Presented at the Department of Philosophy, CUNY Graduate Center, 5 February, 1997
Department of Philosophy U-54
University of Connecticut
Storrs, CT 06269-2054
When I was revising Sensory Qualities there was a period of about a year when I set the manuscript aside and did other things. When I returned to it I found that certain portions of the argument had collapsed of their own weight, like an old New England barn, and could be carted off the premises without compunction. Other parts were wobbling on their foundation, while some had weathered well and seemed nice and solid. My revision strategy was simple: I kept just the nice solid bits, thinking that I could go back and work on the wobbly portions later.
This evening I want to talk about one of those wobbly portions: the problem of how to handle the apparent spatial character of sensory experience. This is a problem about which I have felt considerable confusion for a number of years. It seems to me now as if I feel less confused, though this might be a higher-order illusion that will be dispatched in the course of the proceedings this evening. At any rate I want to sketch the lines of an argument that seems to me to work, and that has made my feelings of confusion subside.
The problem posed by the apparent spatial character of sensory experience can be illustrated in any one of numerous modalities. Consider color vision. There are enormous and widely publicized difficulties in coming to understand how a state of the nervous system could be such as to yield a sensation of something red as opposed to something green. What sort of property of the nervous system, if any, could be such as to give me an impression specifically of red? But I think at least equally pressing is the question: what sort of property of the nervous system could be such as to make the apparent red appear out there, in front of the eyes? A model which suggests that hue, saturation, and brightness exhaust the dimensions of variation in color appearance would be true in a world in which there is one sentient subject, confined to a pitch black room, allowed to see just one visible point, whose color qualities are varied in just those three ways. Add even a second point, or such vagaries as ambient illumination, a surround, shadows, glossy surfaces, mists, or reflections, and one quickly sees that even after the questions about sensory qualities have been answered, there is more work to do. Colors appear to characterize surfaces or volumes, which have relative locations, shapes, and arguably depths as well; and these locations, shapes, and depths are all as much characteristics of how things appear as are colors themselves. It is odd enough that things look red; but at least equally puzzling is the question: how might states of the visual system suffice to make the red appear out there?
Arguably every sensory modality presents appearances that have some spatial character. An ache, cramp, or sprain seems to fill the afflicted body part, while sciatic pain shoots alarmingly from one place to another. Barefoot you feel the textures and temperatures of those portions of the external world lying out there beyond your toes. With warm feet you can still feel a cold floor as something cold, outside of you. Even smells have enough spatial character to be tracked. Consider how hunger can prompt you to follow your nose into the kitchen. Watch a dog sniffing excitedly along a game trail and there can be little doubt that to it smells indicate a direction, and perhaps even a time. Deer went that-a-way, not-so-long-ago.
I think the mystery of these spatial attributes has a particularly compelling form in the case of audition. That we have the capacity to detect the direction of sounds was first demonstrated by the Italian physicist Giovanni Venturi in 1796 (see Murch 1973, 206). He blindfolded subjects, sat them in an open field, and asked them to hold their heads still. He then walked around, piping an occasional note on a flute, and asked them to point in the direction of the sound. Even with more sophisticated methods, errors in localization are surprisingly low, reaching a maximum of about 15 degrees. That is about the angle occupied by your fist and thumb when held out at arm's length, so even the largest errors are surprisingly small.
We sense both auditory direction and depth. Traffic noises provide an example. One can detect not only the direction in which a noisy truck or a siren is traveling, but something of its distance. Somehow one hears a loud siren in the distance as a loud siren, far away. When you think about it, this feat is quite remarkable. After all, a less energetic source of sound nearby--a computer in the next room equipped with a sound card, say--could produce compressional waves that transmit to your eardrums the same amount of energy in the same frequencies. Sometimes indeed one is confused by such stimulus configurations, and one might hear the rattling of the radiators as some tremendous commotion out doors. Yet typically one parses the energy of compressional waves into separate components of loudness and distance. You hear the loud siren as a loud siren far away, and not as a softer simulation from next door.
Spatial perception is an intriguing and mysterious subject. The mystery is that it is possible at all. Your auditory sensations are utterly dependent on events within your cochlear nucleus, inside your inner ear. But what you hear seems to happen out there in space. How is this possible? How can you auditorally locate sounds outside your skin, given that all your auditory information starts with events inside your cochlear nucleus? I think we are here asking for a psychological model, and indeed there are various accounts of the cues used by the nervous system to manage this remarkable feat.
Auditory localization was long thought impossible, on two grounds: first that sounds have no shape, and second that unlike the eyes or the skin the ears do not employ arrays of receptors spread out in space(see Boring 1942, 381-2). It took awhile for psychologists to recognize that the differing locations of the two ears alone might do the trick. Unless the source of the sounds is directly ahead of you or behind you, the compressional waves will strike one eardrum with greater intensity than the other, and the system does use these intensity differences to help determine auditory direction. More importantly, because one ear is slightly further away from the source than the other, there will also be slight differences in the phase of the sounds at the two ears. The waves crest at one ear at a slightly different time than they do at the other. The auditory system actually picks up these tiny time differences-on the order of a few milliseconds--and uses them as cues for azimuth. Given a known distance between the ears and a known time difference, it is a matter of trigonometry to compute the difference in distances from left ear to source and from right ear to source. Presumably the mechanisms which yield a perception of auditory azimuth from phase differences in some sense employ those trigonometric equations, even if their host had conscious access to those equations only briefly and unreliably, during trigonometry class.
