In four patients and one monkey with unilateral visual field defects caused by retro-geniculate lesions we measured forced-choice localization of square-wave gratings as a function of contrast, and compared results from the patients' absolutely and relatively blind fields. In addition, the patients indicated verbally whether they were aware of the stimuli. We then switched to a signal detection task in which the subjects had to signal a stimulus as in the localization task, by touching it, no matter whether it appeared in the good or bad hemifield, and in addition to signal a blank trial by touching an outlined square now constantly present on the monitor, and designated the no-stimulus response area. In this way, we could compare a non-verbal procedure that we had previously used in hemianopic monkeys with a verbal one commonly used to assess visual awareness. The results showed a close correspondence between the two measures of awareness in the human subjects who signalled ‘stimulus' only for targets that also evoked verbal aware responses, validating the non-verbal approach. The hemianopic monkey behaved more like a patient with an absolute rather than a relative defect, and perfectly localized high-contrast stimuli which she nevertheless treated as blanks in the vast majority of presentations in the signal detection task.
Visual awareness of stimuli presented within field defects from retro-geniculate lesions is commonly assessed by asking the patients whether they saw anything. But in patients with whom one cannot readily communicate verbally, like the very young or the severely aphasic, and in animals (with the possible and notable exception of those who have learned to use language to communicate with us, such as Alex, an African grey parrot (Pepperberg, 1991), one can only use non-verbal behaviour to assess whether they have retained or lost conscious vision. In view of the striking dissociations between behaviour and awareness demonstrated in man, here in particular that of blindsight (Pöppel et al., 1973; Weiskrantz et al., 1974), the presence of visually guided behaviour is insufficient to establish whether an animal is still consciously aware of the stimulus that provokes a motor response.
As a loss of conscious vision is a pertinent aspect of many neuropsychological deficits resulting from brain lesions, and animals are frequently used to assess these deficits, two attempts have been made to tackle the difficult question of whether or not monkeys lose conscious vision when their primary visual cortex is ablated. The first attempt used a signal-detection paradigm in hemianopic monkeys who showed almost perfect localization of visual stimuli in their impaired hemifield (Cowey and Stoerig, 1995). Having determined the monkeys' sensitivity in both hemifields, blank (no stimulus) trials were introduced in random sequence with supra-threshold target trials. The monkeys' task was to respond to the targets by touching their position wherever they appeared, as before, but to indicate the absence of a stimulus by touching a no-stimulus response area constantly outlined on the display monitor. The monkeys responded almost faultlessly to targets in their normal field and to blanks, but failed to touch stimuli presented in their hemianopic field, and instead indicated that no stimulus had been presented. As this behaviour mirrored that of blindsight patients who can show high levels of performance in forced-choice tasks but insist that they do not see the stimuli, or for that matter anything in their cortically blind fields, it suggested that the monkeys too lose conscious vision in the field affected by the lesion.
The second attempt used variation of the cueing signal normally given to hemianopic subjects to indicate that a stimulus is or may be presented. Such a signal, usually auditory or visual, informs the subjects when to respond to the ‘unseen' event. If it is omitted, one would expect only those who still retain (or have recovered) a possibly abnormally reduced but conscious representation to nevertheless respond to a stimulus in the affected field. Moore et al. found that monkeys who had suffered ablation of striate cortex as adults no longer responded to targets in the blind field when the signal, here an extinction of the fixation spot, was omitted (Moore et al., 1995). In contrast, infantoperated monkeys did still respond, although not as often or precisely as they did when the signal was present (Moore et al., 1996). The authors concluded that the early as opposed to the later lesion did not abolish conscious vision completely, and that it was only the adult-operated monkeys who showed blindsight rather than residual conscious vision.
Although both findings converge in their conclusion, there are still points of critique that can and have been raised. Regarding in particular the question of whether the paradigms capture the difference between aware and unaware, between phenomenal vision and blindsight, one must ask whether alternative hypotheses can explain the data as well or better. If any stimulus in the normal hemifield was to capture and dominate attention, as happens in some cases of visual hemi-neglect, the monkeys' failure to signal detection of targets in the impaired field could result from their inability to disengage attention from the fixation spot in one, and the prominent and permanently visible no-stimulus response area in the other paradigm (Moore et al., 1998). Also, the finding that infant-lesioned monkeys are less impaired could result from their better functional recovery allowing ‘better blindsight' (note that there are marked differences between human subjects in the level of performance; see also Results) rather than residual or recovered awareness. Lastly, the target stimulus in the hemianopic field, although equally supra-threshold to the one used in the normal field, might simply be perceptually more similar to a blank than to a stimulus in the normal field, prompting the animal to touch the ‘no' area rather than signal a stimulus (Macphail, 1998).
