Featured Video Play Icon

Colour layering and colour relationalism

Derek Brown, Brandon University

[PDF of Derek Brown’s Paper]

[Jump to Mazviita Chirimuuta’s commentary]
[Jump to Jonathan Cohen’s commentary]
[Jump to Derek Brown’s replies]

Abstract

Colour Relationalism asserts that colours are non-intrinsic or inherently relational properties of objects, properties that depend not only on a target object but in addition on some relation(s) that object bears to other objects. The most powerful argument for Relationalism (Cohen 2009) infers the inherently relational character of colour from cases in which one’s experience of a colour contextually depends on one’s experience of other colours. Experienced colour layering – say looking at grass through a tinted window and experiencing opaque green through transparent grey – demands a contextual interdependency of one’s experience of one of these colours on one’s experience of the other. However, most if not all colour ontologies, and core perceptual experiential mechanisms like acquaintance and representation, can accommodate colour layering. It follows that experienced colour layering is consistent with colours being non-relational – this contextual interdependency of colours does not entail the constitutive dependency of one colour on the other. I seek to bring out this tension between a well-known argument for Relationalism and colour layering, and argue that our justification for Relationalism is far weaker than Relationalists suggest. I first introduce readers to colour layering (§1), then to Relationalism (§2), and following this focus on the intersection of these topics (§3).  [1]

1. A primer on colour layering

The aim of this section is to briefly overview how colour layering connects to experience, ontology, and colour constancy. This is essential to bolster the impact layering has on Relationalism (§3).

Pretheoretically, we think of and talk about transparent things or objects as existing and often having their own colours: liquids (beer), solids (glass, plastic), and gases (mist).[2] We arguably draw at least some of these thoughts and utterances from colour experience. Pretheoretically, there is no difficulty with experienced colour layering, that is, with experiencing two colours along a line of sight, one opaque colour through a transparent one. When one looks at grass through the tinted window of one’s car, by hypothesis one’s experience can be of opaque green through transparent grey. In this way there is a contextual interdependency of one’s experience of one of these colours on one’s experience of the other. This is consistent with each colour, and one’s experience of each colour, being distinct – this interdependency does not entail conflation.

Some rough distinctions help sketch the space of possible experiences one might have when looking in a direction that contains (for simplicity) two objects, a proximal and distal object:

  • Colour layering: when one experiences distinct colours belonging to each object along the line of sight.
  • Colour fusion: when one experiences a single colour that mixes or fuses together the contributions of the objects located along that line of sight.
  • Single, non-fused colour: when one experiences a single colour that only reflects the contributions of one of the things located along that line of sight. Here the other object makes no contribution to one’s colour experience – it is in this way invisible. This can occur for example because the proximal object is completely invisible to the subject (e.g., a perfectly transparent sheet of glass whose edges are hidden from view), or because the proximal object is occluding the distal one.

My primary interest is in colour layering and its distinctness from colour fusion (I will largely set (3) aside), concepts and a distinctness I take to be pretheoretically straightforward and uncontroversial.

When we shift to colour theory, matters need be no different. I briefly embed colour layering within different colour ontologies and connect the discussion to colour constancy.[3] The injection of colour fusion into the remainder of this section is not difficult and will be left to the reader.

Theories of colour ontology can accept transparent colours and colour layering, though naturally differ on how to analyze it. Objectivists like Byrne and Hilbert (2003) argue for a generalized notion of colour productance that spans not only surface colours, but also the colours of filters, volumes and light sources. Although they don’t discuss contexts involving potential colour layering in detail, their ontology supports its occurrence. Mentalists – those holding that colours are purely properties of minds – can accept that there are transparent coloured things and colour layering. They will deny that any objective things have colours and attribute colours instead to whatever mental entities they take to bear them. For example, if sense-data are mental objects and the bearers of colours, then one should admit the possible existence of transparent and opaque coloured sense-data, and of contexts in which they co-occur and thus are layered. Eliminativists – deniers of any instantiators of colour in our world – can allow for the possible (i.e., instantiated in some world) existence of transparent colours and colour layering, and for example utilize representation to explain our seeming experiences of such things in our world. Even relationalists, broadly conceived, can countenance transparent colours and colour layering, provided that all colours are defined relative to something. Thus if one experiences green grass through a grey window, Relationalists can hold that the greenness of the grass is not intrinsic to the grass but is a relational property of the grass plus other factors, and so also for the greyness of the window. I will return to the relationship between Relationalism and colour layering in §3. The general point is that no colour ontology need antecedently fear transparent colours or colour layering.

The same can be said for broad approaches to colour experience. Most generally, the two core mechanisms for perceptual experience are representation and acquaintance. Roughly, the acquaintance mechanism explains experience solely by reference to relations between perceivers and perceptual objects, be those objects objective things or subjective entities like sense-data. By contrast representation explains experience most centrally by appeal to contents (e.g., uninstantiated universals). I take it as adequately straightforward that colour layering can be expressed via either mechanism and also by hybrid approaches that utilize both mechanisms. There are thus no antecedent conflicts between colour layering and influential approaches to perceptual experience.

Colour layering directly connects to colour constancy, one of the more difficult issues in colour theory. Roughly speaking, colour constancy involves a stability in the experienced colour of a thing despite the presence of a variation in some perceptual factor such as illumination conditions. For example, when one experiences a partly shadowed object or an object in daylight then under dense cloud, and in some sense experiences that object to have a constant colour, then one undergoes a colour constancy experience. There is thus both a constant and a variable element in colour constancy scenarios, though matters become controversial when one attempts to provide detail. Note that although discussions of constancy typically emphasize illumination conditions as the variable element, the phenomenon is more general than this. Nothing precludes the variable element from being a transparent filter or volume. Further, nothing precludes the variable element from being surface colour and the constant element from being a filter, volume or condition of illumination. Here the “proximal” colour remains constant while the “distal” colours do not. For example we can experience a room as similarly illuminated despite the presence of a host of surface colours before us. This is a kind of experienced colour constancy.[4]

The constancy cases directly relevant to layering are cases in which the variable and constant element are both (in some sense) experienced.[5] Here a means of explaining what is experientially constant that incorporates what is variable into our experiences must be sought, and layering – e.g., experiencing a constant element by looking at it through a variable element, or vice versa – has natural application (see xxx). Suppose the elements are a transparent film in front of an opaque surface. Assume that one experiences colour layering when looking at the scene. If one of these colours remains constant, while the other varies, and we experience precisely that, then one experiences a constant colour despite the simultaneous presence and experience of a variable colour. That is, colour constancy occurs, and it is explained by reference to experienced colour layering. It is (at least logically) straightforward to extend this layering analysis to cases in which a volume is an element instead of a filter, and to cases in which illumination is an element instead of a filter or volume, and of course to cases involving various combinations of these elements. There are many worries one might have about the widespread application of a layering analysis of constancy. In my view such worries are generally misplaced, but detailing why goes outside the scope of this work. For dialectical purposes I ask the reader to take for granted the tenability of a layering analysis of numerous types of constancy cases, including those involving experienced illuminations and surfaces.

2. A Primer on Colour Relationalism

Colour Relationalism proper is the broad thesis that a thing’s colour is not intrinsic to it, but only persists by virtue of relations that thing bears to other things. This alone says nothing about what class of objects (Oc) instantiate these relational colours, nor about the class of relations (Rc) that constitute colours. Regarding the former, for example: Oc may only contain objective things in the world, like dogs and trees; Oc may only contain mental objects like sense-data; Oc may only contain objects that do not exist in our world but do exist in some possible set of worlds; Oc may contain some combination of these or other collections of objects. Similarly, there are many options for the class of relations that constitute colours, including: Rc might only contain relations between objective things; Rc might only contain relations between mental objects; Rc might contain relations between both objective and mental objects.

I will not give names to each of these variants of Relationalism, or explore them in any detail. My point is twofold. First, it is important to emphasize that Relationalism can be articulated within colour mentalism, dispositionalism, objectivism and eliminativism. The above characterization reinforces this. Second, evidence offered in favour of Relationalism must be assessed against options of the above sort to determine the extent to which that evidence in fact supports Relationalism, and if so the extent to which it supports some variant(s) of Relationalism over others. As one might expect, this turns out to be a complicated matter, and it is not one that I will delve into. My interest is instead in a general form of reasoning that has been offered in support of Relationalism and its connection to colour layering. I proceed with the former.

One well-known source of evidence for Relationalism is simultaneous contrast effects involving focal objects and the visible objects and colours above, below and to the right and left of the object.[6] Call this collection of relations the surround of the object and this the class of center-surround instances of the relativity of colour. Note that by stipulation an object’s surround excludes what is in front of and behind the object. The experienced colour of an object can change dramatically when it is surrounded by one set of experienced colours and then another. For example an object can be experienced as blue when embedded in one surround and green when embedded in another. Notoriously, there are surround (or contrast) colours – for example pure black, pure white, brown and olive green – that is, colours that are only experienced with suitable surround contrasts in place. Center-surround contrast phenomena are not only dramatic but now essential to our understanding of colour experience.

The temptation to infer Relationalism is natural.[7] The experienced colour of an object A fundamentally depends on the experienced colours of objects B1-n that surround it (this is an instance of the Relativity of Colour Perception). While one might try to single out a preferred surround for experiencing A’s colour, after inspection there is often no principled, well-motivated reason for doing so (No Privilege). Epistemically, we thus seem obligated to give these different colour experiences of A equal weight (Ecumenicism). Object A seems intrinsically the same across these variations in its surrounds (Object Stability). Therefore, we have grounds to infer (e.g., the best explanation of the above is) not merely that A’s experienced colour depends on the experienced colours of objects surrounding A, but that A’s colour proper depends on the colours proper of objects surrounding it. In such a case I will say that Relationalism has Explanatory Import.

There are various means of resisting this reasoning. One could seek to find a privileged – e.g., “normal” or “ideal” – context in which to perceive A’s colour. One could deny Object Stability, perhaps on grounds that A is a mental object (e.g., a sense-datum) that does intrinsically change across these kinds of contexts. One could generally resist any attempt to infer a categorical conclusion about colour from data drawn from colour experience. Thus, it may be that, because of our visual systems, for us to experience brown suitable contrast must be present, but this does not entail that brown cannot be instantiated without suitable contrast present (e.g., Byrne & Hilbert 2003). This has the potential result of colours being very difficult to learn about via colour experience (Epistemicism).

These disputes are not my concern. For simplicity I will assume that the lesson to draw from center-surround contrast effects is that some variant(s) of Relationalism are true. Which ones? Here are some initial thoughts. If, as is unlikely, the correct explanation of center-surround contrast locates the reason for the effects in objective relations between objective things in our world, then an objectivist Relationalism receives support from the phenomenon, and subjectivist ones do not. If, as is most likely, the reason pertains to peculiarities of our visual system, then some form of subjectivist Relationalism is mandated to the exclusion of objectivist ones. But to this point there is no means of distinguishing, for example, between a subjectivist Relationalism in which the members of Oc are objective things and some members of Rc are subject-involving (e.g., dispositionalism), and a subjectivist Relationalism in which the members of Oc are subjective things like sense-data and the members of Rc are subject-involving. Regarding the latter, the sense-datum theorist can hold that brown sense-data only occur when appropriate contrasting sense-data do too. Thus, center-surround contrast data are at least at first pass equally-consistent with dispositionalist and mentalist forms of Relationalism. Considerably more options are available when one delves with the Eliminativist into possible worlds (which I will not do).

Center-surround contrast effects afford a reasonable quantity of evidence for Relationalism, since these effects occur throughout our colour-visual lives. However, this class of effects is but one of the many classes that fall under ‘the relativity of colour’. Others include variations due to successive contrast effects, to illumination effects, to additional peculiarities of perceivers’ perceptual apparati (e.g., types of cones), et cetera. If there was reason to believe that the above reasoning should be applied to all or at least most instances of the relativity of colour, then we would have a general and powerful argument for Relationalism. Such a generalization of reasoning would follow from a Uniformity Hypothesis to the effect all or most instances of the relativity of colour should be treated as above, namely that each case can be described in a parallel fashion, and that there are grounds to apply No Privilege, Ecumenicism, Object Stability and therefore Explanatory Import. This is what Cohen maintains. At the beginning of The Red and the Real he nicely states the explanatory import of Relationalism for instances of center-surround simultaneous contrast. When concluding the discussion he states:

I’ll be emphasizing that the very same considerations can be applied again and again across different sorts of variations…I take this fact to be significant, since I think the recurrence of identical considerations across different sorts of variation puts pressure on us to favor a uniform reaction in each case – a lesson that, as we’ll see, many philosophers have failed to observe. (2009, 20-21, italics added).

In many respects this remark sets the tone for the book, and I find no wavering from it in the work.

It is worth emphasizing that with the Uniformity Hypothesis Relationalism transitions from an ontological constraint helpful to explain some set of phenomena to the constraint essential to explaining a vast collection of phenomena that is among the most challenging for colour theorists. With the Uniformity Hypothesis Relationalism effectively becomes a default view.

What I wish to challenge via colour layering is this Uniformity Hypothesis. The point is not to argue that there is no evidence for Relationalism – I have, for the purpose of this discussion, conceded that there is. The point is also not to argue that Relationalism is at odds with some set of data – on this I wish to remain mute. The point is instead to argue that it is remarkably premature to postulate the Uniformity Hypothesis, and therefore that we lack a motive to regard Relationalism as the explanatory saviour Cohen presents it as.

3. Colour layering and Relationalism

For simplicity I will split the relativity of colour into three classes. The classes are not meant to be exhaustive, nor need each class be exclusive on further examination. The point is to find a means of isolating the impact of colour layering on the justification for Relationalism. Here is the proposal, noting that terminology will be explicated below:

  • Center-surround relativities: instances in which the experienced colour of a thing varies as experienced objects above, below, to the left or right of the object vary.
  • Line-of-sight (LOS) relativities: instances in which the colour experience along a LOS varies as experienced objects in front of or behind an object vary.
  • Perceiver-object relativities: instances in which the experienced colour of a thing varies as unexperienced aspects of perceivers’ perceptual apparati vary.

Center-surround instances of simultaneous contrast fall under (1). By hypothesis variations in experienced colour due to numbers of cone types and cone sensitivities fall under (3), as do variations due to differences in other aspects of the architecture of perceptual systems. There is a long list of phenomena falling under (2):

LOS relativities can include colour perceptions through: sunglasses, windows and other transparent films; liquids, atmospheres and other transparent volumes; plausibly the vitreous humour and lenses of one’s eyes; the light illuminating a scene; afterimages and other line of sight instances of successive contrast; et cetera.

It is this class that is relevant to assessing the impact colour layering has on the justification for relationalism.

My argument against the Uniformity Hypothesis has two steps. The first is the conceptual claim that experienced colour layering undermines the reasoning that leads to Explanatory Import. The second is the empirical claim that any presumption of the widespread absence of colour layering requires an at present unmotivated bias, for example a bias in favour of Relationalism. I sustain this not by demonstrating the proliferation of colour layering in experience, but by arguing that there may be widespread layering throughout experience, and that there are several factors that make it difficult to assess the extent to which layering in fact occurs. The prudent position is therefore some form of agnosticism and hence not the Uniformity Hypothesis.

3.1 Layering and the explanatory import of relationalism.

In §1 I noted that experienced colour layering – say looking at grass through a tinted window and experiencing opaque green through transparent grey – demands a contextual interdependency of one’s experience of one of these colours on one’s experience of the other. One experiences the transparent colour by looking through it to the opaque colour, and vice versa. In §1 we also saw that most if not all colour ontologies, and core experiential mechanisms like acquaintance and representation, can accommodate colour layering. Hence experienced layering is consistent with each colour being non-relational – this contextual interdependency of colours does not entail the constitutive dependency of one colour on the other.

The basic means by which layering targets the reasoning for Explanatory Import is by offering a scenario in which two colours are uncontroversially intertwined in a context, both ontologically and experientially, but in which there is (I argue) no temptation to infer that, because of this relation, those colours are constitutively dependent on one another. By contrast the reasoning for Explanatory Import roughly recognizes the dependency in a perceptual context of an experienced target colour on another colour (e.g., the dependency of an experienced center colour on an experienced surround colour), and infers from this that the nature of the target colour constitutively depends on the other colour (i.e., infers Relationalism). In short one infers a constitutive dependency from a contextual one. While this inference, though not without its critics, is fairly compelling when applied to simultaneous contrast cases, cases involving colour layering illustrate a novel respect in which the inference is questionable.

