PSY315: Blindsight, V1, awareness and binocular rivalry

Essentially, this lecture will deal with the proposition that, although V1 is necessary for our awareness of visual perception, it is nevertheless not sufficient for awareness. That is, V1 needs to be functioning at least partially in order that we are aware of visual stimulation; but we are not aware of processing which occurs in V1 itself, only of processing in higher, extrastriate areas to which V1 feeds; and which, in turn, feed back into V1.

Blindsight

Stoerig and Cowey (1995) note that there are different forms of what we might call "blindness":

(a) The most complete form of blindness occurs when the eyes are not present - this results in absolute blindness with no processing at all of visual stimuli.

(b) If the eyes are intact with a lesion between the eyes and the LGN but with intact pathways from the retina to extra-geniculate nuclei, then reflexive resonses such as the pupillary resonse to light survive.

(c) If there is a lesion in the optic radiation - the fibers between the LGN and V1 - or in V1 itself, then as Zeki (1993) notes (p. 347) for a long time it was thought that this caused total blindness in the corresponding visual field - hemianopic (left or right visual field) if large and a scotoma (small blind area) if small.

However, Riddoch (1917) observed that some of his patients could see motion in the scotomas. This was taken up again in the 1970s by Weiskrantz and others who showed that other basic discriminations could be made within scotomas, including wavelength discrimination. Yet the astonishing fact was that these patients claimed to have no experience of visual stimulation at all, even though they performed significantly better than chance. Hence the this phenomenon was given the name "blindsight".

(d) Before we consider blindsight in more detail, note that lesions beyond V1 in extrastriate cortex but which spare V1 result in modular deficiencies such as achromatopsia or lack of colour vision if V4 is damaged; or lack of motion perception if V5 is lesioned. Note, however, that these post-striate lesions spare phenomenal vision. That is, even though there is a deficit, the observer is aware of the visual world in the area of the lesion but processes it incompletely. In blindsight consequent upon V1 damage phenomenal experience is absent.

Zeki (p.348) suggests that a likely explanation of bindsight is that there is a weak direct pathway from the LGN to V4 and V5, so that destruction of V1 does not remove all feedforward from the LGN to prestriate cortex: very small cells - smaller than the P cells - found between the LGN layers (intercalated cells) and also beneath the magnocellular LGN layers (koniocellular layer) project to prestriate cortex. Another possibility is that the superior colliculus pathway via the pulvinar (in the thalamus) to V5 with feedback to V4 is involved. In relation to this, it is worth noting that Rodman, Gross and Albright (J. Neuroscience, 1989, 9, 2033-2050) showed that removal of V1, or reversible cooling, made V5 responses weaker but did not abolish directional selectivity or binocularity. These authors suggest, in fact, that V5 (MT) is involved in blindsight. They also showed that cells in MT were completely silenced when the superior colliculus was lesioned as well as V1, consistent with the idea of multiple pathways to extrastriate cortex. A third possibility is that blindsight involves retinal a and ß cells projecting not only to LGN but also to the retina-recipient part of the pulvinar in the thalamus and thence directly to prestriate cortex. Zeki says (p. 349):

The facts given above imply that, to gain a conscious awareness of having seen ... signals must be processed in V1 first, before they are relayed to V5. alternatively, whatever operations V5 may undertake, the results of that operation must be re-entered into V1. Of course, both processes may come into play, and both may be critical for the conscious awareness of having seen and hence of the acquisition of knowledge.

The re-entry - feedback - into V1 is interesting because, as Zeki notes, without it we would have conflicting signals. Van Essen has noted that there is feedback to all cortical areas from above:

It is believed, for example, that subjective contours are processed first by cells in V2 and not by cells in V1. If this is true, then, without feedback, V1 would be coding no contour but V2 would be saying there is one. So feedback may avoid this conflict. Feedback may also be the mechanism which signals the synchronous firing of cells believed to be involved in the binding problem. Until fairly recently, it was believed that higher order, more "cognitive" processes like attention affected neural responses only in extrastriate cortex. Increasingly, however, it is being reported that such effects are also found in V1, possibly implicating feedback effects.

