Brain (1999), 122, 537–559 Memories are made of this: the effects of time on stored visual knowledge in a case of visual agnosia M. Jane Riddoch, Glyn W. Humphreys, Theresa Gannon, William Blott and Victoria Jones School of Psychology, University of Birmingham, Birmingham, UK Correspondence to: Professor M. Jane Riddoch, School of Psychology, University of Birmingham, Birmingham B15 2TT, UK We report the effects of the passage of time on the longterm visual knowledge for objects in a patient with visual agnosia (H.J.A.). The naming of real objects was found to have improved, although this was not associated with any change in H.J.A.’s basic perceptual abilities which were stable over a 16-year period. The improvement in object naming was attributed to better use of non-contourbased visual information (such as surface detail and depth cues). In addition, we demonstrate a deterioration in H.J.A.’s long-term memory for the visual properties of objects, and argue that this has occurred as a result of his having impaired perceptual input. The deterioration was only apparent in drawing from memory and in the verbal descriptions of items; with forced-choice testing, H.J.A. operated at ceiling; we propose that current tests of visual imagery may not be sufficiently sensitive to detect subtle impairments of visual memory. Our findings can be taken to indicate that perceptual and memorial processes are not functionally independent, but are linked in an interactive manner. Keywords: agnosia; stored visual knowledge; visual imagery Abbreviations: BORB 5 Birmingham Object Recognition Battery; HF 5 high frquency; LF 5 low frequency; SD 5 structurally dissimilar items; SS 5 structurally similar items Introduction Visual agnosia is a modality-specific disorder where recognition of visually presented objects is severely compromised; recognition via other modalities (e.g. touch) can be relatively unimpaired. The condition was first described by Lissauer (1890). Lissauer originally distinguished between two forms of agnosia: apperceptive (a disorder resulting from impaired perceptual processing) and associative (a disorder resulting from impaired access to memory, in the presence of intact perception). More recently, it has been shown that there are further fractionations within both apperceptive and associative forms of agnosia (see Humphreys and Riddoch, 1987a; Farah, 1990; Humphreys et al., 1992a). Cases of agnosia typically have been investigated during the months immediately following the lesion, and there have been relatively few reports on the effects of the passage of time on the visual abilities in such patients (see below). However, the effects of time are of interest for at least two reasons: (i) to assess the degree to which there is spontaneous recovery and/or the development of compensation strategies; and (ii) to assess what effects there are on long-term knowledge that was unimpaired at the time of the original insult. For example, in at least some cases, long-term visual knowledge appears to be relatively © Oxford University Press 1999 spared even when perceptual processes are severely compromised (e.g. Riddoch and Humphreys, 1987a; Behrmann et al., 1992). Is this spared visual knowledge retained when perceptual input is impoverished? Effects of time on visual recognition There are three reports in the literature on the effects of time in patients with visual agnosia; H.C., initially reported by Adler (1944, 1950), and then reassessed 40 years later by Sparr et al. (1991); J.R., initially reported by Davidoff and Wilson (1985) and 10 years subsequently by Wilson and Davidoff (1993); and the case reported by Kertesz (1979), where there was again a 10 year gap between different test periods. H.C.’s initial visual agnosia (for objects, pictures and faces) was reported to have recovered very quickly. At 6 months post-injury, visual object agnosia could be demonstrated only when pictures were briefly presented tachistoscopically, although she continued to have significant difficulty in the recognition of familiar faces (Adler, 1944). When she was reassessed 40 years later, H.C. was reported as having no difficulty in recognizing visually presented Downloaded from http://brain.oxfordjournals.org/ at New York Medical College on August 22, 2014 Summary 538 M. J. Riddoch et al. in drawing items from memory, or in providing verbal descriptions of items) was impaired. Thus, in drawing an elephant from memory, she omitted the trunk. Her ability to retrieve functional/associative knowledge appeared to be relatively intact. When asked to describe a skunk she said ‘It’s a small animal and when it’s defending itself it turns its bottom to you and exudes a revolting smell’ (Wilson and Davidoff, 1993, p. 534). The case described by Kertesz (1979) also showed some improvement over time. The patient had suffered extensive brain damage as a result of a road traffic accident which had caused fractures to the right frontal, parietal and temporal bones. The CT showed decreased absorption from the right frontal horn, and some irregular activity in the right parietal region. The patient had a severe visual agnosia. Real objects were named better than line drawings (35% versus 10% correct, respectively); retesting 10 years later resulting in scores of 45 and 9% correct for real objects and line drawings, respectively. Thus, like H.C. and J.R., the ability to name real objects improved relative to the ability to name line drawings. While copying was poor, drawing from memory was reported to be somewhat better although still far from normal. Most of her drawings were recognizable, but there was evidence of neglect of left side details, and disconnected lines. Her ability to write was better than her ability to read but here she superimposed letters, wrote multiple loops at the ends of words, made excess loops for letters such as n, m, w, and made a few spelling errors and omissions. In summary, two of the three patients showed substantive improvement over time, at least in their ability to recognize visually presented objects. This may reflect better use of depth and surface characteristics (such as colour, reflectance, texture, etc.), since the ability to recognize items depicted in line drawings was still poor (for J.R.), and H.C. remained impaired under conditions where greater demands were made on visual processing (e.g. when overlapping figures were used). There was no attempt to assess the integrity of stored knowledge in H.C., and, with Kertesz’s patient, detailed examination of either visual processing or stored knowledge was not performed. In J.R., the data suggest some underlying impairment in stored knowledge in addition to any visual processing deficits. She made semantic errors in picture– word match tasks, and a large number of errors when asked how many legs an item had, when given its name (she