The spatial order intrinsic to sounds has some peculiarities, however. At best the cues yield a ratio of the distance from left ear to source and from right ear to source. Imagine a triangle whose base is the line between the two ears and whose third vertex is the source of the sound. If there is a point in front of you that satisfies that ratio, there is also one behind you that satisfies the same ratio. Just flip the triangle over. It turns out that subjects quite often make front-back reversals in locating a sound, pointing behind themselves when the sound is actually presented in front (and conversely). They will however point more or less precisely at the reversed spot, so that (for example) a sound 30 degrees off the midline in front will sometimes be localized as 30 degrees off the midline in back (and conversely), with an average error of 15.6 degrees in either case. The two points have the same distances to both ears, so are liable to be confused (see Gulick et al. 1989, p. 320).
I am sure you have experienced such auditory spatial illusions. Telephones in offices open to a hall provide a nice example. Standing in the hall, it is sometimes easy to be fooled into thinking yours is the one ringing. I find quite often this is a front-back reversal. The rings are brief enough, and the hall smothers enough other cues, that it sounds as if a telephone up the hall is ringing when in fact it is one down the hall. You rush to your door, only to discover at the next ring that it wasn't yours ringing at all.
So suppose we admit that any complete catalog of the properties of appearance must include spatial properties and relations. Along with hue, saturation, and brightness we need visual azimuth, altitude, and depth. Now I want to argue that any account of the organization of sentience must treat the spatial properties of appearance in a very different way than it does the others. In particular, the "standard model" we use to understand qualities such as hue, saturation, or brightness does not work when applied to visual locations. Although visible hue and visible depth both characterize how things appear, they play different roles in the organization of sentience.
What I call the "standard model" proceeds as follows. First (to use Galen Strawson's useful 1989 terminology): we distinguish between "phenomenal properties", which characterize appearances, and "qualitative properties", or those properties of sensation in virtue of which things appear as they do. Phenomenal properties are attributed to things in front of the sense organs--the sunset that looks red, the floor that feels cold, and so on--while qualitative properties are properties of sensory states, which if they have location at all certainly do not have it out in front of the eyes or beyond the toes. The idea is that the variations in how things appear are to be explained by positing qualitative properties--properties of internal states, in virtue of which the things in front of the sense organs appear as they do. These are qualia. So we get a theoretical correspondence between phenomenal properties and qualitative properties, with the latter posited so as to explain the former. (Endnote 1)
This maneuver is I think strikingly successful in many domains. We first identify colors as properties of appearance. Differences in color characterize some dimensions of variation across which one can discriminate among things. You find that arrangements of colored chips--physical pieces of paper--require a three dimensional ordering, with dimensions (for example) of hue, saturation, and lightness. Naturally we suppose that there must be properties of states of the system sufficient to explain its capacities for discrimination. The properties on which the discriminations hinge are the qualitative properties, properties of sensation. If our descriptions are sufficiently rich we gain the wherewithal to identify the actual neural mechanisms responsible for these capacities of discrimination. When that is done, we will have closed the explanatory gap and can all go home. As you might know, I think in principle this can be done, and that we will not all have to stay on the job forever.
But there is a problem applying this model to the spatial characteristics of appearance. On its terms we should posit, for each dimension of variation in the spatial appearances of things, spatial qualia, or additional properties of sensation, in virtue of which things appear where they do. Nelson Goodman is surely correct that location and other spatial characteristics are properties of visual appearance just as much as color (Goodman 1977, 138-9). If we think of "qualia" as phenomenal properties, it is fine to include among qualia colors, times, and visual-field places, as Goodman did. But suppose we employ the standard maneuver, and hold that for each of these dimensions of variation in appearance there is a corresponding qualitative property, or property of a sensory state. This clearly departs from Goodman, but many of the followers of "local sign" theories in perceptual psychology, from Wundt onwards, do precisely this. Nevertheless there is something very fishy (and, I think, rather funny) about this idea.
What is funny are the vistas it opens into varieties of mental disorganization that were previously unimaginable. For example, it is possible to see two distinct things that appear exactly the same shade of red. On the standard model those things both cause visual states that have the same qualitative properties, and in virtue of that, they look the same. But now we propose to handle visual field locations in the same way. This opens the logical possibility that I have two distinct visual states that both have the same place quale. It is logically possible. So I could be presented with two distinct places that appear at the same place. How unsettling. Even worse, on this model, the place at the center of the visual field appears to be there because it causes a state with the appropriate place quale. But sometimes, alas, things aren't as they seem. Something which is not "really" red might occasionally look red. It causes a state with the same qualitative property as something which is red. On the standard model we would have to admit the possibility that places might also cause states with inappropriate place qualia. A place which is not at the center of the visual field might appear to be there. It causes a state that sports the appropriate spatial qualia. We might have several contenders simultaneously. How very disconcerting. One would require tests to ascertain whether or not the visual field places really are where they appear to be. One would need to proceed gingerly. Is that the center of my visual field, or does it just appear to be there? Until such tests are complete, standing up, walking, grasping a coffee cup would become activities fraught with peril.