Back to blindsight. One of the real difficulties in claiming that blindsight is a genuine phenomenon is that there is the possibility that despite their massive V1 lesions, such patients in fact have islands of V1 which are spared from damage. Wessinger, Fendrich and Gazzaniga (1997, J. Cognitive Neuroscience, 9, 203-221) made a strong case for this explanation. They reported on several patients and noted that one usually finds that each patient will have certain residual abilities in just particular small areas of the visual field and that these abilities and areas differ widely between patients. They note that this is consistent with sparing but not with processing by secondary pathways - the latter would predict residual abilities across most of the visual field and similarity between observers. Wessinger et al. also make the very valid point that normal subjects in psychophysical threshold experiments usually feel that they are guessing despite performing at very high levels. That is, registration and awareness are dissociable and registration without awareness is commonplace. Patients with blindsight may, for whatever reason, have very weak, near threshold signals to stimuli which normal observers would perceive quite easily, and so actually process them but without phenomenal experience.

Cowey and Stoerig (Nature, 373, 1995, 247-249) conducted experiments on monkeys, aimed at answering possible alternate explanations of blindsight. They note that Klüver, in 1941, demonstrated that monkeys with V1 lesions exhibit residual visual functions. But, they say, this in itself does not tell us whether the monkeys, like humans, lack phenomenal representation (awareness) or whether they have degraded but nevertheless phenomenal representation. Four monkeys, one normal (Rosie) and 3 with hemianopias were first shown to be able to detect light increments in both half fields (Figure 2).

The monkey pressed the fixation light and then one of the 4 peripheral lights came on and the monkey was rewarded for touching it. Next (Figure 3) the test lights were in the good half field of all monkeys but there were 3 types of trial: (1) stimulus trials when one of the five test lights came on and the monkey had to touch it; (2) blank trials when nothing came on and the monkey was rewarded for touching the large rectangle at the top of the screen; and (3) probe trials on which a probe light came on in the hemanopia half-field.

The result (Figure 3) was that the blindsight monkeys got the probe trials incorrect: although the light was exactly the same as the one they had detected in the hemanopia almost 100% of the time in the first experiment, they now responded to indicate a blank trial when allowed to do so: they showed the blindsight dissociation between response and awareness.

Cowey (Current Biology, 1996, 6, 45-47) summarises that work but also work on a human patient, JY, reported by Weiskrantz et al (1995) which also uses trial-by-trial reporting of awareness. As this Figure shows, the patient reports less and less awareness of the direction - up or down - of a moving spot of light as the background

becomes brighter and brighter (contrast decreases) yet performance remains at almost 100%. As Cowey points out, this patient has almost all the left side of V1 missing so that residual vision is just not plausible.

In another paper, Kolb and Braun (1995) report what they term "blindsight in normal observers". They used two different displays in Experiment 1. In the first (Figure 1a) single dots moved diagonally back and forth but in one quadrant - at random - a set of four dots moved orthogonally to all the rest. The observer's task was to indicate the quadrant in which this happened (chance = 25% correct). The second display used paired dots (1b) with equal and opposite motion.

Here, it is impossible to see the quadrant with the different orientation at the 250 msec presentation time used. Paired and unpaired trials were randomly interleaved. Subjects performed equally well in the two conditions, with 69.9% correct in unpaired displays and 74.4% correct in paired displays, both >> 25%. Nevertheless, (Figure 1 c, d, e) whereas confidence ratings increased monotonically with % correct for the unpaired dots (1c), there was little confidence and no correlation for the paired dots (1d). The ROC curves, generated by the confidence ratings, showed a d' of essentially zero for the paired dots (i.e no awareness) but large d' for the unpaired dots.

Two other displays were used in Experiment 2. In one, oriented texture elements were orthogonal in the two eyes, but within one eye four elements in one quadrant were at right angles to the rest so that the two eyes saw cross-like elements in rivalry. Here also, subjects could not consciously see the different quadrant. The other displays had identical orientations in the two eyes and subjects could see the different quadrant. As Figure 2 shows, the results parallelled those of the first experiment. Rivalrous and non-rivalrous displays nevertheless gave 75.2% and 83.3% correct responses, respectively.

