Motto : Crapula ingenium offuscat. Traduction : "le bec du perroquet qu'il essuie, quoiqu'il soit net" (Pascal).

Ce blog est ouvert pour faire connaître les activités d'un groupe de recherches, le Séminaire de métaphysique d'Aix en Provence (ou SEMa). Créé fin 2004, ce séminaire est un lieu d'échanges et de propositions. Accueilli par l'IHP (EA 3276) à l'Université d'Aix Marseille (AMU), il est animé par Jean-Maurice Monnoyer, bien que ce blog lui-même ait été mis en place par ses étudiants le 4 mai 2013.


Mots-clefs : Métaphysique analytique, Histoire de la philosophie classique, moderne et contemporaine,

Métaphysique de la cognition et de la perception. Méta-esthétique.

Austrian philosophy. Philosophie du réalisme scientifique.

jeudi 11 janvier 2018

The construction of perceptual experience : what does it mean ? Gary Hatfield



[Gary HATFIELD devrait être mieux connu du public français
- bien qu’il ait publié à ma connaissance un article récent dans Les Etudes Philosophiques en 2017, « L’attention chez Descartes : aspect mental et aspect physiologique » - principalement à cause de son commentaire suivi des Méditations métaphysiques de Descartes qui a connu de nombreuses ré-éditions : et de fait, ce Routledge Philosophy Guidebook paru en 2002 (347 p.) contient un synopsis des arguments présents dans les Six Méditations et n’a pas son équivalent en langue anglaise, pour sa pertinence et sa clarté, dans la mesure où les Objections sont aussi intégrées au commentaire. Professeur de philosophie moderne à l’Université de Pennsylvanie (Philadelphie), Gary HATFIELD est surtout très respecté dans le monde académique pour son livre Perception and Cognition : Essays in Philosophy of Psychology (2009), et pour : The Natural and the Normative : Theories of Spatial Perception from Kant to Helmholtz (1990), devenu classique, encore qu’il ait  aussi traduit et présenté Les Prolégomènes de Kant dans une version révisée en 2004. Il a également contribué à une publication anthropologique remarquable avec Holly PITTMAN : Evolution of Mind, Brain, Culture, et produit depuis 1977 près d’une centaine d’articles. HATFIELD jouit d’une grande notoriété à travers son groupe de travail : le Visual Studies Program, qui connaît une audience internationale par ses productions, ses dissertations et contacts divers avec S. Palmer et le centre d’études sur la vision cognitive de Bielefeld. Dans ce même ordre d’idées, le volume publié avec Sarah Allred chez Oxford, consacré aux constances physiologiques et cognitives : Visual Experience (Oxford UP, 2012) a justifié pleinement son invitation à Aix-en-Provence lors de la soutenance de Ninuwé DESCAMPS le 15 décembre 2017. A cette occasion, Gary HATFIELD a bien voulu nous communiquer une intervention inédite qu’il a faite récemment, présentée ici sous forme de draft, dont le caractère formel illustre tout l’esprit du SEMa, confortant le lien entre
l’histoire de la philosophie moderne et les études plus récentes du groupe sur la perception et la géométrie de l’espace visuel, en rapportavec la métaphysique réaliste. Nous l’en remercions vivement]



The Construction of Perceptual Experience: What Does It Mean?[1]


Gary Hatfield


Abstract. Classical constructivists such as Rock and Hoffman contend that the processes of perception are intelligent and construct perceptual experience by going beyond the stimulus information or by creating a percept that deviates from the physical properties of the object. On these terms, Gibson’s theory of perception is anti-constructivist. After reviewing classical constructivism, this article maintains, first, that the phenomenology of visual space shows a deviation from physical spatial properties, by being contracted in depth, even under full-cue conditions, a fact that makes trouble for Gibson’s version of direct realism. Second, independently of the first argument, it contends that perception is pervasively constructed in the sense that stimulus information must be transformed to yield perception. Accordingly, perception is radically constructed in its very bones.
Keywords: construction in perception; visual space; Gestalt psychology; James J. Gibson; Donald D. Hoffman; radical constructivism



Introduction
To say that experience is constructed can mean many things. One prominent conception stems from Immanuel Kant, who developed an elaborate analysis of how categorial concepts interact with sensory perception to construct the experience of a world. This was a way of world-making long before Nelson Goodman (1978) made that phrase famous. Although Kant held that sensory perception is crucial in the construction of experience, he gave greater attention to the categories and the experience of an objective world. In contrast, this paper focuses on the construction of basic perceptual experience, which, for what it’s worth, could be nonconceptual (presenting basic properties like color and shape, independently of the concepts of color and shape). It examines construction in spatial perception, especially vision, finding that construction is pervasive. A radical constructivism is called for.
The notion that perceptual experience is a construction arises from two main sources. First, the belief that perceptual systems must cleverly manipulate information to produce perception is taken to indicate that constructive processes underlie perception. For vision, this means that the visual system creates the world of experience from ambiguous stimulus information. Second, the fact that this constructed perceptual experience sometimes, or perhaps even always, deviates from physical reality is itself taken as a sign that the experience is constructed. The well-known Necker cube illustrates both points (Fig. 1). It is two dimensional but we tend to experience it in three dimensions. So the third dimension is added by construction. 