Let me explain in more detail by focusing on the following scenario[8]:

Layering Scenario. Suppose there is an opaque white object Owhite, a transparent grey object Tgrey, and a transparent yellow object Tyellow. At first you are presented with Owhite behind Tgrey, and then with Owhite behind Tyellow. Suppose that in both contexts you experience colour layering, so that at first you colour experience Owhite-through-Tgrey and then Owhite-through-Tyellow.

Consider two potential objections to this description of the scenario.

One might ask whether describing your colour experiences via iterations of the complex Ox-through-Ty can be simplified, say to merely describing your experiences via iterations of Oz. That is, aren’t you just experiencing the opaque thing to be differently coloured across these contexts? This is not a friendly amendment, but instead an attempt to eliminate experienced colour layering. There is no such simplification: experienced colour layering can only be adequately described via a complex such as Ox-through-Ty. In a similar vein one might ask: Does the colour of O look different across these contexts? In some sense yes and in some sense no. I suggest that in this discussion focusing on “looks” locutions, well-known troublemakers in perceptual theory, is liable to mislead our analysis. Let me therefore attempt to keep us on track.

What is important about this Scenario is that we may additionally suppose that the object colours are constitutively non-relational, that our three objects are intrinsically white, grey and yellow, respectively. This is reasonable because if these objects do have intrinsic colours, and you are first presented with Owhite behind Tgrey, and then with Owhite behind Tyellow, then, if you experience colour layering, we would expect you to first experience Owhite-through-Tgrey and then Owhite-through-Tyellow. This is to say that there is absolutely no pressure to infer from the contextual interdependence of one’s experiences of these colours to the constitutive dependence of one of these colours on another. There is not only no pressure to infer Relationalism from this scenario, doing so would be illegitimate.

For concreteness, consider how the Scenario impacts the reasoning for Explanatory Import. Object Stability is reinforced, since putting Tgrey and then Tyellow in front of Owhite suggests no intrinsic change to any of these objects. No Privilege and Ecumenicism are also reinforced, since there is no reason to treat either the colour experience Owhite-through-Tgrey or Owhite-through-Tyellow as somehow privileged over the other, and hence both should be treated equally. And the Scenario is clearly an instance of the relativity of colour perception. So why does the explanatory import of Relationalism not follow?

The reason is because the argument for Explanatory Import demands a specific interpretation of the relativity of perception, and that interpretation does not apply in layered scenarios. Here is Cohen’s description of the relativity of colour perception: “There are multiple, psychophysically distinguishable perceptual effects (in respect of color) of a single color stimulus” (2009, 24). I will take his ‘effects’ to be interchangeable with my ‘experiences’. Clearly Owhite is causally involved in bringing about the experience/[effect] Owhite-through-Tgrey, and in that sense Owhite has a role in generating these distinct colour experiences in the Scenario. But it would be erroneous to hold that Owhite itself generates these distinct colour experiences, or to hold that these distinct experiences are simply experiences of the colour of Owhite. The latter in particular is why the reasoning for Explanatory Import does not go through. In order for No Privilege, Ecumenicism, and Object Stability to create a rationale for Relationalism, the variable colour experiences drawn from the relativity of colour must be experiences of the target object’s colour (i.e., “of a single color stimulus”). But in our Layering Scenario the difference between the experiences Owhite-through-Tgrey and Owhite-through-Tyellow is in a fundamental sense not a difference in the experienced colour of Owhite, it is a difference in the experienced colour of something else along the LOS on which Owhite is currently located. Thus while, as it were, the total colour experience along this LOS changes, and Owhite is the terminal visible object along that LOS, the experienced colour of Owhite is stable, and there is no motive to infer that Owhite’s colour is relational.

This is not to say that layered contexts and LOS relativities more generally can all be treated similarly. Most importantly, this class of relativities contains instances of experiential colour fusion. Recall that during fusion one is presented with multiple objects along a LOS, but one experiences a single colour that mixes or fuses together the contributions of the objects located along that LOS. If I put a blue sheet of cellophane a few centimeters in front of a yellow book, I (by hypothesis) can experience Oyellow-through-Tblue. However, if I put the cellophane directly on the book (by hypothesis) the only colour I experience is green. In the latter case the experienced colour is a fusion of the colour contributions made by the cellophane and book. Experienced colour layering is not an adequate tool for describing such a case. By contrast, hypothesizing that the experienced green colour is inherently relational – that it depends on colour contributions from two distinct objects, and on those objects standing in a particular relation to one another – is, though not forced, at first pass credible. However, it is worth gaining some clarity on the significance of this by returning to the reasoning for Explanatory Import.

In the above fusion case Object Stability is reinforced, neither the book nor the cellophane are inherently changed in the scenario. However, the stabilities of the “experiential objects” are arguably not preserved. When there are a few centimeters between the cellophane and the book, both are experienced as distinct objects, and the distinctness of their colours are (by hypothesis) experienced. However, when the distance between the two object bounds toward zero, their experienced colours are fused together. Perhaps this is in part because the objects are no longer experienced as distinct: one doesn’t experience a sheet of cellophane on a book, but a single object with a single facing surface to which green is attributed. If so then there is a single experienced colour that is attributed to some single experiential object. One could debate about how this connects to Cohen’s interpretation of the relativity of colour, but I wish to set that aside. What is important is the impact this fusion scenario has on our analysis of No Privilege.

If one’s aim is to discern the colour of the book, then a colour experience in which the book’s contribution to colour is distinctly experienced should be preferred over one in which the book’s and cellophane’s contributions are fused together. Fusion experiences yield information, not about the book’s colour simpliciter, but about a relation between the book’s colour and the cellophane’s colour. This relational information is valuable, but if one is seeking information about the book’s colour, the information is arguably inferior to information in which the book’s contribution to colour experience is distinctly packaged, or separated out from the contributions of other objects. The contrast with layered experiences is instructive: the layered experience Oyellow-through-Tblue distinguishes the book’s contribution to the experience from the cellophane’s; the fusion experience Ogreen does not. If one is seeking to learn of the book’s colour, it is reasonable to privilege the former over the latter. Privilege should also be granted to experiences in which only the book is contributing to the colour experience along a LOS.[9] Call experiences in which only one object is contributing to the colour experience along a LOS, LOS1-object experiences. Thus, to use fusion cases to ground Relationalism one must exclude these alternatives, for with these alternatives in play, No Privilege and then Ecumenicism fail.

We should therefore resist the reasoning for Explanatory Import in LOS scenarios in which layering or fusion occurs. If experienced layering occurs, then the kind of perceptual relativity at play is not the kind needed to sustain Explanatory Import. If one’s aim is to learn about an object’s colour and experiential fusion occurs involving that object and another object along the same LOS, such fusion experiences are epistemically inferior to LOS1-object or layering experiences involving that object.[10]

In addition a host of intermediary cases between “perfect” experienced layering and “perfect” experienced fusion are possible. There might be cases close to but note quite achieving perfect layering, where the opaque and transparent object colours are almost perfectly distinct in experience but a perceptible degree of mixing occurs. On the other extreme there might be cases close to but note quite achieving perfect fusion, where the opaque and transparent object contributions to experienced colour are almost perfectly fused, but a residual layering effect remains. And so on (see xxx for some additional options). Such variations do not challenge the above analysis. To my mind the only means of undermining the above analysis is via a demonstration that experiences of colour layering and of LOS1-object are impossible or bound toward being so. I know of no attempt to provide this demonstration, but welcome any such efforts.

3.2 Layering: its potential reach and confounds.

Given the sheer size of the class of LOS relativities (see the beginning of §3), the above argument marks a potentially wide-reaching means of undermining the Uniformity Hypothesis. Its potency hinges on the proliferation of non-fusion (LOS1-object or layering) colour experiences within the LOS class. If this is the case, why don’t we start scrutinizing? Unfortunately, assessing instances of LOS relativities for layering/LOS1-object/fusion yields several confounds. Here, in brief, are four of them. For simplicity I focus on layering and fusion experiences, and remark on LOS1-object experiences at the end of the section. It will help to roughly categorize the many LOS relativities into three classes, colour perceptions involving contributions from transparent films, volumes, and illuminants.

Heterogenous classes. We should be wary of any attempts to conclude that a particular class contains experiences of layering to the exclusion of fusion, or vice versa. There is nothing to prevent one context involving a transparent film from yielding an experience of colour layering, and another from yielding experiential fusion. We know that our perceptual systems will have much to say about whether experiential layering or fusion occurs (see, e.g., references in note 3). Stated most generally, there are various factors within a complex scene that prompt, to some extent or other, one’s visual system into a state involving experiential fusion or layering. Nothing as simple as “perceptions involving transparent films yield layered experiences” is antecedently credible. We instead must be open to the possibility that some members of each perceptual class contain experiential fusion and some others layering.

Perceptual ambiguity. A stimulus in one of these classes might give rise to a perceptual ambiguity that “flips” between experiential fusion or layering. For example a partly shadowed wall might at one moment be (correctly) experienced as differently illuminated and similarly painted, and at another moment (incorrectly) as similarly illuminated but differently painted. In the first experience the illuminant and surface contributions to colour are experientially distinct, and in the second at least some of the illuminant and surface contributions are fused together. Arguably experiential layering occurs in the first instance in ways that are replaced with experiential fusion in the second.[11]

There can also be ambiguities across classes, say in which a volume perception (e.g., looking at a scene through thick smog) is experienced as a member of that class then as a member of the illumination class (e.g., as a scene under low levels illumination). In principle there is nothing to prevent a single stimulus from being involved in both types of ambiguity.

First-personal limitations. There are cases in which it is fairly obvious that experiential layering versus fusion is occurring. However, the boundaries of these kinds of experiences is undoubtedly vague, yielding a range of cases with only subtle experiential differences. Further, subjects are in general not accustomed to making judgements about layering/fusion. This is a troublesome combination. Imagine asking a subject: “Are all colour elements experienced along this LOS attributed to that opaque thing (fusion) or are some colour elements instead attributed to a transparent intermediary (layering)?” We cannot antecedently expect straightforward data from subjects presented with a range of subtly difference cases. To take a concrete example, when one’s experience when looking at the world gets a yellowish tinge due to the yellowing of the lenses of one’s eyes with age, is the experienced yellowing an instance of attributing the yellowness to objects in the world (fusion), or of looking at those objects through a yellow lens (layering)? If the former, then Relationalism is potentially critical to explaining the phenomenon, but if the latter then it is otiose. A subject’s merely indicating some form of “things look more yellow” does not decide the matter. Our access to the fine-grained nature of our own colour experience is not adequately reliable to straightforwardly assess the matter.

Colour constancy. The potential impact of colour constancy on this discussion is immense. When colour layering occurs, and one of the layers varies while the other remains constant, and the subject experiences just that, then a form of colour constancy obtains. For ease of reference call it layered constancy. As noted in §1 (and discussed at length in xxx) this model for constancy has potential instances in constancy cases involving opaque surfaces and variations in filters, or volumes, or illuminants (and converse cases where the opaque surfaces vary and the other elements remain stable). Given that constancy obtains throughout our daily lives, the potential number of instances of layered constancy is vast. Each such instance reduces the extent to which Relationalism should be used as an explanatory tool.

Cohen (2008) details a Relationalist-friendly account of colour constancy aptly deemed a counterfactualist account. According to it, the contributions that the constant and variable elements make to colour experience are fused together to yield a relational colour that is both experienced by the subject and in the world. Thus, during constancy cases there is in fact no constant colour, only colour variation. Take for example a partly shadowed blue wall. The surface and illuminant contributions to experience are fused together on both the shadowed and unshadowed parts thus yielding two different colours, a lighter and darker shade of blue. While there is blue on both parts of the wall, and thus a constancy of colour category, there is no constant fine-grained colour. Since these distinct fine-grained colours of the wall are defined by reference to both surfaces and illuminants, they are relational in at least this regard.

That is the “occurrent” (to use Cohen’s term) experiential and ontological reality of constancy cases. To make this an account of colour constancy, however, something has to be added to explain those aspects of subjects’ behaviours that indicate that there is more to these cases than merely colour variations, that there is a sense in which the relevant colour samples are constant. For this purpose Cohen postulates the existence of counterfactual colour contents, contents to the effect that the colours in these two parts of the world would be the same were the perceptual conditions the same. For example the blues on these two parts of the wall would be the same were illumination conditions the same. The thought is that our visual-cognitive systems are making these counterfactual judgements whenever colour constancy obtains (which is throughout our lives), and our commitment to constant colours in these cases is an expression of these contents, not an expression either of a constant experienced colour or of a constant instantiated colour. I will make two remarks.

The first remark is methodological. A counterfactualist analysis of constancy is not empty, on the contrary I suspect it has instances. However, one’s goal should not be to try to force all instances of colour constancy into one model (be it a layering, counterfactualist or some other model) to accommodate one’s broader objectives (be them a defense of colour objectivism or Relationalism). Our goal should be to examine various instances of colour constancy with various models in mind to try to uncover which one or more of them best explain individual or classes of cases. The potential result is that colour constancy is a varied phenomenon, having some instances of layered constancy, others of counterfactual constancy, and so on.

My second remark is the now-familiar one that, with regard to colour constancy, Relationalism only has explanatory import where instances of counterfactualism obtain, and has no explanatory import otherwise. Since colour constancy is such a widespread phenomenon, occurring throughout scenarios involving LOS relativities, there is no simple, straightforward way for Cohen to argue that Relationalism explains these cases, and likely no way in general, since layered constancy plausibly has instances.

Here is the general lesson: there are some and potentially numerous instances of experienced colour layering; determining their true extent must be done not by asserting the Uniformity Hypothesis or any such theoretical principle, but by careful examination of relevant cases; due to various confounding factors that examination will not be straightforward. In other words, we cannot assert the Uniformity Hypothesis.

Note that this conclusion follows regardless of the prevalence of LOS1-object experiences. Recall that an LOS1-object experience includes a case in which a surface is illuminated by an illuminant, and viewed through some air, but in which only the surface contribution to colour is experienced along that LOS. In one sense the independence of the above conclusion from the prevalence of LOS1-object experiences is fortunate, because I am doubtful that LOS1-object experiences are prevalent. I instead suspect that illuminants and transparent volumes make regular contributions to the colours experienced along LOSs. However, if it turns out that LOS1-object experiences are prevalent, then our confidence in the Uniformity Hypothesis is further eroded.

4. Conclusion

The most compelling argument for Relationalism infers the inherently relational character of colour from cases in which one’s experience of a colour contextually depends on one’s experience of other colours. I have argued that this reasoning has limits, that in cases involving experienced colour layering, the contextual interdependency of one’s experience of one colour on one’s experience of another does not ground an inference to the relational nature of colour. However, I have not argued that this argument for Relationalism is generally bankrupt. The prevalence of a rich phenomenon like center-surround contrast effects should tempt us all to be Relationalists of some sort. These two conclusions are a way of expressing what results from denying the Uniformity Hypothesis. The relativities of colour, rather than yielding “identical considerations” that put “pressure on us to favor a uniform reaction in each case”, yield differing considerations, some of which pressure us to become Relationalists, others of which do not (Cohen, 2009, 20-21). The global pressure toward Relationalism is illusory, and to the extent that we succumb to it, we risk being blind to surprising possibilities, like the potential ubiquity of colour layering.

 

References

Anderson, B.L. (2008). Transparency and Occlusion.  In Basbaum, A.I., Kaneko, A., Shepherd, and Westheimer, G., eds., The Senses:  A comprehensive Reference, Vol. 2, Vision II, Albright, T.D., & Masland, R. (volume eds.), pp. 239-244. San Diego, USA: Academic Press.

Brainard, D., W. Brunt & M. Speigle (1997). Color constancy in the nearly natural image. I. Asymmetric matches. J. Opt. Soc. Am., 14(9): 2091-2110.

Brainard, D. & Wandell, B. (1988). Classification measurement of color appearance. Investigative Ophthalmology & Visual Science, 29: 162.

Byrne, A. & D. Hilbert (2003). Colour realism and colour science. Behavioral and Brain Sciences, 26: 3-64.