Even when long exposures were used (with orientations flipping every 250 msec to prevent scrutiny) the same results were obtained (Figure 3). Kolb and Braun note that other work indicates that whereas opposite motions and rivalrous textures are coded independently (i.e. by different cells) in V1, in MT and V4 opposite dot pairs and rivalrous stimuli are mutually inhibitory for single cells and so reduce activity markedly. This suggests that the experiments here mimic V1 damage in that the tasks impede activity from V1 to extrastriate cortex. They conclude:

Note that, even if paired and rivalrous displays would completely block the direct striate-extrastriate projection, activity in striate cortex could still affect activity in extrastriate areas indirectly by way of the superior colliculus, the pulvinar, or other subcortical nuclei. In this way, a texture contrast registered in striate cortex could guide behaviour without entering subjective awareness (p.338).

Since the appearance of this report, which seemed very exciting, a failure to replicate has been reported by Morgan, Mason & Solomon ("Blindsight in normal subjects?", Nature, 385, 1997, 401-402). They used only the rivalrous stimuli but reported that correlations between confidence ratings and correctness were extremely high. They also reported that they were able to detect the key quadrant even in trials masked by rivalry because although they could not "see" the reversed orientation lines, they had experiences of brightness or depth in that quadrant i.e. there were residual cues. They suggested that Kolb and Braun's subjects may have been reluctant to give a high confidence rating to detections using resudual cues and so may have given random ratings. This would account for the low correlation and also for their finding that observers sometimes gave high confidence ratings but made errors (the reverse of blindsight!). For a detailed analysis of this and other interesting experiments, see

http://www.perceptionweb.com/perc0197/editorial.html

All of the above data seem to suggest that one has to be very careful in claiming blindsight. Although the Stoerig and Cowey experiments seem quite conclusive, it is also clear that there are many cases in which V1 damage makes stimuli which are quite salient for a normal observer near threshold for the observers with damage. In such cases, it is very important to distinguish between lack of detection of anything at all and lack of confidence or certainty.

Azzopardi and Cowey (Consciousness and Cognition 7, 1998, 292-311) note that "In practice, blindsight is revealed as a dissociation between visual performance in two different paradigms, namely yes-no and forced choice tasks" (p.293). In a yes-no task, the subject has to say whether a stimulus was or was not presented on any trial - stimulus or blank. In, for example, a two alternative forced choice (2AFC) experiment, a stimulus is always presented in one of two spatial or temporal positions and the subject has to say which one contained the stimulus. Now whereas the 2AFC method is bias free, the yes-no procedure is not - recall the centre figure below from signal detection lectures - one can

easily manipulate the proportion of "yes" responses by changing the subject's response bias, which is independent of sensitivity. And if in fact a signal is presented on every trial (as it is when perimetry is used to map scotomas in blindsight) then proportion of yesses converts to proportion of correct responses. What this in turn means is that a dissociation between yes-no and 2AFC performance in blindsight may not indicate that the subject is detecting the stimulus (2AFC) without being aware of it (yes-no); rather, perhaps the subject can see the stimulus but has a bias to say that he/she cannot see it.

Azzopardi and Cowey (1998) tested a well-knownblindsight subject, GY - a 40 year old man - with static and moving stimuli using signal detection and both yes-no and 2AFC methods.They found that whereas GY was more sensitive to static stimuli when the 2AFC method was used, with moving stimuli there was no difference between the 2AFC and yes-no results. They interpreted this as being consistent with response bias contributing to the performance with dynamic stimuli. Hence, they concluded, the static data are true evidence for blindsight. This seems not terribly convincing, however, because even normal subjects in threshold experiments usually feel that they are guessing when they are doing better than chance. Also, Vision (Philosophical Psychology 11, 1998, 137-159) has suggested that blindsight could be mediated by the relatively unimpaired dorsal (magnocellular) pathway and that normal subjects are usually unaware of activity in that channel. If this were the case, then blindsight would indeed be "visual capacity without awareness" but it would not be so interesting as many suggest. By analogy, in the 1950s many people were very excited about "subliminal perception" (sub - below; limen - threshold); the idea that stimuli below threshold could be perceived. Advertisers inserted "invisible" messages like "Drink Coca Cola" into movies just before interval, and so on. The mystery seemed to be that people were perceiving without awareness. But signal detection theory and methods soon showed that there was no paradox - it was simply unconscious registration (not unconscious seeing) which is not paradoxical at all.