Even at the level of sensory experience we insist on the distinction between qualities and the locations they occupy, or between sensory qualities and the places that those qualities appear to characterize. A sensory quality is something of which there could be multiple simultaneous instances. An instance of given quality can move, and come to characterize different places. Finally, there might be times during which no instance of that quality is present. You listen intently before the concert starts, and silence seems to fill the hall. The experience has a certain voluminous character to it; you can hear that you are in a large hall. But the places that are mentioned in characterizing these appearances cannot be treated in the same fashion as sensory qualities. On pain of unimaginable mental disorganization, one cannot have two numerically distinct instances of the same place in the visual field. A place in the visual field cannot move relative to its neighbors. And even though there may be no instance of a particular quality occupying a given place in a sensory field, the place itself cannot be absent. Otherwise in the pre-concert experience, there would be nothing for the silence to fill!
We have to handle the spatial character of sensory appearance in a different way than we handle its other qualities. Here is a simple argument for this claim. Suppose you agree to participate in an experiment by a latter-day Giovanni. You are blindfolded like the 18th century participants. This Giovanni however has instruments that can reliably produce pure brief tones--beeps, cheeps, or clicks. As he moves about, you are asked two questions: have the auditory qualities of the tone changed at all, and: where is it now? I submit that you could answer these two questions separately. In a slight variant, one can imagine hearing two simultaneous beeps or clicks in different locations. (Endnote 2.) Out in the woods one might hear two apparently identical bird cheeps, one off to the left and one overhead. Perhaps the only thing that appears to differentiate the cheeps is their location. But if this is so, then differences in apparent location do not preclude the possibility that the cheeps themselves are qualitatively identical. It follows in turn that whatever properties of sensation account for differences in apparent location, those properties of sensation are not among those counted when we consider the question of whether those two sensations are qualitatively identical. We treat pitch and place in very different ways.
I said before that we seem to have the capacity to "parse" the energy of compressional waves striking our eardrum into separate components of distance and loudness. "Parse" is not quite the right verb; it is better to think of this in terms of partitioning the variance (of transducer energies) into separate components (of loudness and distance). I am thinking here of the statistical notion of analysis of variance. Such analysis proceeds as follows. Suppose members of a population vary in some attribute of interest. If we know that those members belong to two distinct groups, we can treat the overall variation from person to person as a resultant of two components: the variation within groups, and the variation between groups. You can look at the proportion of variation that each explains and ask whether the "between group" variation is "statistically significant". (Endnote 3.)
What this argument shows is that the variation in phenomenal appearance in a sensory modality is partitioned into two components: variation in location, and then variation in the qualities at those locations. Two presentations might differ only in apparent location, as with the two qualitatively identical cheeps, one heard overhead and one to the left. (Endnote 4.) This argument applies to any sensory modality in which it is possible to have two distinct, simultaneous, qualitatively identical presentations. You hear two qualitatively identical cheeps, you see two qualitatively identical red spots, you feel two qualitatively identical pin pricks. Any such modality presents a manifold of appearances differentiated spatially. We could not distinguish two qualitatively identical presentations as two unless presentations in that modality have a spatial character. It seems we treat the spatial character of appearance as something other than attribution of qualities.
Now let me try to explain why we partition variation in phenomenal appearance into these two components. I will argue that the distinction between the spatial and qualitative components of variation in phenomenal appearance is at root the distinction between reference and predication, or between singular term and general term. I want to put forth a general hypothesis: that we treat sensing as a primitive variety of mental representation--the most primitive variety--and treat those spatial attributes of appearance as products of the proto-referential component of that system of representation.
Mechanisms to account for the spatial character of sensory experience have a referential or proto-referential function. They function something like singular terms. Mechanisms that serve to characterize the qualities that appear at those locations have a predicative or proto-predicative function. I say "proto referential" and "proto predicative" because these functions are similar to reference and predication in a natural language, but lack most of the sophisticated capacities found in reference (or predication) proper. Nevertheless there is an ancestral kinship and the homologies are striking.
Here is the general hypothesis. I call it the "feature-placing" hypothesis. Sensory representation, when successful, proceeds by identifying places and characterizing qualities that appear at those places. Typically, when the identification succeeds, these places are physical places, physically in front of the sense organs. Visually the root form is something like "here it is red; there it is green", where the "here's" and "theres" identify the places in front of the eyes where the red and the green appear to be. For "red" and "green" substitute any of the visible qualities we find to characterize those places. In audition: "here is a cheep, there is a honk", where "here" might specify a direction overhead, and "there" to your left. In bodily sensation we have "here it tickles, there it hurts" where again the here and there latch onto particular bodily places, and the tickles and hurts stand for any of the somesthetic qualities one experiences as occupying such places. Crudely, when you have a pain in your foot, your state of sensing pain-in-the-foot does something like refer to your foot (it identifies that place, as opposed to your other foot or any other place) and it attributes to it a particular somesthetic quality (pain as opposed to ache or chill).
Now why should we think of sensory experience as having anything like a referential character? There are some good reasons. Sometimes standard varieties of reference in a natural language rely on the receipt of sensory experience; they do not succeed unless one has the appropriate sensory experiences. And experiences could not do this unless they have at least a proto-referential character. They could not secure the "direct reference" of a designator unless we can find some of the capacities of reference in sensory experience itself.