also made errors with the ‘How many legs?’ question even when presented with a picture of the item). Imagery ability was impaired in all three instances (as assessed by drawings from memory or verbal descriptions of items). These cases, then, are not appropriate to assess the effects of perceptual impairment on the maintenance over time of long-term visual knowledge of objects. To address such effects, patients must be assessed who show relatively preserved visual knowledge on initial testing; this is the case here, for patient H.J.A. The dissociation between relatively preserved visual knowledge and impaired perception, apparent in patients such as H.J.A., also suggests that perceptual and memorial processes may Downloaded from http://brain.oxfordjournals.org/ at New York Medical College on August 22, 2014 common objects (Sparr et al., 1991); however, under conditions where greater demands were made on her visual processing abilities, such as recognizing faces from photographs, distinguishing the individual items in the Poppelreuter Test (Poppelreuter, 1917) or reading handwriting, H.C. had limited success. Imagery ability was not assessed in any great detail, although it was noted that her initial drawing from memory was very impaired (e.g. in drawing a picture of a girl she drew the nostrils above the eyes, she omitted the mouth and the arms appeared to originate from the head). There was some improvement over time, and her drawings were described as ‘crude’ 40 years after her accident. Sparr et al. (1991, p. 395) state that when she was asked to imagine and describe an object ‘. . . her response was to decompose the object into a few isolated component parts. She provided no description of the relationship or interaction of these parts’. Her spontaneous writing showed an initial deterioration. Mistakes were often made in writing the letters: l, v, w, z and G, W, M and Z (i.e. distortions, eliminations or duplications of parts of the letters). These mistakes led Adler (1950, p. 44) to suggest that ‘. . . the seemingly automatic writing in the adult is influenced by correct visual impressions’. Reassessment of H.C.’s handwriting 40 years later showed that it had stabilized at the 1947 level, with no evidence of further deterioration. Davidoff and Wilson (1985) reported the results of assessments 7 months apart for the case of J.R. who sustained a head injury as a result of a riding accident at age 19 years. Neuroimaging had shown marked dilation of the whole ventricular system, especially of the left occipital horn which extended almost to the tip of the occipital lobe. Initially, J.R.’s recognition of real objects was poor (26% correct), but improvement was shown on reassessment 7 months later (38% correct). As is typically the case with patients with visual agnosia, the naming of line drawings was performed more poorly than the naming of real objects (J.R. named 3% of the items from the Graded Naming Test at the initial assessment, and 10% of the same items 7 months later) (McKenna and Warrington, 1983). J.R.’s recognition impairment was not confined to the visual modality. She was not able to name any tactually presented objects at the initial assessment. Drawing from memory was also poor and inaccurate (an elephant was drawn with a mane and prick ears). Davidoff and Wilson (1985, p. 130) state that ‘. . . She did not seem to have a clear idea about the appearance of objects, she could describe their function but not their physical appearance (e.g. an eagle was defined as a carnivorous bird but was drawn as a carnivorous animal with four legs and teeth. A pistol was defined by its function and drawn to resemble a cannon)’. J.R. was assessed further by Wilson and Davidoff in 1992 (Wilson and Davidoff, 1993). At this stage, J.R. had no difficulty in naming real objects (100% correct), and her ability to name line drawings had also improved to some degree (50% correct). As had been observed in 1985, her ability to retrieve the visual characteristics of items (whether Stored visual knowledge in visual agnosia be somewhat distinct in vision. This suggestion does not mesh with more interactive accounts of visual object processing, for instance as captured in adaptive connectionist models (see Ellis and Humphreys, 1998), where an impairment in perceptual processing ought to lead to some problems in recovering long-term visual knowledge (see also Farah, 1990). Also, within the framework of such models, some degeneration of long-term visual knowledge may be expected when that knowledge is not constantly up-dated by intact perceptual processing. Below, we consider the relationships between visual perceptual processing and long-term knowledge. A number of patients has been reported with relatively preserved visual imagery following visual agnosia and, of these, four have been described in some detail (see Riddoch and Humphreys, 1987a; Behrmann et al., 1992; Jankowiak et al., 1992; Servos et al., 1993; Young et al., 1994; Servos and Goodale, 1995). C.K. (Behrmann et al., 1992) was thought to have sustained bilateral thinning of the occipital lobes as a result of a head injury; M.D. (Jankowiak et al., 1992) suffered bilateral occipitotemporal lesions as a result of a gunshot wound; and D.F. (Servos et al., 1993; Servos and Goodale, 1995), as a consequence of exposure to carbon monoxide, was damaged in the ventral portion of the lateral occipital region (bilaterally in areas 18 and 19, with sparing of areas 17, 20, 21 and 37). The fourth patient is H.J.A., the subject of the present study, whom we discuss following the first three cases. Patients M.D., D.F. and C.K. all showed different degrees of severity of visual agnosia, with M.D. (Jankowiak et al., 1992) showing the greatest sparing of visual recognition (91, 75 and 48% correct for naming visually presented objects, for M.D., D.F. and C.K., respectively). All three patients were more impaired at naming line drawings than at naming real objects [M.D. scored 70% correct in the Boston Naming Test (Kaplan et al., 1982); D.F. failed to name drawings selected from the Snodgrass and Vanderwart (1980) norms (Milner et al., 1991); C.K. named 30% of line drawings from the Boston Naming Test correctly (Behrmann et al., 1992)]. The ability to draw items from memory was assessed for each patient, and in each case it was claimed that the drawings were good and indicated intact visual imagery. M.D. was asked to draw a total of 42 items from memory (20 inanimate and 22 animate objects); analysis of the drawings by naive observers resulted in better identification scores for inanimate relative to animate objects. D.F.’s ability to draw from memory was assessed on two different occasions (total: 32 drawings). Drawings from memory were reported as being much better than her copies of visually presented real objects or line drawings. C.K. was given the names