Erkenntnis 17 (1982): 171-211.


R. C. Richardson and G. Muhlenberg

1. Much of contemporary Anglo-American philosophy in the tradition of logical positivism can be broadly characterized as 'reductionist'. Among the premier convictions of this tradition is a belief in the primacy of science and in its ultimate substantive and methodological unity: science if not as we know it today, then Utopian science provides us with the total and correct account of the world. Wilfrid Sellars is one heir to this tradition who has attempted to propound a philosophical system based on such convictions. We will focus on Sellars' treatment of sense impressions (raw feels), and will be primarily concerned with the account of these non-conceptual elements of perceptual propositional attitudes within the Sellarsian Scientific Image. The crucial issue, of course, will be the introduction of what Sellars has called "sensa".

It is Sellars' contention that no straightforward reduction of sense impressions, or sensations is possible. In Science, Perception, and Reality, Sellars argues in the following way:

Where both thoughts and sensations are conceived by analogy with public observable items, in the former case the analogy concerns the role and hence leaves open the possibility that thoughts are radically different in their intrinsic character from the verbal behaviour by analogy with which they are conceived. But in the case of sensations, the analogy concerns the quality itself.... The crucial issue then is this: can we define, in the framework of neurophysiology, states which are sufficiently analogous in their intrinsic character to sensations to make identification plausible?

The answer seems clearly to be "no". This is not to say that neurophysiological states cannot be defined (in principle) which have a high degree of analogy to the sensations of the manifest image.... The trouble is, rather, that the feature which we referred to as "ultimate j", and which characterizes the perceptible qualities of things, e.g. their colour, seems to be essentially lacking in the domain of the definable states of nerves and their interactions....

Nor do we wish to say that the ultimate homogeneity of the sensation of a red rectangle is a matter of each physical particle in the appropriate region of the cortex having a colour: for whatever other difficulties such a view would involve, it doesn't make sense to say of the particles of physical theory that they are coloured. And the principle of reducibility... makes impossible the view that groups of particles can have properties which are not "reducible to" the properties and relations of the members of the group (Sellars, 1963, p. 35)

The above 'grain argument' has been misunderstood and underestimated by Sellars' opponents. This derives, in large part, from a failure to appreciate the extent to which the argument is embedded in the context of Sellars' thought. This has resulted in the obscuring of two crucial facts: first, that the argument gains its power by being generated from a relatively small number of highly plausible premises; and. second, that Sellars is not, in the long run, pressing for irreducibility but providing just one element of an extended argument designed to accommodate sense impressions within a reductionist program. This has made it all too easy, on the one hand, to point to what appear to be implausible consequences and think that Sellars' argument has thereby been dealt with, or, on the other, to think that the argument can simply be ignored.

Sellars' presentation is dialectical, involving a series of successive approximations and revisions. He begins with the Aristotelian conception of persons as single logical subjects which stand in relations to physical objects having occurrent qualities, including coloredness. Sellars' first revision leads us to a conception of Aristotelian persons standing in relations to non-colored complexes of physical particulars. The apparent color of these complexes is explained in terms of sense impressions. His second revision yields a conception of persons as complexes of physical particulars standing in relations to other complexes of particulars. These persons can correctly be said to sense, but sensing must be taken as an emergent (i.e., non-reducible) state. The final sophistication -- the Scientific image -- is again a conception of persons as complexes standing in relations to other complexes. Persons, thus construed, sense but their sensings are reducible states, and among their constituents are sensa.

There are, in fact, two 'grain arguments' (cf. Sellars, 1971, p. 408). The first 'grain argument' dictates the first revision, by showing that being colored, taken as an attribute of physical objects. would be nonreducible. Similarly, the second dictates the second revision, by showing that sending (i.e., having a determinate sense impression) would also be non-reducible. In both cases, what precludes reduction is the property of homogeneity -- in the first case an attribute of colored objects and in the second as applied to sense impressions.

Just as invoking a primary secondary quality distinction disolves the first barrier to reduction, so by the introduction of sensa (as theoretical constructs rather than as 'sensory objects') Sellars hopes to circumvent the second. We propose to reconstruct Sellars' argument for the introduction of sensa. Parts II, V, and VI contain the substance of this reconstruction. Parts I, III, and IV contain more global considerations designed to provide the framework in terms of which the substantive argument can be and must be understood.


2. The Manifest Image, construed in broad terms, is the world as conceived in ordinary experience. Sellars' most illuminating characterization of the Manifest Image is that of a refinement, both categorial and empirical (cf. Sellars, 1963, p. 7), of what he terms the "Original Image". The universe of the Original Image is completely anthropomorphized; i.e., all the parameters of description and explanation applicable to persons apply as well to 'inanimate' objects. Accordingly, the Manifest Image is the result of a gradual depersonification of the Original Image. Inanimate objects are no longer thought of as acting (thinking, perceiving, etc.) and, consequently, material objects emerge as a distinct category of entities.

3. It should be emphasized that the Manifest Image is an idealized conceptual framework intended to exhibit a way of thinking common to us all. It is an idealization in that it is an oversimplification: in even moderately reflective moments, we are surely aware that the world does not consist simply of Aristotelian particulars and persons. But, Sellars reminds us, this should not lead us to suppose that we ".. . must therefore perceive the world in terms of the conceptual framework of atoms in the void." (Sellars, 1967, pp. 297-298). We operate with a sophisticated schizophrenia: in our guarded and reflective moments we acknowledge the world of science; yet in everyday action and perception we proceed as if the world consisted of Aristotelian particulars and persons.

The Manifest Image should also be viewed as pseudo-historical. It is dubious (to say the least) that man at any point in history has operated with merely this framework. Thus, in a sense to be elaborated below (section 6), the Manifest Image serves as no more than a convenient heuristic fiction.

4. Contrasted with this common sense framework is the Scientific Image. It too is an idealisation, though in a dual sense. Like its Manifest counterpart, the Scientific Image is an ideal type, and, hence, an oversimplification. In addition, it is an idealization in that "... it is an integration of a manifold of images..." (Sellars, 1963. p. 20). We are confronted with a plurality of relatively autonomous sciences. Strictly speaking, there are as many scientific images as there are sciences. Sellars' Scientific Image is a vision of a coherent unitary system for the explanation and description of the world.

In short, the Scientific Image is the Sellarsian conception of Utopian science. It is this Utopian science which Sellars, following Peirce, claims would be the final arbiter of truth. A familiar objection commonly raised against the contention that the scientific enterprise will yield a unique theoretical framework which is totally adequate for explanation and description is Quine's contention, as against Peirce, that

it seems likelier ... that countless theories would be tied for first place. Scientific method is the way to truth, but it affords even in principle no unique definition of truth (Quine, 1960. p. 23).
Quine has since become less sanguine about the underdetermination of theories, and no longer regards it as obviously true that there will be non-equivalent theories capable of accounting for all the observational evidence.1 It might be thought that this would relieve us of the need to respond to Quine's objection, but this is not so: for it can still be pressed that Sellar's supposition that there is an ideal limit to science is unfounded.

Such a contention would be well-taken. But even though Sellars seems at first blush to be subject to such an attack, further consideration shows this stand -- and, indeed, even a stronger underdetermination thesis -- to be entirely amenable to Sellars' views. Sellars, despite showing some ambivalence, need not be committed to the view that the Scientific Image is attainable. This image should be taken to be, in such the Kantian sense, a regulative ideal. With this in mind, Sellars claims that a philosopher should

. . . attempt to envisage the world as pictured from that point of view -- one hesitates to call Completed Science -- which is the regulative ideal of the scientific enterprise. (Sellars, 1959, Chapter V, p. 335; Sellars 1968, Chapter V. p. 75).

The Scientific Image is not a piece of prophesy. The vision of a completed science is simply that assumption which is the conditio sine qua non for making sense of scientific progress and the scientific enterprise in general.

5. We seem, then, to be confronted with two distinct conceptual frameworks, both public and non-arbitrary, each claiming to be correct and, in principle, complete visions of man-in-the-world. But are we in fact now confronted with a plurality of conceptual schemes? A literal reading of much of Sellars would yield an affirmative reply. Yet we think such a view would militate against the spirit of the Sellarsian enterprise. The scientific view of the world is not so disjoint from our 'naive' conception as such a reading might suggest.2 Sellars says that if

. . . scientific discourse is but a continuation of a dimension of discourse which has been present in human discourse from the very beginning, then one would expect there to be a sense in which the scientific picture of the world replaces the common sense picture; a sense in which the scientific account of "what there is" supersedes the descriptive ontology of everyday life. (Sellars, 1963, p. 172).
Thus, there is an intimate connection between the frameworks of science and common sense, and the former in some sense 'replaces' or "supersedes' the latter.

The sense in which the framework of science replaces or supersedes that of common sense is not one which requires the simultaneous presence of fully developed alternatives. The replacement is not replacement in toto but replacement piecemeal. Conceptual replacement is an evolutionary process. Replacing elements of our conceptual framework piecemeal, like replacing the planks of Neurath's ship individually, will eventually yield something qualitatively different. In short, alternative conceptual schemes are presented only across spans of time,

6. Although we have claimed that the simultaneous presence of fully developed alternative conceptual schemes is not required for Sellars' theory, he often speaks as if we are confronted with such alternatives. Thus, he characterizes the Manifest and Scientific Images as "rival", "conflicting", and "clashing" images. Such expressions, and the model of the images in which they apply should be viewed as essentially metaphorical and designed as convenient heuristic fictions. The 'dialectical' portrayal of the relation of the 'two images' leads Sellars to depict diachronic relations as synchronic; that is, to treat historical relations as contemporary. Sellars' dialectical portrayal has, in fact, led critics astray. Thus, having pointed out quite correctly that the Manifest Image "is to a certain extent a myth, like the 'average man'" and that "It would be a miracle if any one person's conceptual scheme conformed to it". C. A. Hooker goes on to urge:

Moreover, once one admits that the Manifest Image is an idealized construction of a certain sort extracted from a changing kaleidoscope of conceptual schemes the neat separation between Manifest and Scientific Images is broken down. There is, and will be multitudinous transitional conceptual schemes ... All of these choices represent significant differences in both Images with some transitional stages already evident. It is not at all clear, therefore, that a suitable 'reading' of the Manifest Image will not reconcile it ... to the Scientific Image (Hooker, 1977, 342).
Having seen clearly that Sellars' talk of Images is not a piece of armchair anthropology, one would think the next step would be to expose what Sellars hopes to convey by such talk. After all, Sellars says that his construct of Images is " . . . designed to illuminate the inner dynamics of the development of philosophical ideas . . . ". (Sellars. 1963. p. 5). It in fact illuminates the dynamics of the development of ideas. Dynamics invoke change, and change is not synchronic, but diachronic. Of course, to acknowledge this and to explain in any detail how the static, synchronic, model is to be related to a dynamic, diachronic, model are quite different tasks. To accomplish the latter task, one must come to see the centrality of Sellars' views on reduction, explanatory coherence, and analogical concept formation; furthermore, these must be understood as they pertain to the dynamics of theory construction and evolution. Dealing with the Sellarsian expression of these three views is the task of Parts II, III, and IV (respectively). Once this is done, we will be in a position to see how they constitute the motive force behind Sellars' views on sense impressions.