Casati, R. (2009). Are shadows transparent? An investigation on white, shadows and transparency in pictures. Res, 56: 329-335. My source is the Author Manuscript ijn_00171296, pp. 1-12, Sept 2007.

Cohen, J. (2008). Colour constancy as counterfactual. Australasian Journal of Philosophy, 86: 61-92.

—. (2009). The Red and the Real: An Essay on Colour Ontology. Oxford, UK: Oxford University Press.

Jameson, D., & Hurvich, L. (1989). Essay concerning color constancy. Annual Review of Psychology, 40: 1-22. Reprinted in Byrne & Hilbert, eds., 1997, Readings in Colour, Vol. 2: The Sciene of Color, pp. 177-98. Cambridge Mass: MIT Press.

Khang, B.-G. and Q. Zaidi (2002a). Accuracy of color scission for spectral transparencies. Journal of Vision 2(6): 451-466S.

—. (2002b). Cues and strategies for color constancy: perceptual scission, image junctions and transformational color matching. Vision Research 42(2): 211-26.

Kingdom, F. A. A. (2011). Lightness, Brightness and Transparency: A quarter century of new ideas, captivating demonstrations and unrelenting controversy. Vision Research, 51: 652-673.

Kitaoka, A. (2005). “A new explanation of perceptual transparency connecting the X-junction contrast-polarity model with the luminance-based arithmetic model”. Japanese Psychological Research, 47: 175-187.

Land, E. (1986). Recent advances in Retinex theory. Vision Research, 26: 7-21. Reprinted in Byrne & Hilbert, eds., 1997, Readings in Color, vol. 2: The Science of Color, pp. 143-160. MIT Press: Cambridge, Mass.

Land, E. & McCann, J. (1971). Lightness and Retinex theory. Journal of the Optical Society of America, 61: 1-11.

Metelli, F. (1974). The perception of transparency. Scientific American, 230(4): 91-98.

Shevell, S. K. & F. A. A. Kingdom (2008). Color in Complex Scenes. The Annual Review of Psychology 59: 143-166.

Wandell, B. (1989). Color constancy and the natural image. Physica Scripta, 39: 187-92. Reprinted in Byrne & Hilbert, eds., 1997, Readings in Color, vol. 2: The Science of Color, pp. 161-75. MIT Press: Cambridge, Mass.

Wollschläger and Anderson (2009). The role of layered scene representations in color appearance. Current Biology, 19: 430-435.

 


Notes

[1] For the final version of this paper please see:

Colour layering and colour relationalism. Mind and Machines: A Special Issue on the Philosophy of Colour ed. by M. Chirimuuta, 25(2): 177-191.

For helpful comments on earlier drafts of this paper, many thanks to Jonathan Allan, Mazviita Chirimuuta, Jonathan Cohen, Wayne Wu, an anonymous referee at Minds and Machines, and audiences at the Pacific APA 2015, the University of Manitoba, and Towards a Science of Consciousness 2015.

[2] Setting aside colours for a moment, here are rough experiential and causal characterizations of ‘transparent’, ‘transluscent’ and ‘opaque’. Experientially, to be transparent is to be something that can be clearly seen through, causally it is to transmit (as opposed to reflect) light without distortion. An object is transluscent to the extent that things are seen less clearly through it, and as it begins to distort (e.g., scatter) light that it transmits. Opaque objects cannot be seen through, and transmit no light. In this context ‘thing’ or ‘object’ must apply at least to liquids, solids and gases.

[3] In psychology, layered visual experiences – scissions – have been of particular interest since the 1970s and are actively studied today. Metelli (1974) is a classic work. Kitaoka (2005) corrects some deficiencies, and Anderson (2008) offers an alternative and summarizes some seemingly fatal problems for Metelli’s view and the views based on it. Kingdom (2011) is an excellent review article of relevant phenomena, and Casati (2009) provides an good summary and connects the issue to some philosophical debates. See also, e.g., Khang & Zaidi (2002a,b) and Wollschläger and Anderson (2009). It is fair to say that there is no general account of scissions that dominates contemporary empirical literature. Instead, we should expect that scissions will remain an active area of empirical research at least in the near-term.

[4] This suggestion goes against the spirit of some constancy models, where preference is given to the distal or opaque surface colours over the contributions filters, volumes and illuminants make to visual stimuli. However, this tension is not a legitimate philosophical reason to exclude this kind of case from colour constancy.

[5] These cases are to some extent in conflict with a traditional computational approach to colour constancy which hypothesizes that the variable element is discounted by our visual systems, and so not experienced by us, or at minimum that this is the relevant goal or ideal of our visual systems. Land & McCann (1971) and Land (1986) are well-known examples, and Brainard & Wandell (1988), Wandell (1989) Brainard, and Brunt & Speigle (1997) offer different approaches. Jameson & Hurvich (1989) is one well-known response to “discount the illuminant” approaches. Shevell & Kingdom (2008) is a worthwhile recent review of the wider literature.

[6] In this context the relations above, below, to the right and to the left of are defined relative to the perspective of the perceiver. Thus, what is to the right of an object for one perceiver might be to the left of an object to another perceiver.

[7] What follows roughly accords with Cohen’s Master Argument (2009, p. 24). Space prevents a thorough comparison.

[8] For the purposes of this section it is helpful to ignore the impact of non-LOS factors on colour experience.

[9] For example perhaps along a LOS is the book, which is illuminated by an illuminant I1, and viewed through some air V1. Consider a context in which only the book’s contribution to colour is experienced along that LOS. If one seeks to learn of the book’s colour, this context should be privileged over a context that fuses together contributions from the book, illuminant and/or air.

[10] Note that even with fusion experiences there may be a means of “inferring” an object’s non-relational colour, namely, by trying (intellectually) to isolate the various contributions to experienced colour that an object makes across a host of fusion experiences.

[11] There is much more to be discussed about this kind of case, but space demands a brief treatment.

15 thoughts on “Colour layering and colour relationalism”

  1. BROWN AND WHITE’S EFFECT

    1 Simultaneous Contrast and White’s Effect

    Derek Brown’s work on colour layering explores an aspect of visual experience which has been wrongfully neglected by philosophers writing on colour, and offers fascinating proposals as to its implications for debates over colour constancy (Brown, 2014), and here, for arguments over the relational theory of colour ontology.

    Brown’s specific target is Jonathan Cohen’s “Master Argument” for relationism in which one, “infers the inherently relational character of colour from cases in which one’s experience of a colour contextually depends on one’s experience of other colours” (Brown, 2015, 1). The core cases are center-surround effects like Figure 1, where the circles have the same physical properties (size, reflectance, etc.) but vary in their apparent colour across the two backgrounds. Assuming that neither background is the better one for viewing the colour of the circle, and that we have no other reason for giving one appearance of the circle epistemic weight over the other, a reasonable conclusion is the relationist thesis that the circle’s colour (not just the colour appearance) depends on the colours surrounding it. Thus relationism, in Brown’s words, has “explanatory import”.

    Cohen (2009, 20-21) urges that we treat other cases of colour variation due to perceiver differences, and changes in the colours of lighting and transparent objects in front of the target object (line-of-sight or LOS relativities), in the same way. The point of Brown’s article is to argue that the evidence base for relationism is much narrower than Cohen has claimed because it does not, without prejudice, include line-of-sight relativities. As an alternative analysis of such cases, such as looking at a scene of post-industrial decay through rose tinted sunglasses, Brown proposes colour layering. Rather than saying, with the relationist, that the cold greys of the abandoned warehouse become warmer, pinker shades when we don our rose tinted spectacles, we more accurately describe our visual experience if we say that we experience the same cold, opaque grey of the concrete through the soft, transparent pink of the glass (Ogrey − through − Tpink).

    Two things strike me about Brown’s exposition of layering. First is the treatment of colour vision as a process independent of the perception of opacity and transparency. One implication is that the same green colour (characterised in a three dimensional colour space by a specific hue, saturation and lightness, or HSL), may be experienced both when we look at a transparent perspex filter or at an opaque piece of ceramic.

    Second, Brown does not consider whether perceptions of opacity and transparency are them- selves subject to context effects. It is easy to show that they are. Figure 2 is a close cousin of the simultaneous contrast effect—White’s effect. The short grey bars have the same physical reflectance/luminance properties in both contexts, but differ in apparent colour. The effect is known as colour assimilation because the short grey bars are said to appear nearer in colour to the vertical bars flanking them. More important for our discussion is the observation that the upper bars, as well as seeming darker, also appear opaque, whereas the lower bars appear lighter and transparent. In what follows, I aim to show how both of these points cause trouble (or at least introduce complications) for Brown’s case against the explanatory import of relationism.

    Figure 1: Simultaneous contrast. The two circles have the same luminance (on screen) or reflectance (if printed) but because of the difference in background they appear to be at different grey levels. From Kingdom (2011). Cf. Cohen (2009).
    Figure 1: Simultaneous contrast. The two circles have the same luminance (on screen) or reflectance (if printed) but because of the difference in background they appear to be at different grey levels. From Kingdom (2011). Cf. Cohen (2009).

    Figure 2: White’s Effect. The two sets of short bars have the same luminance (on screen) or reflectance (if printed) but because of the difference in background they appear to be at different grey levels, and to differ in their opacity/transparency. From Kingdom (2011).
    Figure 2: White’s Effect. The two sets of short bars have the same luminance (on screen) or reflectance (if printed) but because of the difference in background they appear to be at different grey levels, and to differ in their opacity/transparency. From Kingdom (2011).

     

    2 Brown’s Case against Explanatory Import

    Brown (2015, 9-10) writes that:

    The basic means by which layering targets the reasoning for Explanatory Import is by offering a scenario in which two colours are uncontroversially intertwined in a context, both ontologically11 and experientially, but in which there is (I argue) no temptation to infer that, because of this relation, those colours are constitutively dependent on one another.

    The key claim here is that when layering occurs, it is readily apparent to us that our colour experience of one object is contextually related to the colour of another object, and yet there is no pressure to infer from this that the colour itself is dependent on the context. Here is an analogy which shows how intuitive the layering proposal is: my experience of the images on the movie screen happen to be contextually related to the beehive hairstyle of the movie goer in the row in front of me, which is partly blocking my view; yet I would never infer that the film images themselves are constitutively dependent on the offending hairdo.

    Furthermore, Brown (2015, 11-12) tells us that there is a basic invariance (constancy) in our experience of the different colours which appear in a Layering Scenario:

    The difference between the experiences Owhite−through−Tgrey and Owhite−through− Tyellow is in a fundamental sense not a difference in the experienced colour of Owhite, it is a difference in the experienced colour of something else along the LOS on which Owhite is currently located. Thus while, as it were, the total colour experience along this LOS changes, and Owhite is the terminal visible object along that LOS, the experienced colour of Owhite is stable, and there is no motive to infer that Owhite′s colour is relational.

    Here I would like to complicate things. One consistent finding from the empirical literature is that the interpretation of a scene as layered, does affect the perceived colour of terminal and intermediary illuminants and objects, in particular with respect to the lightness or brightness dimension. Kingdom (2011, 655) writes of the,

    mounting body of evidence that brightness depends on perceived illumination or on transparency. For example, shadows appear brighter than their equal-in-luminance reflectance counterparts …and surfaces appearing to lie in shadow or behind dark transparencies appear brighter than if presented on equivalent reflectance backgrounds.

    Adelson’s snake is a dramatic illustration of this phenomenon. All of the small diamonds in Figure 3 have the same physical luminance/reflectance, but appear to have very different grey levels. The left hand side of the image shows the effects of transparency. The top left row of diamonds, seeming to be covered by a dark transparency, appears a much lighter grey than the bottom row, which seems to be strongly illuminated. The right hand side of the image is a straightforward case of simultaneous contrast. Without the additional transparency factor, the apparent difference in the grey levels between the right top and bottom rows of diamonds is not so great. In short, perception of transparency intensifies the simultaneous contrast effect.

    One may object that Adelson’s snake is no counter-example to Brown’s point about the con- stancy of the apparent colours of terminal objects in layering scenarios because it is a computer generated mock up of a scene, where one physical stimulus (the diamond) is designed to convey two layers of a 3D scene (both terminal object and the transparency or illumination it is viewed through). However, that would be to miss the more general point that in the real world, perception of colour, transparency and illumination interact and impact on each other in surprising ways. For example, interpretation of a grey expanse as shadow rather than surface will make it appear lighter, as will interpretation of an object as being perceived through a dark filter (Adelson, 2000).

    Figure 3: Adelson’s Snake. All of the small diamonds have the same luminance (on screen) or reflectance (if printed). But the top left row of diamonds, which seems to be behind a dark trans- parency, appears much lighter than the bottom left row, which seems to be strongly illuminated. The image on the right hand side (the ‘anti-snake’) shows the basic simultaneous contrast effect for comparison. From Kingdom (2011).
    Figure 3: Adelson’s Snake. All of the small diamonds have the same luminance (on screen) or reflectance (if printed). But the top left row of diamonds, which seems to be behind a dark trans- parency, appears much lighter than the bottom left row, which seems to be strongly illuminated. The image on the right hand side (the ‘anti-snake’) shows the basic simultaneous contrast effect for comparison. From Kingdom (2011).

    Thus we cannot assume at the outset that apparent colours are stable with LOS changes. It is fair to say that sometimes they are and sometimes they are not. So the kinds of layering scenario that Brown argues cause trouble for the Explanatory Import of relationism, where stability holds, may be the minority of cases; or they may be the majority. People are free to squabble over this though I doubt it would be a very fruitful way to proceed. I believe some more interesting, albeit more speculative, points can be made in favour relationism. I will sketch them out in the remainder of my comments.

     

    3 The Colour Brown and Context Dependency of Transparency

    The White Effect (Figure 2) shows us that the perception of the same stimulus as transparent or opaque is subject to context effects. I would now like to point out that whether we see anything as apparent or opaque depends on their being the right kind of context. For a surface to appear opaque, we need a sense of there being something behind it which it occludes. For a filter to appear transparent, we need to see through it to a background of other things.

    For this reason, appearances of opacity and transparency are like the surround (contrast) colours, pure black, pure white and brown, which Brown (2015, 6) notes can only be experience when a suitable context is in place. Brown agrees that these surround colours are part of the reasonable evidential base for colour relationism. So should the context dependency of experiences of opacity and transparency tempt us towards a relational ontology of these qualities?

    Well, there is a quick non-relationist (and physicalist) response here, which is to say that transparency just is the physical property of being able to transmit light, and opacity the disposition only to reflect light. The background contexts are just the circumstances needed for us to detect these non-relational, physical properties. It is tempting to think that there is a better correspondence between physical light transmission and perceived transparency than there is between spectral surface reflectance and colour, making this physicalist proposal more compelling for transparency than for colour. But we should also note that the relationship between the perceptual dimensions of transparency and their physical basis is still a matter of scientific controversy (Anderson et al., 2006) (Albert, 2008). It seems that perceived transparency does not just reduce to physical light transmittance. This should not come as a surprise because, as White’s Effect indicates, perception of contrast, colour, and the interpretation of what is there in a scene (e.g. solid rectangle behind bars vs. a rectangular shaped patch of light), all impact on the perception of transparency and opacity.

     

    4 Transparent White?

    One reason to think that these effects of contrast, colour, etc. are not merely side-effects, superfluous to the main business of transparency perception (as the physicalist understands it), is the observation that transparency does not come in all colours. In his Remarks on Colour Wittgenstein (1977) puzzles over such questions as, ‘why can’t there be a transparent white?’ and ‘can a transparent piece of glass have the same colour as an opaque piece of paper?’ He looks for answers in the grammar of colour terms whereas I think it is more profitable to consider if the answers lie in the fact that colour is more than hue, saturation and lightness (HSL).

    It has been commonplace to think of each colour as defined by the three HSL parameters but the lesson from more detailed work on colour appearance modelling is that at least two additional parameters are needed, characterising the illumination on a coloured surface (Fairchild, 2013). Of greater importance here is the relationship between colour appearance and the perception of other qualities associated with the way that the material surface interacts with light, like whether a surface is shiny or matt. Yellow and gold are not the same colour. But the difference is not in the hue, saturation or lightness, but in the lustrousness of gold surfaces.2

    Now back to the impossibility of transparent white. A radical proposal—which I float for the purpose of discussion—is to treat opacity and transparency as forming another dimension of colour. The hue white has a very restricted range of possible brightnesses (only the highest) and a somewhat restricted range of transparency (never completely transparent, but sometimes translucent and normally opaque). The answer to the question, ‘can a transparent piece of glass have the same colour as an opaque piece of paper?’ must then be ‘no’ because the colours of two items which differ in transparency/opacity will occupy different points in our multi-dimensional colour space, even if their hues are the same. If this is right, then we can see an opaque colour through a transparent colour, but—given the context dependency of perceived transparency and opacity—each of those complex chromatic properties will be contextually dependent on one another.