of the 30 items he had failed to name in the Boston Naming Test (Kaplan et al., 1982) and was asked to draw pictures corresponding to the names; Behrmann et al. (1992) state that he performed this task well, failing on only one item (a sea horse). The three patients (M.D., D.F. and C.K.) also performed a number of other tests of imagery (including object imagery, space imagery and mental rotation) (although not all patients were tested on all of these). The three patients generally scored well relative to control subjects on all the tests that were performed. For instance, all three scored highly on various tests of size estimation (e.g. asking whether a horse is larger than a camel). C.K. and D.F. also successfully performed letter imagery tests (e.g. imaging letters to determine whether they had curves or not) and the ‘animal ears’ test [e.g. deciding whether an animal’s ears stand erect above the skull or whether they hang down (see Kosslyn et al., 1985)]. C.K. also performed a number of other imagery tasks (such as supplying the colour associated with an object given its name; determining whether an animal had a long or short tail relative to overall body length; estimation of relative distances between named cities, angle estimation given clock times, and the Brooks letter task) (Behrmann et al., 1992). The additional imagery tasks performed by D.F. included motor imagery, derived from Goldenberg et al. (1989) and image construction tests (see Finke et al., 1989). Since the three patients performed well on all the above tests, claims of intact imagery and long-term visual knowledge for objects were made in each case (see Behrmann et al., 1992; Jankowiak et al., 1992; Servos et al., 1993; Servos and Goodale, 1995). The three cases are significant because the data suggest that the notion of a shared substrate between imagery and perception may be incorrect (Kosslyn, 1994), and that the perceptual impairment does not necessarily lead to poor recovery of long-term visual knowledge. H.J.A., whose case we follow-up here, is profoundly agnosic like M.D., D.F. and C.K. When initially tested, he named only 62% of common real objects from vision relative to a level of 86% naming from touch. Similarly to other patients, he too was worse with line drawings than with real objects (averaging ~40%, correct depending on the particular stimuli). His deficits were exacerbated under conditions that stressed processes of perceptual grouping and integration, e.g. with overlapping figures, with line drawings relative to silhouettes, with fragmented forms and with multiple similar forms in visual search tasks (Riddoch and Humphreys, 1987a; Boucart and Humphreys, 1992; Humphreys et al., 1992b). Riddoch and Humphreys (1987a) proposed that the deficit lay in integrating form information in an efficient, spatially parallel manner. Despite these perceptual problems, H.J.A. demonstrated good long-term knowledge about objects. He produced detailed definitions that typically specified visual as well as functional properties of objects, and he was judged to produce accurate drawings of items from memory. The only test of long-term knowledge on which he showed a clear deficit was when reporting the colour associated with objects (though verbal knowledge of colour was good; he knew that green was the colour associated with envy but not that grapes are green). H.J.A. has cerebral achromatopsia Downloaded from http://brain.oxfordjournals.org/ at New York Medical College on August 22, 2014 Long-term visual knowledge and imagery 539 540 M. J. Riddoch et al. Shared substrate between imagery and perception? Some evidence for there being a shared substrate between perceptual and imagery processes comes from studies of functional brain activation and from studies of patients with cortical blindness. For instance, there have been a number of reports of activation of even the primary visual cortex during imagery tasks, SPECT (Goldenberg et al., 1989), PET (Kosslyn et al., 1993) and EEG (Farah et al., 1988b). However, data obtained in other PET studies have shown activation in occipitoparietal and temporoparietal regions rather than early visual areas in visual imagery tasks (Decety et al., 1992; Roland and Gulyás, 1994). It has been argued that this may be due to differences in the visual imagery tasks (see Policardi et al., 1996). If the primary visual cortex is critical in visual imagery, then loss of it (in cases of cortical blindness) should result in a similar loss or severe impairment to visual imagery. Support for this proposal comes from Policardi et al. (1996) who describe a case (T.C.) who became cortically blind following a road traffic accident. Neuroimaging studies showed bilateral metabolic reduction in calcarine and associative occipital areas, and extending to mesial and temporal cortex. Ophthalmological examination showed that T.C. was completely blind in both eyes (although he denied this deficit); in addition T.C. demonstrated a severe impairment of visual imagery which was tested extensively in a number of different ways (including topographical imagery, symbol imagery, animal and object imagery, and colour imagery). The impaired visual imagery did not result from a degradation of stored knowledge, as T.C. performed well on tests of stored visual and associative/functional information. However, while the case of T.C. supports the contention that there is a common substrate for imagery and perception, other cases of cortical blindness do not. Chatterjee and Southwood (1995) report intact visual imagery in a case of a patient with cortical blindness. The case (Patient 1) suffered a subarachnoid haemorrhage as a result of a right frontal arteriovenous malformation, which caused infarctions of the posterior cerebral arteries. Her imagery abilities were assessed in some detail and she was shown to perform at (or above) control levels in all tests bar one (a test of spatial imagery). A similar case of preserved visual imagery in a case of cortical blindness has been described by Goldenberg et al. (1995). The case, H.S., suffered a bilateral posterior cerebral infarction resulting in destruction of Brodmann’s areas 18 and 19, and atrophy of areas 28, 35, 36 and 37. There was bilateral damage of area 17. H.S. was blind, although she denied this (like T.C., see above). Goldenberg et al. (1995) argued that this was because she confused her mental visual images with real percepts. Assessment of H.S.’s imagery included tests of her ability to access knowledge of the shapes of letters and the shapes and colours of objects. In all instances, she performed well (see Goldenberg and Artner, 1991). Thus, data from cortically blind patients, like the data from agnosic patients (Riddoch and Humphreys, 1987a; Behrmann et al., 1992; Jankowiak et al., 1992; Servos et al., 1993; Servos and Goodale, 1995), do not unequivocally