The first of these points can be refuted by the fact that black outlines around the target positions were also constantly present in both normal and impaired hemifields in the localization task in which the monkeys performed almost without fault in both fields (Cowey and Stoerig, 1995; Stoerig and Cowey, 1997). If any object constantly present in the good field engaged attention too comprehensively for the animals to respond to events in their hemianopic fields, they should have failed to respond in the localization task as well as in the signal detection task. And with respect to the last point, Cowey and Stoerig actually presented low contrast targets in the good and the highest possible contrast targets in the hemianopic field in an attempt to make them equally salient if they are phenomenally represented in the bad field at all. Nevertheless, in view of the potential importance of being able to assess at least a lesion-induced loss of visual awareness if not visual awareness per se, we here compared directly the signal detection paradigm outlined above and previously used with monkeys with verbal reports of awareness in patients with hemianopic visual field defects. In addition to fields of absolute blindness (three patients), in two of the patients we tested fields of relative blindness in which conscious vision is depleted but present, to learn whether stimuli which are verbally reported as faintly visible are treated as targets or as blanks. In a complementary task addressing the possibility that stimuli in the impaired field may appear more like a blank than a target in the normal field, we used stimuli of threshold contrast in the good, and of maximum contrast in the blind field of one patient and a hemianopic monkey so that we might compare directly her performance with that of the patients.
Materials and Methods
The human subjects were tested at the Institute of Experimental Psychology in Düsseldorf between 1999 and 2001. They had suffered a retro-geniculate lesion of traumatic origin in FS, and vascular in WF, HK and GY, in whom the vascular incident followed upon a traffic accident. Only GY suffered his lesion early, at age 8. Overall, the lesions (see Fig. 1) are quite different. In FS, the primary lesion affects predominantly the left temporal lobe where it damaged the optic radiation in the vicinity of the dorsal lateral geniculate nucleus. In WF and GY, extra-visual structures were damaged by additional lesions in the territory of the medial cerebral artery (WF) and the parietal and frontal lobe (GY). The damage to the temporal lobe produced aphasia in WF and FS, but WF in whom it was caused by a previous incident has recovered to a point in which it is barely noticeable, and FS fully understands what is said although he still experiences trouble in understanding or uttering not spoken sentences but isolated words. In contrast to WF and HK, who had no prior experience with tests of their residual visual functions, FS and GY are the two most thoroughly studied patients with cortical blindness [for example, see for FS: (Pöppel, 1985, 1986; Stoerig, 1993; Stoerig and Cowey, 1997; Stoerig et al., 1998; Goebel et al., 2001); for GY: (Barbur et al., 1980; Blythe et al., 1986, 1987; Barbur et al., 1993; Brent et al., 1994; Weiskrantz et al., 1995; Morland et al., 1996; Azzopardi and Cowey, 1997; Sahraie et al., 1997; Benson et al., 1998; Marcel, 1998; Zeki and ffytche, 1998; Baseler et al., 1999; Kentridge et al., 1999; Morland et al., 1999; Weiskrantz et al., 1999; Stoerig and Barth, 2001)]. All subjects gave informed consent to participate in the study.
The monkey, Rosie, previously took part as an unoperated control in experiments on residual sensitivity and blindsight in three other monkeys (Cowey and Stoerig, 1995, 1997; Cowey et al., 1998) and therefore had extensive practice with both localization and detection tasks before her left striate cortex was removed ~6 months before the present experiment. Details of the method used to remove the entire striate cortex on one side have been given before (Cowey and Stoerig, 1997). The monkey is still taking part in experiments but the excised occipital lobe which contains striate cortex representing at least the central 40° was histologically sectioned to show the excised cortex (see Fig. 2). Even if a little of the remaining striate cortex in the calcarine fissure, removed by aspiration after the lobectomy, was left intact in the most rostral part of the sulcus the subsequent field defect would be an almost complete hemianopia, and the stimuli used in the present experiment would be confined to her field defect.