     

    5 Preliminary Conclusion

    To conclude I will draw together a more moderate line of thought. The perception of object and illuminant colour, material qualities (e.g. lustrousness, mattness, opacity and transparency), and the parsing of objects and shadows in a scene, are complex and interconnected processes. If we have reason for thinking that colour is a relational property, then with further investigation we will probably uncover grounds for taking the perceived material qualities to be relational as well, because these show the same kinds of context dependencies that are often observed for colour. They do not reduce to their putative physical correspondents.

    So what then of colour layering? One thing to say is that layering is perfectly consistent with the relationist treatment of object colour, illuminant colour and other material qualities. This in fact would be a version of the “intrinsic images” model of scene perception, itself a descendent of the Gestalt tradition.3

    The visual system is conceived as parsing the world into“layers”ofmaterial variation (opaque surfaces and filters) and illuminant features (e.g. shadows). The perceived hue or lightness of a material surface is not fixed until a guess is made about the illuminant hue and intensity, and vice versa. And context matters: perception of a scene is not built up piecemeal from the recovery of local features but instead vague guesses about larger features (e.g. the extent of a shadow or transparency) influence perception of more localised features. In short, the relationalist treatment of LOS relativities does not entail that variation in illuminant chromaticity is perceptually “fused” to surface colour variation.4

    A proper development of the relationist version of colour layering would go beyond the scope of these brief comments so I apologise for ending here with so many loose threads at hand. I thank the author for providing much material for reflection and look forward to future discussion.

     

    References

    Adelson, E. (2000). Lightness perception and lightness illusions. In M. Gazzaniga (Ed.), The New Cognitive Neurosciences (2nd ed.)., pp. 339–351. Cambridge, MA: MIT Press.

    Albert, M. K. (2008). The role of contrast in the perception of achromatic transparency. comment on singh and anderson (2002) and anderson (2003). Psychological Review 115, 1127–1143.

    Anderson, B. L., M. Singh, and J. Meng (2006). The perceived transmittance of inhomogeneous surfaces and media. Vision Research 46, 1982–1995.

    Brown, D. (2014). Colour layering and colour constancy. Philosophers’ Imprint 14(15), 1–31. Brown, D. (2015). Colour layering and colour relationalism. final version published in Minds and

    Machines .

    Cohen, J. (2009). The Red and the Real. Oxford: Oxford University Press.

    Fairchild, M. D. (2013). Color Appearance Models (3rd ed.). New York: Wiley.

    Kingdom, F. (2011). Lightness, brightness and transparency: A quarter century of new ideas, captivating demonstrations and unrelenting controversy. Vision Research 51, 652–673.

    Pastoureau, M. (2009). Black: the history of a color. Princeton, NJ: Princeton University Press. Wittgenstein, L. (1977). Remarks on Colour. Oxford: Blackwell.

     


    Notes

    1. I.e. they are both present in the same scene.

    2. We tend to think of shiny black and dull black as being ‘the same colour’ but an interesting historical fact is that European languages until recently (i.e. the time of Shakespeare) used two colour words here: in English, swart for dull black and black for the lustrous shade (Pastoureau, 2009).

    3. See Adelson (2000, 342) and Kingdom (2011, 661-663) for discussion.

    4. See Brown (2015, 2). Brown intends the colour layering thesis to be neutral regarding colour ontology. So it should be no surprise that it is possible to develop it along relationist lines. That said, I accept Brown’s point that Cohen’s version of relationism does invite the analysis of LOS relativities as instances of fusion rather than layering.

  2. CHROMATIC LAYERING AND COLOR RELATIONALISM

     
    Herzog was all too well aware of the layers upon layers of reality — loathsomeness, arrogance, deceit, and then — God help us all! — truth, as well.


    — Saul Bellow, Herzog.

     

    Brown (2015) highlights cases of “chromatic layering” — scenarios in which one perceives an opaque object through a transparent volume/film/filter with a chromatic or achromatic content of its own — as a way of reining in the argument from perceptual variation I’ve used to motivate a relationalist account of color properties (Cohen 2004, 2009). That argument (about which more anon) generalizes widely across a very wide range of circumstantial and perceptual parameters which, when they take on different values, lead to a difference in the experienced color of a fixed stimulus. Such parameters include, e.g., viewing distance, viewing angle, retinal photoreceptor type ratios, chromatic surround, and so on. The intended conclusion of the argument is that color properties are best understood in terms of relations to the relevant parameters. But Brown urges that the argument in question does not generalize smoothly to all types of perceptual variation — in particular, that it fits poorly in layering cases in which there is either experiential fusion or scission. While he doesn’t want to reject the relationalist description of variation in all cases, he does suggest that there is an alternative, and preferable, description available for layering cases, and that the availability of this alternative puts limits on the scope of both relationalist arguments and relationalist conclusions.

    In what follows I want to suggest that the obstacles for the argument from perceptual variation (hence for relationalism) posed by cases of layering are merely apparent. I argue that relationalism can be extended smoothly to cases of layering, and indeed that denying this extension, as Brown proposes, contravenes ordinary and more or less universally accepted canons of rational inquiry. I’ll begin by reviewing the argument from perceptual variation, and showing how it motivates color relationalism (§1). Next, I’ll present Brown’s layering cases, and explain how he thinks they pose challenges to the argument from perceptual variation (§2). Then I’ll attempt to defuse these challenges: I’ll argue that there are no grounds for regarding either fusion or scission as problems for the argument from perceptual variation or color relationalism (§§3–4). Finally, I’ll reply to Brown’s methodological objections; I’ll show that, contrary to his accusations, the relationalist treatment of layering rests on (and Brown’s preferred analysis depends on flouting) a very weak and well-motivated conception of uniformity that we all have reason to accept (§5). I’ll conclude that the challenge from layering is a challenge relationalists can meet.

    1 Color relationalism and the argument from perceptual variation

    In earlier work (esp. Cohen 2004, 2009) I have argued for color relationalism — the view that objects’ colors are constituted in terms of relations those objects bear to perceiving subjects and perceptual circumstances. The thought is that, just as things count as instances of being a teacher or being humorous not by virtue of their intrinsic makeups, but by standing in the right (viz., pedagogical/amusing) sort of relation to relata (viz., students/appropriately equipped listeners), so, too, things count as exemplifying colors by standing in the right relation to perceivers and perceptual circumstances.

    My case for relationalism revolves around a traditional, non-deductive, empirical argument based on perceptual variation.1 Roughly this non-deductive argument begins with the empirical claim that there is significant inter- and intra-personal variation in representational responses to a given color stimulus, and the thought that, on standard assumptions, each variant represents the color of the stimulus. The next step is a symmetry claim — viz., that in such cases, it is hard to imagine what, other than ad hoc stipulation, could make it the case that just one of the variants is uniquely veridical (i.e., veridical at the expense of the other variants). The argument then appeals to the principle that we should avoid ad hoc stipulation where possible, and takes this as reason for preferring an ecumenical view on which the ostensibly competing variant representations of the stimulus’s color can be simultaneously veridical. Finally, the argument involves an abductive inference to the relationalist view that colors are constituted in terms of relations to perceivers and viewing conditions on the ground that such a color relationalism gives us the best way of implementing the desired ecumenicism.

    Though the argument form just given generalizes widely, it is perhaps easiest to appreciate its force by considering its application to a single, simple instance of perceptual variation, such as that involving simultaneous contrast in a traditional center-surround configuration such as figure 1.

    Figure 1: A traditional center-surround configuration. The two central patches are intrinsically qualitatively identical, but the patch set against a less yellow/more blue surround on the left appears more yellow/less blue than the patch set against a less blue/more yellow surround on the right.
    Figure 1: A traditional center-surround configuration. The two central patches are intrinsically qualitatively identical, but the patch set against a less yellow/more blue surround on the left appears more yellow/less blue than the patch set against a less blue/more yellow surround on the right.

    Though the two central patches in this configuration are qualitatively identical in their non-relational properties, they look different (in color) depending on the surround against which they are placed. Assuming, standardly, that the way the patches look represents their colors, it follows that the visual system represents the colors of the patches differently as a function of the surround. If so, we can ask: which, if any of the (psychophysically distinguishable) representations of the patches is veridical?2 The logically possible answers are: neither, the first to the exclusion of the second, the second to the exclusion of the first, or both. I suggest that the first option (neither right) is unacceptably skeptical/revisionary, and that the second and third answers (one representation exclusively) are unacceptably arbitrary — that it is hard to imagine a well-motivated, principled, and non-question begging reason to believe that either representation is uniquely veridical. Assuming we want to avoid both revisionary skepticism and ad hoc stipulation when possible, this suggests that, ceteris paribus, we should prefer the view that both representations are veridical. My further suggestion is that we should endorse color relationalism because it gives the best way of understanding how this answer could be true. For relationalism allows us to say that the patch has two compatible colors: it is (simultaneously, all over) light yellowish green with respect to the first perceptual circumstance, and a slightly darker bluish green with respect to the second.3 Color relationalism is attractive because it allows us to avoid a hard choice between unpalatable alternatives.

    2 Layering: scission and fusion

    Brown is interested in the challenges posed to the argument from perceptual variation (and therefore to color relationalism) by instances of chromatic layering. He centers his case against the argument on two types of scenarios: cases of “scission,” in which visual experience separates out multiple objects located along the same line of sight and assigns colors to each of them, and cases of “fusion,” in which experience fails to distinguish between multiple objects along the line of sight, and consequently fails to attribute separate colors to each of them. I’ll follow Brown in discussing these in turn.

    2.1 Scission: Variation without relationalism?

    We can begin with the scission case (what he calls “the Layering Scenario”) represented in figure 2, where the perceiver perceives an opaque object O first through a transparent filter F1 (figure 2a) and then through a second transparent filter F2. (figure 2b).

    Figure 2: (a) Perceiver views an opaque object O through a transparent filter F1, and then (b) through a different transparent filter F2.
    Figure 2: (a) Perceiver views an opaque object O through a transparent filter F1, and then (b) through a different transparent filter F2.

    Now, this sort of case may initially appear to be grist for the relationalist’s mill — yet another instance of perceptual variation to which we can apply the argument from variation, as outlined above. After all, just as in the case of the center surround configuration depicted in figure 1, and used as a paradigm for explaining the argument from perceptual variation, here we have a case in which the subject undergoes psychophysically distinguishable experiences of the color of the object O, each of which purports to represent O’s color, each of which can be assessed for veridicality, and between which there is no obvious way to make a principled and exclusive choice.

    However, Brown urges that we should not assimilate the layering scenario to standard cases of perceptual variation in this way. He accepts that, in the layering scenario, there is a difference in the total color experience involving one target object (“Owhite”) and a numerically distinct transparent object (“Tgrey”) along the line of sight between the observer and the target object.4 But, Brown advises, we shouldn’t take this to indicate that the cross-contextually varying experience of the subject is varying in what color it attributes to the single target object, as needed by the argument from perceptual variation:

    Clearly Owhite is causally involved in bringing about the experience . . . Owhite – through-Tgrey, and in that sense Owhite has a role in generating these distinct colour experiences in the [Layering] Scenario. But it would be erro- neous to hold that Owhite itself generates these distinct colour experiences, or to hold that these distinct experiences are simply experiences of the colour of Owhite . . . . In order for [the argument from perceptual variation] to create a rationale for Relationalism, the variable colour experiences drawn from the relativity of colour must be experiences of the target object’s colour (i.e., “of a single color stimulus”). But in our Layering Scenario the difference between the experiences Owhite-through-Tgrey and Owhite-through-Tyellow is in a fundamental sense not a difference in the experienced colour of Owhite, it is a difference in the experienced colour of something else along the LOS [=line of sight] on which Owhite is currently located. Thus while, as it were, the total colour experience along this LOS changes, and Owhite is the terminal visible object along that LOS, the experienced colour of Owhite is stable, and there is no motive to infer that Owhite’s colour is relational (11–12).

    According to Brown, then, since the successive and psychophysically distinguish- able experiences in question (call them e1 and e2) do not attribute their chromatic content entirely to the very same individual, O, they should not be construed (as in standard instances of perceptual variation) as putatively rival representations of O’s color. True, e1 and e2 differ in their overall chromatic content; but that is because of the differential contributions made to the overall chromatic contents of those experiences by objects other than O — namely, F1 and F2. As far as O itself, e1 and e2 agree in the color they attribute to O (“the experienced colour of Owhite is stable”). As such, there is no reason for thinking we face any hard choice of which of the two veridically represents O’s color. They agree, so we can unproblematically regard both as veridical without any need for treating the properties they attribute as relativized or relational.

    In short, Brown’s contention is that the best description of the psychophysical variation in layering scenarios obviates the hard choices to which relationalism was proposed as an escape, and thereby leaves relationalism unmotivated (at least in these cases).

    2.2 A retreat to fusion?

    Given the problems just sketched, Brown imagines that relationalists might prefer to run the argument from variation only on cases of fusion. But he objects that in that case the argument from perceptual variation will fail for another reason.

    After all, the argument from variation crucially turns on the premise that there is symmetry between perceptual variants — that there is no principled, non-arbitrary, non-question-begging reason for treating just one of the variants as uniquely veridical. But Brown worries that this symmetry premise is plainly false in fusion cases. For, he urges, fusion is confusion. And, as such, the standard epistemic value of avoiding informational conflation counts as a principled, non-arbitrary, non-question-begging reason to prefer over fused variants either (i) a layering/non-fused experience that separates out the contribution of O from other objects on the line of sight, or else (ii) variants in which there is no other object making a contribution to our chromatic experience:

    If one’s aim is to discern the colour of the book [O], then a colour experience in which the book’s contribution to colour is distinctly experienced should be preferred over one in which the book’s [O’s] and cellophane’s [F1’s] contributions are fused together. Fusion experiences yield information, not about the book’s colour simpliciter, but about a relation between the book’s colour and the cellophane’s colour. This relational information is valuable, but if one is seeking information about the book’s colour, the information is arguably inferior to information in which the book’s contribution to colour experience is distinctly packaged, or separated out from the contributions of other objects….If one is seeking to learn of the book’s colour, it is reasonable to privilege the former over the latter. Privilege should also be granted to experiences in which only the book is contributing to the colour experience along a [line of sight] (12–13).

    The upshot, then, is that the argument from perceptual variation fails for fusion cases, so that restricting its application to such cases won’t save the argument, hence won’t allow it to do its work in motivating color relationalism.

    3 Fusion revisited

    Before I respond to Brown’s (as I read him, more central) criticisms involving scission layering scenarios in §4, I want to address more briefly his considerations about fusion cases.

    As we have seen, Brown thinks the argument from perceptual variation breaks down in such cases because ordinary epistemic norms break the symmetry between perceptual variants needed by the argument, and breaks it in favor of variants that avoid informational conflation. But there are two reasons for thinking that this point is ineffective against the argument from perceptual variation.

    First, though the objection turns on thinking of fusion as a kind of informational conflation that is in conflict with epistemic norms, that characterization is unsupported and close to question-begging. It is is true that the fusion experience yields information about O in relation with F1, rather than about O simpliciter. However, there’s no reason to assume that that the former information is less epistemically valuable than the latter — not even when one is “seeking information about the book’s colour.” That conclusion only follows if there is some sort of epistemic preference for experientially representing objects via their non-relational features. But why accept that? In the envisaged case, O and F1 do have a joint causal effect on the perceiver, and the fusion experience under consideration (veridically) represents O in its relation to F1. That content is information about the book and its color. To insist that the content of a fusion experience is epistemically inferior to that of a non-fused alternative experience is just stipulative, and certainly won’t be accepted by Brown’s relationalist opponent.