Consider, for example, demonstrative identification. Suppose we want to learn the Oklahoma terms for native Oklahoma fauna, and to do so we travel to the Oklahoma savanna--the tall grass prairie--and hire a guide. As we sit, watching the tall grass waving in the endless wind, our guide gestures and says "That is a prairie dog". A moment later, "That is a prairie chicken". We want to walk away knowing what is a prairie dog and what is a prairie chicken, and as you might guess, these are not peculiar varieties of dogs or chickens that happen to live in Oklahoma. (A prairie dog is a species of marmot--a member of the same genus as woodchucks--and a prairie chicken is a species of grouse.) What is required for such demonstrative identifications to succeed?
Our guide produced two tokens of the demonstrative pronoun "that". They have differing reference--first to the prairie dog, then to the prairie chicken. You have to identify the two referents of the two tokens. Clearly you could know all the linguistic rules governing the use of demonstrative pronoun "that" and still not be able to do this. Those linguistic rules presumably do not change between the first tokening and the second--the demonstrative has what Kaplan calls a fixed "character"--yet the reference of the two tokens differs. You might have a perfect grasp of whatever meaning the language associates with the pronoun "that" yet still fail utterly to discern its shifting reference.
So what is required for such demonstrative identifications to succeed? You need to determine what each "that" refers to. Evidently the identification relies on capacities that are extra-linguistic. You must see--or somehow perceive--what the guide is talking about. If you are blind, distracted, or looking in the wrong direction, for instance, the guide's utterances will be totally ineffectual. But even if you have normal vision, and you are looking attentively at the waving grasses, the identification still fails unless you can see which of the simultaneously visible items the guide is attempting to identify on your behalf. Prairie chickens in particular can be hard to spot. You might be looking straight at it but fail to discern it among the grasses and hummocks. A minimal condition for a successful identification is that you have some capacity to discriminate, from among the simultaneously visible regions, some that are currently occupied by prairie chicken and some that are not. This condition does not necessarily require a sharp visible border between them, but if you cannot make any such distinction--any distinction between regions occupied by prairie chicken, and those that are not--then clearly you have not picked out what the guide is trying to identify, and the demonstrative identification will fail. Seeing an undifferentiated blur of co-present items will not suffice. "I don't see what you mean" you might say. You must discriminate among those items, and see which one the guide means.
Not only are some such sensory processes necessary for the success of the demonstrative identification, but in some cases they are sufficient: they make the difference between success and failure. With all the other stage-setting fixed, it is these sensory processes that make the last critical step to securing an identification. Suppose you are having trouble picking out the prairie chicken. The guide can of course provide descriptions that help you to see what he means. He might say "Do you see the three hummocks with the rock on the left? Ten feet in front of them is something that looks like a shadow. That's a prairie chicken." But this description does not itself provide you the wherewithal to use the term appropriately thereafter. It only works if it helps you to see which thing is the prairie chicken. If you see it, then and only then do you understand what the guide's demonstrative "that" refers to, and only then do you learn the reference of the term "prairie chicken". So it is those visual processes--seeing which or seeing what--that make the difference. Your ongoing sensory discrimination of the object secures the reference of the demonstrative pronoun. (Endnote 5.)
But sensory processes can in this way secure an identification only if they themselves have some capacity to refer. They can secure the reference of the demonstrative pronoun only if they can provide a route to identifying its referent. You see what the guide means. Any system that can do that has capacities at least akin to those of reference. So sensory systems do something like refer to things. They have proto-referential powers.
I admit this suggestion is initially outlandish. The difficulty in clarifying it lies in clarifying the sense in which sensing achieves proto-reference, but not reference proper; or clarifying the sense in which having a pain in your foot does or does not do "something like" refer to your foot.
One notion I have found to be extremely useful in clarifying these claims is Peter Strawson's notion of a "feature placing language". (This is why I named the general hypothesis about the organization of sentience the "feature placing hypothesis".) Strawson is well aware that the capacity to refer is not an all-or-none affair, but that it comes graded, in rather intricate hierarchies of levels of complexity. Recall his description of a hierarchy of language types in his Subject and Predicate in Logic and Grammar (1974). Only at the very top do we have the full resources of the apparatus of individuation found in a human natural language. Nevertheless in the middle portions of the hierarchy, and even lower, we find capacities and accomplishments that are clearly necessary for full blown reference, and in many ways resemble the most sophisticated versions, but which, just as clearly, lack some critical ingredients.
Of particular interest to those of us mired in the muck and goo of raw sentience is the language-type at the very bottom of all the hierarchies. This is what Strawson calls a "feature placing" language. (He mentions the notion also in Individuals (1963) and in a 1954 Proceedings of the Aristotelian Society. See also O'Leary Hawthorne & Cortens 1995.) Such a language has an extremely simple lexicon, as it includes only indexicals and deictical demonstratives, the present tense of the verb "to be", and what Strawson calls "feature" terms: non-sortal universals or mass terms. Since the language makes no tense distinctions, the verb is actually redundant. Some examples of feature-placing sentences are:
This is coal. That's gold.
Snow is falling.
There is ice here.
Now it is raining.
These demonstratively identify a place, and ascribe to it a feature. Feature-placing sentences name no particulars, and often have the form of a subjectless Lichtenbergian proposition: "It is raining. It is cold."