Visual Field Perimetry
The visual fields of the patients were measured in a Tübinger Perimeter (Oculus, Wetzlar, Germany) with a combination of static and dynamic perimetry. A 116′, 320 cd/m2 circular stimulus was moved slowly (3–5/s) from the periphery towards the fixation spot (30′, red), or from the blind into the functional field. The background was white and had a luminance of 10 cd/m2. The subject's task was to maintain fixation, which was assessed with the help of an infrared sensitive camera projecting an enlarged image of the eye onto a monitor in the experimenter's field of view, and to press a bell button as soon as he detected the target. Within the outlined defect, the same stimulus was then presented for 200 ms at closely spaced positions, to see whether islands of seeing could be delineated. Plots of the central 60° visual fields are shown in Figure 3. Note that black regions correspond to those part of the field in which the stimulus was not detected, while the grey regions in the plots of FS and GY are those in which the stimulus was often reported, but appeared smaller, dimmer, weaker and more transient.
All tests on the human subjects were presented on a VDU (Phillips Mb107) controlled by a computer, with the patient seated in front of the monitor with the head supported by a chin rest, and monocularly fixating a small fixation cross. Viewing distance was 39 cm, at which distance the screen subtended 47° × 39°. The eye with the defect in the temporal hemifield was used; the other eye was covered. Fixation was monitored on-line with an infrared-sensitive camera which was positioned above the monitor and projected the image of the eye onto a video display. The monitor was fitted with a touch screen (élo, Ottobrunn, Germany), and all (non-verbal) responses were made by touching the screen with the index finger. The software was the same as that used in the original experiment on monkeys, although the parameters varied. Each trial started when the subject pressed a start light, a 6° × 6° square of 81 cd/m2 which appeared on a uniformly grey background of 13 cd/m2 (Fig. 4A). This immediately produced a 180 ms 6° × 6° horizontal square-wave grating of 0.57 cpd at 13 or 28° respectively from fixation; this value refers to the stimulus border closest to the vertical meridian. The gratings were separated from one another by 9°. To test the field of relative blindness in FS, the stimuli were shifted downward with respect to the fixation cross so that lateral eccentricity to the vertical meridian remained constant. Stimulus contrast was varied while its mean luminance was kept equal to that of the background. Eight grating contrasts, from 9 to 99%, were used.
For the signal detection task, a rectangular frame subtending 7° × 7° was permanently outlined above the now centrally placed fixation cross (Fig. 4B). The same 6° × 6° square-wave grating stimuli were again triggered by the subjects touching the start light. They now appeared at one of four positions in the normal hemifield or at one 9° eccentric horizontal position in the affected hemifield. Overall probability for the normal field was 60% throughout, and all four positions were equiprobable. Probability for the impaired field was either 5 or 20%; the remaining 35 or 20% consisted of blank trials. Spatial frequency was 0.57 cpd as before, and while the mean contrast was still identical to that of the background, stimulus contrast was low in the good and maximal (99%) in the absolutely blind field. In the fields of relative impairment three different contrasts were used, and again the target was displaced downward for FS to fall into his lower right quadrant.
Monkey Rosie was tested while squatting in a primate chair with her head restrained by baffles. Both her eyes were uncovered and were monitored continuously by CCTV that provided a picture of her face that filled a TV screen and showed on both eyes the specular reflections of three infrared lights around the stimulus display. The stimuli were presented on a Phillips VDU (UP2799, 17′′, vertical refresh rate = 16.6 ms) at a viewing distance of 28 cm, at which the screen subtended ~80° × 60°. She started each trial by reaching out and touching the centrally placed, 3° × 3°, white, 40 cd/m2 start light. The background luminance was 3 cd/m2. Figure 5 is a diagram of the displays used for the localization (Fig. 5A) and the signal detection task (Fig. 5B). In the former, a 14° × 14° 0.3 cpd horizontal square-wave grating appeared at one of four possible positions. Presentation time was 100 ms, too brief for the monkey to move her eyes away from the start light and onto the target before it disappeared. The position of the stimulus was randomly determined by the computer with the constraint that the total number for each position was the same in each session. Luminance contrast was varied from session to session.