    Second, we should recall just what sort of symmetry is at issue. The targeted symmetry premise of the argument from perceptual variation is the claim that there is no well-motivated, principled, and non-question begging reason for singling out one representation as uniquely veridical. But even if (contrary to what I just said) experience did come with an epistemic preference for non-relational feature representations, that wouldn’t by itself break the symmetry on which the argument from perceptual variation is grounded. It wouldn’t be a reason for believing one of the variants is veridical to the exclusion of others. After all, even Brown’s own explicit description of the case has it that the fused color experience of O-cum-F1 is no less veridical than a non-fused color experience of O. Thus, even granting Brown’s assumptions about the epistemic norms governing experience does nothing to undercut the symmetry premise on which the argument from perceptual variation depends.

    In short, I don’t see that what Brown has said about fusion cases is in any way damaging to the relationalist’s attempt to motivate her view by appeal to the argument from perceptual variation.

    4 Layering for relationalists

    But I take it that it is the case of scission, rather than fusion, that lies at the heart of Brown’s criticisms of the argument from perceptual variation and color relationalism.

    I now want to argue that cases of scission (henceforth, cases of layering) are similarly ineffective in limiting the scope or force of the case for color relationalism.

    The first point to make is that (as Brown is well aware) relationalism does make room for a description of layering cases. We have already seen that, for relationalists, object colors are relational properties containing a parameter for perceptual circumstances, which they think must be specified in terms of whatever features of the circumstance make a psychophysically identifiable difference to experience (but where the variants have an equal claim to veridicality). Viewing an opaque object O through a transparent filter, as in the cases depicted in figure 2, certainly appears to make a psychophysically identifiable difference of this sort to our experience of O. So it is entirely consistent with relationalist methodology to describe cases of layering by the same means. A relationalist will say that the relational color O manifests when viewed through filter F1 is, as it might be, grey to S in perceptual circumstance C1, where C1 involves (among other possibly relevant parameters) the presence of F1 along the line of sight. And she will claim that the color O manifests when viewed through filter F2 is, as it might be, yellow to S in perceptual circumstance C2, where C2 involves (among other possibly relevant parameters) the presence of F2 along the line of sight. As usual, this maneuver will allow the relationalist to hold that both of the experiences veridically represent a (relational) color that the cup has (all over, simultaneously).

    As I say, Brown is aware that the relationalist has this story to tell about layering cases; I take it this is why he denies explicitly (8) that they are straightforward counterexamples to relationalism. Rather, his complaint is that the available relationalist account is less than satisfactory, hence that it would be “premature” (8) to accept it. His stated reason for dissatisfaction with the relationalist’s alternative description of layering cases is that it wrongly assigns chromatic content to just the one perceptual object O (albeit O in a circumstance involving mediation by the distinct object F1/F2) rather than to the two distinct objects O and F1/F2, as he prefers. But, of course, this observation, by itself, merely pushes back the question. Now our question is: why should we prefer a description that attributes colors to two objects (as Brown prefers) over a description that attributes color to one object in a circumstance that involves mediation by a second object (as the relationalist prefers)?

    Though Brown is less fully explicit on this point than one might have hoped, I find in his paper a few possible, and somewhat overlapping, answers.

    4.1 Obviously wrong?

    At some points he suggests that it is just obvious his own description of the case is the uniquely correct one. Thus, consider this passage in which he considers and rejects the relationalist redescription:

    . . . aren’t you just experiencing the opaque thing to be differently coloured across these contexts? This is not a friendly amendment, but instead an attempt to eliminate experienced colour layering. There is no such simplification: experienced colour layering can only be adequately described via a complex such as Ox-through-Ty (10).

    But this is unpersuasive: Brown’s insistence that the phenomenon “can only be adequately described” in the way he favors is question-begging in the context of our search for a reason to prefer one of the two descriptions on offer.

    4.2 Collapses onto fusion?

    A second answer, possibly offered in connection with the first (perhaps this is why he characterizes the relationalist description as “an attempt to eliminate experienced colour layering”) involves the worry that the relationalist description wrongly reduces all instances of scission to cases of fusion. But this charge is inaccurate. It’s true that the relationalist description of a layering experience involves the attribution of a color to one object O (better: to one object in a circumstance involving the mediation of a second object, such as F1) rather than two. But there’s no reason the relationalist can’t also recognize that, in scission cases, experience attributes a different (relational) color to the filter F1. And this can provide the basis for a principled relationalist distinction between fission and fusion cases.

    To see the point, notice that a relationalist has no reason to reject the idea that visual experience at a time can attribute colors to more than one object in a single stimulus configuration. After all, that’s exactly what she will say about a center- surround configuration like the one depicted in figure 1.5 She’ll hold that visual experience attributes two relational colors. First, she’ll say, it attributes to the central patch a relational color involving the presence of the surround as part of the perceptual circumstance; second, she’ll say, it additionally attributes to the surround a relational color involving the presence of the central patch as part of the perceptual circumstance. Moreover, this approach can be carried over straightforwardly to cases of scission such as that depicted in figure 2a. First, she’ll say, experience attributes to O a relational color involving the presence of F1 as part of the perceptual circumstance; second, she’ll say, it attributes to F1 a relational color involving the presence of O as part of the perceptual circumstance.

    But now we are in a position to see that this consistent extension of ideas already available to relationalists (and, indeed, already exploited in their account of center- surround variation) provides a simple way of distinguishing cases of scission from cases of fusion. She will say that fusion cases are those in which experience attributes just one relational color to an object along the line of sight (here O), while in fission cases experience attributes relational colors to multiple objects along the line of sight (here O and F1).6

    4.3 Neglects constancy?

    A third possible reason for favoring Brown’s preferred description of layering is connected with his reasonable suggestion (15–17) that there is a kind of color constancy associated with our experiences of O through different layering filters such as F1 and F2. As he says, there is plausibly some respect of chromatic sameness between two such experiences of O, and we should want a conception of the cases that makes room for this fact. One might fear that the relationalist descriptions of the two layering cases (e1 represents O as grey to S in a condition involving the mediation of F1; e2 represents O as yellow to S in a condition involving the mediation of F2) unacceptably makes no room for the (as yet unanalyzed) chromatic sameness between the cases.

    Again, this complaint is unpersuasive; for, as Brown points out, there are indeed relationalist resources for accounting for the desired respect of chromatic sameness. Thus, I have proposed in other work (Cohen 2008) that the visual system achieves color constancy by effectively representing, over and above the occurrent colors of perceived items A and B, counterfactual verdicts about whether A and B would match if perceived under a shared perceptual circumstance that only one of them actually occupies. On this proposal, the chromatic sameness the visual system represents between e1 and e2 is the (true) content that there would be a match between two chromatic experiences generated by the perception of O under the same perceptual circumstance — either two experiences of O in a circumstance involving mediation by F1 but not F2, or two experiences of O in a circumstance involving mediation by F2 but not F1.7

    Brown objects that, while this account may be acceptable in at least some cases, it may be false of others, and that the best methodology requires following the data where they lead:

    . . . one’s goal should not be to try to force all instances of colour constancy into one model (be it a layering, counterfactualist or some other model) to accommodate one’s broader objectives . . . . Our goal should be to examine various instances of colour constancy with various models in mind to try to uncover which one or more of them best explain individual or classes of cases. The potential result is that colour constancy is a varied phenomenon, having some instances of layered constancy, others of counterfactual constancy, and so on (16–17).

    I confess that if there’s an objection here, I don’t see what it is. Of course it’s true that theories should be independently motivated and constrained by the data. In the present case, however, Brown has already accepted that relationalism is independently motivated and correct about many cases; moreover, it appears to provide a description of color constancy that is entirely consistent with the data on hand. Unless and until someone identifies some difficulty (empirical or otherwise) with that description , all we can ask from the relationalist (or any other theorist) is that she provide consistent, well-motivated, and empirically adequate accounts of relevant phenomena. So far as I can tell, the relationalist is as far along in meeting this demand as anyone else.

    4.4 Forced?

    Brown’s final complaint against the relationalist description of layering is that (as he complained about the relationalist description of color constancy just reviewed), he finds it forced. Thus, he urges that “one’s goal should not be to try to force all instances . . . into one model . . . to accommodate one’s broader objectives ” (16–17), and that “succumbing” to a uniform treatment that applies to both center-surround and layering cases risks making us “blind to surprising possibilities” (18).

    Once again, I find this objection puzzling. I agree that the relationalist descriptions at issue are motivated by the aim of making relationalism consistent with the data, and that proponents of other views may prefer their own descriptions. But the mere fact that these descriptions are enlisted in the defense of a theoretical commitment — even a prior theoretical commitment — is no strike against them. What would count as a strike against such a description would be a demonstration that it is empirically inadequate or otherwise objectionable; but no such demonstration has been provided.

    5 Uniformity

    As we have seen, Brown disagrees with relationalists in finding relevant differences between traditional center-surround configurations and layering cases. He accepts that there exist ways of construing both kinds of cases in the very same relationalist-friendly terms; but he thinks that, while such descriptions are appropriate for center-surround cases, they are in various respects unsatisfying when applied to cases of layering. What these considerations suggest to him, then, is that theorizing about the range of cases is best carried out in a case-by-case fashion, rather than by insisting on a uniform treatment:

    Here is the general lesson: there are some and potentially numerous instances of experienced colour layering; determining their true extent must be done not by asserting the Uniformity Hypothesis or any such theoretical principle, but by careful examination of relevant cases; due to various confounding factors that examination will not be straightforward. In other words, we cannot assert the Uniformity Hypothesis (17).

    Thus stated, it’s hard to disagree with Brown’s reasonable and attractively modest sentiment: open-mindedness is better than closed-mindedness, and uniformity is at best a defeasible theoretical virtue that can be applied only after ruling out potentially confounding differences. If the “careful examination of relevant cases” that he recommends did indeed reveal such differences, then it would be absurd to insist on treating those cases uniformly. It’s just that, as I have argued (§4), the alleged symmetry- breaking disanalogies that Brown offers are illusory. And in this situation, where prior careful examination of relevant cases fails to uncover any reason for treating the cases in different terms, it is appropriate to invoke uniformity to choose between descriptive alternatives.

    I regard uniformity as a theoretical virtue, not an explicit premise, and surely nothing as strong as Brown’s Uniformity Hypothesis (“…that all or most instances of the relativity of colour should be treated as above, namely that each case can be described in a parallel fashion” (7)). I agree with him that, like other theoretical constraints, we should invoke uniformity only after careful consideration of the empirical facts on the ground, and only ceteris paribus. But that’s just where we now are. We have considered the cases carefully and found that cetera are indeed pariba —that the available broadly empirical, theory-independent, well-motivated constraints don’t choose between the descriptive alternatives.

    On the contrary, what we know about the cases leaves open the very same theoretical descriptive options with respect to cases of scission and traditional center- surround cases (and, for that matter, many other cases involving variation in the chromatic effects of an object as it moves through different circumstances). On the one hand, we can characterize such cases by saying that experience attributes a net chromatic content to just the one perceptual object O (albeit O in a circumstance involving the influence of a distinct object F1 in the perceptual circumstance). Or, on the other hand, we can say that experience parses out its net chromatic content over two distinct objects O and F1.

    There appear, then, to be three options left open after careful examination of relevant cases, i.e., after the recognition that broadly empirical constraints fail to choose between the descriptive alternatives. First, one could accept the two-object description uniformly for both center-surround and layering cases. Second, one could throw uniformity to the wind, and accept different descriptions of each, as Brown suggests.8 Or, third, one could adopt a consistent one-object description of both cases, as I prefer.

    Since Brown and I both reject the first option (we both accept the one-object description for center-surround cases), I will put it aside.9 This leaves the second and third choices on the table.

    It’s worth being clear that either one of these choices is a choice: electing either is making, rather than avoiding, a decision about how to theorize appropriately. But given that we have to make a choice, and that broadly empirical constraints leave open these two options, it’s hard to see why we should not at that point invoke the theoretical virtue of uniformity as a reason for preferring the third choice (the consistent-one object description) over the second (different descriptions of the cases). After all, given that we’re in a situation in which there’s no compelling, theory-independent reason for treating the (otherwise structurally similar) cases in dissimilar ways, differential treatment of the cases would be arbitrary. It would be unconstrained by reasons in a way that, I take it, we don’t want our choices in theory-building to be.10

    My suggestion, then, is that uniformity considerations do give reason for choosing between alternatives left open by other broadly empirical constraints, and that, when applied to the cases under discussion, they tilt the balance in the direction of the consistent one-object description. Perhaps needless to say, this invocation of a ceteris paribus preference for uniformity allows that inquiry might one day reveal salient differences that make Brown’s non-uniform option preferable. But until such relevant disanalogies are revealed, it’s hard to see why we should prefer to resolve the standoff in a way that needlessly and positively flouts the standard norm of uniformity (hence, therefore, normal canons of intellectual inquiry).11

     

    References

    Brown, Derek (2015), “Colour layering and colour relationalism,” Minds and Machines, Also appears for Minds Online Conference.

    Cohen, Jonathan (2004), “Color Properties and Color Ascriptions: A Relation- alist Manifesto,” The Philosophical Review, 113, 4, pp. 451–506.

    — (2008), “Color Constancy as Counterfactual,” Australasian Journal of Philosophy, 86, 1, pp. 61–92.

    — (2009), The Red and The Real: An Essay on Color Ontology, Oxford University Press, Oxford.

    Friedman, Michael (1974), “Explanation and Scientific Understanding,” The Journal of Philosophy, 71, 1, pp. 5–19.

    Kitcher, Philip (1989), “Explanatory Unification and the Causal Structure of the World,” in Scientific Explanation, ed. by Philip Kitcher and Wesley Salmon, 8, Minneapolis: University of Minnesota Press, pp. 410–505.

    Tye, Michael (2012), “Cohen on Color Relationalism,” Analytic Philosophy, 53, 3, pp. 297–305.

     


    Notes

    1. Since Brown (2015) discusses this argument, I give only a compressed characterization here. For a fuller presentation and much more discussion and defense, see Cohen (2009).
    2. The question here is metaphysical, not epistemic: it is not ‘how do we know which of the perceptual effects veridically represents the patch’s color?’, but ‘what makes it the case that one of the perceptual effects (as opposed to others) veridically represents the patch’s color?’.
    3. Analogy: a single individual can be a teacher of you, and, simultaneously, not a teacher of me.

    4. Brown’s labels for objects (‘Owhite’ for the opaque object terminating the line of sight, ‘Tgrey’/‘Tyellow’ for transparent filters between the perceiver and target object), which he uses to pick out these objects across variations in perceptual circumstances, encourage the assumption that these objects have stable colors that are independent of (and persist across variations in) perceptual circumstances. Since that assumption is crucially under dispute, I have chosen more neutral labels for the objects (‘O’, ‘F1’, ‘F2’) in my own description of the case, but will leave Brown’s nomenclature as it stands when quoting him.

    5. Note that Brown accepts (6–7, 18) the relationalist description of ordinary center-surround perceptual variation.

    6. Note that this description of fusion has a parallel in traditional center-surround cases as well — in such cases, typically labeled ‘color induction,’ the surround influences the chromatic experience of the central patch but is not assigned a chromatic content of its own.

    Indeed, in view of the far-ranging structural parallels between layering and center-surround cases, it’s tempting to think of the former as a special type of center-surround configuration — viz., one where the surround happens to be located at a different depth from the central patch (relative to the perceiver). In principle, the choice between Brown’s favored two-object description and the relationalist’s favored one-object-in-a-circumstance-involving-a-second-object description should be available for any of these cases (cf. §5). In this respect, it seems to me that Brown’s challenge is not ultimately about the phenomenon of layering, per se, but about the much more general questions of whether, how, and in what conditions color experience should be described as distributing the chromatic contents of single intentional objects over multiple entities in the environment. It will come as no surprise that I think color relationalism offers a productive and general set of answers to these questions.

    1. Some (e.g. Tye 2012) have objected that this “counterfactualist” account of color constancy is unacceptably revisionary in only recognizing a non-occurrent respect of similarity. But since the psychophysical data about color constancy (typically, subjects’ matching judgments) are entirely agnostic on this point, surely the onus is on Tye and others who would reject the counterfactualist view to show that the needed respect of similarity must be an occurrent rather than non-occurrent respect. As matters stand, this burden remains unmet.