Figure 1:  The Necker cube. The drawing is initially perceived as a spatial cube, with either the square containing the upper-left corner as the front face, or the square containing the lower-right corner. (Wait a few moments, and which cube is perceived will switch.) With practice, the lines can be perceived as a flat hexagon, containing internal triangles, squares, and other figures.



But, since the image on paper is flat and the experience is three dimensional, the experience deviates from physical reality, suggesting that the experience include constructed elements.
This paper first articulates these conceptions of construction more fully. Subsequently, it argues that spatial perception is deeply constructed, and should be so regarded even by Gibsonians. It offers a notion of radical construction that does not require inferential elaboration of stimulus information. This conception can be neutral about the mechanisms by which the construction occurs. Construction can be pervasive, even if perceptual processes are not clever.
      


1. Conceptions of construction in perception
The first conception holds that perception must be cleverly constructed out of impoverished information. Such construction is usually understood to be a cognitive or inferential process (even if the perceptual product is nonconceptual). A motivation for invoking construction is the view that information for perception is skimpy or inadequate. To construct a perception, an act of cognitive reconstruction is needed, a kind of best guess about what is present.
Accordingly, as I look out upon a scene, the basis for my perception is two impoverished images on my retinas (perhaps supplemented by oculomotor sensations). The constructivist psychologist Irvin Rock (1982, 1983) proposed that the basic processes which allow me to see the shapes, sizes, and colors of things involve the intelligent application of rules to this skimpy information, so as to yield a perceptual presentation of the spatial and chromatic layout. (We leave aside here the harder problem of classifying and identifying the kinds of things and the individuals in the scene.) Rock allows that this reconstruction might be quite good, that it might be completely accurate or veridical. However, skeptics charge that if perception is a reconstruction from poor information, it is subject to mistakes and distortions in the same way as other reconstructive processes are fallible and prone to error. Memory is one such reconstructive process, as psychologists understand it (e.g., Neisser and Libby 2000). As regards vision, the danger is that constructive perception might be a mere fabrication or concoction.
This first conception is motivated by the discrepancy between the impoverished two-dimensional retinal image and three-dimensional perceptual experience, so that construction makes up the difference. The second main conception works the other way around, starting from a mismatch between perception and the external physical scene and concluding from there that perception is a construction in the first sense. Donald Hoffman (1998) promotes this line. He emphasizes mismatches such as those described by the Gestalt principles of organization. These include figure-ground relations, in which figures become more salient than ground even though, physically, they have the same reality as surfaces, as with the black lines and white surface of the Necker-cube line drawing. They also include cases in which the same physical shape takes on different organizations in the absence of any intrinsic change, as when the cube shifts orientation (a “Gestalt switch”). The perceived change in the orientation of the cube is a subjective addition with no corresponding physical reality, yielding two different three-dimensional misperceptions of the same group of two-dimensional black lines. Hoffman concludes that, because perception is constructed in these respects, it might be a mere construction more generally, that is, a fabrication or concoction (1998, ch. 8).
Traditionally, the position opposed to constructivism in the psychology of perception is Gibsonian direct realism. James J. Gibson (1950, 1966, 1979) argued that rich information is available to visual systems in ordinary conditions of viewing, with both eyes open, good lighting, and a mobile perceiver. This information is sufficient to specify the spatial layout. There is no need for “clever” or “intelligent” reconstructive processes. We perceive the world veridically simply by picking up the rich information. (On Gibsonian theory versus constructivism, see Norman 2001.)
A philosophical counterpart to Gibson’s direct realism is the naïve direct realism now popular in parts of England, California, and New Jersey. On this view, no construction is needed because the world with its physical properties is directly touched by the conscious mind, through a relation of unmediated acquaintance. I have elsewhere detailed the shortcomings of this form of nonconstructive theory (Hatfield 2016), and so I put it aside to focus on the ideas of Rock, Hoffman, Gibson, and, as needed, the Gestalt psychologists, especially Kurt Koffka (1935).
To recapitulate, the first two authors hold that perception is a construction but disagree on whether it is therefore suspect as regards veridicality. Rock (1981, p.502; 1983, p.302) leaves the matter open, holding that perceptual constructions may be veridical or not. Hoffman (1998) contends that because perception is a construction that deviates from physical reality, as in Gestalt effects, it is suspect. It might be a deeply faulty reconstruction of what is really there. Gibson (1966, 1979) argues that because visual information, when properly analyzed, is able to specify the physical properties of the scene, there is no need for construction; usually, perception veridically presents us with the physical properties of a mind-independent reality.
Hoffman and Gibson agree that any deviation of phenomenal experience from the external physical environment is evidence of construction (even if Gibson holds that such deviation is aberrant). We can call this shared conception of construction PDE:

PDE (Phenomenal deviation from physical environment) The deviation of phenomenal experience from the external physical environment indicates perceptual construction.

The Gestalt psychologists, whom Hoffman and Gibson each invoke, agree with PDE in principle, without using the terminology of “construction.” That is, they recognize that the experienced behavioral environment deviates from the physical or geographical environment due to perceptual processes (Koffka 1935, pp.27–41). Hoffman and Gibson both think of such deviation as a kind of misperception. Here I disagree, as would the Gestaltists.
Because I disagree, I want to draw a different map of the relations among these positions and of the implications that follow from accepting that perception is constructive. To do this, I must first broaden the notion of construction. I want to include all cases in which the visual system must transform its input in order to produce perception. I thus endorse CT:

CT (Construction as transformation) Any aspect of perception that differs from proximal stimulation as received at the sense organs is to be deemed a transformation and hence a construction.

