    Very interesting post Pete. Thanks for bringing this to our attention. I think I’m going to use this example/paper for the final paper in my “Language, Perception, and Mind” course, where we talk about, among other things, that Fodor-Churchland debate on observation.

  2. patrick says:

    … the later.

    perhaps the activation of high-level visual memory is comparable to
    ivan pavlov’s “conditional reflex”. by our everyday experiences we’re conditioned to judge something looking like a banana to be yellow… we’re used to these stimuli like pavlov’s dogs and thus we show some reactions as if the banana was yellow… martians used to blue bananas would wonder about our yellow bananas, too.

  3. Pete Mandik says:

    Thanks, Ken.

    That sounds like a pretty interesting class.

  4. Joe Paxton says:

    I’m sympathetic with the phenomenal hypothesis. There are good empirical reasons to believe that conceptual information is not available to our perceptual systems. For example, in the Müller-Lyer Illusion, we can’t help but see one line as longer than the other. Even after we measure the two lines, and thus know, at a conceptual level, that the lines are the same length, we still see them as differing in length. Examples of this pheonomenon are legion, and not limited to vision.

    Another, more indirect, argument against the conceptual hypothesis might go like this: if conceptual information effects vision in such a thoroughgoing sense, then most of science becomes difficult or impossible. The reason for this is that no two people share a complete set of concepts. So, for example, my conceptual repertoire is relatively impoverished relative to Professor Mandik’s, in virtue of numerous factors, including differences in education and upbringing.

    So, if the conceptual hypothesis is true, then we are going to literally see the world in different ways, due to our conceptutal idiosyncrasies. The relevant consequence of this fact is that science would have a hard time getting off the ground, since there would be no objective, factual basis for our observations — no way to navigate our conceptuaul idiosyncrasies to get at the underlying objective facts.

    But if the phenomenal hypothesis is true, we could, at least in theory, get past our conceptual idiosyncrasies, by pooling our observational resources. The noise created by our conceptual differences would eventually cancel out, given repeated observations made by different people in different times and places. Of course, this assumes a more or less normal distribution of conceptual idiosyncrasies. But this is certainly a reasonable assumption, since most other properties of human beings — both inherited and acquired — are normally distributed (intelligence, height, political views, hours spent watching television, etc.).

    Where am I going wrong, Professor Mandik?

  5. Ken Aizawa says:

    Ok. I had a chance to read the paper.

    I had a slightly different take on the potential import of this example than did Pete. In the Fodor-Churchland debate on perceptual plasticity, and in Modularity of Mind, I read Fodor as saying that “high-level” scientific concepts do not causally influence our perceptions. In support of this, he alludes to visual illusions such as the Muller-Lyer illusion. They do not disappear when we know the facts about the line lengths. Now we have a case, unlike the familiar illusions, in which it looks like “high-level” scientific knowledge about bananas influences our perceptions.

    This is somewhat different that the issue Pete raises. Pete’s question is whether perceptions are causally dependent on high-level memories or whether perceptions are constitutively dependent on high-level memories. Pete seems to take for granted that Fodor is wrong about his hypothesis of informational encapsulation.

    To me, it looks pretty bad for Fodor. I’m not yet sure what to say about Pete’s question.

  6. Pete Mandik says:

    Ken and Joe,

    Ken represents accurately my take on this and the relevance of Muller-Lyer type stuff. Thanks, Ken.

    Joe, expanding on this further, the M-L illusion shows at best that there are some cases of percepts that are not influenced by concepts (I don’t actual buy that, but I’ll grant it for sake of conversation). The banana case shows that there are some cases of percepts that are influenced by concepts. The conceptual hypothesis need not be committed to the view that every percept is influenced by concepts, nor that every concept has an impact on perception. The conceptual hypothesis is saying simply that when concepts influence percepts it is because they are a part of the percept. The phenomenal hypothesis is saying that even though the concepts sometimes influence the percept, they merely influence it but are outside of it, not part of it.

  7. Ken Aizawa says:


    Just to perseverate on the Fodor-Churchland debate a bit, Fodor and Churchland both appear to think that belief alone does not influence perception. Churchland thinks it is belief plus “long experience” influences perception. Fodor, I think, is silent on this last view. This suggests a variant of the experiment.

    Take a fruit that is unfamiliar to subjects, maybe a starfruit. Then have subjects make it gray. Then show subjects the true color of the starfruit (i.e., show it in yellow). Then have subjects adjust the image of the starfruit to make it gray. The idea would take out the “long experience”.


  8. Ken Aizawa says:


    I’m still not to the point of having an answer to the question, but merely a comment.

    The issue of what constitutes a perceptual experience as opposed to causally influencing a perceptual experience is one of the issues at stake in (a special case of) the debate over extended cognition. Extended cognition applied to perceptual experience maintains that bodily and (perhaps) environmental processes constitute, and not just cause, perceptual experiences.

    So, how to detemine causal versus constitutional determiners of perceptual experiences is a central question.

    So, again, this is a very intriguing example. But, again, this is just a comment. Having gotten that off my chest, maybe I can figure out how to answer the initial question.