    2. It’s worth adding that, whatever its other virtues and vices, the non-uniform strategy Brown is offering does not avoid commitment to relational colors. Brown is prepared to admit that colors are constituted in terms of relations to at least those parameters (such as, in his view, chromatic surround) with respect to which the one-object description of variation is appropriate. So he’s already willing to that colors are relational with respect to at least some parameters. But if so, then the colors attributed by experience even in scission cases, where Brown is urging us not to adopt a relational description, will in any case be relational with respect to other parameters. For this reason, the non-uniform strategy on offer will be unattractive to those who (unlike Brown) had hoped to avoid a relationalist conception of color.

    3. I can’t help making the further remark that this uniform two-object description appears much less plausible when applied to cases of interpersonal perceptual variation — viz., cases where one stimulus O in one perceptual circumstance C varies in perceptual effects on different perceivers. In such cases, it is difficult to describe the chromatic effects as varying as a function of differences in two perceived objects: it is in the nature of such cases that the variation occurs without a difference in any perceived objects. What varies, of course, is the perceiver/visual system. But it is far less plausible to construe experience as ascribing part of its chromatic content to the visual system itself.

    4. I take it this point lies at the heart of the attraction of so-called unificationist theories of scientific explanation (see, e.g., Friedman 1974; Kitcher 1989). One doesn’t have to be a full-fledged explanatory unificationist to see value in theoretical uniformity (when other things are equal), as I’m suggesting.

    5. Thanks to Matthew Fulkerson for discussion of these matters.

  3. Colour layering and colour relationalism – Reactions to Chirimuuta and Cohen

     

    Many thanks to Mazviita and Jonathan for graciously taking the time to read and engage my article, and for putting forth so many interesting ideas in response to it. My aim in what follows is equally to address their challenges (in some cases begin to) and to highlight points/questions that might be valuable for discussion in this forum.

     

    Reaction to Chirimuuta

    Chirimuuta brings to the fore a number of important empirical results/issues that I did not delve into in my paper. I thus welcome the opportunity to briefly remark on them here. At the outset she identifies two central items for discussion:

    1. Non-modularity of colour & transparency/opacity. Chirimuuta claims that I treat “colour vision as a process independent of the perception of opacity and transparency” (1). She believe this is a mistake.
    2. Relational nature of experiences of transparency/opacity. “Brown does not consider whether perceptions of opacity and transparency are themselves subject to context effects” (ibid). And according to her “[i]t is easy to show that they are” via White’s Effect and related phenomena (ibid).

    The rough idea behind (A) is that our vision system processes colour in part by utilizing data from transparency/opacity processing and vice versa – these two kinds of processing influence and assist one another. The rough idea behind (B) is that the kinds of contextual effects colour relationalists use to bolster their view (e.g., center-surround colour effects) are also found in perceptions of transparency/opacity.

    I have no qualms with either of these claims. Chirimuuta is right that I didn’t discuss them explicitly: (A) was not alluded to; and, while empirical work pertaining to (B) was referenced (see esp. p. 2 n. 3), the matter was not brought into the piece. The more important issue is the significance of (A) and (B) for my argument and for colour theory more generally. I begin with (B).

    §1 Contextual nature of experiences of transparency/opacity.

    Set aside colour for a moment and suppose (B) is true, as White’s Effect makes seemingly undeniable: our experience of the transparency/opacity of a stimulus S hinge on contextual effects such as what kinds of stimuli surround S. Does this undermine my argument? I don’t think so. My argument is about how to interpret the import of colour layering/non-layering experiences for colour ontology and in particular colour relationalism. So far as I can see the contextual nature of transparency/opacity experiences on their own do not speak to this. If, for example, this contextuality could be used to bolster the inference from the contextual interdependency of layered colours on one another to their constitutive interdependency on one another (see also §3 below), then I should look closely at how my argument might be impacted. However, at present I don’t see how to do that, and in fact there is an important disanology between the transparency/opacity case and the colour case (remarked on briefly by Chirimuuta, p. 5 top).

    The contextual nature of transparency/opacity doesn’t tempt many of us to adopt a relationalism about the ontology of transparency/opacity, even if to some it suggests a relationalism about experienced transparency/opacity. Indeed, an anti-relationalist about colour might argue, by analogy, that because the temptation isn’t there, center-surround and similar colour contrast effects shouldn’t tempt us to adopt a relationalism about colour ontology, even if it suggests a relationalism about colour experience.

    To be sure, the analogy runs into immediate difficulties. The move from colour contrast effects to relationalism is bridged by additional premises such as No Privilege, Ecumenicism, and Object Stability, and the abductive reasoning that links them to relationalism. For the analogy to hold the contextual nature of transparency/opacity experience must contain a means of undermining this reasoning. It is not clear how to achieve this, and in fact one reason we aren’t tempted to infer a relationalism about the ontology of transparency/opacity these contextual effects is because we feel fairly confident in independent measures of transparency/opacity and so can independently adjudicate the veridicality of different experiences of transparency/opacity. In short, No Privilege is difficult to sustain. By contrast in center-surround colour effects No Privilege is plausible. Hence the analogy breaks down.

    All of this is to say that I don’t see a clear route from the contextuality transparency/opacity experiences to a response to my argument, or to colour relationalism. But I am very interested to hear what others think.

     

    §1 Discussion Questions

    1. Are there examples of the contextual nature of experiences of transparency/opacity that you find particularly interesting?
    2. What do you think we should learn from these transparency/opacity contextual effects?

     

    §2 Non-modularity of colour & transparency/opacity.

    Suppose that the mental processes for colour and transparency/opacity are intertwined. How does this connect to my argument and colour relationalism more generally? If these processes were encapsulated from one another, then we would expect our experiences of colour and transparency/opacity to be non-relational with respect to one another. By contrast, if they are interdependent on one another then the door is at least open for the resulting experiences of each to be fundamentally interdependent on one another and hence experientially relational. For example, perhaps the colour you experience a surface to have depends on what kind of surface your “surface/object processing” decides you are looking at, and vice versa. Maybe a surface deemed “matte” will be attributed a different colour than a surface deemed “glossy”, even when the incoming “colour information” is held constant. And vice versa.

    There is now decent evidence that these sorts of interdependencies in fact hold (Chirimuuta cites some relevant articles in her commentary and discusses the issue at length in her new book Outside Color, MIT Press). A thorough treatment of how they connect to our discussion falls well outside the purview of this reply. However, here is one worry one might think I should have. Consider these interdependencies applied to layered scenarios. Suppose the colour attributed to the proximal layer depends on the kind of surface/object it is assessed to be, and vice versa; and suppose the colour attributed to the distal layer depends on the kind of surface/object it is assessed to be, and vice versa. To this point there is at best evidence for surface/colour relationalism about experience, but as yet no compelling evidence for surface/colour relationalism about ontology, or for colour relationalism proper about experience or ontology. To move toward evidence for colour relationalism (about experience or ontology) the processing of these two layers of objects/colours would have to be interdependent on one another. Roughly, the colour attributed to a proximal layer would depend not only on the determined nature of that surface/object, but also on the determined nature of the distal surface/object, and vice versa for the colour attributed to the distal layer. And I think this is what Chirimuuta suggests there is evidence for (citing Kingdom 2011). The worry as I see it is that this makes attributed colours along a line of sight [LOS] either directly or indirectly interdependent on one another, which seems like a solid base for at least a colour relationalism about experience.

    This is a base for the relationalist to work from, but I suggest that getting from here to a compelling argument for colour relationalism requires considerable work. One reason why is that I suspect various people won’t be surprised if these kinds of processing interdependencies obtain. On some level it would make sense if the human (or some) visual system uses colour information to help discern surface/object identifications and vice versa. But at first pass this will at most yield direct evidence for a kind of relationalism about aspects of experience (assuming some kind of determination relation between perceptual processing and experience). The real challenge is to use this to yield relationalist conclusions about ontology. I admit that this is underexplored territory. However, I also submit that the relationalist owes us some labour, to begin with perhaps a sketch of how to apply propositions like No Privilege, Ecumenicism and Object Stability, and how to motivate the abductive reasoning Cohen uses these propositions for.

    Beyond this Chirimuuta asks us to consider the possibility that, because of these processing and experiential interdependencies, transparency and opacity are not merely features instantiated by objects but more precisely are features of colours. Like her I am interested to hear what others think.

    §2 Discussion Questions

    1. Do you see a path from interdependencies in the processing of colours and surfaces/objects to colour relationalism (about ontology)?
    2. How should we understand the ontology of transparency and opacity?

     

    Reaction to Cohen

    §3 Fusion re-revisited.

    Cohen’s first substantive objections occur in his ‘§3 Fusion revisited’. The core issue concerns the significance of fusion cases and in particular their significance when contrasted with layering cases. Underlying this are fascinating questions about the relation between colour phenomena, epistemic norms, conceptions of veridicality and colour ontology – a cluster of issues that to my mind are ripe for discussion!

    In brief, I argue that fusion cases on their own provide reasonable justification for relationalism, but if layering is available as a contrast, then that justification is reduced. The thought is that in fusion cases there is a conflation in informational contributions from multiple sources, for example an experienced surface colour that reflects contributions from the surface and the incident illuminant. All else being equal, an inference to relationalism is tempting: if experienced surface colour were always a conflation of surface and illuminant contributions then surface colour would seem to be not intrinsic to surfaces but constitutively dependent on relations they bear to illuminants. However, in layering cases these contributions are separated and experienced as contextually related, namely as one behind the other. Here a simple inference to relationalism seems ill-advised. Thus to the extent that fusion on its own provides some justification for relationalism, if layering is available as a contrast then that justification becomes importantly limited.

    §3.1 The objection from norms. 

    Cohen claims that:

    though the objection turns on thinking of fusion as a kind of informational conflation that is in conflict with epistemic norms, that characterization is unsupported and close to question-begging…to insist that the content of a fusion experience is epistemically inferior to that of a non-fused alternative experience is just stipulative, and certainly won’t be accepted by Brown’s relationalist opponent (6).

    Here is my reply.

    Unfortunately, I didn’t insist (or intend to imply) that the content of an experience of fusion is epistemically inferior simpliciter to that of a non-fused (e.g., layering) alternative (see esp. p. 13 of my piece). An experience of surface-illuminant fusion is epistemically valuable, arguably superior to say an experience of layering, if for example one seeks to experience a conflation of surface-illuminant contributions to colour, or one seeks to gather evidence for relationalism, or if it is simply an available experience of a surface (as compared to not experiencing that surface), et cetera. However, if one seeks to experience a surface contribution distinctly from an illuminant one, then an experience of fusion is less valuable than an experience of layering. And if one seeks evidence in a quest to avoid relationalism, then the availability of experiences of layering to contrast with experiences of fusion is welcome. Thus, my point wasn’t to stipulate that “fusion is inferior” or that “layering is superior” or whatever. It was that the norms an individual applies to experiences of fusion, particularly when layering is available as a contrast, will likely if not inexorably hinge on other norms that individual holds (e.g., “I wish to experience a distinct surface contribution to colour”). So far as I can see, that point still stands, though I welcome any reactions to this.

    Beyond this there is the question of how this connects to the argument for relationalism and in particular to the No Privilege premise. This turns out to be a difficult issue that I hope others will wade into. Here is a brief comment to help prompt discussion. A core idea of No Privilege is that there is no non-question-begging reason to epistemically prefer one candidate experience over another. On one extreme, a norm such as “I seek evidence in favour of relationalism” would be as question-begging a norm as one could muster. On the other extreme, one may seek to draw norms from the experiences of (fusion or layering) themselves, without appeal to any commitments held by the perceiver. I suspect that this goal cannot be met. Perhaps in this case all colour experiences are on the same epistemic footing and we’ve secured No Privilege for free. Do others agree?

    What of the norm “I seek an experience of a distinct surface contribution to colour”? This one is tougher to assess. All else being equal it would generate a preference for experiences of layering over fusion, and thus give reason to reject No Privilege. But does this make it question-begging in the relevant sense? For starters, the relevant sense must tie to veridicality. In this case one could argue that an experience of layering is more veridical of a “distinct surface contribution to colour” than is an experience of fusion. But perhaps the relevant sense of ‘question-begging’ goes beyond tying to veridicality simpliciter and instead ties to veridical of colour. That is, is the norm “I seek an experience of a distinct surface contribution to colour” question-begging toward one conception of the veridicality of colour experience? At first pass I suspect the answer is “no”, for while this norm states a preference for a kind of colour experience it says nothing specific about the nature of colour. However, I admit that this is underexplored territory. Others please weigh in.

    §3.2 The objection from symmetry. 

    Cohen’s second objection regarding fusion moves past the general issue of epistemic norms and specifically targets veridicality:

    we should recall just what sort of symmetry is at issue…the claim [is] that there is no well-motivated, principled, and non-question begging reason for singling out one representation as uniquely veridical. But even if (contrary to what I just said) experience did come with an epistemic preference for non-relational feature representations, that wouldn’t by itself break the symmetry…[i]t wouldn’t be a reason for believing one of the variants is veridical to the exclusion of others. After all, even [for Brown] the fused color experience of O-cum-F1 is no less veridical than a non-fused color experience of O (6).

    Let me focus on a consequence of this.

    Suppose we have various experiences of fusion and of layering and we treat all of them as equally veridical. Cohen claims that it would be “unacceptably skeptical/revisionary” (3) to treat them as both falsidical, and thus that we should treat both as veridical. The question that drives the conclusion of the argument for relationalism is “What best explains this?” and Cohen’s answer is “relationalism”. But it’s hard to see why we should accept this answer. A decent explanation of veridical experiences of surface-illuminant[/filter] fusion is that colours are relational properties of surfaces defined by reference to illuminants[/filters]. However, we wouldn’t then explain veridical experiences of surface-illuminant layering (where there is a distinct surface and a distinct illuminant colour presented in a layered manner) by saying that colours are relational properties of surfaces defined by reference to illuminants. Postulating relational colours affords no explanation of this phenomenon. By contrast we might for example explain it by postulating that colours are properties of surfaces and are also properties of lights. If we then return to the general question “What best explains veridical experiences of layering and fusion?” the answer to this point looks like “sometimes colours are relational (e.g., surface-illumiant fusion) in ways that they are sometimes not (e.g., surface-illuminant layering), and when they are not relational in those ways they can be properties of lights and filters in addition to properties of surfaces.” Do others find this kind of pragmatic approach attractive?

    Interestingly, if one is looking for another, less messy and perhaps less ad hoc option, treating fusion-layering cases as equally non-veridical may start having unexpected appeal. Roughly, a colour projectivist can and should allow that sometimes experienced colour correctly and distinctly informs the subject about surface information in the world, sometimes it correctly and distinctly informs her about illuminant information, and sometimes it correctly and distinctly informs her about the result of conflating these two sources (and conflating them in various ways). These colour experiences can “correctly” make these informings because they are informative of surface and/or illuminant information (e.g., spectral power distributions, spectral reflectance functions), but not of worldly colours. By hypothesis of the latter there are none to be correctly informative of. From this perspective, Cohen’s rejection of the idea that both types of experiences are equally non-veridical on grounds that it is “unacceptably skeptical/revisionary” is too quick, and might mask an explanation that is at least on par with relationalism in this context, if not superior to it.

    There is much more to discuss about this fascinating cluster of issues, but suffice it to say, I remain unconvinced of Cohen’s efforts to rebut my charge.

     

    §3 Discussion Questions

    1. Can we glean any epistemic norms from colour experiences themselves?
    2. Suppose there are veridical experiences of fusion and layering. Does this help/hurt relationalism or other colour ontologies?
    3. How heavily should we weigh the charge that error theories of colour are “unacceptably skeptical/revisionary”?

    §4 Layering and relationalism revisited.

    A core part of Cohen’s commentary concerns a central question regarding the interpretation of layering scenarios: “Why should we prefer a description that attributes colors to two objects (as Brown prefers) over a description that attributes color to one object in a circumstance that involves mediation by a second object (as the relationalist prefers)?” (7). Let ‘cx’ denote a colour. When presented with opaque object O behind transparent object T, is the resulting colour experience best described as:

    1. Oc1-through-Tc2, or
    2. Oc3?

    One way to address the question is to again appeal to the fusion/layering distinction: if one has an experience of fusion then (2) is a more accurate description, and if one has an experience of layering then (1) is more accurate. But this isn’t what Cohen is asking. At this point he has set aside fusion cases and is focusing only on layering/scission cases (6-7). If one is presented with O behind T, and one has an experience of colour layering, is (1) or (2) a better description of one’s experience? According to Cohen: “At some point [Brown] suggests that it is just obvious [that] his own description of the case [i.e., (1)] is the uniquely correct one” (7). I feel the same way now. Yet according to Cohen “this is unpersuasive…[and] is question-begging in the context of our search for a reason to prefer one of the two descriptions on offer” (ibid). Where did our paths diverge?

