In a review of John Haugeland’s Having Thought, Dan Dennett called Haugeland’s essay “Representational Genera” a “stunning piece” and a “display of philosophical move-making of the highest order” (Journal of Philosophy, December 1998). I, too, admire this piece by Haugeland, but it seems to have gotten little attention (the piece was originally published around 1990). In the spirit of paying homage to Haugeland, I’d like to (briefly) sketch the main thesis of the essay. I’d also like to know whether others are familiar with the piece and, if so, what they think of it. My reaction to the piece was that if Haugeland is correct, then this would force us to restructure significantly the ways discussions of representation have been traditionally framed (and could potentially help resolve many thorny problems in the philosophy of mind and psychology).
As the title of his essay suggests, Haugeland wants to explore the distinguishing features of representational genera. Though he ultimately wants to consider whether distributed representation is a genus, he begins with the familiar distinction between language-like or logical schemes of representation and image-like or iconic schemes of representation. According to what Haugeland calls the “canonical” account of the distinction, the distinction is not to be found in the contents represented but rather in relationship between representations and their contents. So, for example, the canonical account has it that a sentence describing a dog can have the same content as a picture of a dog.
Haugeland spends a great deal of time trying to show why the canonical account of the distinction cannot be right. Haugeland goes about this in a number of ways, but the perhaps most powerful route—rhetorically and probatively—is his debunking of several “outlandish theses” concerning the alleged ease with which we can “translate” iconic representations into equivalent (content-wise) logical representations (and vice-versa). For example, one might be tempted to think that by taking a picture of a written copy of a description, one can produce an iconic translation of the description. Not so, says Haugeland, and to think otherwise is to confuse representing with recording. The photo of the description records the description, and if the photo is clear, one may be able to recover the content of the description from the image. But the content of the photo is not the content of the description.
So what distinguishes genera? Haugeland argues that it is content: iconic representations have different contents from linguistic representations. The former represent what Haugeland calls “relative” elements, whereas the latter represent “absolute” elements. The following passage may help convey both the basis of the distinction and why it is important:
Insofar, however, as contents are worldly, it may seem that logical and iconic contents can overlap or coincide. We can, for instance, say—it is a fact—that the Earth is round. Is this not the very same “structure or feature of the world” that would be represented also by a silhouette of the Earth against a bright background? I don’t think so. In the first place, the sentence identifies (the fact comprises) and particular object and a specific property of that object, whereas the silhouette identifies no object or property: its skeletal content is just the overall pattern of light and dark from some perspective. But further, though the sentence is entirely compatible with the Earth being transparent or just as bright as the background, the silhouette is not; and so on.
The thesis that representational genera are distinguished according to the structure of their contents yields an unexpected dividend: it explains and therefore supports the observation made earlier that “translating” from one genus to another requires wits. If the skeletal contents of two generically different representations—say, a picture and a description of the scene of the crime—differ qualitatively, even in their basic structure, then, in particular, neither includes the other. That is, much (or all) of what the one representation represents is simply not represented at all by the other. Thus, a description of a situation does not “say” how the light values vary with angle of view, any more than a photo of those values “graphs” what objects are present with what properties. Hence, a witless conversion is not possible, simply because the content is not there to convert. On the other hand, a system with wits of its own—background familiarity with the world and the circumstances—might be able to “tell” or to “see” what that missing content would have to be, and fill it in. Often, for instance, a person, relying on knowledge and experience, could tell that a certain light pattern would normally issue only from an object with a certain property, and could thereby supply the logical content needed to “translate” an image into words (pps. 193-194, Having Thought).
Obviously, a lot more could be said here about the views and arguments Haugeland advances, and I have completely ignored some important parts of the essay (I find that it resists easy summary). However, my hope is that the above is enough to at least begin a discussion of his position. I would be especially interested to hear what others think about the claim that iconic and linguistic representations represent different contents and, if you think there is merit in this claim, what the consequences might be for various positions on, e.g., perceptual content, the epistemological relationship between perception and belief, the knowledge argument, and so forth.
Martin, thanks for this interesting post. I probably read Haugeland’s essay a long time ago; in any case I don’t remember it at all.
That being said, it seems to me that a lot here hinges on what we mean by content. If “content” means referent, then the received view seems to hold in at least a modest form: both iconic and linguistic representations may represent the same objects and their properties (albeit they may well represent these objects and their properties in different ways).
But if “content” means something else, such as modes of presentation or something like that, than I think I can make sense of Haugeland’s thesis–i.e., senteces and images have different kinds of modes of presentation, which are mutually untranslatable.
Does this distinction between referents and modes of presentation help to clarify what Haugeland is trying to say? Can you say more about what Haugeland means by “content”?
Thanks for asking this question, Gualtiero; I meant to say a bit about it in the original post but then forgot to do so. I’ll post more later, but for now, here is a footnote Haugeland inserted in 1997:
“This use of the term ‘content’ is not altogether standard. Most contemporary authors (and I, in the other essays in this volume) mean by the “content” of a representation something distinct from the object it represents, and which determines that object (as sense determines referent, for instance). Here, however, I mean by ‘content’ that which the representation represents–the “object” itself–but as it is represented to be (whether it is that way or not). Thus, it is a possible object–which may in fact be actual, or similar to something actual, or neither.”