It is startling to see how much introspective content can be cast in this form. We have not only the Lichtenbergian "It thinks", but the colloquial "It hurts" and "It stinks." Thanks to Sellars, we also have "It pinks". My "red here, green there" is already in feature-placing form. The sentence demonstratively identifies places and attributes features to those places. The sensation of a red triangle next to a green square does something analogous; it latches onto places and attributes features to them. But the latter does not require or proceed in a natural language; any creature that can sense a red triangle next to a green square has the requisite representational capacity.
The feature term is replaced by some property of central states (probably some determinate value of an activation pattern) whose function is to represent a dimension of variation in phenomenal appearance. The demonstrative is replaced by some determinate value of activation patterns of mechanisms of spatial discrimination--the ones in the given modality responsible for tracking where the feature in question is to be found. So instead of a demonstrative term like "here" or "there" we have the activation of mechanisms of spatial discrimination sufficient to track the place in question. They identify the place not with a word, but in the same way that a setter or bloodhound can point to the location of the fox or of the drug cache.
Or, to vary the example, they pick out the location in the same way that a lioness lying in the grass of the savanna tracks the location of a particular gazelle. Even though the lioness has no words, she definitely has something targeted, something in the cross-hairs, whose position and distance is gauged very carefully indeed. A philosopher set loose on the savanna would be unwise to think "she cannot identify my location, since her lexicon lacks demonstrative terms". The feature-placing content operative in the lioness might be something of the form "prey there" (or perhaps "food there"). Instead of the demonstrative "there" we have mechanisms of spatial discrimination which are currently activated so as to track the location of that very same (and we hope tasty) academic, soon to be valued not for his insights, but for his protein content.
I should admit that there is a lot of neuroscience behind this hypothesis as well. Let me lift the veil briefly. When I talk about "feature-placing" sensory representation, I hope to be talking about the very same thing that neuroscientists refer to as "feature maps" in primary and secondary sensory areas of neocortex. What neuroscientists mean by a "feature" is roughly a "sensory quality": some determinate value along a dimension of variation in phenomenal appearance. Some areas of cortex are called "feature maps" for two reasons: first that all the cells in the region seem to be responding to some particular dimension of variation in phenomenal appearance (and states of the cells are states registering those variations); and second that those cells are arranged in roughly somato-topical order, so that adjacency relations are more or less respected. Cells next to one another in the retina typically project to cells next to one another in the lateral geniculate, which in turn usually project to cells next to one another in striate (or primary visual) cortex (area V1). So cells in V1 are in that sense a "map" of retinal receptors. This is by no means a road map, nor is it a straightforward topological mapping, since there are many tears and discontinuities in the projection. The most notable one is that retinal cells in both eyes are divided down the mid line, and cells in the right hemifield of both retinas project to the left cortex, while those in the left half of both retinas project to the right cortex. Furthermore the mapping is non-isotropic; areas better populated with receptors are better represented in the cortex, even if this means that the spatial metric is stretched or compressed proportionately.
Now if you find a feature map, you can always ask two distinct questions about any portion of it. (1) What territory (what spatial region) is mapped by this portion? and (2) What features does this portion of the map attribute to that territory? These are the proto-referential and proto-attributive components of the representation.
How common are these feature maps? In vertebrate sensory physiology they seem to be everywhere. You will not go far wrong if you summarize the neuroscience in a series of headlines: Layered Somato-Topical Distributed Feature Maps (see Arbib 1972, 1989). In fact typically there are many feature maps per modality; in vision one has separate maps for hue, lightness, local motion, shape, spatial frequency, retinal disparity, and so on. Even within one dimension of variation in phenomenal appearance such as color, one finds multiple feature maps at different levels within the neuroanatomical hierarchy: blob cells in V1, thin stripes in V2, projections to V4, and so on (see Davidoff 1991, 20-25). By one count, in the macaque monkey we have now identified 32 distinct visual feature maps (see Felleman & Van Essen 1991). 32! Either the design works or Mother Nature is a wastrel.
Now it is quite likely that a feature map, or something like it, is the neural instantiation of what I am calling a "feature placing" representation. I think also that the sensory features I am talking about are exactly the ones that cognitive psychologists talk about in their "feature integration" theories of attention (e.g. Treisman & Gelade 1980). But with that brief glimpse, I let the veil drop.
I said that the referential character of sensory representation is similar to that of a feature placing language. Now why call either of them proto-reference? Strawson says the placing of a feature is the simplest possible analog for the attribution of a property to an individual. It includes primitive analogs of both reference (to a place) and predication (of a feature). But the analogs are quite primitive; these languages are at the very bottom of the hierarchy. Strawson says:
If any facts deserve, in terms of this picture, to be called ultimate or atomic facts, it is the facts stated by those propositions which demonstratively indicate the incidence of a general feature. These ultimate facts do not contain particulars as constituents but they provide the basis for the conceptual step to particulars. The propositions stating them are not subject-predicate propositions, but they provide the basis for the step to subject-predicate propositions. (Strawson 1963, p 218).