For detection (Fig. 5B) the same screen, background luminance and start light were used. But the latter was placed slightly lower on the screen in order to make way for an omnipresent black outlined rectangle, 11° wide by 13° high, above it. In addition, the 14° × 14° grating stimulus of 0.3 cpd appeared in only one position on each side of the start light. If a target appeared on either side, the monkey was rewarded for touching it. If no target was presented (blank trials) the monkey had to touch the outlined square. The probability of a target in the left normal field was 45%, like that for blank trials; targets in the impaired right field made up the remaining 10% of trials. These ratios were changed in control tasks.
We first measured the patients' ability to localize the stimulus grating in the hemianopic field as a function of grating contrast. The subjects were instructed to touch the position at which the stimulus had been presented, and always to guess if uncertain. When we tested the absolutely blind region of subjects WF, HK and FS, but also in GY, we initially gave verbal feedback after each response, saying ‘Yes' for correct and ‘No' for incorrect localization. Testing continued without feedback once they could localize the 99% contrast grating; no feedback was given when FS's field of relative blindness was tested. Trials were given in blocks of 200, with each response triggering the next stimulus with a delay of 2 s. In the fields of conscious residual vision, in the lower quadrant of FS and in GY, localization was assessed in the same way. Note that only FS was tested in a region of absolute blindness as well as in one of relative blindness; the latter testing was done after the former was completed.
In a separate series, the patients said on each individual trial whether they had been aware of an event — any kind of event — in the impaired field. A minimum of 20 such trials were given per contrast provided the responses were consistent and the subject, after each 200 trial localization series, had when questioned said that he had not been aware of the stimuli. While the incidence of aware response was consistently 0 in the fields of absolute blindness, it was variable in the relatively blind fields, where up to 270 responses were collected per contrast. In addition, in several series of trials, following each manual localization response an aware/unaware judgement was required.
Unlike the human subjects, monkey Rosie was rewarded on each trial for touching the correct stimulus position by automatically delivering a preferred food item into a food well beneath the screen. Trials were self-paced and each session contained 120 or 200, depending on the particular stimulus values. Each trial ended when the monkey had responded or after 3 s if she failed to respond, which was rare. Following an incorrect response there was no food reward and the screen went black for 1.5 s. Rosie had received many thousands of trials on the localization task before her operation when she was an unoperated control for other hemianopic monkeys [see Cowey and Stoerig for details (Cowey and Stoerig, 1995)]. After her operation the grating targets were initially presented for 1 s until she began to respond to them again in her impaired field. They were then progressively reduced in duration to 100 ms. In addition, they were also initially presented at a mean luminance of 20 cd/m2, 17 cd/m2 above mean screen luminance, and were only reduced to equal mean luminance after she was responding at better than 90% correct in her impaired field. Contrast was then systematically reduced, while preserving mean luminance, as follows. Each day the contrast was reduced from its value on the previous day until performance in the impaired field declined to ~60% correct. The contrast in the impaired field was then restored to 1.0 — in order to maintain a high level of performance — while the contrast in the normal field only was systematically reduced from session to session to determine the threshold for ~70% correct in the normal hemifield. She was also tested for two sessions with the proportion of targets in the impaired field at only 10% in order to see whether their greatly reduced probability altered performance in a major way. This was an important control for the subsequent signal detection task where targets in the impaired field were necessarily rare in order that overall she was rewarded sufficiently often to maintain her performance.
With the human subjects, a visual stimulus was presented in the normal hemifield on 60% of all trials. Here, stimulus contrast was low at 3, 5 or 9% depending on the patient's sensitivity, and the grating could appear with equal probability at any of four positions. A no-stimulus response field was permanently present above the fixation point (see Fig. 4B). In a small number of trials (5 or 20% in different sessions), a stimulus was presented in the impaired field. Its contrast was 99% in the absolute field defects, 5, 9 or 20% contrast in the amblyopic field of FS, and 30, 46 or 99% contrast in GY. A range was used in the relative field defects to include contrasts that often evoked ‘aware' responses in the preceding task and low contrasts that rarely, if ever, did so. Blank trials made up the remaining 35% and 20% of the trials, respectively. Again, each trial started when the subject touched the start light. He was instructed to touch the stimulus position ‘wherever it appeared' if there was one, and to touch the no-stimulus response area whenever no stimulus was presented. We did not alert any of the subjects to the fact that on a small proportion of trials there would be a stimulus in the impaired hemifield because we specifically wanted to see how they would respond to them.
