John Kulvicki is Associate Professor of Philosophy at Dartmouth College. He works on representation in philosophy of art and philosophy of mind. He is the author of two books on images, one if which is actually called On Images: Their Structure and Content (Clarendon 2006). The other is cleverly titled Images (Routledge 2014), because he’s gotten into the whole brevity thing. He is hard at work on papers on analog and digital representation, and hopes very soon to write more about the philosophy of perception.
I just published a book as part of Routledge’s New Problems of Philosophy series, Images(2014). While I spend a good deal of time discussing theories of pictorial representation, my overall hope is to refocus philosophers on a broader class: the images, generally speaking. You all know what this class is like, at least extensionally speaking: photographs, line drawings, radar images, x-ray images, diagrams, graphs, and so on.
I’d like to share one thought about what makes the images, generally speaking, rather interesting. The thought cuts across philosophy of art, philosophy of science, and philosophy of mind. Understanding this point requires what Oliver Scholz (2000) has called “a solid sense of syntax.” My overall sense is that philosophers of art, but also philosophers of mind, do too little to keep syntax in sight, and as a result they miss some very cool phenomena. The point I want to make is, it seems to me, implicit in a number of discussions of representation. We find it in the mental imagery debate, in some work on pictorial representation, even going back to Alberti, in discussions of analog and digital, and in the recent literature on scientific representation. Chris Swoyer’s (1991) discussion of “structural representations” comes closest to making this point explicit. I want to convince you that making it explicit is worth the time and effort.
The idea has three parts, one semantic, one syntactic, and one an amalgam of the two. Keeping these three parts separate is essential to seeing the point, even though they very much want to collapse into one another. In brief, the idea is this: abstractions from an image’s syntactic details map readily onto abstractions from its semantic details. There’s a more wearable slogan for this claim, but I’ll save it until the end, when the details are clear.
We need an example, so consider the photo of part of Soo Sunny Park’s Unwoven Lightinstallation, which was recently on show at the Rice Gallery in Houston. The photo represents a fairly complex array of colors in space, and it represents them at high levels of detail.
This is a point about the picture’s content: it represents highly specific colors arrayed in a highly specific fashion. A related point about the picture’s syntax is that the properties relevant to it representing what it does are highly specific colors arrayed in a highly specific fashion. The picture has a lot of properties, but only some of them are syntactic. Its absolute size has nothing to do with its content, for example, but the pattern of colors on its surface does. This is all very obvious and simple, but things get interesting once we consider the relationship between the photo’s syntactic qualities and the features it represents.
What happens if we systematically ignore some of the photo’s syntactic qualities? Remarkably, we wind up with an interpretable representation. And even more remarkably, we wind up with a representation whose content is like the original but for the fact that we are ignoring some of it. I’m not saying it’s easy to ignore the syntactic colors in the photo of Park’s installation. If you do, however, the result is a representation of her work that says nothing about color. It’s a representation with content very much like this photo’s:
It’s easy to generate this greyscale image from the original photo, because all we need to do is discard the hues in the original. Those discarded hues are syntactic, but discarding them has an obvious semantic result.
This point doesn’t just apply to color. Disregard the high-frequency spatial features of the photo, and the result is a representation that fails to depict high-frequency spatial features of its object. In other words, blur the photo and you get a representation that says less about the spatial features of its object:
We can disregard the low-frequency spatial features of the picture, too. The result of this high-pass filtering is a photo that only represents the high-frequency aspects of the scene:
One other thing we can do is disregard whole regions of the photo, and focus just on a spatial part of it. The cropped result is a photo of a limited portion of the scene that the whole depicts:
The point here is NOT that pictures’ colors represent colors, shapes represent shapes, and so on. They do, but the point here is about a syntactic-semantic pattern. Ignoring some syntactic features altogether, or ignoring their determinate details, leaves one with interpretable representations. And the interpretations of these representations ignore details in the contents of the originals. Abstractions over the syntactic features of an image readily map onto abstractions over its content. This works for graphs of temperature over time, even though time and temperature play no syntactic role in such graphs, so it’s not a point about resemblance. These graphs aren’t pictures, either. They are part of the genus, to borrow John Haugeland’s (1989) term, of the images, generally speaking (Haugeland called them ‘icons’).
Contrast this with a paradigmatically linguistic representation. Dürer includes an inscription in his engraving of Erasmus (1526), part of which is reproduced here (original photography by Jeffery Nintzel for Dartmouth’s Hood Museum of Art):
Send this inscription through a high-pass filter, and the result has much the same content as the original. It is not an abstraction from it:
Send the inscription through a low-pass filter, or crop it at random and you get two kinds of nonsense:
Some lucky croppings, and some abstractions over linguistic representations’ syntactic qualities yield interpretable representations whose contents are abstractions over the originals, but this phenomenon is a pale shadow of what we can do with the images.
The relationship between syntax and semantics in images tells us something interesting about their syntactic and semantic qualities. Specifically, the photo of Park’s installation represents highly specific colors, but it also representsrather indeterminate colors. It represents things as being red, blue, and yellow, just as it represents things as being magenta, aquamarine, and chartreuse. It represents high spatial detail, and it also represents general spatial patterns. No such thing is true of the lexeme ‘chartreuse’. Descriptions deploying it do not represent their objects as being yellow, even though all chartreuse things are yellow. The photo’s syntactic features exhibit a similar pattern. Highly specific colors and shapes are syntactically important, but so are unspecific colors and shapes. This claim requires careful unpacking, I realize, and I’ve tried to do that elsewhere (2007, 2010, 2014).