provide support for the proposition that imagery and perception are necessarily coupled. As a result, Goldenberg has argued that image generation results from activation of stored visual knowledge, rather than the activation of early visual areas (Goldenberg and Artner, 1991; Goldenberg, 1992, 1998). Object recognition also requires access to stored visual knowledge, but may not require the same degree of precision of information. For instance, visual recognition of a fox will require the Downloaded from http://brain.oxfordjournals.org/ at New York Medical College on August 22, 2014 (Humphreys et al., 1992c), and his deficit with naming the stored colours for objects suggests that, for colour at least, perceptual information might be drawn on when retrieving information from long-term memory. A further suggestion for a link between perception and long-term visual memory comes from a study of H.J.A.’s memory for facial information. Following his stroke, H.J.A. was profoundly prosopagnosic, failing to identify any famous faces by sight and often failing to identify even the gender of the person (Humphreys et al., 1993). When asked to make judgements about particular facial attributes of people (e.g. ‘Did Harold MacMillan have a moustache?’), H.J.A. performed well (Young et al., 1994). However, he performed less well when asked to make judgements about the configural properties of faces (e.g. ‘Who looks more like Elizabeth Taylor: Joan Collins or Barbara Windsor?’). When identifying faces from vision, H.J.A. shows a piecemeal approach in which he uses individual features rather than configural information (Humphreys and Riddoch, 1987a). Thus H.J.A. was poor at making judgements based on long-term memory for facial properties that he was impaired at using in on-line visual identification (Young et al., 1994). Again, there is some evidence of linkage between perceptual processing and longterm visual memory. It may be that the tests probing longterm visual memory for objects (Riddoch and Humphreys, 1987a) were not sensitive enough to detect a subtle impairment in retrieval, or that, when these tests were given, H.J.A. maintained sufficient visual knowledge to enable him to perform adequately (especially when performance was also supported by excellent semantic knowledge). Performance might then be reasonably good even if our retrieval of long-term visual knowledge does normally recruit perceptual as well as memorial information. This last proposition is assessed in more detail here, where we evaluate his long-term visual knowledge across a broader range of tests than before, and where we evaluate memorial retrieval up to 16 years post-lesion. It is possible that, during this time, there has been some degradation specifically of longterm visual knowledge, in the absence of up-dating from intact perceptual information. Stored visual knowledge in visual agnosia 541 processes play a larger role in the identification of some categories of object than others. Case history matching of a visual percept with stored knowledge about the prototypical shape of a fox; it may not be necessary to activate more specific stored visual information (such as its colour, its size, etc.). However, in order to answer imagery questions (such as ‘Can a fox be described as reddish brown or as yellowish brown?’), it may be necessary to activate more specific visual attributes. Goldenberg uses this account to accommodate cases of intact object recognition but impaired visual imagery (e.g. Farah et al., 1988a). Goldenberg argues that if perceptual processing is disrupted (causing visual agnosia), imagery may still be intact if stored knowledge is unimpaired (e.g. Behrmann et al., 1992; Jankowiak et al., 1992; Servos et al., 1993; Servos and Goodale, 1995). This proposal is subject to a more stringent test here, where we assess the effects of the passage of time on H.J.A.’s long-term visual knowledge for objects. If perceptual and memorial processes are functionally independent, then there should be no necessary effect of H.J.A.’s perceptual deficit on his longterm knowledge for the visual properties of objects. On the other hand, a deterioration in this knowledge would be consistent with perception and memory being linked in a more interactive manner. We also test whether any degeneration in stored knowledge affects some categories of object more than others. Any differential impact of time on particular categories of object will be relevant to the debate concerning category-specific deficits in neuropsychological patients and whether visual memory Downloaded from http://brain.oxfordjournals.org/ at New York Medical College on August 22, 2014 Fig. 1 MRI scan demonstrating lesions of the inferior gyrus, lateral occipital gyrus, fusiform gyrus and lingual gyrus. H.J.A. was aged 77 years at the time of the present investigations. He had suffered a posterior cerebral artery stroke perioperatively in 1981, resulting in a dense visual agnosia, prosopagnosia, alexia without agraphia, achromatopsia and topographical impairments. Neurological and psychological investigations have been reported in detail previously (Humphreys and Riddoch, 1984, 1987b; Riddoch and Humphreys, 1987a). The stroke resulted in bilateral lesions of the occipital lobe, extending anteriorly towards the temporal lobes. An MRI scan in 1989 showed bilateral lesions of the inferior temporal gyrus, lateral occipitotemporal gyrus, the fusiform gyrus, and the lingual gyrus; see Fig. 1. H.J.A. had markedly poor visual recognition affecting objects from all categories (although this was most pronounced with animate objects). Yet, despite these impairments, he had considerable insights into the nature of his disorder; he was able to provide a detailed written account of the difficulties he experienced in everyday life which was included in a book on visual agnosia published in 1987 (Humphreys and Riddoch, 1987b). As noted in the Introduction, he was able to draw from memory objects that he could no longer recognize visually (see Fig. 2), and he was able to provide precise definitions that included details about visual features such as size and shape. H.J.A. showed no indication of a general intellectual deficit. Like Adler’s patient H.C. (Adler, 1944, 1950), H.J.A. is now largely independent in everyday life, although he lives in sheltered accommodation. This report is divided into two main sections. In the first section, we evaluate whether there was any recovery from agnosia. We contrast the result of tests reported in 1987 (Riddoch and Humphreys, 1987a) with those collected on similar tests performed in 1997. We initially report his performance on ‘diagnostic tests’ used to assess and classify his visual object recognition deficit. These tests evaluated (i) whether his disorder was specific to the visual modality (visual versus tactile naming test) and (ii) whether there was impaired access to stored semantic knowledge, and not just name information from vision (using function match and association match tests). In this section, we also report on H.J.A.’s ability to name line drawings and his performance on a number of more detailed perceptual tests, to confirm a diagnosis of a form of apperceptive agnosia (a failure to achieve a coherent percept) as distinct from associative agnosia (a failure to access stored knowledge along with intact perception) (see Lissauer, 1890). In the second section, we go on to assess H.J.A.’s longterm visual knowledge for objects, again contrasting data reported in 1987 with those collected in 1997. Here we 542 M. J. Riddoch et al. Experiment 2: item match test Fig. 2 Drawings from memory contrasting performance on the same items in 1982 (left-side drawings) and 1997 (right-side drawings). This test involved being presented with line drawings, two of physically different instances of the same base-level item (e.g. two horses) and a third, different distractor item. The task was to judge which two items had the same functional identity. This test has been used previously to probe access to stored functional knowledge about objects (Warrington and Taylor, 1978; Riddoch and Humphreys, 1987a). In 1987, H.J.A. scored 12/20 (60%) correct. In 1997, he scored 27/ 32 (84.4%) correct on a standardized set of items from the Birmingham Object Recognition Battery (BORB) (Riddoch and Humphreys, 1993). The improvement in performance over time was just significant [χ2(1) 5 3.8, P , 0.048]. [A small subset of the items used in 1997 was identical to those used in 1987. H.J.A. scored 6/8 correct for these items on both occasions. He made an error with one of the items in both 1987 and 1997.] His performance in 1997 fell within the range of agematched control subjects (see Riddoch and Humphreys, 1993) (range 24–32, mean 30/32, standard deviation 2.2). Experiment 3: naming line drawings test whether there has been degeneration in long-term visual memory, linked to the perceptual deficit. Recovery from visual agnosia? Diagnostic tests Experiment 1: visual versus tactile naming When originally tested, H.J.A. was able to identify 28/45 (62.2%) of real common objects from vision as against 36/ 42 (85.7%) of the same objects from tactile presentation. Performance was better with tactile presentation. When reassessed in 1997, he named 34/39 (87.2%) of a sample of visually presented objects in contrast to 33/39 (84.6%) tactually presented objects. Five age-matched controls performed at ceiling on both tests. [Unfortunately, it was not possible to match items exactly over time. However, 15 of Both in 1987 and in the re-test, H.J.A. was given sets of line drawings of both living things and artefacts to name. On average, the living things tended to have higher levels of perceptual similarity relative to other category members than the artefacts [using measures of overall contour overlap and listed number of parts in common, derived from the Snodgrass and Vanderwart (1980) set of pictures; see Humphreys et al. (1988)]. The 1987 stimuli were Snodgrass and Vanderwart pictures; the retest stimuli were taken from the same set of items (but redrawn for inclusion in BORB). Although the line drawings used in 1997 were slightly modified depictions of the Snodgrass and Vanderwart items used in 1987, these modifications made no difference to control performance which is at the same level for Snodgrass and Vanderwart and BORB depictions. The pattern of performance across time was similar in that H.J.A. was more likely to make errors when naming items from living categories, with high degrees Downloaded from http://brain.oxfordjournals.org/ at New York Medical College on August 22, 2014 the same objects were tested for visual naming on both occasions. H.J.A. operated near ceiling for these items, scoring 14/15 in 1987 and 13/15 in 1997. No item was failed in both 1987 and 1997. Eight of the same objects were tested for tactile naming on both occasions. Again, H.J.A. operated near ceiling for these items, scoring 8/8 in 1987 and 6/8 in 1997.] On re-test, there was no significant difference in his ability to name visually and tactually presented objects [χ2(1) 5 0.02], unlike in 1987. Unfortunately, it was not possible to match the two sets of real objects across test sessions, although overall H.J.A. performed better on visual identification of the items in 1997 compared with 1987, treating each item as an independent sample [χ2(1) 5 6.7, P , 0.01]. Stored visual knowledge in visual agnosia Discussion H.J.A.’s ability to name real objects showed some improvement over time, though his naming of line drawings remained stable. H.J.A. also showed better performance across time on the item match test. However, it was clear from his verbal comments that item match judgements were not being made on the basis of functional/categorical knowledge. For instance, one trial consisted of two different pictures of lamps with a distractor item of a flag. H.J.A. recognized the flag, but not the pictures of the lamps. While he did not know what the lamps were, he decided that neither of them contained the visual features of a flag, and that the flag must be the distractor item in the triad. Other matches were performed correctly based on visual characteristics of the stimulus items (e.g. given two different pictures of squirrels and one of a monkey, he classified the two squirrels as a pair on the basis of their smaller paws). H.J.A. may have improved at using perceptual features to constrain his guesses rather than improving in his ability to recognize line drawings. We therefore administered a further, more stringent test of visual access to semantic knowledge from line drawings. The association match test (Test 12, BORB) (Riddoch and Humphreys, 1993) consists of three pictured items, two of which are related associatively but not visually. The task is to match related items. For a correct match to be performed, at least two of the three items must be recognized. matched control subjects reported by Riddoch and Humphreys (1993) (control range 21–30, control mean 27.5, control standard deviation 5 2.4) and was no better than chance. Discussion The data from the association match test indicate that H.J.A. remains profoundly impaired at recognizing as well as naming line drawings. These findings are consistent with his verbal reports. In a recorded interview in 1997, he stated that ‘. . . my actual recognition has never got any worse or better. My ways of getting around it have improved—but not my recognition, at least that’s what I believe . . .’