With Rosie, the stimulus was presented at only one position in each hemifield (Fig. 5B). For 2200 trials the contrast was kept at 30% in the normal hemifield, and at 99% in the impaired hemifield. Forty-five per cent of the trials were blanks and the correct response was to press the outlined rectangle; on 45% a visual target was presented in the normal, left, hemifield, and on 10% the target was in the impaired right hemifield. Subsequently she was tested for five sessions in which the contrast of the grating in the good field was just above the previously determined threshold for localization, at 0.04, whereas it remained at 0.99 in the impaired field. In addition the proportion of trials with target left, target right, and blanks was changed to 60, 20, 20 (two sessions) or to 20, 60, 20 (one session) or equiprobable (one session). These four sessions were controls for the possibility that in the earlier sessions, detection was poor for targets in the impaired field because their probability was low and/or they may have looked too dim to be categorized like the targets in the good field. For control sessions 3 and 4, where the probability of a target in the blind field was now higher, the darkening of the screen that signalled an erroneous response was removed in order not to upset the monkey if she was forced into making a high proportion of errors. Finally, she received a fifth control session in which there were no targets in the blind field, to control for false positive responding, and the signal for errors was restored.
The software recorded the position of responses on the touch screen, classifying each one as correct or false. With the human subjects the distribution of correct and false responses over positions and contrasts was assessed with the χ2 test, and standard errors were estimated on the basis of the binomial distribution. For the verbal aware responses, individual test series differed in length between 20 and 100 presentations, and accordingly weighted mean and standard error values are given. Pearson's correlation coefficients and P-values (two-tailed) were calculated with the SPSS statistical package. Slightly different analyses were performed with the monkey's data, as described in the results, since the number of possible response positions was smaller, very few sessions with the same stimulus parameters were usually given, and there were no verbal responses.
In Figure 6, localization (filled symbols) and aware responses (empty symbols) are given for the absolute defects of WF, HK and FS. HK scored significantly above chance levels at and above a contrast of 46%, but WF was significantly better than chance at only 79 and 99% contrast, and FS only at the highest level of 99%. No subject ever reported awareness of the stimulus.
In the fields of relative blindness of FS and GY, localization performance increased steeply with stimulus contrast. FS's localization was excellent down to a contrast of 20% and still statistically significant at 9%. Importantly, at the latter contrast he still reported awareness of the stimulus on half of the trials. GY's performance correlated significantly with stimulus contrast (R2 = 0.93, P ≤ 0.000), and from 20% contrast he performed above chance. His localization improved with practice over the four series of sessions which were separated by several months; failing to reach the 5% level of significance in the first series of tests even at 38% contrast, it was statistically significant at the 0.001% level at and above 20% contrast at the end (χ2 test). His ‘aware' responses varied less consistently with contrast — and performance — but were low below 38% contrast, and never exceeded 75%, not even when localization was ~90% correct (see Fig. 7). Note that GY almost as often reported awareness at 9% contrast, where his localization was at chance, and at 30% contrast where it was highly significant, showing that ‘aware' responses are not necessarily a good predictor of successful localization. Indeed, when the lower contrasts (<50%) and higher contrasts (>50%) are analysed separately, they do not correlate significantly with the incidence of aware responses (R2 = 0.29, P ≤ 0.6 for low, R2 = 0.88, P ≤ 0.11 for high contrasts). Over all contrasts, the incidence of aware responses increases with contrast (R2 = 0.89; P ≤ 0.001), and correlates significantly with localization performance (R2 = 0.77; P ≤ 0.016).
Rosie's results in the localization task are also shown in Figure 7. Note that the curve is similar to that for the patients' relative defects, reaching >70% correct at 30% contrast. The corresponding value for the normal hemifield was 3–4%, as can be seen on the curve for the normal hemifield (empty circles). Her performance in the blind field was analysed by Pearson χ2, which revealed a significant association between percentage correct and contrast (χ2 = 78.96, d.f. = 8, P ≤ 0.001). The strength of the association was also highly significant (Ø = 0.296; P ≤ 0.001). As awareness could not be verbally assessed, the right-hand scale is omitted.