7. The goal of the Sellarsian enterprise is to outline a unified, consistent conceptual framework adequate for explanation and description. Falling back again to Sellars' heuristic model, we find ourselves confronted with the pretensions of common sense and contemporary science to be correct accounts of the world (cf. Sellars, 1963, 25). In evaluating these conflicting claims, we are faced with three alternative lines of thought:

  1. Manifest objects are identical with the systems of imperceptible particles countenanced by micro-theory;
  2. Manifest objects are real, while the entities postulated by science are merely 'calculaljonal devices', and
  3. Scientific entities are real, while Manifest objects are merely appearances to the human mind.

The second alternative is extrinsic to the issues with which we are concerned, and we will not deal with instrumentalism but will assume it to be inadequate. For the remainder of this part we will be concerned to evaluate the first alternative and Sellars' reasons for rejecting it. This will take us some distance toward an understanding of Sellars' principle of reducibility (PR).

8. Sellars states his PR as follows:

If an object is in a strict sense a system of objects, then every property of the object must consist in the fact that its constituents have such and such qualities and stand in such and such relations... (Sellars, 1963, p. 27).
Put in a more perspicuous fashion, the PR states that
Given an object O such that O is identical with a system of objects (O1, O2, . . . On), then for every property P correctly attributable to O, O's having P must consist in the properties of and the relations between O1, O2 ..., and On.
The PR is obviously concerned with the reduction of properties. In this context, we can delimit two mutually exclusive and exhaustive categories of properties in terms of their modes of reducibility.3

What we shall term ''structural properties" can he defined in the following way:

Given an object O and a property S correctly ascribable to O, S is a structural property if for some system of objects (O1, O2, . . . , On) identical with O. S is correctly ascribable to O without being correctly ascribable to any of (O1, 02, . . . , On).
Structural properties are unproblematically reducible. To take one of Sellars' examples, that a ladder is identical with a system of objects (i.e., the rungs and the frame) is evident from the fact that for this system as a whole to be a ladder is for its parts to be of a certain kind (e.g., wood) and for these parts to stand in certain (spatial) relations to one another. However, not all properties fall into this category. Thus, while temperature is patently a structural property, mass is not.

Mass appears to be what we will term a "content property", and content properties can be defined as follows:

Given an object O and a property C correctly ascribable to O, C is a content property if for any system of objects (O1, O2 ..., On) identical with O, C is correctly ascribed to O only if C is correctly ascribable to some (perhaps all) of (O1, O2 . . . , On).
For example, consider a red brick wall. That the brick wall has the content property redness and that the brick wall is identical with a system of objects (the bricks) entails that each of the bricks has the property redness.4

9. Given this principle of reducibility, we can now turn to an evaluation of the first option stated above, viz. Manifest objects are identical with systems of imperceptible particles. The crucial problem for this thesis arises when we acknowledge that objects as conceived in the Manifest Image are colored. It is, according to Sellars, an essential categorical feature pertaining to manifest objects that they have color in literally the same sense in which they have shape (cf. Sellars. 1963, p. 60; Sellars, 1971. p. 394, p. 395n).

This is the 'Direct Realism' which Sellars holds to be characteristic of the Manifest Image. And, moreover, the manifest object is conceived of not only as colored but as colored in a particular way. Sellars tells us that

Pink does not seem to be made up of imperceptible qualities in the way in which being a ladder is made up of being cylindrical (the rungs), rectangular (the frame), wooden, etc. The manifest ice cube presents itself to us as something which is pink, through and through, as a pink continuum, all the regions of which, however small, are pink. It presents itself to us as ultimately homogeneous (Sellars, 1963, p. 26).
Suppose that, in accordance with alternative (i) manifest objects are identical with systems of imperceptible particles. Now, since manifest objects are homogeneously colored, these systems of imperceptible particles must be homogeneously colored. Since to say that, e.g., a pink ice cube is homogeneously colored is equivalent to saying it is "something which is pink, through and through. . . . a pink continuum, all the regions of which, however small, are pink", it follows that, in accordance with the definition of "content property" (section 8), being colored is a content property. (Nothing actually depends on the ice cube being homogeneously pink: it might, so far as Sellars' point is concerned, appear homogeneously pink and actually be variegated in color. What is important (cf. section 11 below) is that it be homogeneously colored.) Since it is a content property, we can infer, by virtue of the PR, that some of the constituents of the system are coloured. In short, the assumption of alternative (i) commits us to the claim that (some of) the micro-particles countenanced by modern science are colored. But that these particles are not colored surely needs no argument.

The most common objection to this view is to press that color, as we naturally conceive of it, is whatever property objects have that makes them capable of reflecting determinate wavelengths of light. What color may be, in itself, is not revealed from within the Manifest Image. Hooker remarks: "When I turn my mind in upon my own conceptual scheme... I do not find here any concepts of simple, homogeneous properties. What I do find instead is a yawning gap bespeaking ignorance as to the true natures of these properties.'' (Hooker, 1977, p. 341).

The reaction does not lack attraction. But, without even attempting a full scale evaluation, it can be seen to be misguided. In the first place, Sellars is not actually committed to this being part of 'our' conceptual scheme. Galileo took us far beyond the Manifest Image. Under the impact of science, our conception of physical colors is much as Hooker suggests. To take this as an objection is, again, to take Sellars much too literally (see section 6 above). In the second place, the conception of color Hooker finds when he looks within is, in a sense, precisely the one Sellars' subsequent argument is driving for. Hooker's conception of physical color is a dispositional concept (and this is why we have a "yawning ignorance" of its "true nature"), whereas the color Sellars is speaking of at this early stage of the argument is conceived of as an occurrent property. What Sellars' first grain argument is pressing for is the need to abandon such occurrent properties. Once done, we could (though Seilars does not) go on to talk of the remaining concept of physical color qua a dispositional property. But to do this requires that we introduce or have at hand the concept of a sense impression; for these physical colors are, after all, those properties which enable objects to reflect determinate wavelengths of light; and if we are to explain why objects appear to be colored in the occurrent sense, this evidently demands the light reflected produce, in perceivers, the appropriate sense impressions. The objection here to thinking of colors as homogenous, occurrent, properties already grants Sellars what he hopes to gain through the first "grain" argument.

10. Reconstructing Sellars' first 'grain argument' in the above manner should clarify the structure of the argument by emphasizing its dependency on the PR. We take this argument to be quite straightforward, but there is a need to reiterate one facet of the argument which has been overlooked by commentators. Being colored is a content property. The first application of the 'grain argument' revolves around the manifest conception of coloredness qua homogeneous-coloredness. Thus, we take C. F. Delaney to be incorrect in claiming that homogeneity is the content property which is decisive for the argument.5 To say merely that a thing is homogeneous is misleading at best. More properly, we should say that a thing is homogeneously-thus-and-so (e.g., homogeneously-colored). Homogeneity simpliciter should not be thought of as a property of individuals, though it could be thought of as a property of properties of individuals. Sellars does not, after all, claim that homogeneity is a quality of things, but that it "...characterizes the perceptible qualities of things" (Sellars, 1963, 35). The force of the 'grain argument' above rests on the fact that the (equivalent) propositions

  1. "x is homogeneously P" and
  2. "x is P and P is homogeneous" entail the following (equivalent) propositions:
  3. "x is P and some (perhaps all) of the constituents of x are P" and
  4. "x is P and P is a content property".
To say that manifest objects are homogeneously-colored is to say, at least, (a) manifest objects are colored, and (b) color is a content property. The point of Sellars' claim that manifest objects are homogeneously-colored is that color, as conceived in the Manifest Image, cannot be construed as a structural property.

11. In order to further clarify the concept of homogeneity as employed in the argument, we would like to entertain one objection. Reinhardt Grossman comments:

The physicist does not claim that the spatial structures which are tables are not colored. He merely maintains that certain very small spatial parts of these structures, elementary particles, are not colored.... But this should not come as a philosophical shock to anyone. We do not have to consider such esoteric entities us elementary particles in order to see that the spatial parts of a perceptual object may not have the same properties as the perceptual object of which they are parts (Grossman, 1975. p. 131)
Grossman is, evidently, right to think such a possibility is not an occasion for bewilderment: structural properties generally exhibit this trait. Similar concerns have been raised by Hector-Neri Castaneda (in correspondence) and, perhaps, by Keith Gunderson (in 'The texture of mentality'). If we turn to the example used by Gunderson, we can come closer to understanding Grossman's worry, and dispensing with it. Gunderson considers a poster of Dick Tracy's face, which when viewed from a distance appears homogeneous but when viewed from close up is seen to consist of a series of dots. Drawing on this, he distinguishes between "perspectival non-identity'' and "non-perspectival identity". Tracy's face and the system of dots are perspectivally non-identical; i.e., from different perspectives. Tracy's face appears to have different characteristics. Yet they are non-perspectivally identical; i.e., it is the same thing viewed. The example, when generalized, leads to the thesis that " . . . items which are perspectivally non-identical can be non-perspectivally identical."