. H.J.A.’s apparent improvement on some tasks (when naming of real objects and on the item match test) but not others (naming of line drawings) suggests that his ‘ways of getting around the problem’ include using depth and texture information, available in real objects but not in line drawings. His improvement on the item match test suggests increased attention to detail and the development of hypothesis testing to deal with his visual information. In circumstances where matching based on physical similarity is difficult (on the association match test), H.J.A.’s visual impairment is clearly apparent. The selective improvement reported here for real objects is similar to that reported for other agnosic patients (Adler, 1944, Adler, 1950; Kertesz, 1979; Davidoff and Wilson, 1985; Sparr et al., 1991; Wilson and Davidoff, 1993). Tests of the nature of the deficit Clinical tests commonly used to assess whether perceptual processing is intact in agnosia include (i) the ability to copy line drawings and (ii) tests of the ability to segment figure from ground. Of the visual agnosic patients where followup studies have been performed, only J.R. is reported to have intact copying (Davidoff and Wilson, 1985; Wilson and Davidoff, 1993). In the cases described by Kertesz (1979) and Adler (1944, 1950), copying performance was reported to be poor. On formal tests of figure–ground segmentation, H.C.’s performance was poor (Sparr et al., 1991). There are no reports of formal tests of figure–ground segmentation for either Kertesz’s (1979) patient or for Davidoff and Wilson’s patient (1985), although both patients were reported as having difficulty in the interpretation of complex pictures. In our original experiments in 1987 (Riddoch and Humphreys, 1987a), H.J.A. was reported as having a problem with the perceptual integration of form information. He was able to copy items, showing reasonable perception of the basic components of form, but failed when perceptual integration was stressed—with overlapping figures, when object decisions to line drawings were compared with those to silhouettes, and when presentation times were reduced. To test perceptual integration, these tests were repeated here. Experiment 4: association match test Experiment 5: copying This test was administered in 1997, but not in 1987. H.J.A. scored 18/30. His score fell outside the range of the age- In 1987, we reported that H.J.A. had an intact ability to copy line drawings although his method did not appear to be Downloaded from http://brain.oxfordjournals.org/ at New York Medical College on August 22, 2014 of within-category exemplar similarity [structurally similar items (SS)], than when naming artefacts from categories where category exemplars share little visual similarity to each other [structurally dissimilar items (SD)]. In 1987, H.J.A. scored 12/35 and 24/40 for SS and SD items, respectively; in 1997 he scored 17/38 versus 24/38 for SS and SD items, respectively (these last items being matched for name frequency across the categories) (overall, the age-matched control range was 64–76 years, the mean score was 70.3/78, with a standard deviation of 3.2). [A subset of the same objects was used in both 1987 and 1997 (n 5 34). H.J.A. scored 14/34 and 19/34 in 1987 and 1997, respectively (McNemar χ2(1) 5 2.3). He scored 8/24 and 11/24 on SS items in 1987 and 1997, respectively (the same items were named incorrectly 11 times on both occasions). He scored 6/10 and 8/10 on SD items in 1987 and 1997, respectively (he named one item incorrectly on both occasions).] A fully saturated log-linear analysis was conducted, with the factors being time (1987 versus 1997), object type (SS versus SD) and number correct relative to number wrong. There was no effect of time, nor were there any interactions involving this factor (all χ2,1.00). There was a reliable interaction between object type and number correct to wrong [χ2(1) 5 7.24, P , 0.01]. There were proportionally more correct to error responses for SD items. 543 544 M. J. Riddoch et al. Fig. 3 Copies of line drawings. The target items are from the Snodgrass and Vanderwart (1982) picture norms. We contrast copying performance in 1982 (central drawings) and 1997 (rightside drawings). entirely normal (he adopted a slow and apparently slavish line-by-line approach). On reassessment, our observations were the same. His copies are good, but his performance was very slow and laboured. Figure 3 illustrates copies of line drawings in 1982 and 1997. In other unpublished work, we have found that H.J.A.’s copying is often based on incorrect parsing of the elements present. For example, given overlapping figures and instructions to draw the different figures in different colours, he often assigns different colours to common parts, particularly when they are intersected by other edges. This suggests that his drawing is not based on achieving normal perceptual processes, even though the final product has the correct appearance (see Riddoch and Humphreys, 1987a). Experiment 6: overlapping figures In 1987, H.J.A.’s ability to segment visual information was assessed using an overlapping figures test (Riddoch and Humphreys, 1987a, 1993). In 1997, we reassessed performance using letters as stimuli. Sets of letters were presented on a page (either 18 single letters, 18 pairs of letters or 18 triplets of letters). H.J.A. was asked to name each letter on each page as quickly as possible. Naming latencies for each page of letters were timed with a stopwatch. In the paired and triplet conditions, letters were either placed alongside each other (non-overlapping in pairs and triplets) or they overlapped. The non-overlapping conditions served as baselines for the overlapping letter conditions. Performance was assessed by computing the ratio of reaction times (RTs) to name non-overlapping letters relative to overlapping letters. Figure 4 shows H.J.A.’s mean RT per item in 1987 and 1997, together with the means of eight control subjects (Riddoch and Humphreys, 1993). The performance of the age-matched control subjects did not vary as a function of the experimental condition; in particular, naming performance was not slowed in the overlapping relative to the non-overlapping conditions. Relative to the control subjects, H.J.A. showed increased latencies in the all conditions in both 1987 and 1997, and this was particularly marked for the overlapping letters conditions. In both sessions, there was no effect of the proximity of additional items (i.e. naming letters presented in isolation did not differ from naming letters presented in pairs or triplets), and H.J.A.’s RTs fell within the control range; however, when letters overlapped, there was a marked effect on H.J.A.’s latencies. The mean naming latency per letter for H.J.A. was 0.6 s in the single letter, paired nonoverlapping and triplets non-overlapping conditions (the mean naming latency for the control subjects was 0.4 s with a range of 0.3–0.6 in each of the three conditions). However, in the paired overlapping condition, the mean naming latency per letter was 1.5 and 0.4 