The patients' responses in the signal detection task in which they had to touch a target wherever it appeared, and to signal ‘no stimulus' by touching the appropriate outlined area on the monitor when there was no target, are shown in Figure 8. Data for the absolutely blind fields are given in the left half of the figure (Fig. 8A–C). Throughout, performance was perfect for stimuli in the normal hemifield (striped bars) and for blanks (white bars). But when a 99% contrast stimulus was presented in the blind field, regardless of whether its probability was 5 or 20%, no patient responded by touching its position, and all three instead signalled ‘no stimulus'.
The data from testing the relatively blind fields of FS and GY are shown in Figure 8D,E. Again, both subjects performed faultlessly with targets in their normal field. In line with his 90% incidence of aware responses at this contrast, FS correctly classified all 20% contrast targets regardless of their probability. With the 9% contrast stimulus, of which he reported to be aware in 50% of trials, his responses depended on probability; he correctly touched 40% in the 5%, and 90% in the 20% probability condition. At the lowest contrast, 5%, he paradoxically scored better at the lower than the higher target frequency, yielding 40% correct in the low and 0% in the higher probability condition. For GY, performance depended both on stimulus contrast and on presentation frequency. When 5% of stimuli of 30 or 46% contrast appeared in the impaired hemifield, GY responded to them as if they were blanks, touching the No area. However, at 99% contrast, he never made this mistake, but appropriately touched the stimulus position. When the occurrence of stimuli in the impaired hemifield was increased to 20% the results were nearly identical, the difference being that the target of 46% contrast, which provoked verbal aware responses on 18% of presentations, was now correctly detected on 10% of its presentations.
In HK the signal detection test was repeated several months later. The procedure was as before, but the stimulus contrast in the normal hemifield was reduced so that he verbally reported seeing the stimulus on less than 50% of its presentations. The contrast of the stimulus in the blind field was kept at maximum (99%). The results (Fig. 9A) show that he now missed up to 72% of targets in the good field. Still, this did not alter his response to the stimuli in the hemianopic field which he still did not signal as targets.
Rosie's performance in the signal detection task is plotted in Figure 8F. It is based on a total of 2200 trials in which the target in the blind hemifield occurred on 10% of the trials at a contrast of 99%. Targets in the normal hemifield (contrast 30%) and blanks were equiprobable on the remaining trials. The trials on which she failed to respond (usually blank trials or when the target was in the blind field) reduced the actual number of responses to 2024. Like the patients, she performed nigh on faultlessly with targets in the normal field and with blanks. However, the high contrast target in her impaired field was correctly categorized on only 6% of its presentations. Her performance is much worse than that of both FS and GY in their fields of residual vision, but better than that found in the patients' fields of absolute blindness.
That she performed much worse in the detection than in the localization task where she made very few errors (95% correct) at this contrast cannot be explained by the low incidence of targets in the impaired field for two reasons. First, with the same incidence in the localization task and at a contrast of 99% she scored 92% correct over 240 trials in two sessions. Second, when she was tested for four control sessions in which targets in the blind field were much more probable than before and the targets in the good field were close to threshold contrast, her percentage correct response in the blind field was only 2.8%, despite the fact that her detections in the good field were only 66%. Although most of her false negative responses were by signalling blank, 1 out of 71 errors was a false positive; she signalled a target in the blind field although a blank trial was given.
Overall, in the first four control sessions, 2.2% of her positive responses to targets in the blind field were false positives; she responded on the right although the target was in the good field or, more often, a blank. In addition, on the fifth and final session of control trials where the targets on the left were again close to threshold, targets on the right were assigned 0% contrast so that they were effectively blanks. All trial types were equiprobable on this session. She scored 70% correct on the left, 92.5% correct on blanks, and 5% correct on the right. Obviously, the latter were all false positive responses since the targets on the right were invisible.
The results show that in the absolute regions of their cortically blind fields, where the patients could localize stimuli of sufficient contrast, they not only verbally claim never to be aware of them but consistently categorize them as blanks in the signal detection task as well. In contrast, in the regions of relative cortical blindness, stimuli that yielded verbal aware responses were correctly classified as stimuli, and touched accordingly. The percentage of the non-verbal no-responses reflects that of the verbal ones even when the contrast of stimuli in the normal hemifield is at threshold. Hemianopic monkey Rosie, whose localization of targets in the impaired field was almost perfect at high contrasts, correctly indicated ‘stimulus' on a very small percentage of trials in the signal detection task. Several aspects of these results require discussion.