CT broadens the notion of construction to include as a transformation and hence a construction any instance in which what is perceived regularly differs from the energy distributions received by the sense organs. Rock agrees already. The Gestaltists (Koffka 1935, pp.129–41; Köhler 1947, pp.132–3) happily recognize  that perceptual processes in vision respond to the two-dimensional stimulus mosaic by producing an experience in three dimensions, which presents the world under a specific aspect (including organizational features such as figure-ground). Gibson might say that he is not involved in any kind of perceptual construction, except in deviant cases.
CT may seem so broad as to be weak. My job is to show that, in fact, under this broad notion, striking but usually unnoticed instances of construction reveal themselves. I will show that such transformations and constructions must occur even for Gibson. Indeed, it is only from the implausible perspective of naïve realism that perception just copies, or directly encompasses, physical things. Any description of perceptual processes that pays attention to the characteristics of optical stimulation will note that what the eyes receive, no matter how rich it may be in information, does not copy the world and so must be transformed to yield perception.
I will argue further that perceptual deviations in presenting external physical properties need not be characterized as misperceptions, or even as non-veridical. Here, the standards of veridicality track presentations of the environment in a way that is adequate for the guidance of action:

AA (Aspect for action) Perception presents the environment under an aspect in a manner suitable for action guidance; deviation from mind-independent physical properties need not be a “mistake.”

I will offer phenomenal grounds for believing that ordinary spatial perception does not match the physical scene, but presents it under an aspect. To follow the Gestalt psychologists for a moment, there is a difference between a phenomenally described behavioral environment and the geographic or physical environment that phenomenal experience presents (Koffka 1935, pp.27–31). The behavioral environment, although not a literal copy of the physical environment, nonetheless presents that environment in a way that can be relied upon for acting within it.
Consequently, perception is more deeply constructed than the usual notions of constructivism tell us. Perception includes a phenomenal construction that differs from a bare representation of mind-independent reality, which need not be a bad thing. Indeed, for the purpose of guiding action, a subject-dependent, environment-tracking phenomenal construction may be just what is needed.


2. Constructivism, Gibson, and Gestalt on phenomenal experience
Gibson (1966, chs. 10, 11, 14; 1979, ch. 14) considered the accepted theory of visual perception, the traditional constructivist account, to consist of these key points: the stimulus for spatial vision is impoverished, failing to specify the third dimension; this stimulus produces “sensations” that mirror the proximal stimulus (the retinal image in the case of vision); in order to compensate, computations or inferences based on memory or past experience are required, which yield a constructed perception that may or may not be veridical.
The constructivist view can be illustrated by considering these diagrams of the size-distance invariance hypothesis (SDIH). The physical relation between the height of objects, their distance, and their visual angle (which correlates directly with their retinal projections) is shown in Figure 2. Equally tall objects at different distances produce retinal projections that diminish as the distance of the object increases (Fig. 2a). A tall object (Fig. 2b, A) and shorter near one (C) can produced the same visual angle (retinal projection). But how are these perceived? Rock held that perception, at least for near objects, often achieves full phenomenal size constancy (1975, pp.31–2; 1982, pp.528, 532). The perceived size (in this case, height) depends on both visual angle (which is constant for an object at a given physical distance) and perceived distance. If the distance is accurately perceived, objects are perceived at their true locations with their actual sizes (A and C), a result known as size constancy. If the distance is under-perceived or over-perceived, the objects are seen as nearer and smaller (at B and E) or farther away and larger (A perceived as at D), respectively called under-constancy and over-constancy.


Figure 2: Part (a) shows poles of equal height at different distances. The poles farther away produce corresponding smaller visual angles (retinal images). If perceived distance matches physical distance, then, in accordance with the SDIH, perceived size matches physical size. Part (b) illustrates the SDIH for two objects, A and C, that are at different distances but project the same visual angle. Depending on the perceived distance of the objects, the perceived size varies.