This is obviously true, since it amounts to nothing more than the claim that one and the same thing can appear differently from two perspectives. Superficially, it might seem that the 'grain argument' could be 'analyzed sway' via this distinction. In short, one might claim that

(1A) From P1, O appears homogeneous;
(2A) From P2, O appears non-homogeneous;
(where P1 and P2 are different perspectives and "O" designates the object) are consistent. But (as was remarked in section 10) to speak of homogeneity simpliciter is confused. Amending (1A) and (2A) accordingly, we have:
(1B) From P1, O appears homogeneously colored;
(2B) From P2, O appears non-homogeneously colored.
The claim, of course, would be that (1B) and (2B) are consistent. Unfortunately, there is an ambiguity. If (1B) and (2B) are interpreted as
(1C) From P1, O appears colored and all the regions of O appear colored;
(2C) From P2, O appears colored and at least one region of O does not appear colored;
respectively, the claim to consistency is reasonable. Following Gunderson's example, the regions of Dick Tracy's face would include both the dots comprising it and the spaces between the dots. Put intuitively, anything is a region of a two-dimensional surface if it can be defined by any grid square of any grid size. (It is a matter of indifference to us whether or not a partially colored region is labeled "colored" or "non-colored", since, for any partially colored region, there will be a sub-region which is a region and which is not even partially colored.)

However, if (1B) and (2B) are interpreted not as (1C) and (2C), but as

(1D) From P1, O appears colored and all the constituents of O appear colored.
(2D) From P2, O appears colored and the constituents of O do not appear colored;
respectively, they cannot be jointly maintained. Again following Gunderson's example, the constituents of Dick Tracy's face would include only the dots (not the spaces between the dots). Clearly, the dots are not un-colored.

A criticism of Sellars following Gunderson's analogy must fail simply because Sellars' appeal to the ultimate homogeneity of colors is a claim not that the regions but that the constituents of colored objects must be colored. It follows, then, that to argue from the consistency of (1C) and (2C) is to commit an ignoratio. If Gunderson's claim is to legitimately apply to Sellars'argument it must be interpreted in the (1D) (2D) fashion. But (1D) and (2D) are inconsistent.

If this interpretation of Sellars' 'grain argument' is correct, the claims that (a) being colored is a content property, and (b) objects which are colored consist of (i.e., have as constituents) micro-particles, do not commit us to the further claim that the spaces between the microparticles are colored. They do commit us to the claim that the microparticles are colored. But, of course, this is the absurdity the grain argument is designed to point out. Gunderson's criticism of Sellars relies on the consistency of maintaining objects to be colored while the spaces between the constituents of the object are not colored -- a claim Sellars would surely not deny. Gunderson's attack would be potent only if we could sensibly maintain that Dick Tracy's face could consist of a system of uncolored dots -- a claim (Gunderson would surely not make. If Grossman is not beset by the more grave error of overlooking Sellars' distinction between content and structural properties, he too may have failed to see what is demanded in order to reduce content properties.


12. The first 'grain argument', using Leibniz's law and Sellars' PR, is designed to reveal the unacceptability of alternative (i); i.e., the view that the Manifest and Scientific Images are consistent systems operating at distinct levels of analysis. The remainder of this paper will evaluate and examine the claim that science provides a superior account of the world. It is Sellars' view that insofar as we have good reasons for espousing a theory which countenances entities of a given sort, we have good reasons for claiming entities of that sort really exist (cf, Sellars, 1963, pp. 95-96). With reference to our concerns, this is tantamount to the ontological claim that scientific objects are real. And this, in turn, given the previous argument, gives us good reason to maintain that the macro-entities of the Manifest world do not 'really' exist (cf. SPR. p. 96). For, assuming scientific realism, if manifest objects are real, they are identical with complexes of scientific objects. And since Manifest objects are essentially colored, it would follow that these scientific objects are also colored. But this claim abrogates the reductionist program.

13. It is therefore a version of the third alternate above -- scientific realism -- which Sellars advocates. Sellars says, in the statement of his Scientia Mensura (cf. Sellars, 1971, p. 396), that in matters of description and explanation "... science is the measure of all things, of what is that it is, and of what is not that it is not" (Sellars, 1963, p. 173). This is simply to accord ontological priority to science. Sellars is careful to divorce his claim for ontological primacy from any further claim to the methodological or epistemological primacy of science. Sellars denies the sciences the latter primacy for the reason that they are dependent upon the framework of common sense for observations, (cf. Sellars, 1967, chapters XIII and XIV; Sellars, 1963, chapter 4).

14. Sellars' commitment to the Scientia Mensura should be viewed as the claim that science provides us with a greater degree of explanatory coherence. Accordingly, the Scientific Image would embody the greatest possible amount of explanatory coherence. This leads us to a formulation of what might be termed an "inter-framework decision procedure". In crude fashion, that framework which serves to most adequately describe and explain the phenomena concerned (i.e., the observable behavior of objects and persons, particularly the perceptual functioning of sentient organisms) is the framework which must finally be taken the measure of what is real.

It is far from clear how such a criterion should be elaborated, and such an attempt would demand that we go far beyond the scope of the present enterprise. It is evident, however, that 'explanatory coherence' is to be multidimensional. Maximizing 'explanatory coherence' would mean, at a minimum, maximizing the integration of theoretical principles, the accuracy of prediction, and the overall scope of the theory in question. Adequacy of description would be evidenced by adequacy of explanation and prediction. What we have termed Sellars' "inter-framework decision procedure" demands maximization on at least the first two parameters.

But how is the claim to the maximum explanatory coherence of science to be justified? What seems to be lacking in Sellars' exposition is a criterion on the basis of which we can judge which framework presented is superior for 'explanation' and 'description'. Keith Lehrer attempts to provide such a criterion for Sellars when he says

The methods of explanation in theoretical science are a paradigm of explanatory methodology and, hence, are best suited to maximizing explanatory coherence (Lehrer, 1966, p. 269).
This is inadequate. The difficulty rests in the fact that there may be many ways of describing and explaining the world, each of which is successful relative to the purposes and ends it is designed to serve. Surely, we cannot take Sellars to be saying that insofar as the Manifest Image gives a scientific explanation and description of the world, the Scientific Image can do this more adequately. For this is tantamount to making the trivial claim that if our task is one of scientific explanation and description, theoretical science is supreme. But if this is not what is intended, what is the paradigm for explanation and description? Indeed, since there do seem to be different forms of explanation, why should we even suppose that there is just one paradigm for 'explanation'? Without some further support and elaboration, Lehrer's proposal cannot do the job it is designed for.

A related reply6 would base the primacy of the Scientific Image on a difference in its methodology: unlike the Manifest Image, the Scientific Image is the product of postulational techniques for theory construction (see Sellars, 1963, 19). Insofar as postulational techniques give more penetrating and general explanations than simple correlational and inductive approaches, they accord the Scientific Image explanatory primacy. Aside from reasons of a more systematic nature -- e.g., that this threatens a genuinely evolutionary approach to scientific nature -- there are reasons to think this an untenable position. Sellars does not maintain the methodological distinction between the Manifest and Scientific Images consistently. When Sellars worked out the analogy theory -- in constructing part of the very heart of the Manifest Image -- both thoughts and sensations are introduced as postulated, unobserved, entities. Thoughts, to be sure, come to acquire the status of 'observables' in that they are subjects of non-inferential reports (in a rather Rylean manner); but, as introduced, they are postulated and unobserved. Sensations by contrast are never given the status of observables: they are always inferred entities postulated to explain the qualitative content of sense experience. But in both cases, the 'inner occurrences' are postulated at the outset; consequently, the use of a postulational, rather than merely correlational, methodology lies at the very center of the Manifest Image. (For more discussion, see sections 20, and 21 below.)

It is a mistake to retort that this is not a genuinely postulational approach. It is, of course, true that thoughts are postulated in such a way -- or so Sellars holds -- that it leaves their ultimate nature untouched (see Sellars, 1963, p. 33). This is a function of the fact that the use of formal analogies (see section 17 below) is central in 'introducing' thoughts (and sense impressions). Seliars holds, though, that the use of formal analogies is central to the introduction of any theoretical entities, in some cases -- e.g., the Mendelian gene -- that will tell us only the function and not the actual realization -- the 'ultimate nature' -- of the postulated entity. The methodology of the Manifest and Scientific Images are one.

The result is a less than comfortable one. We are left with two different approaches: first, we might admit that the 'inter-framework decision procedure' is inappropriate as a justificans for the Scientia Mensura; or, second, we might search for a reformulation of the 'inter-framework decision procedure' which would evade the problem of circularity.

With regard to the second alternative, it should be noted that the statements of the 'inter-framework decision procedure' and of our objection were couched in the metaphor of "Images". This suggests that a non-metaphorical restatement of the principle might alleviate the difficulty. According to the interpretation in Part I (sections 5 and 6), such a non-metaphorical restatement would utilize the notion of alternative modes of reconstruction or revision of conceptual frameworks. What is then required is a criterion for choosing between such alternative reconstructions. Following the terminology introduced, such a criterion might be: given two or more alternative modes of reconstruction or revision, that alternative is best which (a) is the simplest, and (b) maximizes explanatory coherence.

The same problem of circularity persists. Even in the non-metaphorical form, no 'decision' procedure has been provided which can justify the Scientia Mensura, since there is again an assumption that all explanations are of a kind. But even if scientific explanations were all of the same form there is little reason to think that all explanations are of the same form. If the 'inter-framework decision procedure' is to be taken as a justification of the Scientia Mensura and if "explanation" is interpreted as "scientific explanation", then the same trivial circularity results. On the other hand, if "explanation" is not so interpreted, how could it possibly provide any support for the Scientia Mensura?

We conclude that the relation of the 'inter-framework decision procedure' to the Scientia Mensura is not that of justificans to justificandum. The Scientia Mensura functions, rather, as an assumption in Sellars' philosophy. As such it must be regarded as an assumption to be vindicated, if at all, by the acceptability of the results obtained by assuming it. The 'inter-framework decision procedure' is. on this view, based on appeal to the unity of scientific methodology and that, in turn, is based on the Scientia Mensura. As long as we acknowledge the 'inter-framework decision procedure' does not justify the Scientia Mensura, we can, without circularity, take "explanation" in its statements to mean simply scientific explanation. In this light, the 'inter-framework decision procedure' is best seen as a restriction on the construction (or reconstruction) of conceptual frameworks which is grounded on the assumption of the Scientia Mensura.