1. Do the Results from the Signal Detection Task Match the Patients' Verbal Aware Responses?
In view of the perfect correspondence between verbal and non-verbal aware responses in the absolute defects, we can safely conclude that the signal detection paradigm captures the complete loss of conscious vision that characterizes a field of absolute cortical blindness. In the relative defects, both FS and GY correctly touched targets they verbally reported to be aware of, but the incidence of aware responses was not identical for the two tasks. As the verbal assessment appears to be more sensitive than the non-verbal one at low (30 and 46%) but not high (99%) contrasts in GY, and less sensitive in FS (at 5, 9 and 20% contrast), we assume that the difference in results does not indicate a failure of either paradigm but rather the variability of sensitivity across different testing sessions in the same part of the impaired field. This conclusion is supported by the range of verbal aware responses collected at different times. With the 30 and 46% contrast targets, GY reported awareness in 7 and 18% (mean) of presentations, but the range was 0–30% for the 30% and 7–35% for the 46% contrast target. The signal detection based values of 0 and 10%, respectively, therefore fall well within these ranges. As they are based on just one series of trials, the correspondence is no worse than to be expected and, as shown by FS's data, neither method of assessment is more conservative.
Whilst all patients were able to localize targets in their impaired field, albeit to a different degree, it was only GY who provided aware responses to targets he could not localize better than expected by chance, the 9% and, in the early sessions, also the 20 and 30% contrast ones. For the 9% contrast target that he failed to localize correctly until the end of extensive testing, he verbally indicated awareness on a mean of 6% of presentations, with a range from 0 to 51%. At such low contrasts, and with the mean luminance equal to the background, GY's awareness responses are unlikely to reflect the detection of straylight, which might otherwise have accounted for aware responses in the absence of localization (Zihl and Werth, 1984). However, in the detection task, even stimuli of 30% contrast were not detected but signalled as blanks. We therefore think it is more likely that GY gave false positive responses in order to optimize his performance on the assumption that a stimulus was presented. This hypothesis is supported by the responses he made in the final series of detection trials, which was not included in the analysis because GY was inadvertently made aware of the fact that there were targets in his blind field as well as in the good field. This produced a radical change in his behaviour in that he now actually responded to blanks as if they were targets in his impaired field, and signalled only 6 of the 25 blank trials correctly. Moreover, he distributed his 19 false positive responses as he did in his good field, touching symmetric positions on both sides even though the stimuli actually appeared only at one central position (Fig. 4B). While distributing the touched positions symmetrically to those visible in the normal field is indicative of top-down expectation rather than bottom-up processing, the false positive responses to blank trials that are reminiscent of Rosie's, cannot be accounted for by detection. Together, these observations make it implausible that in the series of verbal ‘aware' responses, he was occasionally aware of low-contrast stimuli which he nevertheless could not localize, a phenomenon for which Zeki and ffytche (1998) coined the term ‘agnosopsia'.
Overall, these results validate previous conclusions on blind-sight in hemianopic monkeys (Cowey and Stoerig, 1995) by demonstrating a close correspondence between verbal and signal detection based assessment of visual awareness in human subjects. Regarding the three points of critique that have been raised (see Introduction), we shall address them in turn. (i) It is important to highlight that the presence of a constantly present fixation spot did not prevent localization in the human subjects in the present task; nor did the constantly present outlines around the four possible stimulus positions in the previous and present experiment on the monkeys prevent them from scoring highly in the localization task. That a visible feature on the screen engages attention too selectively for the subjects to respond to stimuli in their impaired field, as suggested by Moore et al. (Moore et al., 1998), is thus not possible unless the additional presence of yet another feature — the frame of the no-stimulus response area — makes all the difference. (ii) It seems more likely that the fact that stimuli could appear in either hemifield in the signal detection but not in the localization task as given to the human subjects could account for the difference in the results yielded by the two tasks. However, the fact that FS and GY did correctly respond to stimuli they (verbally) reported as visible indicates that the occurrence of stimuli in both hemifields does not prevent correct categorization of target lights as stimuli provided they are in fact consciously detected. In addition, localization of stimuli is possible in both patients and monkeys even when the stimuli appear at random in either hemifield, as shown by our monkey data [see (Cowey and Stoerig, 1995, 1997) and present study] as well as by independent reports on patients (Weiskrantz et al., 1974; Perenin and Jeannerod, 1975). That there was only one stimulus position in the signal detection task in the impaired fields of our human and monkey subjects should have made detection easier rather than harder, and thus makes it even more likely that by almost always indicating ‘No' they were not visually aware of the stimulus. (iii) Finally, regarding the possibility that a visual stimulus in the cortically blind field appears just more like a blank than like a stimulus (Macphail, 1998), we tried to find a situation in normal vision in which this could apply. In normal vision, stimuli presented at detection threshold or in masked conditions may be described as ‘more like nothing' [although they can be used to guide localization (Meeres and Graves, 1990)]. In additional series of the signal detection task, we therefore used stimuli of threshold contrast in the normal hemifield of HK, to see whether stimuli that might appear more like a blank would alter the response to high-contrast stimuli in the cortically blind field. If faintly perceived stimuli were more like a blank than like a stimulus in the normal field, this manipulation should make the subject signal real but faint stimuli as stimuli in both normal and impaired visual fields. However, the results show that this was not the case. HK did not touch the 99% contrast target in his blind field any more often than when the stimuli in the good field yielded 100% correct responses, but remained at 0% correct.