Feature-placing sentences do not introduce particulars because they lack the identity predicate, they lack counting criteria, they lack the wherewithal to make any sort of distinction between "another one" and "the same one again". They lack negation and all other truth functional connectives. They make no tense distinctions. Similarly, the features attributed to places are not properties. Two instances of the same feature are at best qualitatively identical--there is nothing that can distinguish them --but feature-placing language lack identity, so they lack the means to say that there is anything strictly identical in the two instances. (Endnote 6.)
According to Strawson, these languages do not contain sufficient resources with which to frame notions either of re-identifiable particulars or of properties. They lack the means to introduce the notion of an individual. Feature-placing sentences "provide [some of] the materials" for the introduction of particulars, but do not (and cannot) themselves make that introduction (Strawson 1954, p. 244.) The difficulties are instructive. The closest we can get to an individual is a locus at which multiple features can be placed:
The individual instance .... emerges as a possible location-point for general things other than the feature of which it is primarily an instance... (Strawson 1954, p. 249-50).
Similarly distinctness of placing underwrites distinctness of instances:
The considerations which determine multiplicity of placing become, when we introduce particulars, the criteria for distinguishing this patch of snow from that, or the first fall of snow from the second. (Strawson 1954, p. 245).
To modify one of Quine's examples, suppose we have a white cat facing a bristling dog (Quine 1992, 29). To say "white cat" is not just to say white and cat, since the latter might be satisfied by a white dog facing a bristling cat. Conjunction is too loose. Whiteness must characterize the same place as that occupied by the cat. We want the features to coincide or at least amply overlap. (Endnote 7.) Predication serves admirably: there is something which is both white and a cat. What underwrites this introduction of a particular--this x that is both white and a cat--is ultimately the overlap of features at a place. (Endnote 8.) To use Strawson's memorable metaphor: you put pressure on the idea of a particular, trying to reduce it to its ultimate logical constituents--and at the limit of pressure, it unfolds into a feature-placing fact (see Strawson 1963, p. 218-19).
One might say that feature-placing languages have a proto-referential character, but they lack the means for full blown reference. The formulation sounds familiar. In fact if sensory representation in general proceeds by characterizing or filling out space in various ways, then sensory representation fits into feature-placing language like a hand in a glove. There could be no more natural expression of how sentience represents. I can now give content to the suggestion that sensory representation is proto-referential: it has exactly the referential character of a feature-placing language.
I hope that pointing out some of the regrettably primitive characteristics of feature-placing removes the sting from the claim that an extra-linguistic mental capacity has something like the power to refer. But as a final step, I want to argue that this claim is one that we have to accept anyway, for other reasons. Or at least we do if we accept any version of the causal theory of reference. If you accept that theory, the moral of our attempt to spot the prairie chicken can be generalized, perhaps alarmingly.
According to the causal theory, the reference of a designator is established by a causal connection between the term and the referent itself. The connection can become remote and indirect, as one can pick up the use of a referring term from other people. But ultimately even such "chains" of assisted reference must be "grounded" in the object itself. (I take the term "grounding" from Michael Devitt's 1981 book.) It is very interesting to think about what has to happen for such a grounding to take place. The paradigm is a "baptism" or any introduction of a new proper name in the presence of the person or thing named. If the term is a term for a natural kind, then its reference is secured only if one can trace the reference back to some episode in which an instance of that kind is present and can be identified demonstratively. One might learn how to use a sortal by a demonstrative identification of an instance. And demonstratives themselves require a rather immediate link to extra-linguistic reality. In all these cases of "direct reference" we typically trace the chain back to some direct face-to-face perceptual encounter with the referent of the designator. Something has to happen during that encounter sufficient to fix the reference of that designator to that referent. What has to happen? Here's what Devitt says:
In a grounding a person perceives an object, preferably face to face . . . The grounding consists in the person coming to have "grounding thoughts" about that object as a result of the act of perceiving the object. A grounding thought about an object is one which a speaker of a public language would express using a demonstrative from that language... However, I see no reason to deny that beings which do not speak public languages could have grounding thoughts. If they do have them, then they must have appropriate mental representations. Call all these representations of the object in grounding beliefs "demonstrative representations". The act of perception leading to these representations defines a mode of presenting the object. It leads to an ability to designate made up of these grounding thoughts. (Devitt 1981, 133)
These "demonstrative representations" are required for the grounding to succeed; without them the name fails to gain a reference. Devitt describes the naming of his cat Nana:
The chain underlying my first use of 'Nana' begins with Nana at her naming ceremony; it runs through my perception of that ceremony; from then on it is my ability thus gained to use 'Nana' to designate her. (Devitt 1981, 29)
The grounding "runs through" the perception of the thing named. Suppose chains of indirect and assisted reference for proper names and natural kind terms are all ultimately grounded in face to face perceptual encounters with the objects designated, and those chains must "run through" particular demonstrative representations, caused in the language user by those encounters. All subsequent reference depends upon those demonstrative representations. It is their perception of the baptism that gives attendees the power thereafter to use the new name to designate, and to pass that ability along to others.
But now it is hard to see how those grounding representations could do their job if they themselves failed to identify what is designated. And notice that these mental representations are by hypothesis extra-linguistic. It would be fatal to the theory if one had to know a language in order to have the requisite grounding representations. So on this account, direct reference ultimately relies upon some extra-linguistic capacities to pick out what is meant.