Binocular rivalry

Binocular rivalry occurs when the two eyes are presented with dissimilar stimuli, such as orthogonal gratings. If the gratings are flashed for 100-200 msec they will fuse into a checkerboard or plaid and rivalry will not occur (Wolfe, 1983). For longer durations, and if the patterns are small (~1°) then complete rivalry will occur: the subject will report seeing the complete grating in one eye for a few seconds, then the complete grating in the other eye, and so on. However, for stimuli larger than this, rivalry is more local and different regions of the percept reflect one orientation or the other (Blake, O'Shea and Mueller, 1992).

Logothetis and Schall (1989) recorded from single cells in STS and MT while rhesus monkeys, known to experience binocular rivalry, viewed horizontal gratings in both eyes but the gratings could be both drifting up or both down or in opposite directions. They found virtually all possible cell types. [1] 25% of cells were non-directional on both the non-rivalrous (NR) and rivalrous (R) trials; [2] 21% became directional only on R trials; [3] 11% were directional and responded on R trials only when they were seeing that preferred direction; [4] 32% became non-directional on R trials; and [5] 11% were directional but responded on R trials only when the monkey was seeing the non-preferred direction.

Leopold and Logothetis (1996) found similar cells in V1/V2 and V4 using stationary orthogonal gratings for rivalry or parallel ones in congruent trials. They used 3 methods to ensure monkeys were doing what they were supposed to do. First, there were catch trials in which new orientations were faded in and the monkey was expected to respond or not: these were 95% correct. Second, rivalry phase durations were similar to those in humans. Third, monkeys showed the same effect as humans when contrast was varied in one eye - the variable contrast grating did not significantly increase in number of dominance phases as its contrast increased but the fixed grating's dominance phases decreased.

About one third of 101 neurons modulated their activity during rivalry, 6/33 in V1/V2 and 26/68 in V4. In V4, 12 fired best when their preferred orientation was perceived, 6 fired best when it was supressed and 8 had no preferred orientation during congruent stimulation. On the basis of all their data - with correlates of rivalry occurring in several extrastriate areas where cells are mainly binocular - Leopold and Logothetis concluded that Blake's (1989) model of rivalry in which it occurs as a result of reciprocal inhibition between monocular cells in V1, is wrong. As we shall see later, van der Zwan and Wenderoth (1994) reached the same conclusion for other reasons.

Wenderoth (1994) criticised the conclusions of Logothetis and Schall (1989) based on their 5 cell types. He stated (p.3-4):

Thus, Logothetis and Schall found just about every conceivable possibility of direction tuning with and without rivalrous stimuli. What can be concluded? Careful analysis showed that the neural responses were not due to eye movements but apart from that the conclusions drawn, as the authors admit, were pure speculation. They suggest that the class [4] cells might be first order neurons that are not involved in rivalry. The class [2] cells might be "dynamic and adaptable to the perceptual requirements" (p.763) because though they were non-directional to non-rivalrous gratings moving up or down, they had horizontal or oblique preferred directions. This conclusion thus merely restates the observation. Class [3] cells "could mediate the perception of motion that was expressed in the behavioral response of the monkeys" (p.763), another restatement of the data. Finally, the cells which fired when the non-preferred direction was seen in rivalrous conditions ([5]) "might provide the inhibition to lower or higher visual centers to suppress the view of one eye during rivalry" (p.763). So might class [3], of course: while it is true that "the results of this study suggest the possibility of experimentally relating the activity of single neurons... to the internal perceptual state of the subjects" (p.763), this particular experiment really allows no conclusions at all to be drawn about the site of rivalry or its mechanism.

In their most recent study, Logothetis, Leopold and Sheinberg (1996) very cleverly showed that rivalry is not between the eyes but between some more central representation of the stimuli. Observers viewed orthogonal sine wave gratings, one in each eye, which flickered on and off at 18 Hz. under control conditions, the same grating was presented to the same eye throughout. In the experimental condition, the gratings swapped eyes every 333 msec. So if the observer had shut one eye, he/she would have seen a grating which changed orientation by 90° three times a second. The fascinating result was that subjects did not see the reversals and the measured rivalry, in terms of phase duration and varying contrast of one grating was exactly the same in the two conditions.