s for H.J.A. and the control subjects, respectively (control range 0.3–0.7). The nonoverlapping : overlapping ratio was 1 : 2.0 and 1 : 1.2 for H.J.A. and the control subjects, respectively. In the triplets overlapping condition, the mean naming latency per letter was 1.1 and 0.4 s for H.J.A. and the control subjects, respectively (control range: 0.3–0.7). The non- Downloaded from http://brain.oxfordjournals.org/ at New York Medical College on August 22, 2014 Fig. 4 Histograms illustrating the reaction times (per letter) in the overlapping figures test. Contrasts are made between H.J.A.’s performance in 1987 and 1997 and that of eight control subjects. Stored visual knowledge in visual agnosia overlapping : overlapping ratio was 1 : 1.7 and 1 : 1.1 for H.J.A. and the control subjects, respectively. The only evidence for any improvement over time for H.J.A. was in the overlap condition with two letters, but this may have been due to a speed–error trade-off. In 1987, H.J.A. made no errors in letter report. Here he made 12 errors overall, nine with overlapping items (four with paired items, five with triplets). We conclude that H.J.A. remains impaired with overlapping letters, and that the deficit does not seem to have receded in any significant way. In 1987, H.J.A. appeared unable to access stored knowledge of object shape from vision, since he was impaired even on object decision tasks which do not require either naming or even access to semantic knowledge (Riddoch and Humphreys, 1987a, b). In the earlier study, the non-objects were constructed in each of two ways: (i) by replacing a feature of the line drawing of one object with a feature of a different object (feature-replaced condition) and (ii) by adding a feature to a whole object from a different line drawing (featureadded condition). The original version of the test consisted of 30 feature-replaced and 30 feature-added non-objects together with 60 unadulterated line drawings. In 1997, a reduced set of items was administered (28 feature-added items, 12 feature-replaced items and 40 unadulterated line drawings). [Unfortunately, it was not possible to contrast performance in 1987 and 1997 with the matched subset of items. However, five controls performed at the same level with this subset as did the controls given the full set.] In 1987, H.J.A. scored 69/120 (57.5%) correct (in contrast to the controls scoring a mean of 96.4% correct, with a range of 112–118), and in 1997 he scored 52/80 (65%) correct (control mean: 96.3% correct, range 73–79). For H.J.A., performance did not differ across time [χ2(1) 5 1.1]. As was reported in 1987, the rejection of non-object stimuli was slightly better if they were feature-added (24/28) than if they were feature-replaced (8/12); however, this difference was not significant [χ2(1) 5 1.9]. We conclude that H.J.A.’s ability to access stored knowledge of object form from vision remains impaired and has improved little over time. understood if H.J.A. tends to be disrupted by the internal segmentation cues in line drawings which weigh against any benefit of extra features present. The effect of using silhouettes was again assessed using a subset of the line drawings (as in 1987, stimuli were omitted if they could only be recognized on the basis of internal details). H.J.A. scored 48/67 (71.64%) correct for the silhouettes and 43/67 (64.18%) correct for the same items presented as line drawings. There was again a trend for better performance with silhouettes, though this difference did not prove to be significant [McNemar χ2(1) 5 0.14]. His performance with silhouettes was not significantly different from that reported in 1987 (63/88, 71.59% in 1987 versus 48/67, 71.64% in 1997) [χ2(1) 5 0.04]. As suggested by Riddoch and Humphreys (1987a), H.J.A.’s lack of a silhouette deficit is consistent with him tending to use global outline shape for object recognition. Again, these data point to the fundamental perceptual deficit in H.J.A. remaining quite stable over time. Experiment 8: tachistoscopic identification As noted in the Introduction, on first testing, H.J.A. showed strong effects of exposure duration on object naming. This was retested here. Twenty line drawings were presented tachistoscopically each at 100, 500, 1000 ms and unlimited exposure. The line drawings were selected as those that H.J.A. had identified correctly on at least one occasion in previous testing sessions. Stimuli were presented blocked for duration and starting with the longer exposure durations via a Macintosh Powerbook using VScope software (Rensink and Enns, 1992). Drawings were centrally presented and subtended a visual angle of ~2° in height and 2° in width. Each drawing was preceded by a fixation cross for 150 ms. H.J.A. was asked to name each depicted object. He named 15/20 of the items given unlimited exposure, and 14/20, 13/ 20 and 14/20 items at 1000, 500 and 100 ms exposure durations, respectively. In 1987, he scored 16/20, 9/20, 8/20 and 3/20 at the same durations of exposure. His performance in 1997 did not differ from that reported in 1987 at unlimited exposures or at the 1000 ms exposure [Fisher’s Exact Probability 5 0.7, and χ2(1) 5 0.04, respectively]; however, performance in 1997 was significantly better for the 500 and 100 ms exposures [χ2(1) 5 4.4, P , 0.04, and χ2(1) 5 12.4, P , 0.0004, respectively]. Five age-matched controls performed at ceiling with line drawings presented for 100 ms. (B) Object decision with silhouettes One interesting pattern of performance shown by H.J.A. in 1987 was that his object decision performance was little affected by making the stimuli into silhouettes. In contrast to this, normal subjects find silhouettes more difficult [in Riddoch and Humphreys (1987a) controls scored 96.4 and 88.3% correct for line drawings and silhouettes, respectively; Lawson and Humphreys (1998) report a similar result but using a picture–word verification task]. This pattern can be Discussion H.J.A.’s performance with the perceptual tests reported here generally differed little from that reported initially in 1987. Performance remained poor with overlapping figures and on the object decision test, and he showed a trend again for better object decisions with silhouettes than with line drawings (the opposite effect to normal). The main change we observed was some improvement in his ability to identify line drawings Downloaded from http://brain.oxfordjournals.org/ at New York Medical College on August 22, 2014 Experiment 7: object decision (A) Object decision with line drawings 545 546 M. J. Riddoch et al. from relatively brief exposures, so that, in 1997, there was little effect of exposure duration on identification. It is possible that this reflects the fact that H.J.A. has become a very experienced experimental subject over the time course of his visual agnosia, and that he is better able to rely on