I think the required capacities might be as minimal as the bare discrimination of differently occupied regions of space. The ability to designate might be grounded in the capacity to discriminate between regions occupied by the referent in question and others simultaneously present that are not so occupied. In short, the grounding might proceed through feature-placing. It probably requires nothing more than feature-placing for terms such as "col", "cornice", and "couloir", or other terms for morphological or geographical features. Typically though your success in grounding episodes is much greater than mere spatial discrimination. You can not only discriminate some region where there is prairie chicken from some region where there is not, but actually see the outline or borders of the animal, and discern its shape and size. Visual processes of edge detection, segmentation, grouping, and size and shape constancy all kick in; you see that region as a bounded volume having a particular shape and size. This is typical, and helpful, but not necessary: the identification can succeed even if one cannot pick out the precise borders of the target or discern its shape. The target may lack precise borders, size, or shape, as in our mountaineering terms. Such spatial discrimination is sensory, and it is sometimes sufficient to secure the identification. What I have called "proto-reference" in feature-placing reduces to these capacities of spatial discrimination. (Endnote 9.)
Perhaps the tie between reference and these extra-linguistic capacities of spatial discrimination runs even deeper. Consider what is necessary to move upwards in Strawson's hierarchy, to transcend the limits of mere feature-placing. The decisive step is the dividing of reference: the step achieved when one can treat "more cat" as either "another cat" or "the same cat again". For this one needs the identity predicate, counting and count nouns, sortals, singular terms: what Quine calls the "apparatus of individuation". Without the capacities of spatial discrimination it is very hard to see how one could learn--or how one would have any use for--that apparatus.
Suppose you lost all spatial discrimination in all of your sensory modalities. You could never subsequently have the experience of simultaneously present regions several of which manifest the same feature. Our qualitatively identical bird cheeps from overhead and from the left would fuse into one. You might still have occasion to use the one word observation sentence "cat", but as far as I can see you would never have--and could never have--an experience that would lead you to say "not just one cat, but two". You might count a series of cat cries, but not simultaneously sensible cats. One needs the capacity somehow to discriminate among distinct simultaneously present instances. Without it, experience might present episodes of "more cat", but never force upon us a choice between one cat and another. There is no dividing of reference without dividing space.
Consider again Quine's "white cat facing the bristling dog". One thing which is white and a cat is facing another thing which is bristling and a dog. In one word observation sentences this might be "White. Cat. Bristling. Dog." But such observation sentences fail utterly to distinguish Quine's scene from one in which a white dog is facing a bristling cat. We can distinguish these two scenes because we can see that the cat's place is the place that is white, and it is bristling where it is dogging. But if we lost all capacity for spatial discrimination then we could not experience such spatial coincidences and divisions of features. We would have four simultaneously present features, but no capacity to divide the scene among them. In such a world it is indeed hard to see how one could ever make the "decisive step" to cat particulars.
Kathleen Akins has recently (1996) described sensation as "pre-ontological". The label is apt. If sensation has an ontology, it is an ontology without objects, without particulars. The same is true of feature-placing languages. I think the congruence is exciting: working upwards from sensory capacities, and downwards from subject and predicate in logic and grammar, we meet at feature-placing. The representation of features in space is arguably the most sophisticated of sensory capacities. From my point of view, starting down in the muck and the goo of raw sentience, such spatial representation is a complex and sophisticated achievement. Yet here it is sitting just below the least sophisticated of linguistic capacities, those sufficient for a feature-placing language. And if we can make that one tiny upwards step, Strawson and his successors can take it from there.
Back to Austen Clark online papers.
Back to Uconn Philosophy home page.
1. This maneuver is an old one. Sellars 1963 described it with his notion of the "counterpart properties", ascribed to sense impressions as part of an explanatory theory. These are not the same properties as the red, green, blue (and so on) ascribed to manifest objects, but their relations of similarity and difference are modeled on the relations among sensible qualities, and are isomorphic to them. (Back)
2. Ernst Weber pointed this out in 1846, using the example of two watches whose ticking noises could not be discriminated from one another. With the watches placed on either side of the head, one can hear two distinct instances of qualitatively the same ticking noise. See Boring 1942, 383. (Back)
3. What is mysterious about auditory localization is that the auditory nervous system does not know the "groups" beforehand, yet it still manages to partition variance in transducer energies into separate components of loudness and distance. (How does it assess the loudness of the sound without knowing its distance, or its distance without knowing its loudness? That's the mystery.) (Back)