Hi Martin – I remember liking this article very much, though my memory of the content is a bit foggy! Here’s a stab at what my reaction was (and still is, I think):
I like looking at this as a continuum of representational formats/contents, where fully detailed full-scale models and sentences are at the two ends of the continuum. The fully detailed scale model represents everything via isomorphism and similarity (iconicity), whereas the sentence’s isomorphism to a fact is very rarified and abstract, if it’s present at all. The fully detailed scale model has elements that stand in for objects, and those elements also display (iconically/isomorphically) all the features of those objects. (So the object is represented “relatively” – perhaps this means it’s not “identified” in Haugeland’s terms.) In a photograph, some of the isomorphism is lost, namely the 3D structure – but every object still has iconic representational structure to it. In a drawing, even more isomorphism is lost – a nose might be represented as a simple dot, where none of the intrinsic properties of the nose are iconically represented; the only representational isomorphism concerns the location of the nose with respect to the rest of the face. Same with a map. On the London Underground map, the relations represented by isomorphism become even more abstract, and fewer of them are represented; with graphs even fewer, and more abstractly. Now the elements (e.g. the dots on the graph) become more “absolutist” in Haugeland’s terminology. They stand in for objects, where the structure of the rest of the representation depicts (via abstract isomorphism) only a few facts about the objects. In a sentence, this process reaches its highest degree of abstraction. Maybe the only relation represented iconically here is object/property structure, for instance. The words stand in for objects and properties, and they’re set into a subject-predicate structure that mirrors object-property structure. (Wittgenstein-Tractatus, Sellars, and Millikan, yup.)
So, my reaction to Haugeland: relative and absolute representation are two ends of a continuum. Is this a format or a content difference? Well, I’m not sure what to say (“content” is rather a term of art, after all), but I’m inclined to think this picture lies closer to the traditional view that it’s a format distinction. Translation from a single sentence to a relative/iconic representation will often be impossible, but that’s just because iconic representations always say too much about the relevant objects – they can only express *many* sentences. As soon as they report only a single fact, they’re thereby rendered absolutist. (I suppose you could represent a single fact by multiple iconic representations that overlap in content *only* for that particular fact?)
Thanks for the response, Dan; I think your proposal for how “relative” contents can become “absolute” (via abstraction) is interesting, and I wonder whether you think it can be extended in the following way: we can always introduce a name to refer to the content of an icon, so if the content of the name is just its referent, then icon content can get expressed by sentences that use the name.
My worry, though, is about the kinds of contents that get generated by, e.g., atomic sentences. You mention predication and the Wittenstein/Sellars view on “picturing” (Haugeland discusses this, too). As Haugeland notes, the problem here is that photographs are supposed to be paradigm cases of iconic representations, but photographs do not have subject-predicate structure (the photo “of” my cat Sketchy does not “say” that Sketchy is a cat). So if we want to say that both photographs and sentences represent by picturing, then won’t we still have to introduce different kinds of structured contents? If so, then it may be incorrect to say, of any photograph, that there is some set of sentences that is equivalent (content-wise) to that photograph–more sentences just yield more contents of a different kind.
What do you think?
Your proposed extension is interesting, and sounds right to me – a foolproof way to get a translation from an icon to a sentence. But it also sort of seems like cheating. 🙂
Suppose you’re right that the photo of Sketchy doesn’t say that Sketchy is a cat. Would you agree it says other things, e.g. that a certain pattern of light occurred at a certain 2D position relative to the camera (or a certain range of 3D positions)? If it doesn’t say that, why not? And if it does say that, can it say: there was something that looked like furriness (from a certain angle) at a particular location? There was something that looked like a cat (from a certain angle) at a particular location? And that these things all enter into certain relations?
I agree that it is cheating, but it is bound to look like fair play if you think that referring is representing (this is related to Haugeland’s point about recording being distinct from representing. If the two are a collapsed, then translation will be easy).
Haugeland has to avoid a trap here. If I ask him what the content of a photograph is, and he proceeds to tell me, cannot I say “Aha, you just represent the photo content using sentences!” But this reaction rests on the cheating you mentioned. Sure (Haugeland might say), I can use sentences to say all sorts of true things about the content of a photo, but it does not follow that any of those sentences represent the content of the photo. So perhaps Haugeland would say that the photo represents a patten of light from some perspective, but the photo doesn’t say THAT a certain pattern of light occurred from this perspective. Furthermore, in characterizing the content as I just did, I did not represent the pattern
Of course, Haugeland wants to allow that if a photo is accurate, then a person can be in a position to figure out that certain non-iconic representations are accurate, e.g., ‘there is a cat’ is true.
Is the view sounding more plausible now, or less? ;
fta: Insofar, however, as contents are worldly, …
I have the book on the shelf, and find no marginal comments on the pages of this article. Haugleand is in the Philosophy of Mind Hall of Fame for his “GOFAI” neologism, and is a clear writer. However, I’m of the school that believes a Newfangled AI may be just around the corner.
Where I depart from Haugeland is right there at the beginning. Who says that contents are worldly? Not me. The world is worldy. Contents are contents. I’m very big on deflation! If there is a relationship between contents and the world, that does not make contents worldly, nor does it make the world content – or wouldn’t we all be omniscient?
Which leads us to the idea of representational genera. Who says there is a cannonical form? What if, say, language, is more normative than absolute? I suppose one can taxonomize the contingent and ephemeral, like collecting snowflakes, but it may not have the direct significance that one would no doubt like to acquire.