15. In addition to the 'inter-framework decision procedure', there is a further restriction on the reconstruction of conceptual frameworks which we shall call the "principle of comprehension". It is Sellars' conviction that nothing in the phenomenal field is to be completely repudiated, but rather relocated, and, hence, in an adequate explanation, accounted for. The force of this principle becomes apparent once we contrast Sellars' position with that of, e.g., J. J. C. Smart. Smart also takes it to be (at least part of) the task of the philosopher to adumbrate a synoptic view of the world. However, Smart claims

We can see . . . that colour is in an anthropomorphic concept. . . . If our philosophical task is, in part to see the world sub specie aeternitas, to see the world in such a way as to discount our idiosyncratic human, terrestrial perspective, then we must eschew the concepts of colour and other secondary qualities (Smart. 1963, p. 84).
Sellars would find such a position unconscionable. He writes
. . . what I find objectionable in the views of my tough-minded colleagues is not their "extreme scientific realism," but rather their failure to pay serious attention to the problem of specifying the conditions which an adequate scientific account of human behavior must meet. This problem calls for just that careful examination of what we already know about human behavior in terms of existing categories, for which, with few exceptions, they have no patience (Sellars, 1971, p. 399).
It is a perplexing aspect in the thought of philosophers of Smart's general persuasion that because science has found it requisite to dispense with the unacceptably anthropomorphic view of the Original Image in dealing with inorganic matter, we must, in a somewhat reactionary way, dispense with the 'anthropomorphic' aspects of our everyday explanation of human behavior.

Sellars' avowed position is in fact quite extreme; we think unreasonably so. Roughly, Sellars claims that if a statement within the Manifest Image is true (i.e., is correctly assertible on the basis of semantical rules proper to the Manifest Image [cf. Sellars, 1968, ch. 2, §26, and ch. 3, §§ 50, 52, and 53]), then the counterpart of that statement within the Scientific Image is true relative to the Scientific Image (i.e.. correctly assertible within the Scientific Image) and also true relative to the Manifest Image (cf. Sellars, 1968. ch. 3, § 74). (The final conjunct appears superfluous given other explications provided by Sellars [see especially Sellars, 1968, ch. 3. § 53].) A "counterpart" may be taken, simply, as the translation of one statement in another conceptual scheme (see below, section 18, for details). If we may think of a single proposition as expressible in alternative conceptual schemes, then Sellars' claim is simply that if a proposition is held true in the Manifest Image, then it must also be held true in the Scientific Image. Such a position is clearly too restrictive, since it would allow, as viable changes, only additions to, or elaborations on, previous theory.7

As we have employed it, however, the principle of comprehension requires not that everything held true must continue to be held true but that it must be possible, within later theories, to explain why previous theories were held true. Sellars, and many others, would construe this task to be one of showing why previous theories are 'approximately' true. We could, while maintaining the principle of comprehension, allow as well for the possibility of explaining only why previous theories were held true, despite their not being even approximately true. (The shift from holding colors to be occurrent properties to explaining them away in terms of sense impressions is an example of just this.) Even this restriction is harsh, but it is what is demanded by Sellars' claim that science provides a maximum of explanatory coherence. A (re)construction which did completely repudiate some element could not correctly lay claim to a greater degree of explanatory coherence than its predecessor; for the goal of maximizing explanatory coherence commits us to producing not only a maximum of consistency, and unity of explanatory principles but also a maximum of explanatory scope.8


16. As we have repeatedly slated, Sellars is committed to a form of scientific realism. The discussion in the previous part was designed to introduce and collate two elements which function as restrictions on the (re)construction of conceptual frameworks and, as such, have a direct effect on the nature of the Scientific Image. In parts V and VI we will be concerned with the status of sense impressions in the Scientific Image. As we shall see in part V, the concept of a sense impression is an analogical one. Therefore, it is incumbent upon us at this stage to give a general accounting of Sellars' views on analogical concept formation. Following this, we will briefly investigate the closely allied notion of a successor concept.

17. We can begin by distinguishing between formal and material analogies. Roughly, a material analogy is one concerned with similarities and differences of individuals (i.e., whether or not the properties of the individuals are the same); whereas a formal analogy is one concerned with similarities and differences of properties of the individuals (i.e., whether or not the properties of the properties of the individuals -- or the types of properties are the same). Mary Hesse has essentially the same distinction in mind when she says:

We have seen that the analogy is used in two senses in physical theories -- there is the one-to-one correspondence between different interpretations, of the same formal theory, which we may call formal analogy, and there are pretheoretic analogies. . . which enable predictions to be made from a model. Let us call this second sense material analogy (Hesse, 1966, 68).
In short, formal analogies involve sharing properties of properties of individuals (second-order properties). Material analogies are, in addition, explicitly concerned with the sharing of properties of individuals (first-order properties).

Sellars holds that it is formal analogies which are of paramount importance; indeed, this is the type of analogy he employs in analyzing sense impassions and thoughts. A well known example might help in understanding the use of formal analogies. The analogy which obtains between moments in time and points on a line does not involve moments and points sharing first-order properties; rather, moments and points share certain second-order properties. Thus, moments and points are analogous in that certain of their first-order properties share second-order properties (i.e., their first-order properties are similar, though not the same). For instance, "before" and "to the left of" signify first-order properties which share the second order properties of transitivity and asymmetry. These elements are part of the positive analogy. There are also elements of the negative analogy. To hark back again to Kant, points are represented as simultaneous but moments are only successive.

Formal analogies, as Sellars and Hesse employ them, can be seen as idealizations conceived by analogy with set theory. First, formal analogies involve a 'commentary' which specifies the positive, negative, and neutral elements of the analogy. It is as if we were to provide an alternative interpretation for a set of axioms and specify in a commentary that certain of those axioms are to be abandoned. Second, it is an idealization insofar as what would correspond to axioms cannot always be specified as such (cf. Sellars, 1967, p. 347). In employing formal analogies, though we rarely specify second-order properties in true axiomatic fashion, we are none the worse off, for we can settle for a meta-linguistic turn. Instead of

(x) (y) (z) xRy · & · yRz: → · xRz
we can just as well say
R is transitive.
Thus, the demands of analogical concept formation, as conceived by Sellars, require that (1) we provide a model. i.e., specify a familiar domain of objects which have familiar (kinds of) properties; and (2) provide a commentary, i.e.. specify which properties of the model are, and which are not, common to the objects being introduced (cf. Sellars. 1963. p. 182). In the process of constructing an analogy, a second domain of objects is posited and specified as these which share the second-order properties of the revised model (i.e., the model as restricted by the commentary). This second domain of objects can be radically different from the objects of the model to the extent that they are in fundamentally different ontological categories (e.g.. attributes rather than substances) and, thus, share no first-order properties. An intriguing aspect of analogical concept formation is that we are able to generate completely unique concepts and, by sustaining this process, create a totally new conceptual scheme. Analogical concept formation is at the very heart of the evolutionary-revision model sketched above (part I).

18. We see, thus, that new conceptual frameworks can be constructed by the use of analogy. An important problem is the specification of the relationship obtaining between different conceptual schemes. This is a question of the relationship of constituent concepts in different conceptual schemes. In part II, we argued, in a specific case, that manifest objects are not identical a la Leibniz with complexes of scientific objects. It might, however, be thought that there is a different sense of "identity" in which it would he correct to say manifest objects are identical with scientific objects. Sellars, in fact, is committed to conceding the Quinean contention that "...reference, is nonsense except relative to a coordinate system" (Quine, 1969, p. 48). In order to specify the reference of a term, we must utilize a conceptual framework (what Quine colorfully terms a "coordinate system"). Within a given conceptual framework, we can specify the reference of a term which is a constituent of that framework. But if we were to attempt to compare the reference of terms within two different conceptual frameworks, we would need a point of view external to them both. But, as Sellars remarks, ". . . the conceptual framework of common sense has no external subject-matter. . . " (Sellars, 1967, p. 339). Lacking an external standard, we lack any chance of even saying what kind of thing we are referring to, and so lack a categorial demarcation. The moral to be drawn is that we cannot specify for two terms in different conceptual frameworks whether they are not co-referential. And since identity must at least involve sameness of reference, and reference is framework dependent, we do not yet have a clear and applicable sense of identity.

Yet even if identity is only an intra-framework notion and, as such, is inapplicable, there must be some basis for comparison. For although manifest objects are not, strictly, identical with (complexes of) scientific objects, there clearly is some sort of relationship between them. Sellars takes this to be a 'successor' relationship. (This might serve as the basis for a kind of trans-framework identity.) Let us consider two concepts C and C' in two different conceptual frameworks F and F' (respectively) where F' 'evolves' from F. C' will be the successor concept of C if C' is formed by analogy (either material or formal) with C, and C' plays roughly the same role in F as does C in F.

The notion of the role of a concept within a given conceptual framework can, for our purposes, be given in two parts: (1) conceptual roles, and (2) language-entry roles. Conceptual roles are the material and formal rules of inference governing the use of the concepts. The conceptual role of a concept includes the kind of non-logical inference countenanced by the conceptual framework in question. Thus, it is a material rule of inference that from "a is red" we can infer "a is colored"; or, alternatively, "everything red is colored" is a framework principle (see Sellars, 1948, passim.). Language-entry roles can be understood as principles which permit the uttering of (synthetic) proposition on the condition that the speaker is in a certain kind of situation. Thus, it is permissible to respond with "this is green" when presented with a green object. This is not to say it is permissible to respond with "this is green" when one observes that he is in the presence of a green object. This latter would already involve the use of a conceptual framework. The first would not, but the second would, involve the apparatus of reference. Thus, C will be the successor concept of C only if C and C play similar conceptual roles in F' and F (respectively), and the conditions under which we are justified in using C' in F' are similar to those justifying the use of C in F. (If the analogy in question is material, C and C' will have similar conceptual roles. If it is only formal, they may not.) This account of the notion of a successor concept is, admittedly, somewhat imprecise owing to the use of the term "similar"; but it must be on the right track since it is, after all, a difference in use (roles) which makes concepts distinguishable, and therefore identity of role cannot be made into a desideratum (cf. Sellars. 1968. ch. 3. § 38), Some of the linguistic roles of successor concepts will be different and some will be the same. How many of the roles a concept must have in common with another in order to be branded its "successor concept" is left open.


19. We have been concerned to develop the apparatus necessary in order to properly understand Sellars' second 'grain argument'. In this section we resume the substantive considerations and present the central argument in Sellars' critique of physical2 reductive materialism. Sellars draws a now common distinction between two senses of "physical":

Physical1: an event or entity is physical1 if it belongs to the space-time network.