I promised that this point about images would prove interesting in the philosophy of art, philosophy of science, and philosophy of mind. Let me sketch its significance to each area.
First, consider the arts that play with the potential for figuration: drawing, painting, sculpture, and even some sound art and installation. With these art forms, the question, “Does it represent, and if so what does it represent?” is important for understanding a work in all of its artistic, representational, expressive, exemplifying, and even decorative glory. The way we go about answering this question is different for images than it is for written texts. With images we play a game of enfranchising and disenfranchising certain features of the work as syntactic, whereas with texts that’s set beforehand. Does this painting represent the colors of things, or not? The painting is colored, but that’s not decisive. The colors here might serve expressive or exemplifying roles that are important to the work, but irrelevant to what it represents. Similarly, any colored canvas is covered with determinate spatial patterns, but these are only sometimes syntactically significant in all of their detail.
The idea about images gives shape to the process of interpreting and appreciating the non-literary representational arts. Because images are forgiving—you can ignore syntactic detail without producing nonsense—they are open to this kind of searching interpretation. Ignoring some feature as syntactic means freeing it to play some other role. This helps explain why certain kinds of realism are distinctively interesting with respect to pictorial representation (2014 Ch6), and I think this offers richer sense of what makes image interpretation aesthetically interesting than is captured by Goodman’s symptoms of the aesthetic (1976 Ch6). It does flesh out why syntactic and semantic density were so important to Goodman, however. It’s not simply that in such systems the syntactic and semantic detail outstrips what we can ever fully notice. It’s that the number of available abstractions over such details far outstrips what we can ever fully grasp.
The same feature of images helps explain their uses in science. Images present their contents across levels of abstraction, so they are ideally suited to presenting data for those whose goals vary. Some might be interested in highly specific claims about changes in temperature at specific locations, while others are interested in general patterns of warm and cold. The temperature map makes all of this information readily available. We can work with such representations, making judgments about patterns on their surfaces, which map readily onto claims about the temperature at this or that location. Such representations are ideal for surrogative reasoning (Swoyer 1991, Hughes 1997): we can think with them, and thereby think about their targets (see my 2010, 2014 Ch7).
Coming around to philosophy of mind, it’s a commonplace that images are deeply perceptual representations. I think this is true, and I think it’s true for things like graphs of temperature over time as well as for focused color photographs. These representations are perceptual because of how abstractions over their syntactic details map onto abstractions over their semantic details. Why does this make them deeply perceptual? In perception generally we are presented with an array of worldly features that are available for thought and attention, across levels of abstraction. We can scan a scene looking for the green things, or focus exclusively on the aquamarine. We can care about details of shape and shading, or just general patterns of light and dark. I suggest that an important part of the explanation for how we do such things is that perceptual states are images in the sense I’ve articulated here. They have contents that span levels of abstraction, and syntactic features that do the same. This is what makes is easy for perceivers to attend to the features of interest in their environments and ignore the rest (see my 2007, 2014 Ch8).
I also promised a wearable slogan. Jerry Fodor recently suggested a “picture principle”: “If P is a picture of X, then parts of P are pictures of parts of X.” (2008, 173) It’s tempting to call it Fodor’s picture principle (see Balog 2009) but it’s not really Fodor’s (Sober 1976, 124), it doesn’t specifically concern pictures (it’s about images, generally speaking), and absent a careful understanding of ‘part’ it’s not much of a principle (see my 2014 Ch8). Nevertheless, there is a deep truth here, one that might not have been seen by Sober or Fodor. It’s the truth implicit in so much work that I’ve been trying to make explicit. Let’s say that parts of an image are abstractions over the details of its syntactic features, along any of the lines illustrated above, and parts of a represented scene are abstractions over its most specific details. So understood:
Parts of images are images of parts.
It’s not a slogan to rule the tattoo parlors or anything, but it neatly captures the point I’ve been making.
Balog, K. 2009. Jerry Fodor on nonconceptual content. Synthese 170: 311-320.
Fodor, J. 2008. LOT 2: The Language of Thought Revisited. Oxford: Clarendon Press.
Goodman, N. 1976. Languages of Art, 2nd ed. Indianapolis: Hackett.
Haugeland, J. 1989. Representational genera. In Having Thought, Cambridge, MA: Harvard UP, 1998: 171-206.
Hughes, R.I.G. 1997. Models and representation. Philosophy of Science 64: S326-S336.
Kulvicki, J. 2006. On Images: Their Structure and Content. Oxford: Clarendon Press.
——. 2007. Perceptual content is vertically articulate. American Philosophical Quarterly 44(4): 357-369.
——. 2010. Knowing with images: medium and message. Philosophy of Science 77(2): 295-313.
——. 2014. Images. New Problems of Philosophy. London: Routledge.
Scholz, O. 2000. A solid sense of syntax. Erkenntnis 52: 199-212.
Sober, E. 1976. Mental representations. Synthese 33(1): 101-148.
Swoyer, C. 1991. Structural representation and surrogative reasoning. Synthese 87: 449-508.