The authors conclude that it is the stimulus and not the eye that competes for dominance and that rivalry probably represents perturbations of the same neural machinery involved in other multistable phenomena - such as monocular rivalry and ambiguous figures, which have the same temporal dynamics as rivalry.

Interestingly, in his comments on this paper, Wolfe has some cautionary words. He points out that normally, using a stimulus as large as 3° x 3°, the rivalry would be piecemeal rather than complete as the authors report. Also, normally, this piecemeal effect can be eliminated by using very low contrast gratings and flickering them, again as is done here. So maybe, says Wolfe, their conditions favour pattern rivalry with disruption of the normal ocular rivalry and further research is needed to find out if there is just a pattern rivalry mechanism or whether there is also an ocular rivalry mechanism.

In a recent related paper, Kovacs, Papathomas, Yang and Feher (Proceedings of the National Academy of Science, USA, 93, 1996, 15508-15511) presented rivalrous patterns to the eyes in which pattern and eye of origin were completely confounded

red dots in left eye, green in right eye) or unconfounded ( mixed red and green in each eye). They found that subjects still reported all red or all green in the unconfounded condition indicating that the patterns, not the eyes, were rivalling. However, they did find that this occurred slightly less than expected if eye of origin was irrelevant, thus supporting Wolfe's conjecture that eye of origin may play a minor role in rivalry.

In the most recent paper from Logothetis (Sheinberg, D.L. & Logothetis, N. K. "The role of temporal cortical areas in perceptual organization" Proceedings of the National Academy of Science, USA, 94, 1997, 3408-3413) monkeys saw 2.5° patterns and recordings were made from cells in IT. In one eye was the cell's best stimulus, in the other a sunburst pattern. In a control experiment, both monkeys and humans were

shown to have the predominance of the sunburst patterns increase from 0.2 to 0.7-8 as the highest spatial frequency in the sunburst increased from 3 cpd to 10 cpd - that is, high frequencies predominate in rivalry. The take home message from the rest of the experiments was that the IT cells fired whenever only the best stimulus (e.g. butterfly, monkey face) was presented alone or when - during rivalry (grey background) - that

the best stimulus was predominant. The IT cells never fired when the sunburst was dominant or presented alone. Finally, it is known that if a stimulus is in one eye only and a new, rivalrous one is flashed on in the other eye, the new, flashed stimulus immediately suppresses the old stimulus. These IT cells alsways fired when the new

flashed stimulus was the best stimulus; but when the best stimulus was initially presented and then the sunburst was flashed on, the IT cells immediately stopped firing.

So it looks as if, by the time we get to IT, there is an excellent correlation between cell firing and awareness. Sheinberg and Logothetis postualte a close relationship between selective attention and rivalry.

Cortical areas in visual awareness

Crick and Koch (1995) have argued that we are not aware of processing which occurs in V1. I'll go into the rationale for their claim in a moment. Just consider now how nicely this claim fits with some findings on rivalry that I will consider in detail in later lectures. It is believed that some simple aftereffects arise in V1, including the tilt aftereffect induced by first-order (luminance or colour) contours and the linear motion aftereffect (MAE) induced by a drifting grating. Other aftereffects are believed to arise in higher, extrastriate areas. These include motion aftereffects induced by spirals and by plaids and tilt aftereffects induced by illusory contours.

The aftereffects which are believed to arise in V1 are not affected by rivalry. That is, if the adapting stimulus is presented to one eye and a rival stimulus to the other eye, the observer is adapted for X seconds but sees - or is aware of - the adapting stimulus for only about X/2 seconds. Now it is known that most aftereffects take about 60 seconds to reach their full strength so during the first 60 seconds of adaptation they are approaching an asymptote. The statement that V1 AEs are unaffected by rivalry means that the AE is the same size with and without rivalry - it makes no difference that in one case you see the adapting stimulus for 60 sec and in the other for 30 sec. In contrast, AEs which are believed to occur in or beyond V2 are reduced by rivalry.