Physical2: an event or entity is physical2 if ii is definable in terms of theoretical primitives adequate to describe completely the actual states though not necessarily the potentialities of the universe before the appearance of life. (Sellars, 1956. p. 252).

Any event or entity which is physical2 is, ipso facto, physical1, although the converse does not hold. Employing this distinction, the thesis of the physical2 reductive materialist, as restricted to sense impressions, can be stated as follows:
Sense impressions are states of complex systems of physical entities. These states are reducible to states of and relations between the constituents of these physical2 systems.
In other words, the thesis of the physical2 reductive materialist is the visionary claim that, in the Scientific Image, sense impressions will be held to be identical with neurophysiological states (what Sellars elsewhere terms "states of core persons" (cf. Sellars, 1967, 386)). These latter, in turn, are definable in terms of theoretical primitives which are necessary and sufficient for a complete description of items other than sentient organisms. It is this thesis which Sellars denies. Sellars' counterthesis is that physical2 primitives are inadequate for a complete description of sentient organisms. Sellars says he has

. . . rejected "reductive" materialism, if by this is meant the view that reality consists, in the final analysis, of physical2 particulars characterized [only] by physical attributes and relations (Sellars, 1971, p. 403).
The physical2 reductive materialist, then, holds that all physical1 events and entities are physical2 reducible. Sellars, on the other hand, maintains that some such events or entities are not physical2 reducible.

20. The second 'grain' argument aims to show that sense impressions, or sensations, are among those things that are not physical2 reducible. In order to fully grasp the argument, one further dimension of Sellars' work must be grasped: the analogy theory as applied to sense impressions.

There are six elements of the analogy which need to be introduced.

  1. The concept of a sense impression is analogically introduced within the Manifest Image.
The analogy has as its model the particulars of the Manifest Image (i.e., occurrently colored and shaped physical objects), and as its revised model the facing surfaces of these particulars (i.e.. colored and shaped expanses).
  1. The analogy is transcategorial (cf. Sellars, 1963, p. 93).
Whereas the model consists of spatio-temporal particulars, sense impressions are states of perceivers rather than 'inner particulars'. A failure to take into account this transcategorial nature of the analogy will lead us to take (α) "S has an impression of a red triangle" to be of the form "xRy", while it ought to be taken to be of the form "Φ(S)". To have a sense impression is not to stand in a relation to some curious kind of particular. Rather, having a sense impression of a certain kind is to sense in a certain way. Sellars. impressed by the equivalence of "a sensation of warmth" and "a warm sensation", analyzes (α) as (β) "S has an (of a red triangle) impression". In (β), "(of a red triangle) serves as an adverbial modifier specifying a kind of impression. Hence, (β) can in turn be analyzed as (γ) "S has a red and triangular impression" which is equivalent to (δ) "S has a red triangular sensing", or, in the active mode, (ε) "S senses redly and triangularly". On Sellars' analysis, (α) through (ε) are equivalent.9 Sellars' fundamental point is that sense impressions are nominal objects: though sense impressions can be treated as if they were objects, a more adequate portrayal would reveal their adjectival status.

The third relevant principle is:

  1. Sense impressions are introduced to explain the occurrence of certain minimal conceptual representations. (cf. Sellars, 1968, § 42; Sellars, 1967, p. 387).
For example, sense impressions are introduced in order to explain why percipients judge that there is a red rectangular object present when in standard conditions there is such an object present, and when in non-standard conditions there are ". . . objects which have other, but systematically related characteristics" (Sellars. 1968, § 44].

  1. Sense impressions are causal intermediaries (cf. Sellars, 1963, p. 91, p. 94).
Sense impressions serve to mediate between minimal conceptual representations such as "There is a red rectangular object there" and the world, and hence explain the occurrence of the former. A corollary of the thesis that sense impressions are causal intermediaries is the thesis
  1. Sense impressions are not epistemic intermediaries.
That is to say, sense impressions do not mediate by virtue of being known (cf. Sellars, 1963, p. 91). Sellars holds, in fact, that sense impressions cannot, in principle, come to be immediate objects of awareness (see Sellars, 1968, Ch. 1, § 35). As such sense impressions are non-cognitive and lack, therefore, the characteristic of intentionality (cf. Sellars, 1968, §59, Sellars, 1963, p. 73, p. 90; Sellars, 1967, p. 376).

21. The analogy for sense impressions is a formal one (cf. Sellars, 1963, 94). As such, the analogy defines two sets of predicates {Φ1, Φ2,..., Φn} and {Ψ1, Ψ2,...,Ψm} which are, respectively, the successors of the color and shape predicates which apply to physical objects. Φ1, Φ2,..., Φn and Ψ1, Ψ2,...,Ψm are the predicates applicable to sense impressions and, like their counterparts, stand for first-order properties. According to the dictates of formal analogies, sense impressions and (colored) physical objects must share second-order properties (cf. section 17). This is the point Sellars has in mind when he says "Colors as manners of sensing form a logical space modelled on colors as attributes of the physical objects of the Manifest Image." (Sellars, 1971, p. 408; cf. Sellars, 1967, p. 379)

It should not be thought that this commits Sellars to the claim that sense impressions (N.B. these are events) are literally shaped or colored. All the (formal) analogy requires is that the second-order properties of physical objects (the model) be shared by sense impressions. In other terms, what is being said is that even if the properties of properties of sense impressions are the same as the properties of properties of physical objects it does not follow that the properties of sense impressions are the same as the properties of physical objects.

In section 10. we remarked that homogeneity is not a property of individuals but a property of properties of individuals. In short, homogeneity is a second-order property of colored physical objects. Thus, the final and most important element of the formal analogy is:

  1. Homogeneity is shared by both physical objects and sense impressions.
It is natural to inquire why homogeneity should not be purged as part of the negative analogy. It is not clear just what Sellars' response would be to such an inquiry, but Sellars' contention that coloredness is a "categorical feature" (Sellars, 1971, p. 394) of the physical objects of the Manifest Image provides us with a clue (cf. Sellars, 1968, p. 171). Sellars might reply that it is a defining characteristic of Manifest objects that they are colored, and a defining characteristic of colors that they are homogeneous. If this is so, to purge ultimate homogeneity as part of the negative analogy would be to make nonsense of the model and, hence, also of the analogy itself.

22. It should not be thought that the claim that sense impressions are homogeneous is grounded on the claim that sense impressions are unanalyzable or simple, having neither parts nor constituents (see n. 5 above). Though this would be the easier route. Sellars takes a more powerful line in claiming that sense impressions have parts. As he says, ". . . an impression of a complex is a complex of impressions." (Sellars, 1968. p. 27). For instance, consider

S's having an impression of a green square adjoining a red square.
That is,
S's having an (of a green square adjoining a red square) impression.
It is clear that to have such an impression S must have an impression of a red square and an impression of a green square. Moreover, it is most natural to understand these latter impressions as parts of the former (cf. Sellars. 1968, 24ff.). Finally, Sellars suggests, these latter impressions must stand in some kind of relation in virtue of which they are parts of the former. To consider another case, it we are to have an impression of a red rectangle, must we not also have impressions of the parts? And must not these impressions of the parts be 'parts' of the impression of the whole?

This kind of idea is not peculiar to Sellars. For example. Peter Geach remarks that

. . . just as the pictures of parts of the thing are parts of the picture of a thing, so it might well be held that the seeing of a complex whole is itself really complex and composed of parts ... And if so, sensations would be formaliter, really composed of parts ... (Geach, 1957, p. 128).
But, as Geach recognizes, when we apply such concepts as parts
. . . We are not just saying the same thing when ascribing them to sensation as when we apply them in the physical world: even here we have an analogous extension of a concept (Geach 1957, pp. 128-129).
Sellars would definitely concur. The relation terms applicable to sense impressions are the first-order analogues of spatial relations. They might, for example, be functional or causal relations. All that is required of such first-order analogues is that they share second-order properties. Sellars says
Though . . . we conceive of certain of these attributes and relations [of sense impressions] as counterparts of spatial attributes and relations proper, they would not literally be the spatial attributes and relations in terms of which we conceptually represent physical objects and events. (Sellars, 1968, p. 29; our emphasis).
Given this caution, we take it that there is, at least, a strong intuitive base for the claim that sense impressions have 'parts.'. If this is granted, it may be taken as established that sense impressions, as analogously characterized, have the second-order property of homogeneity.10

23. We are now in a position to present Sellars' second 'grain argument' -- his reductio ad absurdum of physical2 reductive materialism. Suppose, with the physical2 reductive materialist, that just as material objects are complexes of physical2 objects, so also persons are (nothing but) physical2 systems. Now, persons sense (i.e., have sense impressions). Therefore, these complex physical, systems must also sense. On the physical2 reductive thesis (cf. section 19), the sensings of these physical2 systems are identical with neurophysiological states. Hence, (again, cf. section 19) the sensings of these physical2 systems are identical with complex states of and relations between primitive physical2 entities. But sensings are ultimately homogeneous (cf. section 21). And this entails (cf. section 10) that sensing is a content property. Thus, if persons sense and persons are identical with complex physical2 systems, then by virtue of sensing being a content property, some of the constituents of these physical2 systems must also sense. If physical2 reductive materialism is correct, primitive physical2 entities must sense. But clearly such entities those very entities appealed to in the first grain argument (section 9) do not sense. Therefore, either physical2 reductive materialism is false or persons do not sense.

24. It should be quite obvious by now that the PR is a crucial element in both of Sellars' two 'grain arguments'. One way, then, to avoid the consequences of the 'grain arguments' is to attack the principle on which they are based. James Cornman objected to the PR on the grounds that it is compatible with a 'reduction' of sense impressions to certain complex brain phenomena. Cornman gives what we take to be a nearly correct formulation of Sellars PR as his

For any object O, that is a system of objects, if O has a property P, of kind, K, then either K is a kind of property the individual constituents of O have, or O having P is reducible to some (at least) of the constituents of O having some other properties or relations (Cornman, 1970, p. 425).
Cornman continues to say that
. . . if "O having P" is true at some non-basic level, and O having P is not only correlated via correspondence rules with, but also reduced to the constituents, C, of O having Q, then the Cs of O having Q would be the basic level counterpart of "O has P" (Cornman, 1970, p. 426)
In terms of our standard example, this proposal can be restated in the following way: If "S has an (of a red triangle) impression" is true at some non-basic level, and if S's having that impression is both correlated with and reduced to the c-fibers of S being in state Φ, then the c-fibers of S being in state Φ would be the basic level counterpart of S's having an (of a red triangle) impression. Cornman may he guilty of either of two errors in his formulation.