4. Suppose we try to collapse this partitioning of phenomenal variation, so that all such variation is counted as qualitative. All variation in apparent location is to be treated as variation in qualitative character. As a simple model for this thought experiment: suppose you think of the visual field as a two dimensional array of minima visibilia, each of a particular color. But now we are to treat the two dimensional location of each point in the field as itself another sensory quality--a qualitative "local sign". We must attach a quale of location to each point. Thankfully, since each point is visible (minimally!), on the backside you could attach a separate color tag, with the hue, brightness, and saturation of the color on the tag used to indicate where in the visual field the color on the front is to go. We tag all the points in this fashion and toss them into the bin. (The tag is the "local sign".) All the variation in appearance is now accounted for by variation in qualitative character. For Wundt and the believers in qualitative local signs, that tagged jumble is what it's like to have a visual field. (Back)
5. See Evans 1985b: The demonstrative identification succeeds only because of an on-going "perceptual link" to its object. At least in some situations sensory processes can make the difference between success and failure of a demonstrative identification. They can secure the reference of the demonstrative; a description of the "completing sense" of the two tokens must mention them. (Back)
6. In a feature-placing language we cannot say of a feature that two places instantiate the very same one, where by "very same" we mean "strictly identical". As Strawson says, "The idea of a property belongs to a level of logical complexity which we are trying to get below" (Strawson 1963, p 209). For a civilized Oxbridge philosopher such downward movement--this shucking of complexity--is indeed a struggle, but Strawson has some success. (Back)
7. "The conjunction is fulfilled so long as the stimulation shows each of the component observation sentences to be fulfilled somewhere in the scene--thus a white pebble here, a blue flower over there. On the other hand the predication focuses the two fulfillments, requiring them to coincide or amply overlap. The blue must encompass the pebble. It may also extend beyond; the construction is not symmetric." (Quine 1992, p. 4) (Back)
8. "The conjunction is too loose. It tells us only that the four things are going on in the same scene. We want them all in the same part of the scene, superimposed. It is this tightening that is achieved by subjecting the four-fold conjunction to existential quantification, thus: "Something is catting and is white and is dog-facing and is bristling". ... An object has been posited, a cat." (Quine 1992, p. 29). (Back)
9. As Ruth Millikan puts it: "The pointing finger is understood only if what it points at is visible or otherwise independently identifiable. Or suppose it is the job of "that" to point out a direction, "that way". ... The interpreter must be able independently to identify that direction, not necessarily with a name ("east", "west"), but, say, via an ability to track it, to know what it would be to continue following that selfsame direction, as opposed to turning away from it. To know what an indexical indexes, to identify the indexed, requires that one have a second route to thinking of it, a route other than the indexical token, and that one grasp this second route as one arriving at the same referent." (Millikan 1993, p. 271). (Back)
Akins, Kathleen (1996). Of sensory systems and the "aboutness" of mental states. Journal of Philosophy, 337-72.
Arbib, Michael A. (1972). The metaphorical brain; an introduction to cybernetics as artificial intelligence and brain theory. New York, Wiley-Interscience.
Arbib, Michael A. (1989). The metaphorical brain 2 : neural networks and beyond. New York: Wiley.
Boring, Edwin G. (1942). Sensation and Perception in the History of Experimental Psychology. New York: Appleton Century.
Campbell, John (1994). Past, Space, and Self. Cambridge, Mass: MIT Press. A Bradford Book.
Clark, Austen (1993) Sensory Qualities. Oxford: Clarendon Press.
Davidoff, Jules. (1991). Cognition through Color. Cambridge, MA: MIT Press.
Devitt, Michael (1981). Designation. New York: Columbia University Press.
Evans, Gareth (1985a). Collected Papers. Edited by Antonia Philips. Oxford: Oxford University Press.
Evans, Gareth (1985d). Understanding demonstratives. In Evans 1985a, 291-321. Originally published In Meaning and Understanding, eds. H. Parret and Jacques Bouveresse. (Berlin: W. de Gruyter, 1981).
Felleman, D. J. and Van Essen, D. C. (1991). Distributed hierarchical processing in the primate cerebral cortex. Cerebral Cortex 1: 1-47.
Goodman, Nelson. (1977). The Structure of Appearance. 3rd edition. Boston: Dordrecht Reidel.
Gulick, W. L., Gescheider, G. A., and Frisina, R. D. (1989). Hearing: Physiological Acoustics, Neural Coding, and Psychoacoustics. Oxford: Oxford University Press.
Kaplan, David (1989a). Demonstratives. In Joseph Almog, John Perry, and Howard Wettstein (eds), Themes from Kaplan. New York: Oxford University Press.
Kaplan, David (1989b). Afterthoughts. In Joseph Almog, John Perry, and Howard Wettstein (eds), Themes from Kaplan. New York: Oxford University Press.
Millikan, Ruth. (1993). White Queen Psychology and other Essays for Alice. Cambridge Mass: MIT Press.
Murch, Gerald M. (1973). Visual and Auditory Perception. New York: The Bobbs-Merrill Company, Inc.
O'Leary-Hawthorne, John and Cortens, Andrew (1995). Towards ontological nihilism. Philosophical Studies 79: 143-65.
Perry, John (1977). Frege on demonstratives. Philosophical Review 86: 474-97.
Quine, W. V. (1992). Pursuit of Truth. Revised Edition. Cambridge, Mass.: Harvard University Press.
Sellars, W (1963). Science, Perception, and Reality. London: Routledge & Kegan Paul.
Strawson, Galen (1989). Red and 'red'. Synthese (78), 193-232.
Strawson, P. F. (1954). Particular and general. Proceedings of the Aristotelian Society 54: 233-260.
Strawson, P. F. (1963). Individuals. New York: Anchor Books, Doubleday & Company, Inc. Originally published by Methuen & Co Ltd, 1959.
Strawson, Peter. (1974). Subject and Predicate in Logic and Grammar. London : Methuen.
Treisman, Anne & Gelade, Garry (1980). A feature-integration theory of attention. Cognitive Psychology 12: 97-136.
Return to Austen Clark's online papers .
Return to the Philosophy Department home page.