2. Is Rosie's Cortical Blindness Absolute or Relative?
Rosie's localization performance was excellent, falling some-where between that of GY and HK: it was worse than GY's in his field of low-level conscious vision but better than HK's in his field of absolute cortical blindness (see Figs 6 and 7). As no verbal responses could be collected in Rosie, we have only her categorization responses to infer whether her performance reflects blindsight or residual conscious vision. Remarkably, it was in only 6% of presentations in the signal detection task that she correctly responded to a 99% contrast target in her impaired field; in 94%, she touched the No area, behaving as if no stimulus had been presented. Her excellent localization can thus not be explained by her correctly localizing targets she is aware of as this should yield a score of ~56% rather than >90% correct. Although the dissociation between localization and signal detection is not as strict as in the patients' absolute defects, the localization must tap processing that does not depend on a conscious stimulus representation.
Her positive responses to the grating in the impaired field occurred in the main as well as control series, and are offset by false negative responses on trials where the target appeared in the good field as well as false positive responses on blank trials, notably in the control sessions where targets in the blind field were more common and her error rate on blanks was 6.4%. All these responses were so rare that their slightly different incidence is insufficient grounds for concluding that she is occasionally visually aware of targets in her blind field.
3. Does Rosie Have Conscious Vision in Her Normal Hemifield?
In operational terms, Rosie shows a dissociation similar to that of human hemianopes (with absolute defects) in that she has learned to localize stimuli in both hemifields but does not respond to perfectly localizable blind field stimuli as stimuli when blanks are introduced and she is given the option to report stimuli as ‘nothing'. For this behavioural dissociation to reflect a loss of conscious vision from the striate cortical ablation, it is necessary that she had conscious vision before the lesion, and thus still has it in her unaffected hemifield. For reasons of evolutionary continuity, the similarity between the nervous systems, and the survival value we attribute to conscious representations, we think that this is indeed the case. Nevertheless, it cannot be proved. Even if it is biologically implausible and socially disastrous to assume otherwise, it is logically possible; as yet, no one has been able to prove beyond a shadow of a sceptic's doubt that any other person, let alone an animal, is indeed conscious.
If the signal detection task, with the non-verbal Yes/No decision it requires, is an effective substitute for the question commonly put to hemianopic patients, namely ‘Did you see the stimulus?', it could be used to assess a possible recovery of residual conscious vision with prolonged training in monkeys, and it could be adapted for use in other neuropsychological syndromes, to non-verbally assess dissociations between awareness and processing. Implicit processes, characterized by a presence of (residual) function in the absence of awareness and known to occur in syndromes such as amnesia (Warrington and Weiskrantz, 1968), prosopagnosia (Bruyer et al., 1983), achromatopsia (Heywood et al., 1991), cortical deafness (Mozaz-Garde and Cowey, 2000) and numb sense (Paillard et al., 1983), might thus be revealed without relying on verbal responses to indicate the lack of a conscious representation. A wide range of implicit processes could therefore be assessed in human subjects who, owing to pathology or pre-language age, have lost or not yet acquired language, as well as in animals.
We thank all the subjects for their helpful cooperation. We gratefully acknowledge the help of Steven Young and Manfred Mittelstaedt in writing and adapting the software program, as well as Carolyne Le Mare and Iona Hodinott-Hill for help in testing the monkey.