The (possible) errors revolve around an ambiguity in the term "reducible". Thus, given an object O to which P is correctly ascribable (i.e., "O has P" is true); and given a system S of objects with properties and relations P1, P2,..., Pn, "reducible" can be interpreted as either of the following:

reducible1: "O has P" is reducible, to the constituents of S having P1, P2,...., Pn iff (1) O is identical with S, and (2) either "P" is definable in terms of "P1", "P2", ..., and "Pn" or the property P is the same as P1, P2, ..., or Pn.
reducible2: "O has P" is reducible, to the constituents of S having P1, P2,...., Pn iff (1) O is not identical with S, but "O" is correlated via correspondence rules with "S", and (2) "P" is correlated via correspondence rules with some "P" such that "P" is correctly ascribable to S and either "P" is definable in terms of "P1,", "P2,..., and "Pn" or the property P is the same as P1, P2,..., or Pn.

Notice that "reducible2" can correctly be said to apply in inter-framework contexts. For correspondence rules are designed to correlate a successor concept in one framework with its predecessor in another (cf. section 18). (The correspondence rules employed here would be substantive and not methodological (cf. Seflars, 1967, pp. 329-330).). On the other hand, "reducible1" can correctly be said to apply only in intra-framework contexts. For "reducible1" requires identity, and identity according to Sellars (cf section 18), is only comprehensible in an intra-framework employment. Sellars' PR is of the reducible1 variety, (cf. Sellars, 1971, p. 411).

If Cornman's use of "reducible" in R3 is that of "reducible2", his attack on Sellars is an ignoratio. That this is not the sense of "reducible" which is employed in Sellars' argument is clear when he states he used his PR

. . . to argue that whatever manifest objects may be correlated with, they cannot literally consist of micro-particles, or literally be identical with wholes consisting of micro-physical particles [Sellars, 1971, p. 412; cf. Sellars, 1971, p. 407, Sellars, 1963, pp. 193, 194).
The emphasized terms here -- "correlated with" and "consist of" -- demarcate the two senses of "reducible" given above. Cornman's apparently innocuous insertion of "reducible" (contrast this with the formulations of the PR in section 8) is the source of the confusion. The R3 formulation leaves it unclear how the PR is to be employed -- whether inter-framework, intra-framework, or both. Sellars, at least, maintains that his principle concerns the internal structure of conceptual frameworks (cf. Sellars, 1971, 411). Thus, the PR is an intra- and not an inter-framework principle. As such, it can apply only to concepts within the same conceptual framework.

If Cornman's use of "reducible" in R3 is that of "reducible1", his attack on Sellars is impotent. Consider Cornman's claim that (as we instantiated it) the c-fibers of S being in state φ is the basic-level counterpart of S's having an (of a red triangle) impression. This claim is consistent with R3 (as amended with "reducible" = "reducible1") and the PR if φ is a state of sensing. But, of course, Cornman would not admit that c-fibers sense. The above claim would also be consistent with R3 and the PR if having a sense impression were a structural property. But that sense impressions are homogeneous entails that having a sense impression is not a structural property. Thus, Cornman may well have missed the pivotal role of the concept of homogeneity in the argument.11

Finally, it might be the case that Cornman has not committed either of these errors. In this eventuality, Cornman would have to deny that sense impressions are homogeneous by claiming that having a sense impression is a structural property. Cornman would, then, be committed to abandoning Sellars' analogy theory for sense impressions. However, he has not even attempted to construct an alternative.

25. The physical reductive materialist could avoid Sellars' grain argument by abandoning the analogy theory. But such a stand would be credible only if accompanied by a comprehensive alternative account of the concept of a sense impression. Barring the appearance of such an alternative, the (second) 'grain argument' yields the following antinomy (cf. Sellars, 1963, p. 36):

  1. The neurophysiology of the Scientific Image is incomplete, and must be supplemented by (emergent) states of core persons which manifest homogeneity and are not reducible1 to the properties of the physical2 system; or
  2. the neurophysiology of the Scientific Image is complete, and the homogeneity of sensory qualities and therefore the sensory qualities themselves do not exist.

The latter leg of the antinomy is precluded on the basis of Sellars' principle of comprehension (cf. section 15). The first 'grain argument' purports to show that complexes of physical2 objects are not colored. Colors, with their ultimate homogeneity, were relocated in the perceiver as modes of sensing. The second 'grain argument' purports to show that if we conceive of persons as merely complexes of physical2 objects, they do not sense. Option (B) would have us purge sense qualities. This, unfortunately, is tantamount to dispensing with colors altogether. This repudiation would be as unacceptable to Sellars and for the same reasons as was Smart's original one.

Thus, Sellars is apparently left with the first leg of the antinomy. But option (A) is also unacceptable. The PR requires that the successor concept of sense impressions be reducible, in the framework of ideal neurophysiology. That is, in the framework of the Scientific Image, for a neuro-physiological whole to have a sense impression of a pink ice cube is for the constituents of this aggregative whole to be in certain states and stand in certain relations to one another. Unless the having of a sense impression which is attributable to the conglomerates of scientific objects (the 'persons' of the Scientific Image) is reducible1 to the constituents of the conglomerate having certain properties and standing in certain relations, the theoretical unity of the Scientific Image, which is so essential to its viability, is lost. Sellarsian (and. it would seem, any non-minimal) scientific realism would then be untenable.

Fortunately, the 'antinomy' does not exhaust the options. Option (A) can be resolved into the claim that " . . . unless we introduce Cartesian minds as scientific objects, individual scientific objects cannot be meaningfully said to sense redly." (Sellars, 1963, p. 411). The introduction of Cartesian minds would clearly unify the Scientific Image. Its unity demands reducibility. By admitting sensing to be a property of minds, and countenancing minds as constituents of Scientific Image 'persons', the reduction would go through in accordance with the schema for content properties (cf. section 8). But we will see in part VI that Sellars has an alternative to such mind/body dualism.


26. We have thus far explained the grounds for Sellars' contention that the ultimate homogeneity of sense impressions cannot be accommodated in a physical2 reduction. Sellars accepts only a weak form of the identify theory (cf. Sellars, 1967, p. 386). On this weak form, sense impressions are states of core persons which are not reducible, to the physical2 constituents of core persons alone. They are, then, emergent states.

Sellars says that "... the logical space of raw feels will appear transposed but unreduced. . . " (Sellars, 1967, p. 386) in the neurophysiology of the Scientific Image. The second-order properties which were, in accordance with the analogy, attributed to sense impressions will be carried over by a second analogy and become second-order properties of a new category of which Sellars terms "sensa". It is sensa which will provide for the unity of the Scientific Image. As preparatory to a more detailed investigation of sensa, we must acquaint ourselves with the notion of emergence.

27. Pepper attacked emergentist doctrines by claiming that ". . . a theory of emergent qualities is palpably a theory of epiphephenomena" (Pepper, 1926, p. 242; cited in Sellars and Meehl, 1956, p. 241). Sellars and Meehl argue that, to the contrary, emergent qualities will be epiphenomenal only if there are not, in the relevant sense to be elucidated, emergent laws. They give an explanation of emergent laws which does not involve " . . . the simultaneous presence in a single situation of two or more levels." (Sellars and Meehl, 1956. p. 247):

What the emergentist says is that there is a region in the fourspace qrst within which f1(q,r,s,t). This region is the "lower level of integration". ... On the other hand, there is a region -- the emergent region -- in which f2(q,r,s,t) holds. f1 ≠ f2 (Sellars and Meehl. 1956, p. 246).
Essentially, the claim is that f1 adequately serves to describe the regularities of a limited region of qrst-space; and f2 describes the regularities within a different region of qrst-space. Now, assuming determinism (in its unproblematic form according to which only predictability is required), the emergent qualities must be lawfully related to the lower level variables, and will, hence, be eliminable. Consider the emergents in question to be a and b such that the formulae "a = g(q,r)" and "b = h(s,t)" hold. We can then form a function which will describe the regularities of the inclusive set {q,r,s,t,a,b}. It is, for the proper E, "E(q,r,s,t,a,b)", or equivalently "E(q,r,s,t,g(q,r),H(s,t))". This is equivalent to "f3(q,r,s,t)" for the correct f3. If f1 and f3 range over exactly the same region of qrst-space, then f1 = f3 = f2. But if, as is supposed above, f3, has a greater range of values than f1, then f3 = f1 only for the limited region of qrst-space in which f1 holds. The function f1 proves to be incorrect when extrapolated beyond the initial region. The same holds true for f2 for the portion of qrst-space in which it holds.

If in accordance with the above procedure, the "emergent" dimensions of E can be eliminated, achieving f3, then what reason could a scientist have for introducing the emergent dualities or emergent dimensions (N.B. Sellars will claim sense impressions are emergent states)? Could the scientist not remain content with f3, knowing that it adequately describes the phenomena? The reply Sellars and Meehl offer to these queries is based on the distinction between the descriptive and theoretical aims of the scientific enterprise. Since f3 is adequate to describe the phenomena, there will be no constraints to introduce a and b from a descriptive standpoint. But there may be theoretical constraints. If we consider a theory T which has a set X of primitives such that each element of X is physical2 and another theory T having as its primitives X U {a, b}, the theoretical motivation for introducing a and b is that f1 can be derived from the postulates of T, but f2 cannot (f3 ranges over the emergent as well as the non-emergent regions of qrst-space). Sellars and Meehl claim that it is the derivability of f1 which is at stake: f3 is derivable only within the theory T (with the suitable postulates). The derivability of f3 therefore provides the theoretical motivator for the introduction of a and b as theoretical primitives. Emergence, thus understood, would entail that emergent qualities (e.g., sense impressions) are physical1 but not physical2.