To replace the constructivist picture, Gibson proposed that the stimulus for vision carries rich information that unequivocally specifies the spatial layout. This information is picked up by the visual system to yield a direct perception of the layout with its objective properties, (allegedly) obviating the need for computations or inferences. We do not, in normal veridical perception, have mediating perceptual experiences, whether of a proximal “sensation” or of the objective visual scene. Under impoverished stimulus conditions, size and shape constancy may indeed break down so as to reveal, according to constructivism, the true elements of perception: more distant objects produce smaller sizes in sensation, pennies seen at a slant produce elliptical sensations. But, according to Gibson, far from being fundamental, these aberrant sensations play no role in the ecologically normal situation of rich stimulation, in which the scene is perceived as it is (1966, chs. 10, 12; 1979, chs. 4, 14).
Gibson often wrote as if impoverished proximal sensations and direct perception of the physical layout were the only choices (1950, ch. 3; 1966, ch. 14). He contended that, in speaking of things looking small in the distance or of a penny at a slant looking elliptical, one is forcing proximal sensations into experience, either through impoverished viewing conditions or by adopting a special attitude, called the pictorial attitude, which yields an experience corresponding to a perspective projection of the scene into a two-dimensional plane (1950, p.27; 1966, pp.235–8, 306–7). Without impoverished conditions or the pictorial attitude, things should not look small in the distance under full-information or full-cue conditions. Indeed, Gibson acknowledged that if full-cue perception did not conform to the “objective facts” (1966, pp.6, 306) of the environment, his theory would be in trouble (1966, p.306; 1979, pp.166–7). Problematic examples would include “incomplete perceptual constancy” (1966, p.306; 1979, p.160), as when a coin looks elliptical or railway tracks converge phenomenally.
If incomplete constancy was found in full-cue cases, Gibson contended either that the described result reflects poor phenomenology, or that taking the pictorial attitude creates sensations that intrude upon experience. He countered the poor phenomenology by saying that the penny looks like a circle at a slant so that, unless viewed at an extreme angle, it looks like a circle. I find this response compelling. He responded to the claim that things look small in the distance by saying that, in fact, they don’t. He cited an experiment with stakes in an open landscape in which average responses of full constancy were achieved even at a distance of 700m (1950, pp.174–87; 1979, p.160). If this means that, under ordinary perceptual conditions, there is no phenomenal diminution at distances of a 200m – or 30m, or 7m – I am incredulous. I don’t believe it. I am not saying that, to an adult observer, things look as if they were diminished in physical size in the distance. Nor am I denying that one can alter the experience of size by taking the pictorial attitude. But I am asserting that, in normal perception, without the pictorial attitude, things are diminished phenomenally in the distance (as Rock 1982, p.533, allows).
Let us return to the phenomenological point anon. Gibson considered the Gestalt tradition to be distinct from standard constructivism (1950, pp.22–3; 1966, pp.266–7). Koffka’s Principles of Gestalt Psychology laid out a theory of perception which accepted the impoverishment of the stimulus with respect to perceptual experience but rejected the idea that perceptual experience results from intelligent operations based on proximal sensations (1935, pp.85–8). The impoverished stimulus can yield a presentation of a spatial and chromatic world because brain processes operate dynamically to produce organized perception, tending in the direction of the constancies of shape, size, and color. Organization includes figure-ground relations (including enhanced contrast at edges), grouping by proximity or similarity, a tendency toward symmetry and simplicity, and so on. (On the relations among Gibson, the Gestalt psychologists, and Rockian constructivism, see Epstein 1994.)
The Gestalt psychologists accepted a distinction between physical objects and the perceptual experience of those objects. The physical objects are not directly present to consciousness, but our perceptual experience “mediates” contact with the physical environment. Koffka called the world of immediate experience the behavioral environment, as opposed to the physical or geographic environment. In a famous example, a horseman on Lake Constance rides over solid ground in his behavioral environment (he perceives the flat snow-covered expanse as a snow-covered meadow), but in his geographical environment he in fact rides over the thin ice of the lake, which happens to bear the weight (1935, pp.27–8). Koffka describes the phenomenally present behavioral environment as achieving “the mediation between geographical environment and behaviour” (1935, p.36).
Although one might take this behavioral environment to be a kind of sense datum, I don’t read Koffka that way. Rather, the behavioral environment is a presentation of the geographical environment under an aspect. It presents the world phenomenally, in a subjectively conditioned way. Further, the Gestaltists did not believe that in order to be correct perception had to present only objective, mind-independent physical structures. Perception presents the world as having figure-ground organization, enhanced edges, groupings, and the like. The Gestaltists did not consider these departures from a neutral physical description to be illusions or mistakes. As illustrated in this photograph (Fig. 3), figure-ground organization can enhance object boundaries and help to sort out overlapping structures in space.




 
Figure 3: Figure-ground organization segregates the various items as phenomenal units, including presentation of surfaces as continuing underneath smaller units, and segregation of various smaller units from one another.