    I suggest that if one has an experience of colour layering, then an accurate description of that experience must (seemingly analytically) reference two experienced layered objects, each of which is experienced to have a numerically distinct colour. In short, something like (1) must be accurate. The point isn’t that if one is presented with O behind T then one must have an experience in accordance with (1). One might have varying experiences when presented with O behind T, including an experience of fusion or of layering or in which T (or O) is invisible and one only experiences O (or T). The point also isn’t that if one is presented with O behind T, and one has an experience of layered objects, then one must have an experience in accordance with (1). Although I think this is generally true, because visually experienced objects are typically experienced to have colour, we should allow that a visually experienced object may not be experienced to have colour (see, e.g., Macpherson 2015), including an experienced layered object. The point is instead that if one is presented with O behind T, and one has an experience of colour layering, then one must have an experience in accordance with (1). If one’s experience does not conform to (1) or something very much like (1), then I do not see how it can be accurately described as an experience of colour layering. (2) is a fine example of a description that, if it accurately describes some experience, by its nature precludes that experience from being an experience of colour layering. Thus, when Cohen asserts that this “is question-begging in the context of our search for a reason to prefer one of the two descriptions on offer” (ibid), I submit that a reason has been given, and with it our search is concluded: it is true, seemingly analytically, that an experience of colour layering must conform to (1) or something close to it. To offer (2) as an accurate description of an experience of colour layering amounts to denying the possibility of such experiences.

    There remains a critically different question: when (if ever) do experiences of colour layering occur? Suppose for humans that such experiences do not occur. Then (2) might be a fine description of all experiences we have when O is experienced and presented behind T. As I discuss in my article, I think it is difficult to assess the extent to which experiences of colour layering occur. There are many cases where most will grant that it does occurs. However, its extent hinges on a number of difficult factors, not least of which is the correct interpretation of the various kinds of colour constancy experiences and a general understanding of how regularly experienced items like illuminations and surfaces are processed by our visual systems and “integrated” in visual experience. Thoughts on this would be most welcome!

    Following this Cohen recognizes (his §4.2), as I did (p. 3 of my piece) that relationalism is consistent with experiences of layering. Here are two (of many) ways: (a) each of the distal opaque and proximal transparent colours can be defined in relation to their surrounds; (b) the opaque colour can be defined relative to the transparent one, and vice versa. Either of these is sufficient to articulate the distinction between experiences of layering and fusion within relationalism, but (b) is of particular interest because it allows for the expression of relationalism within the confines of a single LOS and hence seeks to directly meet my challenge to the view. Note that (b) involves each colour contribution doing double-duty: the opaque object is not only ascribed a colour, but it (or its colour) is impacting the ascribed colour of the transparent object, and vice versa. By hypothesis this impact, as per relationalism, is not merely contextual but constitutive. This is much like in center-surround cases, where the colour ascribed to the center is impacted by the surround (or its colour), and vice versa. Since in the original paper I lacked space to discuss this issue, a brief comment seems appropriate.

    There is nothing problematic about the possibility described in (b), and it may be the norm, but I suggest that it is difficult to muster evidence specifically for its occurrence as opposed the occurrence of an experience of a simple, non-relational colour layering in which the transparent and opaque colour are contextually but not constitutively related. The matter becomes challenging because its assessment hinges in part on interpreting the modal force of experiences of “filter effects” between objects (or their colours) along a LOS. When an opaque object is viewed through a transparent one, the reflectance/transmission properties of the transparent object will “filter” aspects of the opaque surface’s colour that are being transmitted through the transparent object. For example a transparent object that reflects or absorbs (i.e., doesn’t transmit) long wavelengths of light will “filter” the contribution to the opaque surface’s colour made by long wavelengths (since the transparent object doesn’t let those through). This will often impact how we experience these objects’ colours. Do these sorts of filtering effects mark a constitutive relational feature of the opaque surface’s colour, or merely a contextual one? I suggest that our default answer should be the latter, since these effects would arguably be predictable in a close possible world with intrinsic (i.e., non-relational) colours. What do others think?

    All of this is to say that I agree with Cohen and Chirimuuta that relationalism is not only consistent with experiences of colour layering but can articulate a layering/fusion distinction. However, to me the core question is how to draw evidence in favour of relationalism from colour experiences along a LOS, and it strikes me that this is much more difficult to do than is suggested by familiar relationalist treatments of LOS colour phenomena.

    Related replies can be made to Cohen’s interesting remarks in §§4.3&4.4, and I welcome anyone’s reaction to them. Let me conclude by commenting on his claims about the Uniformity Hypothesis (§5).

    § 4 Discussion Questions

    1. How can/should we draw evidence in favour of relationalism from LOS cases?
    2. How should we understand the difference between contextual and constitutive colour relations?
    3. Are there “filter effects” we should pay special attention to?
    4. To what extent do we experience fusion/layering/something else along a LOS? What kinds of cases do you think are of philosophical interest?

    §5 The Uniformity Hypothesis.

    One important motivation for relationalism is its ability to independently resolve/explain various challenges in colour theory by appeal to relational colours, including colour effects due to center-surround variations, illumination and other LOS variations, interpersonal and interspecies variations, et cetera. For Cohen another is that the cumulative effect is an additional virtue, the added benefit of having a uniform explanation of disparate phenomena, something, as Cohen points out (p. 11, n. 10), familiar to us from the philosophy of science. I am very interested to hear what others think about the role such “theoretical virtues” can and should play in philosophy of colour. Here are a couple of remarks to help the discussion.

    I am not alone in suggesting that different explanations may be appropriate for some of the disparate colour phenomena just listed. Thus, for example, one might follow Kalderon in postulating colour pluralism specifically to explain interspecies variations, but then postulate a distinction between colours and how they are presented to explain the difference between metamers and constancy, and use the same distinction to explain some purported cases of colour illusion (though perhaps not others), and so on. The thought isn’t that there wouldn’t be an added benefit to having a single postulate that uniformly explains all/most of these phenomena. Instead the thought is that the pressure toward a uniform explanation is outweighed by the pressure toward different explanations that emerges from scrutinizing details and in particular differences between these phenomena. Perhaps unsurprisingly, Cohen still disagrees (see his §5). Perhaps unsurprisingly, I am unmoved by his steadfastness.

    It is worth trying to tease out – in rough fashion – how he and I conceive of the dialectic, particularly in an effort to generate discussion:

    Brown. In my piece I concede, for the sake of argument, that relationalism provides a good explanation of center-surround variations in colour. Perhaps I wasn’t as explicit about my intentions as I should have been. The point wasn’t to accept relationalism (because of center-surround variation) and then try to fight-off using relationalism to explain LOS variations. The point was to recommend a case-by-case approach to colour phenomena, and that means studying LOS cases after broadly (and only where appropriate) “bracketing off” victory points relationalism might accrue elsewhere. All else being equal, if we study LOS cases what kind of explanatory value would relationalism have, as opposed to other postulates? I re-submit that if it has value for experiences of colour layering this has yet to be shown and its value for experiences of fusion hinges on the unavailability of layering. Outside of this, other possible LOS experience types should be identified and examined.

    Cohen. Regarding (1) and (2) Cohen claims that “broadly empirical constraints leave open these options, [making it] hard to see why we should not at that point invoke the theoretical virtue of uniformity as a reason for preferring [(2) over (1)]” (11). Thus, the value of postulating relationalism in LOS cases is because doing so would allow us to gain the value of having a uniform explanation of colour phenomena. The allure of uniformity is doing the work.

    I hope these quick-and-dirty characterizations stimulate your thoughts and typing. I’m am curious for example about:

    §5 Discussion Questions

    1. What impact does accepting/rejecting a relational analysis of center-surround cases (or other non-LOS cases) have on our analysis of LOS cases?
    2. How much weight should we put on the power of uniform explanations in philosophy of perception?
    3. Do broadly empirical constraints leave open (1) and (2)?
    4. Suppose broadly empirical constraints do leave open both (1) and (2). Should other constraints, beyond uniformity, be considered? For example I claim that it is seemingly analytic of experiences of layering that (1) is preferable to (2). This is obviously not an empirical claim.

    References

    Byrne, A. & D. Hilbert. 2003. Colour realism and colour science. Behavioral and Brain Sciences, 26, 3-64.

    Gert, J. (in preparation). Primitive Colors: A Case Study in Neo-Pragmatist Metaphysics and Philosophy of Perception.

    Hardin, C. L. 1988. Color for Philosophers, expanded edition. Hackett: Indianapolis, ID.

    Kalderon, Mark Eli. 2007. Color pluralism. The Philosophical Review, 116 (4): 563–601.

    Macpherson, F. 2015. The structure of experience, the nature of the visual, and type 2 blindsight. Consciousness and Cognition, 32, pp. 104-128.

    1. I have a couple of comments on Brown’s interesting replies, but will break them up over a few comments in the service of readability.

      First, in the discussion of cases of fusion, Brown says
      “However, if one seeks to experience a surface contribution distinctly from an illuminant one, then an experience of fusion is less valuable than an experience of layering.”

      If this remark was intended as a reason for preferring non-relationalist descriptions of layering cases, it’s worth reminding ourselves that relationalists are not forced to describe cases of layering as cases of fusion. As I pointed out in my comments, a relationalist will say that fusion cases are those in which experience attributes just one relational color to an object along the line of sight (here O), while in fission layering cases experience attributes relational colors to multiple objects along the line of sight (here O and F1). So if one should find an experience of layering more valuable than an experience of fusion (for whatever reason), the relationalist can accommodate the difference between the two, and can allow one to choose whichever of those she prefers. There is nothing here at odds with relationalism in any way.

      The larger issue about fusion cases was, I take it, Brown’s contention that in such cases the relationalist’s symmetry premise (“there is no principled reason for favoring one of the variants as being exclusively veridical, to the exclusion of others”) fails, since the ordinary epistemic value of avoiding informational conflation is itself a reason to prefer experiences that separate out the contribution of O from other objects on the line of sight.

      But I worry that there is no truly general epistemic preference for separating out contributions in the way Brown imagines. (Here I set aside the obviously theory-dependent motivations Brown considers and appropriately sets aside.) It would be absurd to demand an account of whether someone is a, say, sister, or a teacher, in terms that don’t mention (“conflate the contributions of”) other persons. Likewise, it seems inappropriate to insist on applying some putative epistemic preference for “non-conflated” descriptions of our chromatic experience before we have reached a verdict about whether the contents of chromatic experiences are relational. (I’m not begging the question in the opposite way by assuming a contrary epistemic norm; I’m rather suggesting that neither sort of epistemic norm should be invoked at this stage in the dialectic.)

      1. Hi Jonathan,
        Thanks for you thoughts. I think we’ve misunderstood each other somewhat.
        1) I don’t need there to be an “ordinary epistemic value of avoiding informational conflation.” And I agree that imposing such a value over some kind of conflation seems question-begging against relationalism regarding that kind of conflation. Let me try to put the thought this way:

        Suppose both fusion and layering experiences of some stimulus are treated symmetrically regarding veridicality. All else being equal, the fusion experiences seem to (can be used to) justify relationalism, and the layering ones seem to (can be used to) justify non-relationalism. So relationalism cannot be offered as an explanation of both phenomena. This conclusion follows even though we’ve granted No Privilege and the rest of the Master Argument. Relationalism is consistent with both phenomena, but it does not explain both phenomena. To explain experiences of layering veridically we need to postulate layered colours, and in particular illuminant colours in addition to surface, volume and filter colours.

        2) I also agree that “relationalists are not forced to describe cases of layering as cases of fusion” (and I believe I said as much in my original article). The question for me is how to justify relationalism in the first place while remaining focused on LOS cases. And the thought is that if relationalists do describe experiences of colour layering as cases of fusion, then they have a fairly direct route to their view but misdescribe the case. If they don’t describe experiences of colour layering as cases of fusion, then they must find another route to their view. One route (as you point out) is to (a) postulate that even in experiences of layering the transparent colour is defined relative to the opaque colour and vice versa. Another route (as you point out) is to (b) seek justification outside LOS cases. My aim was roughly to bracket off (b) for another time, and I am eager to see some clean evidence of (a). I think Mazviita puts forth some evidence for (a), particularly via the Kingdom quote, but the data needs to be scrutinized more closely (and I happily admit that I haven’t done so yet).

      2. Hi Derek:

        I’m afraid I still don’t understand your objection. You say:
        “Suppose both fusion and layering experiences of some stimulus are treated symmetrically regarding veridicality. All else being equal, the fusion experiences seem to (can be used to) justify relationalism, and the layering ones seem to (can be used to) justify non-relationalism. So relationalism cannot be offered as an explanation of both phenomena. This conclusion follows even though we’ve granted No Privilege and the rest of the Master Argument. Relationalism is consistent with both phenomena, but it does not explain both phenomena. To explain experiences of layering veridically we need to postulate layered colours, and in particular illuminant colours in addition to surface, volume and filter colours.”

        But I don’t see why you say that either (fusion or scission) experience justifies one view rather than another. The experiences are just data, and views of color experience/color properties are required to give adequate descriptions of those data. I claim relationalism makes available descriptions of both sorts of experiential data, and allows one to distinguish between the two kinds. I’m not
        seeing what further kind of justification or explanation you think is needed, or why you think relationalism can’t supply it. In particular, I don’t see why you think relationalism makes it not possible to recognize “illuminant colours in addition to surfaces, volume and filter colors”.

        You also suggest that if relationalists don’t describe experiences of color layering as cases of fusion, then they must find another route to their view. Again, I don’t see why. There is a color experience that prima facie represents a color of the object, the usual symmetry considerations motivate (just as well or as poorly as in center surround or any other kind of chromatic configuration) treating that
        experience as veridical, and relationalist descriptions give us a way of vindicating that verdict: relationalism allows us to say that the surface bears color c under conditions C (which, inter alia, involve the mediation of a filter, which itself bears a relational color as well) for a perceiver S. And the motivation for relationalism here is just that it allows us to be accept the veridicality of the variant representation of the surface, which is what we wanted. That’s just exactly the normal route to relationalism, not some alternative route. No?

    2. Here is the second of my comments on Brown’s interesting replies, which I am breaking up over a few posted comments in the service of readability.

      In this comment I’ll turn to the dispute over cases of layering/scission.

      I had asked whether we should “prefer a description of layering scenarios that attributes colors to two objects (as Brown prefers) over a description that attributes color to one object in a circumstance that involves mediation by a second object (as the relationalist prefers)?”

      Brown renders this question as a choice between thinking of color experiences as (1) or (2):
      (1) Oc1-through-Tc2, or
      (2) Oc3.
      Unfortunately, as written, (2) actually obscures the circumstantial parameters that relationalists think contribute constitutively to colors, and that are crucial to the adequacy of the relationalist conception of layering. In fact, the choice should be between Brown’s preferred (1) and a relationalist description something like (3)
      (3) Oc in a circumstance involving the mediation of T.

      Brown says he thinks description (1) is just obviously superior to description (2) — at one point he says this is even analytic. Maybe so. But the comparison of interest is in fact that between (1) and (3). And here it’s not at all obvious or analytic that (1) is superior.

      For consider what Brown says in support of (1) over (2):

      i. "if one has an experience of colour layering, then an accurate description of that experience must ... reference two experienced layered objects, each of which is experienced to have a numerically distinct colour."

      Response: (3) does reference two layered objects; it also attributes numerically distinct relational colors to each (O will have the relational color c in a circumstance involving T, T will have the relational color c' in a circumstance involving the presence along the line of sight of O).

      ii. "(2) is a fine example of a description that, if it accurately describes some experience, by its nature precludes that experience from being an experience of colour layering.... To offer (2) as an accurate description of an experience of colour layering amounts to denying the possibility of such experiences."

      Response: Perhaps so (though I would like to see more of an argument here). But what rules out what the relationalist is actually putting forward as a description of layering experience, viz, (3)?

      1. Thanks Jonathan,

        I’m not entirely sure how to understand the proposed significance of:
        (3) Oc in a circumstance involving the mediation of T.