The parallel with Crick and Koch's argument is compelling. First, it might be argued that the very fact that V1 AEs are not reduced by rivalry demonstrates that V1 adaptation is occurring but we aren't aware of it - so registration occurs in V1 without awareness. Second, the fact that prestriate AEs are reduced by rivalry might suggest that prestriate AEs require a conscious adapting stimulus for adaptation to occur!

Crick and Koch's arguments are as follows. They make 2 assumptions: [1] For an animal to be aware of a stimulus, there has to be a group of similar neurons located somewhere in the cortex the firing of which is correlated with that visual stimulus; and [2] such neurons must project to frontal cortex. The basis for this second assumption is a little weak - they say that we need to be aware in order to plan for action and we do that in the frontal cortex so anything which does not project to frontal cortex will not be in consciousness.

Crick and Koch then present a series of anecdotal bits of evidence for their claim. First, humans and monkeys exhibit colour constancy. As Land showed, if one sets up a Mondrian display - which is simply a mixture of differently coloured randomly sized bits of paper on a surface - then a red patch looks red even in coloured illuminating light. That is, a red patch may look red even if it is in fact reflecting more green light, say, than red light, unless one views it through a reduction tube when it looks green - one needs to be able to compare it to surrounding patches to discount the illuminant. Land's theory of colour vision explains this. He claims that the three cone types do not register wavelength but rather relative lightness. Thus, a surface which is light in long wavelength light, darker in middle wavelength light and black in short wavelength light is coded as red. Because of this, it doesn't matter what the illuminant is - the brain will still recover the relative lightnesses of all the coloured patches in the scene. Zeki reports a very powerful experiment (p. 252): If one adapts to the patch which is reflecting more green than red light and it still looks red, the afterimage is green not red. Had the afterimage been determined by the wavelength of the adapting light (green) then the afterimage would have been red. So the afterimage is determined not by the wavelength but by the perceived colour - it is calculated after the perceived colour is generated. Now Zeki reported cortical V4 cells in monkey which also responded to a stimulus on a Mondrian if it looked red, even if it was reflecting more green than red. Also, Wild, Butler, Carden and Kulikowski (1985) reported that monkeys with V4 lesions no longer exhibited colour constancy despite the fact that they could still discriminate colours. As wavelength difference - hue separation in nm - increased, both normal monkeys and V4 lesioned monkeys showed gradually increasing % correct, with no difference between the two groups, in signalling the greener of two stimuli under normal illumination. Now, the point that Crick and Koch want to make is that V4 neurons show colour constancy as we do but V1 neurons do not. This shows, they assert, that we are aware of extrastriate V4 processing but not V1 processing.

He and MacLeod (Nature, 411, 2001, 473-476)reported that gratings with such a high spatial frequency that they cannot be seen nevertheless produce adaptation effects on slightly less fine and thus visible gratings. This was demonstrated both with contrast threshold elevation and the tilt aftereffect The argument is that the very fine gratings can only be processed in V1 where receptive fields are sufficiently small and that is why these invisible - non- conscious - gratings can have adaptation effects without conscious awareness of them. Ther authors conclude that "one possibility is that information must be relayed from primary visual cortexto another region of the brain to be represented in conscious experience" (p.476). Simlar kinds of data were later reported by Shady & MacLeod (2002) who found that an illusory colour saturation effect is produced by invisible gratings; and Shady, MacLeod & Fisher (2004) who reported flicker adaptation effects of lights that were flickering above the critical flicker frequency, so that the flicker was not visible.

Interestingly, I made exactly the same argument some years ago when I started working on plaids: recall that neurons in V1 respond only to the plaid's components and not to the pattern. We see the pattern and so do pattern-selective cells in MT. I therefore suggested that perception does not occur in V1.

Psychoanatomy

The formal basis for all of these kinds of arguments originated in Julesz's (1971) book and is nicely spelled out in Blake's chapter which is on the reading list (reference 6) and is called "Psychoanatomical Strategies for studying Human Visual Perception".

Psychoanatomy involves using psychophysical experiments to draw conclusions of the form: "Process A precedes process B". According to Julesz - and quoted by Blake -

A process A is before another process B if B utilizes A. This directionality in the information flow is not affected if process C is after process B and the output of C is utilized by A (feedback). Even if A and B are connected in a closed loop, B thus utilizing A and vice versa, we will regard A as being before B if A (and not B) utilizes a process I that utilizes neither A nor B.