I believe that Gibson did consider these departures from physics as mistakes. His ecological psychology took into account “ecological” facts such as the scale of the environment in relation to the organism, the placement of the organism in relation to a scene, standing features of illumination and of surfaces, and the relation of the animal to media such as air, land, or water. He said that these factors have a “subjective” element, by which he did not a mean mind-dependent aspect, but rather the physical relation of the subject to the scene, including the subject’s physical size in relation to the physical sizes of things. He retained the notion that “objective” means what is determined by ordinary physical properties (1966, pp.6, 306; 1979, pp.8–9, 126, 129, ch. 9). Thus he did not escape, as he might have, the naïve realist preference for an “objective” physical world that perception reveals (albeit at an organismic scale).
Now we can return to size perception and the phenomenology of things looking small in the distance. Recall that Gibson believed that this phenomenology is artificially produced, by taking a pictorial attitude, so as to experience a two-dimensional perspective image of the scene. Our attitude produces an experience of intermediate sizes and shapes (as in a perspective rendering). But, he claims, with full-information and no attitude, we should perceive shapes as and where they are with their true sizes. He held that, in the normal case, there is information to specify the true sizes of things. Examples include the fact that an object on the ground covers the same amount of (homogenous) ground texture independently of the distance. This invariant relation between ground texture and size provides the information for size constancy (Gibson 1979, p.163). Or, for a person standing and looking at a row of light poles of equal height, poles both near and far are cut by the horizon line at the same proportional location; given that eye-height is known, information for the physical heights of the poles is present through the proportional relations (Gibson 1979, p.165). Gibson claimed that such higher-order ratios in the stimulus array directly specify the “objective,” that is, physical, layout (1979, pp.126, 129, ch. 9), thereby avoiding any need for subjectively conditioned representations.
There are things to challenge conceptually in Gibson’s claim. For instance, one might wonder why the receipt at the visual system of information sufficient to specify the layout should produce a direct unmediated acquaintance with objects, as posited (problematically) by naïve direct realism. But I prefer to mount a phenomenological challenge. I believe that Gibson has used binary thinking in attributing the phenomenology of intermediate constancy to intruding sensations. On the contrary, phenomenological reflection, also supported by experimental evidence, suggests that intermediate results are quite normal.
Let’s consider the notion of size constancy and things seeming smaller in the distance. Koffka (1935, p.229) held that the size-distance invariance hypothesis (not his term) covers this case. The SDIH doesn’t itself specify what size something will appear to have, but it says that, holding visual angle constant, the perceived size of an object directly varies with its perceived distance (Fig. 2b). If an object is perceived as being at its true physical distance, then it should be perceived with its true size; if it is perceived as nearer, it will be phenomenally smaller.
The perception of railway tracks extending into the distance, or of a long, straight hallway in a building, offer counter-instances to Gibson’s phenomenology. I perceive railway tracks as converging (Fig. 4). This convergence is not merely, as Gibson supposes, a convergence in perspective. If it were, I would experience the tracks as in a two-dimensional perspective projection, which would mean – for the usual perspective plane perpendicular to the line of sight – that they would appear perpendicular to the ground and would project into an upright, rather truncated triangle (Fig. 4, right). But my experience, looking down the tracks, is of them receding into the distance (in the third dimension). They converge, but they do so as they extend toward the horizon. And, indeed, there is good evidence that, in normal perception, perceived distance falls short of physical distance, the more so the farther away things are (Hillebrand 1902; Blumenfeld 1903; Wagner 2006, chs. 6–7; see also Hatfield 2003, 2016).



Figure 4: Railway tracks perceived as converging into the distance. The photograph (left) gives a sense of depth; on real tracks, the sense of depth and distance is even more salient. If our experience matched a perspective projection, it would be of an upright triangle (right).








In order to understand the regularity of this phenomenology, consider Figure 5, a drawing from Aage Slomann (1968). The drawing shows a rectangular space, say, a hallway, with a physically flat floor Ff running from observer P1’s feet (at d) to staff bc. Slomann reports that we do not experience the floor as “flat” in the sense of forming a ninety-degree angle with the observer’s body (or being perpendicular to gravity), but that the floor phenomenally rises. I agree. And if the floor rises while the directions to locations on the physical floor remain the same (that is, while we continue to perceive directions veridically, which in fact we do), then distance must be under-perceived. For example, location c appears in the direction given by the line ac, but because it is seen in relation to the floor, it appears in location f, at the shorter distance af. Taking this condition of the ground phenomenally rising to be general, then, in accordance with the SDIH, objects at a distance will appear closer and smaller, as physical staff bc appears at phenomenal location gf. The space is contracted in the in-depth dimension (along lines of sight). There is, in fact, no optical reason why the ground couldn’t appear flat and things appear where they are at their true sizes (e.g., staff bc at location bc), but that would be quite strange for us: it would mean that the train tracks don’t appear to converge, or that buildings a kilometer away aren’t phenomenally of less extent than the same building 20m away.


Figure 5: An illustration of the effects on phenomenal visual space if distance is under-perceived. For observer P­1 (ad), the ground (dc) appears to rise (de), and objects appear closer and smaller (bc appears at gf). From Slomann (1968, Fig. 1).






It bears repeating that this diminution or contraction, although related to perspective, is not the same thing as linear perspective, which is the projection of the three-dimensional world onto a two-dimensional plane. This contracted space may be viewed as a phenomenal presentation of the physical or geographical environment under an aspect in which distance contracts in a regular manner. In this visual space, much useful information is phenomenally available: phenomenal direction is congruent with physical direction, objects retain an invariant proportion to their surrounds across different distances and phenomenal sizes, next-to-ness relations on surfaces are preserved, as are ordinal relations along the ground plane, and so on (see Hatfield 2012, 2016). Phenomenal experiences of these relations satisfy a norm for perception in accordance with AA, that perception presents the environment under an aspect suitable for action guidance. Such norms include appearing in the right direction, retaining proportional size, and so on. Deviation from perceiver-independent physical properties need not be a “mistake.” Indeed, it is presumably better that things 100m away do not appear with the same phenomenal size as when they are near at hand.
These findings challenge Gibson’s phenomenology. Size constancy is intermediate (see also Daoust 2017). Shape, at the scale of a hallway, is also affected (floors in a long rectangular hallway appear as elongated trapezoids). As Gibson realized, this favors phenomenal mediation. If the object is not itself directly present in experience, but is presented in manner that is subjectively conditioned, then the natural place to find the object as presented under an aspect is in subjectively conditioned phenomenal experience. The Gestaltists and Gibson agree on this (even if Gibson tries to avoid the conclusion through his phenomenological argument). They accept an age-old logic for drawing a distinction between appearance and reality: that properties as present in experience deviate from mind-independent physical properties. This need not, by the way, commit one to appearances as sense data. Appearances can be conceived as presenting the physical world under an aspect (Hatfield 2016).
Contrary to Gibson’s direct realism, we don’t perceive the world just-as-it-is physically. This raises the following question about size: Even if Gibson is right that in ecologically normal situations information sufficient to specify physical size and distance is present, why should the physical value be the target of perception? What is wrong with contracted space, if it is ecologically adequate, and perhaps preferable, for guiding action? Size perception can be about relations, directions, and proportions, presented perceptually in a way that guides action. It needn’t be about mind-independent sizes and distances, although these can be ascertained from the layout as perceived, through known operations of measurement or estimation.
If we accept the phenomenal intermediacy of size perception, then we should accept that perception is a construction, in accordance with PDE (phenomenal deviation from the physical environment):

PDE: The deviation of phenomenal experience from the external physical environment indicates perceptual construction.