28. Sellars claims that sense impressions can be interpreted in two ways in the Scientific Image:

  1. A sense impression of red is a system of physical2 scientific objects in a state of sensing-redly. This state, being physical1, is not definable in terms of states of and relations between the constituent physical2 objects of the system.
  2. A sense impression of red is a system of scientific objects in a state of sensing-redly and this state is definable in terms of the states of and relations between the constituent objects of the system. Some of the constituents of this system will be physical1 and not physical2 (Sellars, 1971, 393).
Option (1) is equivalent to alternative (A) considered above (section 25). We there claimed that if sense impressions were not reducible1 within the Scientific Image, then the Scientific Image could not be unified. If option (1) is correct, then either sense impressions are epiphenomenal or they are non-epiphenomenal emergent states of holistic organisms. In the former case, non-reducible 'nomological danglers' are unavoidable. This clearly violates the demand for theoretical unity. In the latter case, as we have shown, there can be non-epiphenomenal emergent states on the condition that there are emergent laws as the f2 law above. Then there is a more inclusive law f3, which correctly describes everything described by the emergent law f2 as well as everything described by the non-emergent law f1. Now if, in accordance with (1), only physical primitives are allowed in the Scientific Image, then f3 cannot be derived. And if f3 is not derivable, then the Scientific Image in which it is found is not unified. For this reason, Sellars rejects option (1) in favor of (2).

29. Option (2) can be satisfied only if the following two requirements are met: first, sense impressions must be definable in terms of (i.e., reducible to) the basic entities of the categorial framework -- the framework must be ontologically unified; and second, the laws describing the relationships between the primitive entities of the framework must be deducible from the 'postulates' which constitute that categorial framework -- the framework must be nomologically unified. Option (1) failed to meet the condition of nomological unity. If these two conditions are satisfied, the Scientific Image will constitute a unified categorial framework, and will contain an account of sense impressions which preserves the 'logical space of colors'. In particular, this account will capture the elusive characteristic of homogeneity.

The task is to analogically construct a successor concept of sense impressions (i.e., sense impressions2 which is reducible1 within the Scientific Image. The analogy involves a specification of analogues of the color-predicate analogues which apply to sense impressions. Recall that on the view advocated by the direct realist, color predicates were adjectival. Through the first analogy, adverbial successor predicates were defined which specified modes of sensing. In this second analogy, Sellars closes the circle by again construing color predicates as adjectival (cf. Sellars, 1968, 172).

Although these successor predicates are adjectival, as were their counterparts in the Manifest Image, we must not lose sight of the fact that they are, indeed, quite different predicates applying to radically different particulars. Thus, the analogy also involves a specification of a new domain of particulars which, for want of a better term, Sellars labels "sensa''.12 These particulars share the second-order properties (including homogeneity) which characterized both sense impressions and (colored) physical objects.

30. The most important, and most frequently misunderstood, element in the analogy is that a new category of particulars is introduced (cf. Sellars, 1971, p. 410. p. 428) to which the new successors of color predicates apply (cf. Sellars, 1971, p. 413. p. 416. and esp. p. 421. p. 426). Sensa should not be thought to be the successors of sense impressions (cf. Sellars, 1971, pp. 392-393. p. 415, p. 426). Rather, they are basic particulars which, along with the physical2 entities of neurophysiology, form complex states (cf. Sellars, 1971, p. 404, p. 417); and these complex states are the successors of sense impressions (cf. Sellars, 1971, p. 393, p. 437). Sensa though basic particulars, are not particulate; they are occurrents, or processes, not analyzable into physical2 processes. They are physical1 (but not physical2) ingredients of the complex neurophysiological states which are sense impressions2. Sense impressions2, thus conceived, are reducible1 states of complex systems.

Sensa are introduced to serve as logical subjects for the analogues of color and shape predicates in a theoretical account of what it is for persons to have sensations (cf. Sellars, 1967, pp. 386 ff.: Sellars, 1971, p. 437). As such, they must "...relocate the 'ultimate homogeneity'. . . " (Sellars, 1971, p. 409) which characterized the colors of the Manifest Image. Sensa are certainly equal to the task. Since they are primitive entities having no constituent parts, sensa trivially satisfy the requirement of homogeneity (cf. section 8).

31. Having seen that sense impressionss are reducible1, the Scientific Image will at least be ontologically unified. The question remaining is whether or not it will be nomologically unified. Sense impressions2 are emergent in that sensa exist only in association with certain patterns of activity in the regions of the brain. The laws which will govern the patterns of neural activity (f2) will be emergent relative to the subframework of physical2 theory, but will not be emergent within the entire framework -- the laws governing the activity in the brain will be deducible within the Scientific Image as a whole. Similarly, the laws (f3) governing both patterns of neural activity and physical2 phenomena (f1) will be emergent relative to the physical2 sub-framework, but not emergent within the Scientific Image as a whole. It is the introduction of sensa as basic particulars in the Scientific Image that will provide for the deducibility of these laws (cf. Sellars, 1971, p. 417; Sellars, 1956, pp. 250-251), and thereby eliminate the need to posit emergent states of organisms. Therefore, the introduction of sensa as physical1 particulars produces both ontological and nomological unity in the Scientific Image.

32. It is worth emphasizing the centrality of the doctrine of emergence in justifying the introduction of sensa. Unless the impact of it is properly understood, Sellars' position appears to be confronted with intractable puzzles. William Lycan has posed two clear problems (Lycan, 1979). First: Just why should microphysicists be led to posit sensa? Are they, in so doing, only serving the interests of utopian psychology? Second: Why do sensa occur only within sentient organisms? In this, are they not unlike every other fundamental entity?

In fact, Sellars' views on emergence provide ready, and not implausible, answers.13 We may recall the stage at which we were left with sense impressions as states of core persons. These states were held to be emergent in that the laws which governed interactions in these systems, though expressible in physical2 terms, were neither the same laws as those necessary to explain occurrences outside sentient organisms nor reducible to those laws; that is, though expressible in physical2 terminology, the laws which govern interactions in core persons are not physical2 laws. This would leave the Scientific Image unabashedly bifurcated, with different (and inconsistent) laws governing different domains. Sensa turn out to provide, or create, nomological unity. Their presence means simply that core persons do not consist simply of physical, constituents. The reason microphysicists, by Sellars' lights, should be led to posit sensa should now be clear. It is not to appease the psychologists; rather, it is to unify physical theory in such a way that it is capable of explaining occurrences both without and within sentient systems. The reason laws differ from one case to the other is simply that there are different sorts of entities present in the respective systems. This, of course, takes us to the second puzzle which Lycan labels the problem of "shyness". Here the best appeal Sellars could make is simply that whether sensa occur only within, or also without, sentient organisms is finally an empirical question; it will be found out by finding out if sensa are needed to explain (at the level of microphysics) the behavior of systems other than sentient systems. Sellars has wagered they will not. This will be confirmed or disconfirmed, in turn, by finding out what laws are operative in these systems.

Each of the replies to these puzzles rests on Sellars' projection that the laws governing the interaction of physical2 particles in physical2 systems will be false when applied to core persons. More 'tough-minded' physicalists will, no doubt, regard this as a reductio ad absurdum of Sellars' position. But here it is Sellars who stands as an empiricist, awaiting the verdict from the sciences. If his projection is false, and physical2 laws are adequate for core persons, the motivation for sensa will evaporate. Along with it, however, goes the hope of dealing cogently with sense impressions. Homogeneity would remain elusive and intractable, and the 'Scientific Image' forever incomplete.


33. A recurrent and disquieting feature of Sellars' argument is the persistent assumption of the untenability of dualistic theories. It simply will not do to attack Cartesians as Sellars commonly does with comments such as this:

The pressure for such a conception of substantive dualism comes from ... religious considerations pertaining to life after death, from the atomistic philosophy of nature and from bad philosophy, thus the Cartesian argument for real distinction between mind and body (Sellars, 1968, p. 170).
The climate of current opinion is hardly such as to allow a dualist serious hearing, but the dogmatism of many (though not all) contemporaries hardly excuses Sellars' lack of argument. It is important to the Sellarsian enterprise that some substantial grounds be given for precluding this option. How important we will quickly show.

It is interesting that Sellars' argument can be recast as an argument against materialism. If the argument is sound, then materialism in anything other than the Sellarsian form is incorrect; and if there are no emergent laws (and this is, after all, an empirical question), then the final Sellarsian version is wrong as well. An unfavorable empirical result would leave no recourse but dualism: if the laws governing particle interaction in core persons are physical2 laws (and therefore they are formulable in physical2 terminology), then we will be committed by Sellar's grain argument to endorsing an epiphenomenalist theory; for the qualitative content so integral to sense impressions will have no relevance to the functioning of the organism. All this is amenable enough to Sellars' position but the epiphenomenalist can hardly fail to gain some ground from the realization that at this stage in the development of neurophysiology and physiological psychology there is no reason whatsoever to think (here are emergent laws -- that, in the words of Sellars and Meehl, "the flow of electrons at the synaptic interface 'breaks the laws'" (Sellars and Meehl, 1956, p. 251).

Of course, Sellars thinks, quite rightly, that the theoretical framework which would be adequate to explain perceptual propositional attitudes "is not yet even on the horizon" (Sellars, 1967, p. 387); hence, appeals to the current state of the art are hardly decisive. So let us assume with Sellars that the future development of science will uncover emergent laws -- laws which are not physical2. Would this establish the existence of sensa? It would not. We could just as well introduce Cartesian minds as sensa. To see why, let us contrast two manifestly dualistic accounts of the mind: the first is a standard Cartesian-style theory in holding that the mind is a single logical subject (though temporally varying, it is not spatially complex in having no proper parts) which is the seat of sense impressions (or, if we acknowledge the historical Descartes, is part of a complex -- the person -- the whole of which is the subject of sensory experience); the second is of a more Humean tinge, holding that for each sense impression there is a momentary logical subject which 'has' that sense impression, and the sum of these logical subjects will be all there is to the self. On neither view, it should be noted, is there a commitment to the question of whether the subject needs to be thought of finally as process or 'substance'. Aside from titles, the latter bundle theory is indistinguishable from the more properly Sellarsian sensa: each element in the bundle' interacts with physical2 particles, alters the overall behavior of the system, and is not a physical2 entity. The identity of indiscernibles does the rest: Sellars' theory becomes a sort of bundle dualism where the sense impressions are no longer states of which we are directly aware, and which must be taken, as wholes, to be non-physical2 because they have as constituents elements -- namely, sensa which are not physical2. A dualist theory of a Cartesian stripe will do all this and more: it allows for the unity of the subject across time.