Here's a simple example Blake discusses. We all know that if an oblique grating is drifting within a vertical rectangular aperture, then about half the time the grating appears to be drifting obliquely - orthogonal to its orientation - and about half the time it appears to drift in the direction of the long sides of the aperture - vertically up. The latter percept is called the barberpole illusion. Shimojo, Nakayama and Silverman (1989) showed that if retinal disparity is used to make the grating appear behind the aperture it always drifts obliquely; but if disparity makes the grating appear in front it always drifts vertically. Why? Because when the grating is behind, the line ends (terminators) on the aperture belong to the aperture and not to the grating so they do not influence its movement. The grating is seen amodally completed and its lines go on for ever. But when it is in front, the terminators belong to the grating and they are going up - hence the motion is disambiguated in both cases by using disparity to define the relationship between the two surfaces - the aperture and the grating. Blake says: Surface segmentation and representation of occlusion precede disambiguation of motion in a barberpole animation sequence.

Of course, the main example of psychoanatomy of interest to Julesz was in using RDSs to test for various effects (Necker cube fluctuation, various illusions etc) the argument being that if an effect still occurred in full under RDS conditions then it occurs after the combination of the two eye's views and does not occur prior to the coding of stereoscopic depth.

For the large part of this paper, Blake concentrates on using binocular rivalry to do psychoanatomy, describing mainly his own work. First, it is known that when the rival stimuli are small - circa 1° - rivalry is complete but when they are larger - say 5-10° - rivalry is piecemeal. Blake, Fox and Westendorf (1974) induced afterimages of rival targets which had small or large visual angles and then projected these afterimages on close or far surfaces so that the afterimages appeared large or small (Emmert's Law). They found that complete rivalry occurred if the afterimage was small, regardless of how big it appeared. Thus, they concluded that rivalry occurs before size constancy.

Wiesenfelder and Blake (1991) prepared an apparent motion display in which a bar shaped area of random dots moved either right to left or left to right from one frame to the next. These two frames were presented to one eye. Subjects looked at the first frame while a sunburst pattern was presented to the other eye. They triggered the second frame only when the sunburst pattern was dominant so that the subject could not see the random dot pattern. Nevertheless, apparent motion - left or right - could be seen. So positional information was available despite the invisibility of the first pattern. Wiesenfelder and Blake also noted that the AM sequence terminated the suppression of the first stimulus, just as real motion can trigger a switch from suppression to dominance. This implies, says Blake, that the AM mechanism feeds into the neural site for rivalry suppression.

Finally, Blake with Yu had observers press one key to indicate dominance of a contrast reversing sunburst pattern and another to indicate dominance of drifting dots. When the dots were visible, they saw incoherent motion. When the sunburst was dominant, the incoherent motion was replaced either with linear dot motion - up or down - or circular motion so this coherent motion was never seen. The result was that after the adaptation to unseen linear motion, a full linear MAE was obtained. But the MAE for the unseen circular motion was greatly weakened. This suggests that the linear MAE occurs before rivalry but that the circular one occurs after rivalry. Exactly the same result was reported for linearly drifting gratings and for plaids by van der Zwan, Wenderoth and Alais (1993) - ref 31.

Aglioti et al. (1999) tested a patient with severe visual form agnosia - SF was so impaired that he could not identify single letters, This included the letters R and V which start the Italian colour words "rosso" (red) and "verde" (green). Although SF could not discriminate between the two letters, and felt that he was guessing when asked to name them, he was more accurate and faster when R and V were presented in red and green colours, respectively, than the other way around. In other words, SF displayed a kind of Stroop effect despite claiming no knowledge of the forms of the letters R and V. Aglioti et al. hypothesise that SF's normal colour perception activated weak traces of the colour names but these traces were too weak to reach consciousness.

Finally, Humphrey et al (ref 11) report a patient - PB - who cannot discriminate orientation differences of 90° due to brain damage. Nevertheless, this patient gets a McCollough effect. So it must be the case that at some level orientation is being processed without awareness. This patient has massive extrastriate damage so it is likely that the orientation processing is occurring in V1 - of which the patient is unaware!.