The Gestaltists might embrace this point and see it as an extension of their notion that perceptual processes produce phenomenal organization that is beneficial. Rock and Hoffman already have accepted this conclusion. Gibson wouldn’t want to accept phenomenal deviation in normal cases, but that’s too bad for him. In fact, even worse is in store.


3. Perception as radically constructed
Let us now consider perception as constructed according to CT, construction as transformation:

CT: Any aspect of perception that differs from proximal stimulation as received at the sense organs is to be deemed a transformation and hence a construction.

There are obvious instances of such transformations, as in Rock’s conception that ambiguous two-dimensional stimulus information must be transformed by intelligent perceptual processes to yield the experience of surfaces in three dimensions. Hoffman would accept that construction in perception involves somehow transforming what is contained in retinal images into organized percepts. The Gestaltists would accept something similar, while of course not conceiving the processes, with Rock and Hoffman, as clever or intelligent. Gibson would allow that such transformation occurs in the case of illusions. Otherwise, he would not want to accept that construction occurs in everyday full-cue perception of a three-dimensional world. My first task is to show that even leaving aside the phenomenal criterion for construction from the previous section (PDE), that is, putting aside for now my phenomenological argument, Gibson and his followers should accept this broader notion of construction in perception.
Let us continue to grant Gibsonian claims about rich information. There is still a need for some mechanism or process to transform stimulus information into perception of the visual world. The information available to the visual system does not copy the world, rather it specifies it in relation to a system that is attuned to, and resonates with, the available information.
An example may help. Gibsonians have successfully analyzed optic flow patterns that produce perception of the motion of the observer through the environment (see Warren 2008). When an observer moves forward in a static environment with normal optical texture, that motion produces a flow pattern of outward radial expansion at each eye (Fig. 6). This information results in an experience of forward motion through the environment. There is, however, an obvious difference between expansion within the optical array, which can be represented in two dimensions and is sampled by the two-dimensional surface of the retina, and the visual phenomenal experience of moving forward in depth. Gibson recognized this, and offered a division of labor to handle it. As a psychologist, he was working at the level of perceptual systems and the analysis of stimulus variables. If he found a psychophysical relation between optical expansion and perception of forward motion, he then said that the perceptual system detects the information that specifies the motion. If pressed on how this occurs, he allowed that receptors respond to light energy, but maintained that it was not his job to say how the receptors and visual nervous system integrate the patterns of stimulation. He was ready to move on after finding complex stimulus information that specifies the physical environment (1950, p.viii; 1966, pp.2–5, 167; 1979, pp.53, 251, 263).


Figure 6: An optical flow pattern produced by the forward motion of an observer through a stationary setting with normal optical texture. Optical features expand from the point of fixation in the scene. From Palmer (1999, p.227).









Nonetheless, Gibson allowed that neuroscientists might ask how receptor systems respond to and integrate the information in optic flow (1979, p.251). Neuroscientists answered by seeking neural mechanisms for detecting optic flow. An early effort was van de Grind (1988). He proposed motion-detecting neural mechanisms organized in a ring (see also Duffy 2004). For global optic flow, the detectors would be set in concentric rings centered on the fovea. Motion detectors are velocity and direction specific. These detectors would be oriented so as to respond to local motion of optical features moving outward on radial lines and crossing the concentric rings perpendicularly. Van de Grind suggested a similar mechanism for more localized optical expansions with the information that an object, such as a soccer ball, is approaching the perceiver’s head. These two mechanisms take as input local or global optical expansion and produce the perceptual experience either of a looming object moving toward the perceiver or of the perceiver moving forward. By CT, these are instances of construction as transformation.


Figure 7:  Reflection of light from external points V, X, Y into the eye. Proper refraction of the light rays focuses the light on to points R, S, T. Originally published in Descartes (1637, p.36). Reproduced from Descartes (1692, p.58).








A Gibsonian might object that there is no construction because there need be no underlying “intelligent” or hyper-intellectualized processes. But notice that I haven’t characterized any processes as symbolic, conceptual, inferential, or the like, the usual suite of words associated with Rockian construction. This is because I have sought to render the criterion for construction neutral between Rock and Gibson. The point is to define construction as a perceptual process, however conceived, in which perception results from a transformation of stimulus values. We clearly have a transformation in the case of optical flow patterns, which are not themselves three dimensional but yield the perception of motions in three dimensions (observer motion or ball motion). This sort of construction is needed even if we accept Gibson’s rich information. In accordance with neutrality, any “computations” involved might be effected by nonconceptual connectionist nets or by Rockian intelligent inferences (see Hatfield 1990). CT doesn’t specify. Accordingly, cognitive constructivism, as with Rock or Hoffman, becomes but one subspecies of construction as transformation (CT).
But now I want to convince you that perception is even more radically constructed – that it is constructed in its very bones. Let us begin with a very basic aspect of vision: that we see in straight lines. This has been theorized since ancient and medieval times, in the optical writings of Euclid, Ptolemy, Ibn al-Haytham, and Witelo (see Lindberg 1976). The precise geometry by which light reaches the retina was discovered by Kepler and illustrated by Descartes in his widely known writings on optics and visual physiology, the Dioptrics (1637) and the Treatise on Man (1664). This finding is illustrated here in Figure 7, Descartes’ diagram of 1637. The physical path of light is shown from illuminated points V, X, and Y to retinal locations R, S, and T.