Sellars, of course, claims to be a materialist, and clearly wishes to extend the rubric of "materialism" to encompass his own stance. He remarks that if the

ties of "materialism" with "matter" construed in terms of classical paradigms, are loosened, its usefulness hinges on its standing for a type of world picture which contrasts with non-materialism in a way which preserves the force of the classical materialist claim that the ultimate logical subjects involved in sensation, feeling, emotion, and conceptual thinking are "physical", in a sense of "physical" which does not so trivialize it that it would apply to Cartesian minds (Sellars, 1971, 446)
Sellars has not delimited a sense of "physical" extensive enough to meet this standard. Sellars admits that sensa "are not 'material' as 'matter' is construed in the context of physics with a particular paradigm" (Sellars, 1971, p. 446). But the difference between Sellarsian 'materialism' and more classical versions is not, as one might be led to think by this, that Sellars prefers a 'process' over a 'particulate' paradigm -- though, as is quite clear from his Carus lectures, he does. The difference is, rather, in the types of entities countenanced and the types of interactions allowed: classical materialism eschews all that is not physical2 (see section 19 above for the contrast between the two senses of "physical"). If the 'physical' is extended in scope to mean the physical1, it will include all that interacts with the physical2; for to "belong to the space-time network" it is enough that something interact with what is physical2. In fact, it seems we could with as much justification, extend "dualism" and characterize Sellars' position as a non-substantive dualism. Such banner-waving obscures the issues, and, as Sellars recognizes elsewhere, is bound to create more heat than light. If the Sellarsian argument is correct, classical physicalism is incapable of giving an adequate account of perceptual propositional attitude The task which remains for Sellars is an evaluation of the conflict between his own position and more traditional dualism; if these terms are unacceptable, the task is that of evaluating the conflict between his own position and the physical2ism that is espoused by Cartesians.14

University of Cincinnati
Concordia College


1 The most Quine is now willing to endorse is "that our system of the world is bound to have empirically equivalent alternatives which, if we were to discover them, we would see no way of reconciling by reconstrual of predicates" (Quine, 1975, p. 327). The more sweeping view that "for each theory formulation there will be others, empirically equivalent but logically incompatible with it and incapable of being rendered logically equivalent to it by any reconstrual of predicates". Ouine judges to be moot for lack of knowing whether "we are underestimating the power of reconstrual of predicates" (Qume, 1975, p. 326).

2 Witness the effects this reading would have for reduction and theoretical explanation. After all, scientific theories are, on Sellars' view, methodologically dependent upon the conceptual framework of common sense (cf. Sellars, 1973, p. 20). To construe science and common sense as irreconcilably disparate would make the avowed methodological ties incomprehensible.

3 This can be made even more defensible with some relatively minor alterations and restrictions. Some of the relevant ones are discussed in section 24 below. For an elegant discussion that clearly indicates the potential for elaborating the PR into a model capable of dealing with a good deal of scientific practice, see Schaffner, 1974, passim.

4 An optional formulation, which does not seriously affect the systematic issues, would follow Hooker (1977, p. 337) by saying that C is a content property if C is a determinate of determinable C* and either C or some other determinate of determinable C* is attributable to some of (O1, O2,... On). The definition of a structural property would be similarly altered. This might appear to make a real difference, since, e.g., being pink can be a structural property by our definition in the text provided that the constituents are, say, red, but not very dense. On the optional formulation, it would be a content property. However, since the long range concern is with the status of secondary properties, allowing such systematic deviations will not alter the outcome. The formulation in the text, on the other hand, has the advantage of relative simplicity.

5 In Delaney, 1970, p. 16. By way of contrast. C. A. Hooker is more nearly correct in suggesting Sellars holds that "within the naive conceptual scheme . . . secondary qualities are simple, homogeneous properties" (Hooker, 1977, p. 377). As we shall see, the appeal to secondary qualities being "simple" is superfluous. It appears, in fact, to be what Hooker relies on to drive Sellars' argument; however, since we do not know what a "simple property" is -- whether, e.g., just an unanalysable property or a property of logical simples' -- its function in the argument cannot be assessed.

6 This has been independently suggested by Alan Donagan (in conversation) and William Lycan (in correspondence).

7 Sellars also endorses the view that if a given statement within the Scientific Image is true relative to the Scientific Image, then, for any other conceptual scheme (including the Manifest Image) in which a counterpart to that statement is expressible, that counterpart should be (a) true relative to the Scientific Image, and (b) not false (though perhaps also not true) relative to the conceptual scheme of which it is a part (cf. Sellars, 1968, ch. 3, § 74). This principle, though, we think, no more tenable than that discussed in the text, makes it appear clear that Sellars' system is conservative in that it allows, in much the spirit of Carnap, only for additions to previous theory.

8 For an elegant discussion of Sellars' views on this topic and their relations to inductivist methodologies, see Lehrer, 1973, passim.

9 These formulations are technically incorrect, since "red" and "triangular" are the successor concepts of "red" and "triangular" as applied to physical objects. Hence (α) is properly rendered as (α') "S has an impression of a reds triangles" (and mutatis mutandis for (β) (ε), where the subscript "s" indicates that the term to which it is appended expresses a successor concept). The force of this qualification will become apparent subsequently.

10 This provides the core of the answer to the problem Frank Jackson dubs "the many property problem" (Jackson, 1975, p. 129); viz., how to account adequately for the fact that, e.g., after-images have multiple properties. Jackson alleges that if to sense an Rφ (e.g., a red square) is to sense-Rly and to sense-φly, then no distinction exists between sensing an Rφ and a Gψ and sensing an Rψ and a Gφ. Sellars' solution is obvious: to sense an Rψ is to have a sensing of an R standing in a relation (defined by the analogy) to a sensing of a ψ. See Tye, 1975 and Sellars, 1975 for discussion.

11 In light of subsequent discussion by Cornman, this seems the most likely alternative. Compare Cornman's definition of theoretical reduction with reducible1 and reducible2 (in Cornman, 1971, 60).

12 Though we will follow Sellars' use of this term, it is highly misleading. For anyone brought up in the tradition of classical sense datum theory, the temptation to equate sensa with sense data is almost overwhelming. As an example, see James Cornman, 1971. Part III, ch. 7.

13 In commenting on Lycan's paper, however, he did not employ these replies. Why we cannot say.

14 This paper has gone through many drafts and has been discussed, in whole or part, with many people. It began when we were graduate students at the University of Chicago, and we are grateful to the many fellow students and faculty that discussed Sellarsiana with us there. It has since been read, though not in the complete form, to gatherings at the University of Cincinnati and the University of Notre Dame. It has since been used in seminars at the University of Cincinnati, and to those who discussed it there we are also grateful. The following individuals should especially be recognized for comments or criticisms which improved on the result: Wayne Backman, Hector-Neri Castaneda, C. F. Delaney, Alan Donagan, Michael Loux, Gary Gutting, William Lycan, Kirk Robinson, and William Wimsatt. Last, but not least, we are thankful for the encouragement and comments of Wilfrid Sellars.


Cornman, James: 1970. 'Sellars, Scientific Realism and Sensa'. The Review of Metaphysics 23, 417-452.

Cornman, James: 1971. Materialism and Sensations (New Haven: Yale University Press).

Delaney, C. F.: 1970, 'Sellars' Grain Argument', Australasian Journal of Philosophy 50, 14-16.

Geach, Peter: 1957. Mental Acts (London: Routledge and Kegan Paul).

Grossmann, Reinhardt: 1975. 'Perceptual Objects, Elementary Particles, and Emergent Properties', in Hector-Neri Castaneda (ed.), Action, Knowledge and Reality (Indianapolis: Bobbs-Merrill), pp. 129-146.

Hesse, Mary: 1966. Models and Analogies in Science (University of Notre Dame Press).

Hooker, C.A.: 1977. 'Sellars' Argument for the Inevitability of the Secondary Qualities', Philosophical Studies 32, 335-348.

Jackson, Frank: 1975. 'On the Adverbial Analysis of Visual Experience', Metaphilosophy 6, 127-135.

Lehrer, Keith: 1966. Review of Science Perception and Reality, in The Journal of Philosophy 63, 266-277.

Lehrer, Keith: 1973. 'Reasonable Acceptance and Explanatory Coherence: Wilfrid Sellars on Induction', Nous 7, 81-103.

Lycan, William G.: 1979. 'Sellars on Sensa and Second Guessing', presented at a conference on Sellars' philosophy at Ohio State University, May 10-11, 1979.

Pepper, Stephen C.: 1926. 'Emergence', Journal of Philosophy 23, 241-245.

Quine, W.V.: 1960. Word and Object (Cambridge: MIT Press).

Quine, W.V.: 1969, Ontological Relativity and Other Essays (New York and London: Columbia University Press).

Quine. W.V.: 1965, "On Empirically Equivalent Systems of the World", Erkenntnis 9, 113-328.

Sellars, Wilfrid, and Meehl. P.E,: 1958, 'The Concept of Emergence', in Minnesota Studies in the Philosophy of Science (Minneapolis: University of Minnesota Press). Volume 1, pp. 239-252.

Sellars, Wilfrid: 1948, 'Concepts as involving Laws and Inconceivable Without Them', Philosophy of Science 15, 287-315.

Sellars, Wilfrid: 1959, Philosophical Perspectives (Springfield: Charles C. Thomas Publishers).

Sellars, Wilfrid: 1963, Science, Perception and Reality (London: Routledge and Kegan Paul).

Sellars, Wilfrid: 1967, 'Rejoinder (to Aune)', in H. Castaneda (ed.), Intentionality, Minds, and Perception (Detroit: Wayne State University Press), pp. 286-300.

Sellars, Wilfrid: 1968, Science and Metaphysics (London: Routledge and Kegan Paul).

Sellars, Wilfrid: 1971, 'Science, Sense Impressions, and Sensa: A Reply to Cornman', Review of Metaphysics 24, 391-447.

Sellars, Wilfrid: 1975, 'The Adverbial Theory of the Objects of Sensation', Metaphilosophy 6, 144-160.

Shaffner, Kenneth: 1974, 'The Watson-Crick Model and Reductionism', in Marjorie Grene and Everett Mendelsohn (eds.), Topics in the Philosophy of Biology (Dordrecht: D. Reidel).

Smart, J.J.C.: 1963, Philosophy and Scientific Realism (London: Routledge and Kegan Paul).

Tye, Michael: 1975, 'The Adverbial Theory: A Defense of Sellars Against Jackson', Metaphilosophy 6, 136-143.

Manuscript received 7 November 1980.