        One question of importance to me is: How does “the mediation of T” impact experience? Let us grant that T has a causal impact on the colour ascribed to O in experience. Such an impact is (obviously, I take it) consistent with Oc3 being an accurate description of the resulting experience, even if O’s colour is in fact relational (say defined by reference to T). Hence something more than causal mediation is needed for (3) to yield an adequate description of an experience of colour layering, and I guess I’m not sure what that is in this context? Unless…

        You then say “(3) does reference two layered objects; it also attributes numerically distinct relational colors to each (O will have the relational color c in a circumstance involving T, T will have the relational color c’ in a circumstance involving the presence along the line of sight of O).”

        I agree with the first claim, but (3) alone doesn’t achieve the second. Your parenthetical remark of course does (O’s colour is defined by reference to T and vice versa), but that description adds substantively to (3) and is actually a “relationalized” form of my:
        (1) Oc1-through-Tc2
        To me this is to concede my point: any accurate description of an experience of colour layering must be of this form. It is then technically open as to whether the ascribed colours are to be intrinsically or relationally defined. But it is quite difficult to get evidence for a constitutive relational interpretation, given that there is already an intimate contextual relational between these colours.

      2. Hey Derek:

        You ask some interesting and useful questions about how to understand my proposed relationalist understanding of layering cases, (3):
        (3) Oc in a circumstance involving the mediation of T.

        A first question is how the mediation of T impacts experience. I agree that this shouldn’t be understood as merely a causal impact. Rather, the suggestion is that T is a relevant parameter of the perceptual circumstance that partly individuates the color represented by visual experience, and as such is part of what’s represented in visual experience.

        This is a straightforward extension of what relationalists already say about other psychophysically relevant parameters of viewing circumstances. Eg, in traditional center surround cases, relationalists say that the color of the center patch is partly constituted by, and has a parameter for, hence is partly individuated by, the chromatic properties of the surround. (Of course relationalists will construe the chromatic properties of the surround as relational colors as well.)

        The present point is that relationalists will extend the very same apparatus for describing the parameters such as the surround in traditional center-surround configurations to the description of parameters like filters in scission configurations.

        I pointed out that, so understood, (3) does reference two layered objects, and you agreed. Excellent. I also pointed out that (3) attributes numerically distinct colors to those two objects (O will have the relational color c in a circumstance involving T, T will have the relational color c’ in a circumstance involving the presence along the line of sight of O), but here you disagreed, saying that the parenthetical goes beyond (3).

        But that’s not so — it only appears to go beyond (3) because the property of the circumstance figuring in (3) is characterized in an abbreviated and schematic form that needs filling out. (Sorry if my shorthand schematic characterization was misleading in this respect; certainly my fault.) In order to fill out the parameter T with an adequate description of what about the filter accounts for the psychophysical data, you’re going to have to mention the chromatic features of T. So how should the chromatic features of T be described? Well, note that the chromatic properties of O affect the chromatic perception of T in just the way that the chromatic properties of T affect those of O. (Which is just to say that the simultaneous chromatic perception of O and T is an interaction effect involving contributions of O on T and vice versa.) So consistent application of the very same tools we’re using for describing O’s color in terms that involve T will require characterizing T’s color in terms that involve O.

        Hence there really are two numerically distinct properties attributed in (3): a first relational property (attributed to O) spelled out in terms of T, and a second relational property (attributed to T) spelled out in terms of O.

        If this is to concede your point that “any accurate description of an experience of color layering must be of this form,” then I agree: the relationalist can describe layering scenarios in a way that respects the constraints that you are laying down. It is partly for this reason that, it seems to me, layering creates no special problem for relationalism.

      3. Hey Derek:

        After thinking about the matter some more, I think I may have come down more decisively than was warranted in some of my earlier comments about the interpretation of the relationalist’s description of scission cases, (3).

        You’ll remember that (3) was:
        (3) Oc in a circumstance involving the mediation of T.

        We had come to the point of agreeing that (3) does advert to two distinct objects, and then the question was whether (3) also attributes to those objects two numerically distinct colors. You cared about this because you were saying it is obvious/analytic that any acceptable description of layering has to both (i) recognize two objects and (ii) attributes numerically distinct colors to them.

        And you were worried that, while (3) pretty obviously meets (i), it doesn’t meet (ii) without further supplementation.

        What I said in response yesterday was that (3) should be understood as requiring that further supplementation — in effect, that the schematic letter ‘T’ in (3) has to be filled out with more specificity that would include the attribution of a relational color, and one that is distinct from the one (3) attributes to O. So, I claimed, your constraint is satisfied.

        But then I got thinking: just why does the relationalist, qua relationalist, have to take sides about just how much detail is filled in about the perceptual circumstance she takes to be a constituent of the relational color of O? Wouldn’t you still be a relationalist if you just described the circumstance as one that adverts to a difference made my the filter (inter alia)? Why do you have to go on and say just what it is about the filter — and to what level of detail would you have to do that — in order to count as a paid up member of the relationalism club?

        And my answer is: I have no idea why you should have to say that. So, on reconsideration, I guess I want to say I can imagine relationalists who say what I said yesterday, but I can also imagine relationalists who leave the specification of T pretty much as it looks in (3). For relationalists of the latter kind, the content (3) mentions two objects, but only attributes a color to one of them. And that color is: c in a circumstance involving the mediation of T (inter alia).

        Now, I expect you will say that that’s no good, in so far as then we won’t have a description of scission cases that meets what you take to be obvious/analytic adequacy conditions. I guess I’m worried that those adequacy conditions aren’t obvious. But beyond that, I do want to make it clear that even a relationalist who goes down the route we’re now considering will still be able to say (as you want her to say) that T exemplifies a color numerically distinct from the color of O.

        For even if this relationalist’s description of the color attributed to O doesn’t fully specify the color of the filter, she will still want to say — just not in her description of the color attributed to O — that T has a relational color constituted partially in terms of its relation to O. And to reiterate what I said earlier, that just exactly mirrors what the relationalist will already say about traditional center-surround cases, so there’s nothing new or fancy here. Just consistent application of the same story to another case.

        I guess what is coming out of this is that I think relationalists have both some freedom and some obligations in deciding how to describe the scission configuration. They are free to choose between
        a single more fully specified experiential content that attributes both a first relational color to O and a second relational color to T, and
        a pair of less fully specified experiential contents, one of which attributes to O a relational color partly constituted in terms of T, and the other of which attributes to T a relational color partly constituted in terms of O.

        But, however they shake out on that choice, they are not free to describe the scission configuration in a way that ignores T, or that fails to attribute a color to T (on pain of failing to do justice to the experiential facts).

        What you said earlier in the discussion suggests that you think the first of these options is preferable or even mandatory. But I’m not sure why.

        Given that one of both of these options remains not only available to, but nearly forced on relationalists given their treatment of psychophysical variation, I don’t see that the case of scission presents any special challenge to relationalist resources.

    3. Here is my third comment on Brown’s interesting replies, this time on the topic of uniformity.

      Though Brown presents us as thinking of the dialectic in different terms, I don’t see that there’s any significant methodological difference between us on this front.

      Brown recommends “a case-by-case approach to colour phenomena, [which] means studying LOS cases after broadly (and only where appropriate) “bracketing off” victory points relationalism might accrue elsewhere.”

      I am happy to accept this methodological recommendation. It’s just that, as I argued in my comments on his paper, the case by case inquiry has the result that all of the same descriptive options seem to be on the table for center surround configurations, layering configurations, and many other configurations where there is a psychophysically significant effect of some parameter of the perceiver or viewing circumstance on color perception.

      Brown appears to disagree about this: he thinks layering scenarios are less conducive than other cases like center-surround configurations to relationalist-friendly description. As he puts it, “the pressure toward a uniform explanation is outweighed by the pressure toward different explanations that emerges from scrutinizing details and in particular differences between these phenomena.”

      I’m not sure why he thinks scrutiny reveals such differences, but I suspect it is because he thinks the relationalist description of such cases is what he labels (2), which doesn’t even mention the experienced filter layer, and so is indeed plausibly inferior to his own description, (1):
      (1) Oc1-through-Tc2, or
      (2) Oc3.
      But, as I pointed out in an earlier comment (and in my original commentary on his paper), the relationalist description of layering cases is not (2), but (3):
      (3) Oc in a circumstance involving the mediation of filter
      T
      And, as I went on to note, (3), which does make reference to the
      experienced filter layer T, meets Brown’s own desiderata just as well as (2).

      Thus, what case by case inquiry that brackets off victory elsewhere seems to reveal is that layering cases, no less than center-suround (and other) cases, are amenable to the very same descriptive options: any of them can be described in relationalist terms, or not. As such, further case by case scrutiny of the kind Brown requests is not what we need to help choose between the descriptive options. (A fortiori it doesn’t tell against relationalist treatments.) If we are going to make progress, we will need to appeal to something else.

      I suggest that at this point, after the case by case scrutiny has already transpired and failed to take off the table any of the candidate descriptive options in the cases under comparison (though there are of course differences between them), it is appropriate to invoke theoretical virtues like a preference for uniformity, ceteris paribus, given that the cetera are indeed pariba. In this sense I suppose I am agreeing that “the allure of uniformity is doing the work”; but this is only because nothing else, and in particular nothing that emerges from scrutiny of the cases, turns out to be up to the job.

      1. Hi Jonathan,

        Thanks (again) for your tough and stimulating thoughts. Here are some (hopefully tough and simulating) thoughts in return.

        To begin, here are two misinterpretations I’d like to set aside:
        “[Brown] think[s] relationalism makes it not possible to recognize “illuminant colours in addition to surfaces, volume and filter colors”.”
        “[Brown] thinks layering scenarios are less conducive than other cases like center-surround configurations to relationalist-friendly description.”
        Unfortunately I don’t think either of these things. I agree that Relationalism is, broadly speaking, applicable to (/consistent with/“conducive to”) a vast array of data, as is true of many colour ontologies.

        More substantively, you also say:
        “I [Cohen] don’t see why you [Brown] say that either (fusion or scission) experience justifies one view rather than another…relationalism…allows one to distinguish between the two kinds. I’m not seeing what further kind of justification or explanation you think is needed, or why you think relationalism can’t supply it.”
        This I find a somewhat puzzling. To me the issue is not whether relationalism can distinguish between experiences of fusion and of layering – we both grant that it can. But, the issue is nonetheless whether, in response to these experiences (and these alone – for now), we should postulate relationalism. Let me explain.

        Returning to our schema, suppose, as I think we now agree, experiences of colour layering are accurately described as
        Oc1-through-Tc2.
        Should these colours – c1 and c2 – be understood relationally or intrinsically? Here is one (no doubt imperfect) way to characterize your thinking:
        A relationalist can postulate that these colours are relational colours, and in particular postulate that c1 must be defined relative to c2 and vice versa. But, one might ask, why would we do that? Similarly, a non-relationalist can postulate that these colours are intrinsic such that c1 is not defined relative to c2 and vice versa. But, you may ask, why would we do that? It seems that these data – experiences of colour layering – are mute on the matter. In this case, you may say, all we can do is look outside LOS data, and when we do we’ll find plenty of evidence for relationalism. Hence we should invoke Uniformity and “import” relationalism into LOS cases.

        Consider two reactions.

        1) To me, it is worth trying to generate empirical tests to tease out whether or not c1 is constitutively dependent on c2 and vice versa. E.g., is there reason to think the analogue of center-surround contrast effects happens, and happens just as robustly, in LOS cases? Consider a simplified “yes” and “no” outcome:
         If the answer is “yes” (as Mazviita bets it is), then LOS cases provide additional support to relationalism, and the relationalist needn’t rely on Uniformity alone to extend her view to these cases. Indeed, the relationalist can use LOS cases to help make the case for Uniformity even stronger. Here layering is helping to justify relationalism.
         If the answer is “no”, then we will have learned something important about the extent to which colours are relational: e.g., roughly, they are not relational along a LOS. That is, in this case we wouldn’t say that that c1 is constitutively dependent on c2 and vice versa, even though c1 might be constitutively dependent on (e.g.,) something in its surround, and similarly for c2. Here layering is consistent with relationalism, but justifying one non-relational aspect of colour. And relationalism is justified by appeal to non-LOS cases.

        I’m not sure where we’ll end up on this, but I think it’s fair to say that this debate has not occurred in philosophy of colour, including in relationalist writings that appeal LOS cases. If nothing else I am content to make this “increase in awareness” contribution to the field.

        2) Suppose further tests yield genuinely ambiguous data in regards to the relational nature of colour in LOS cases: we just can’t tell whether or not c1 is constitutively dependent on c2, and vice versa. In some respects this seems to be where you think we are (“nothing that emerges from scrutiny of the cases, turns out to be up to the job [of choosing between colour ontologies]”). And for you, at this point Uniformity or some other theoretical virtue has to do the heavy lifting (“I am agreeing that ‘the allure of uniformity is doing the work’”). But now…

        I must admit, the idea that LOS cases are neutral in regards to relationalism is not what I’m used to hearing relationalists say. As we know LOS cases include for example perceptual variations due to changes in illumination, changes in media (e.g., atmospheric conditions), and so on. These are typically GO TO cases for the justification of relationalism, not cases that should receive relationalist treatment because of the allure of relationalist treatments of other cases. So I’m inclined to see this as a substantive concession, and very happy to accept it.

        This also suggests a more general strategy anti-relationalists can take: the greater the number of perceptual variation cases that can be pushed into “ambiguity between ontologies”, the heavier the burden you have to place on Uniformity, and, in turn, the less bite Uniformity has. I suggest that this is an appealing prospect to your opponents, even if they still have much work to do to see where in the field of perceptual variations it will yield fruit.

      2. Hey Derek:

        Apologies for my misunderstandings of your views, and thanks for the further clarifications. Your last reframing of the dispute was
        especially helpful to me.

        Here’s where I think we are. I had argued that the comparative, down and dirty scrutiny of cases you recommend has so far failed to uncover a decisive reason for thinking the colors attributed in our experience of scission must be construed relationally or non-relationally — on the contrary, such scrutiny leaves both of those options on the table. And so I argued that, given the failure of the (broadly) empirical facts to settle the matter, it is completely appropriate to haul in theoretical virtues like uniformity (here assuming we’re already prepared to accept relationalist descriptions of colors in other kinds of configurations) to tilt the balance in the relationalist direction.

        Of course, I certainly remain open to your suggestion that we should continue to look for broadly empirical constraints that would more decisively push us in one direction or the other — that would be great if we could find such evidence, and could thereby cease to rely so heavily on the (to my mind, weaker) considerations about uniformity and other theoretical virtues. I’m just not holding my breath.

        So, yes, uniformity has a role to play in drawing cases together, and it would limit the scope of relationalist arguments/conclusions if we could show that there are significant disanalogies between the cases. Needless to say (but I keep saying it) I don’t see that that has been shown. I also don’t see that these relationalist arguments/conclusions become any weaker merely from the recognition that they rest (partly) on considerations about uniformity. They would be so weakened if those considerations were empirically or otherwise objectionable, or if we had convincing reasons to prefer non-uniformity to uniformity in our theory-building. As far as I can tell, however, they’re not and we don’t.

  4. Hi Derek,

    I agree that arguments from perceptual experience to ontology are troublesome. But isn’t it the case that your account of layering as colour-constant perception of non-relational colours relies on such a move? Why would perception of (constant) yellow through (constant) grey cause problems for the relational account of layering scenarios unless perceptual experience was assumed to have import for ontology?

    I don’t think we can infer a relationist ontology of layered colours just from the experiential interdependence of colour, opacity and transparency. But I do think that layering scenarios are just as good cases for the relationist as centre surround ones. To argue this I helped myself to some assumptions about the evidential import of experience which I believe are also in the background of your account.

    So on this score relational and non-relational analyses of layering are on a par.

    1. Hi Mazviita,

      Yes my account of constancy as layering does involve an inference from experience (& other things) to ontology. I think we both agree that caution is required in making such inferences, and has been employed in many contexts. In this case you think “layering scenarios are just as good cases for the relationist as centre surround ones.”

      1 Although I didn’t mention it in my reaction to your comments, it would be nice to get clearer specifically on the kinds of data Kingdom is describing in the enticing passage you quote, data which at this point you may have more familiarity with than I.

      2 From here perhaps we can consider whether or not “No Privilege” and the other propositions should be applied.

      best,
      Derek

Comments are closed.