Notice that pencils of light come from each point (or small area) on the object VY. These pencils bathe the entire front of the eye, but, due to the refractive power of the cornea (the clear cover at the front of the eye) and the lens, the rays are focused onto the retina in small regions at R, S, and T (with other points on VY projecting to all points in between).
So far, we have the physical path of the light to the retina. A natural assumption, made by Descartes and Kepler and generally affirmed by subsequent theorists, assumes that we see points of the object in the direction in which they lie, projecting outward from the stimulation of the retina (Hershenson 1999, pp.12–15). Thus, we should experience point Y as in the direction projected back from T, and points V and X along their own lines of projection. In accordance with the deeply held assumption that we see in straight lines directed toward the illuminated regions that send light to the retina, we readily accept that this is how the eye works. If we then take it that these visual directions form the bones of our experience in the three dimensional world, we can use this ray geometry to understand the perception of size and shape. So, accepting that we see the various points of object VY in the direction that they lie, we can see the shape and distance of VY by perceiving the distances to these points.
Here I want to ask a question so basic that it may seem ludicrous. Why should we see the points V, X, Y in the directions in which they lie? This is the same as asking why retinal stimulations at R, S, T should produce the experience of a small region of light localized in directions toward V, X, Y. It can’t be simply a matter of the direction of the light. If it were only that, why wouldn’t we experience all the directions the light takes in approaching each retinal point, as represented in the pencil of rays converging on point R, point S, and point T? In such a case, we might project back from each point along a cone of rays. But we don’t. If, however, the actual physical direction of the impinging light is not sufficient for predicting visual direction (by tracing back), we are free to imagine various relations between the stimulation of retinal points and experienced direction. So, why not projection in other ways (Fig. 8)?
Clearly, our intuition is that the potential phenomenal directions for stimulation of R as shown in Figure 8 would not produce a good result. For well-functioning vision, we should see the light in the direction from which it comes by tracing a single line back through the center of the eye (or the center of focus), such as lines from R to V, S to X, and T to Y. But notice, this is already a transformation. 


Fig 8:  There is no logical (or purely physical) reason why stimulation at R shouldn’t result in an experience localized in the direction of the arrow on the right, or of the arrow on the left, or any other direction.




 The light hitting the retina does not by itself carry the information of 
the direction from which it came. It is only because of the peculiar psychophysiology of our visual system that the eye experiences a luminous point as lying in the (central) direction from which the light came. The point of light on the retina is not a copy of the direction, but merely a physical result of light coming from a direction. Hence, we have here a transformation from the light stimulating the retina to the experience of a visual direction. According to CT, we have a construction.




Figure 9:  Differing focal lengths when the eye is
accommodated near (B0) or far ( B1). From Helmholtz (1867, p.584).





Looking still more closely, we find that this construction is more complicated than advertised. In fact, the optics of the eye changes for near and far vision. This means that the paths taken by light in the eye differ as a result of these changes. If we assume that visual direction from the retina is preserved, the experienced visual direction has first one, then another relation to a central path of light in the eye. The lens in the eye must accommodate, or change its shape, to produce an image in focus for objects at different distances (say, 1m and 2m). Figure 9 shows the focal lengths in the eye corresponding to the two arrows on the left. B1 is the focal length for the arrow at the far distance, B0 for the closer arrow. In order to see them both with the same angle, as is expected on the assumption that the retinal image of the arrow, FG, is the same in the two cases, there must be
differing constructions of visual direction in relation to the central physical path taken by the light within the eye as accommodated for the near and far arrows.
But things get worse, at least for deniers of construction. I have so far been following visual direction for a single eye. But human vision is normally carried out with both eyes, as in Figure 10. The two eyes are directed toward a focal point (fixation point F). 


Figure 10: Binocular vision, showing points in the field of view as projected into real eyes LE and RE. Phenomenal or perceived visual directions proceed from the Cyclopean (virtual) eye, located between the two real eyes. From Hershenson (1999, p.22).

The focal point may be nearer or farther away and the eyes will adjust to the distance. But how do we experience the focal point? Do we experience ourselves as seeing it along two directions, one for each eye (lines FLF and FRF)? This can happen if we close first one eye, then the other, while focusing on a single finger at arm’s length. But what about with both eyes open, as is usually the case? Then we see the object in front of us, in a direction from the center of the face, or from a point between the two eyes, as illustrated in Figure 10 (direction dF). This aspect of human vision is called “cyclopean vision,” after the ancient cyclops (Julesz 1971). But the cyclopean eye of normal human vision is a virtual, or constructed eye. Using two eyes, we see things in a direction from the center of the face (directions dA, dF, and dB), not in two directions, one for each eye (see Hershenson 1999, pp.17–23). This is again a construction by transformation. It is ubiquitous in human vision. The case for this transformation doesn’t depend on impoverished information, a special attitude, or my phenomenal points about contracted visual space. According to this argument, human vision is constructed in its very bones. Which is what I promised to show.    