  9. Pete Mandik says:

    Hi Ken,

    One kind of consideration relevant to this question concerns phenomenal character or what it is like to undergo certain perceptual experiences. Consider two people looking at visually similar insects from similar viewing angles and lighting conditions who differe in what concepts each brings to bear in the perception of the insect. One conceves of the insect as a delightful source of protein. The other conceives of it as the reincarnation of their deceased enemy and a hideous carrier of disease. Plausibly, there’s a big differnece in what it is like to see the insect as lunch and what it is like to see it as a vehicle of supernatural evil. And what constititutes this aspect of what it is like to undergo the different experiences can’t be accounted for in terms of low level representations of perceptible qualities, or even the conceptual influence on low level representations of perceptible qualities. In this sort of case, it looks like what it is like to undergo the respective experiences is partially consistuted by the coneptual contribution.

    Of course, even if the above case is correct, it would not settle the question with regard to the banana example. But it does provide a way of thinking about this.

    And, by the way, I don’t see this kind of line of thought giving the externalist much to go on.

  10. Ken Aizawa says:


    This does not seem to me to settle this, or even weigh in favor of one hypothesis rather than another.

    Grant that the two cases you describe involve differences in what it is like to see the insect. Grant as well that the difference cannot be accounted for in terms of low level representations of perceputal qualities. Event grant that what determines the difference in what it is like is a difference in high level concepts. The question still remains how these high level concepts make a difference, is it by causally giving rise to distinct perpceptual experiences or by constitutively giving rise to distinct perceptual experiences.

    To my mind, this question is still completely open.

  11. Pete Mandik says:

    Hi Ken,

    What I was thinking was along the following lines.

    1. Whatever is constitutive of what it is like to have a perceptual experiences is itself partially constitutive of the perceptual experience.
    2. In at least one case, the insect case, the conceptual content is consitutitve of what it is like is to have the perceptual exeperience.
    .: In at least one case, the insect case, the conceptual content is partially constitutive of the perceptual experience.

    My guess is that you have a problem with 1. Is that right?

  12. Hi Pete,

    I think that experiment provides an argument for the position that the concepts and phenomenal experience don’t exist on separate levels. I would also say that that conceptual is partially constitutive of the phenomenal experience, but only as much that conceptual was based on specific phenomenal experience (e.g. bananas being seen as yellow in first place).

  13. Ken Aizawa says:


    Ok. Now I think I get what you are driving at. Let me withdraw my concessions regarding the insect case. Sorry. You will see why, I think, if we return to the achromatic bananas.

    Let’s look at your premise 1) applied to the achromatic bananas.

    What is it like to see the achromatic bananas? Seeing the achromatic bananas is like seeing slightly yellowish achromatic bananas. (This explains why subjects adjust the images in a slightly blue direction.) So, what concept is constitutive, in part, of the perceptual experience of a slightly yellowish achromatic banana? Maybe the concept of slightly yellow, rather than the concept of yellow. So here, you will just say that the concept of yellow is partially constitutive of the concept of slightly yellow. Maybe not so bad.

    So, consider what it is like when a subject has produced what appears to her to be an achromatic banana, i.e. one that in reality is slightly bluish. What is this like? It is like staring at an achromatic banana. The concept of yellow does not appear to be applied at all to the stimulus. So, what explains why the bluish banana looks achromatic? The constitutive theory seems to me to be at a loos. By constrast, the causal account can explain why the bluish banana looks achromatic. The incoming stimulus, i.e., the bluish banana makes one causal contribution to the perceptual experience and the high-level concept of yellow (or light yellow) makes another causal contribution to the perceptual experience. The “sum” as it were of these opposing causal contributions is that they produce an experience of an achromatic banana.

    So, this last case seems to me to provide some reason to prefer the causal account.


  14. Pete Mandik says:

    Hi Ken,

    I think the gist of the summing story is quite nice, but I’m not sure why the conceptual hypothesis would be incompatible with something like it. The conceptual hypothesis isn’t saying that the percept is wholly constituted by the concepts, but that it is partially constituted by the concepts. As such, there’s room for it to also be partially constituted by nonconceptual elements.

  15. Ken Aizawa says:


    You raise a fair question. I worried about that after I had posted.

    On the causal “summation” story I have in mind, the perceptual experience of an achromatic banana is causally due to the bluish banana and the memory concept of the yellowness of the banana. Having the perceptual experience of an achromatic banana is the application of the concept “achromatic banana” to the current sensation.

    On the constitutive “summation” story, the perceptual experience is due, I’m guessing, to the bluish banana triggering the simultaneous tokening of the concept of a yellow banana (or maybe just yellow) and of the concept of an achromatic banana. So, what is going on in the constitutive story is that the yellow (banana) concept is in some sense latent, i.e. the achromatic banana does not look yellow, even though the concept of yellow is being applied to it. Does this idea of the latent application of a concept to a sensation make sense? Is it plausible? I guess I was tacitly assuming that the application of a concept is always a manifest application. Notice, further, that the concept that is applied manifestly, so to speak, is the concept of an achromatic banana. Why is the application of the concept of the achromatic banana manifest and the application of the concept of yellow (or yellow banana) latent?