References
Blumenfeld, Walter (1913). Untersuchungen über die scheinbare Grösse im Sehraume. Zeitschrift für Psychologie und Physiologie der Sinnesorgane 65: 241–404.
Daoust, Louise. 2017. Seeing Things As We Do: Ecological Psychology and the Normativity of Visual Perception. Ph.D. Dissertation. University of Pennsylvania.
Descartes, René. 1637. La Dioptrique. Maire. Trans: Discourse on Method, Optics, Geometry, and Meteorology, trans. Paul J. Olscamp. Bobbs-Merrill, 1965.
———. 1664. L’Homme. Angot. Trans.: Treatise of Man, trans. Thomas S. Hall. Harvard University Press, 1972.
———. 1692. Specimina Philosophiae: seu Dissertatio de Methodo, Dioptrice, et Meteora. Knoch.
Duffy, Charles J. 2004. The cortical analysis of optic flow. In: Leo M. Chalupa and John Simon Werner (eds) The Visual Neurosciences, vol 2, pp.1260–83. The MIT Press.
Epstein, William. 1994. “Why do things look as they do?”: What Koffka might have said  to Gibson, Marr, and Rock. In Stefano Poggi (ed) Gestalt Psychology: its Origins, Foundations and Influence, pp.175–89. Olschki.
Gibson, James. J. 1950. The Perception of the Visual World. Houghton Mifflin.
———. 1966. The Senses Considered as Perceptual Systems. Houghton Mifflin.
———. 1979. The Ecological Approach to Visual Perception. Houghton Mifflin.
Goodman, Nelson. 1978. Ways of Worldmaking. Hackett Publishing.
Hatfield, Gary. 1990. Gibsonian representations and connectionist symbol-processing: Prospects for unification. Psychological Research 52: 243–52.
———. 2003. Representation and constraints: The inverse problem and the structure of visual space. Acta Psychologica 114: 355–78.
———. 2012. Phenomenal and cognitive factors in spatial perception. In Gary Hatfield and Sarah Allred (eds) Visual Experience, pp.35–62. Oxford University Press.
———. 2016. Perceiving as having subjectively conditioned appearances. Philosophical Topics 44(2): 149–78.
Helmholtz, Hermann von. 1867. Handbuch der physiologischen Optik. Voss.
Hershenson, Maurice. 1999. Visual Space Perception. The MIT Press.
Hillebrand, Franz. 1902. Theorie der scheinbaren Grösse beim binokularen Sehen. Denkschrift der Kaiserlichen Akademie der Wissenschaften Wien, Mathematisch-Naturwissenschaftliche Classe 72: 255–307.
Hoffman, Donald D. 1998. Visual Intelligence: How We Create What We See. W. W. Norton.
Julesz, Bela. 1971. Foundations of Cyclopean Perception. University of Chicago Press.
Koffka, Kurt. 1935. Principles of Gestalt Psychology. Harcourt, Brace and Company.
Köhler, Wolfgang. 1947. Gestalt Psychology: An Introduction to New Concepts in Modern Psychology. Liveright.
Lindberg, David C. 1976. Theories of Vision from al-Kindi to Kepler. University of Chicago Press.
Neisser, Ulric, and Lisa K. Libby. 2000. Remembering life experiences. In: Endel Tulving; Fergus I. M. Craik (eds) The Oxford Handbook of Memory, pp.315–32. Oxford University Press.
Norman, Joel. 2001. Two visual systems and two theories of perception: An attempt to reconcile the constructivist and ecological approaches. Behavioral and Brain Sciences 25(1): 73–96.
Rock, Irvin. 1975. Introduction to Perception. Macmillan.
———. 1982. Inference in perception. In: Peter D. Asquith; Thomas Nickles (eds) PSA: Proceedings of the Biennial Meeting of the Philosophy of Science Association, Volume Two: Symposia and Invited Papers, pp.525–40. Philosophy of Science Association.
———. 1983. The Logic of Perception. The MIT Press.
Slomann, Aage. 1968. Perception of size: Some remarks on size as a primary quality and “size constancy.” Inquiry 11:101–13.
Palmer, Stephen E. 1999. Vision Science: Photons to Phenomenology. The MIT Press.
van de Grind, W. A. 1988. The possible structure and role of neuronal smart mechanisms in vision. Cognitive Systems 2(2): 163–80.
Wagner, Mark. 2006. Geometries of Visual Space. Erlbaum.
Warren, William H. 2008. Optic flow. In: Allan I. Basbaum, Akimichi Kaneko, Gordon M. Shepherd and Gerald Westheimer (eds) The Senses: A Comprehensive Reference, vol 2, Vision II (ed) Thomas D. Albright and Richard Masland, pp.219–30. Academic Press.

Gary Hatfield
Department of Philosophy
University of Pennsylvania
Philadelphia, Pennsylvania
hatfield@sas.upenn.edu





[1] Paper given at the Xth Principia International Symposium: The Construction of Experience (Florianopolis, Brazil, August, 2017). In preparation for the journal Principia (in a translation into Portuguese).

.

.

Pièces mécaniques en chocolat - Lucca

..

..