visual short-term memory and partial visual information for object identification than he was in 1987. He may also have benefited more from contextual priming, since the longer exposure durations preceded the shorter ones. However, with unlimited exposures, his basic problems in identifying line drawings and in dealing with overlapping figures have remained essentially stable. When initially tested after his stroke, H.J.A. demonstrated intact stored knowledge about objects. For instance, he was able to give detailed definitions of objects and he could draw them from memory (see Fig. 2 for examples). Also, when given the names of items, he scored 100% correct on questions probing general knowledge and 97.3% on questions probing detailed knowledge. Performance was significantly poorer in answering questions about pictures (H.J.A. scored 83.3% correct with general knowledge questions and 65.8% correct with specific knowledge questions). In our original report, the assessment of H.J.A.’s drawing from memory was based on a small number of items (n 5 9) (Riddoch and Humphreys, 1987a). Subsequently, he was asked to draw a much larger group of items from memory, and also, on separate occasions, to provide a written description of each item. These data were collected in 1985 (4 years after his stroke and resultant visual agnosia). Ten years later (in 1995), the same data were collected again to determine whether there was a deterioration in stored knowledge over time. These results are reported in this section, and we also report other tests of H.J.A.’s ability to retrieve visual knowledge from memory in 1997. Drawings and definitions Experiment 9: drawing from memory Method The ability to produce an accurate depiction of a named item from memory was assessed in H.J.A. and three age-matched control subjects (one male and two females). Neither H.J.A. nor the control subjects had any special expertise in drawing. For all subjects, 76 A4 sheets of plain paper were provided. At the head of each sheet was the name of an item, the task was to draw each item below its name. The 76 items used were the same set of objects as those employed as line drawings in Experiment 3. They included equal numbers of SS and SD items [see Experiment 3 and Humphreys et al. (1988) for definitions], and equal numbers of high and low name frequency items. Full details of the selected items are given in Humphreys et al. (1988). The drawings obtained Results Two of the drawings were omitted from the 1985 data set as H.J.A. had misread the item name at the top of the sheet (i.e. he misread DOG as JUG and MOUSE as HOUSE). This resulted in 18 items in both the SS high name frequency (SS HF) and low name frequency (SS LF) categories. The missing data were replaced by the mean score for that particular category. There were 19 items each in the SD high (SD HF) and low (SD LF) name frequency categories. No drawings were omitted from the 1995 data set, or from the control subjects sets, resulting in 19 items in each category in all cases. Analysis of the control data. Since drawing ability is fairly idiosyncratic, the data from the control subjects were analysed for differences. The identification scores and accuracy scores for each subject were analysed using independent design ANOVAs, with subject (Control 1, Control 2, Control 3), structure (SS versus SD) and frequency (HF versus LF) being the main factors. The mean identification scores for the 76 drawings were 0.80, 0.72 and 0.65 for Controls 1, 2 and 3, respectively (where a score of 1.0 represents identification of that drawing by all the raters). The mean accuracy scores were 5.40, 4.48 and 4.27 for Controls 1, 2 and 3, respectively (a score of 7.0 would indicate an accurate representation of that item). (i) Identification scores: significant main effects were found for subject [F(2,216) 5 4.59, P , 0.01] and structure [F(1,216) 5 12.22, P , 0.0006]. The main effect of frequency was not significant [F(1,216) 5 0.38]. The second order interaction between structure and frequency was significant [F(1,216) 5 5.72, P , 0.018], but the other second order interactions proved unreliable [F(2,216) 5 1.93 and F(2,216) 5 1.46 for subject 3 structure and subject 3 frequency, respectively]. The third order interaction was also unreliable [F(2,216) 5 0.59]. For the controls, drawings of items with low name frequencies tended to be identified more easily than drawings of items with high name frequencies, particularly for items from structurally similar classes. This ‘negative’ effect of name frequency may be because items Downloaded from http://brain.oxfordjournals.org/ at New York Medical College on August 22, 2014 Intact stored knowledge in visual agnosia? from H.J.A. and the control subjects were assessed in two ways. First, new independent raters were asked what they thought each drawing depicted, and second, given the name of the drawing, they were asked to rate how good a representation had been produced. H.J.A.’s drawings were assessed by 20 raters. The drawings of the control subjects were assessed by a different group of 10 raters. In both instances, a different randomized order of stimuli was used for each rater. The rating scale ranged from 1 to 7, where a score of 1 was indicative of a poor representation and 7 indicated a good representation. An identification score (derived from the proportion of raters who were able to identify the drawing) and an accuracy score (obtained from the average of the rating scores for each picture) for H.J.A. and the control subjects was calculated. Stored visual knowledge in visual agnosia Identification of drawings. H.J.A.’s data initially were analysed in order to contrast his performance in 1985 and 1995. His performance at both these time periods was then compared with that of the worst control subject. Identification scores (H.J.A.’s 1985 versus 1995 data). The data were analysed using a mixed design (one within, two between) ANOVA, with time (1985 versus 1995) being the repeated measures factor and structure (SS versus SD) and name frequency (HF and LF) being the between-items factors. Significant main effects of structure [F(1,72) 5 59.6, P , 0.0001] and time [F(1,72) 5 5.2, P , 0.03] were found. The main effect of frequency was not significant [F(1,72) 5 0.6]. The two-way interaction between structure and frequency was significant [F(1,72) 5 9.5, P , 0.003], but neither the other two-way interactions [F(1,72) 5 0.79 and F(1,72) 5 0.04 for structure 3 time and frequency 3 time, respectively] nor the three-way interaction [F(1,72) 5 0.24] were reliable. The data are presented in Fig. 5. The structure 3 frequency interaction occurred because, in the SD set (of objects), items with high name frequencies tended to be identified better than those with low name frequencies; for the SS items (natural objects), low frequency items tended to be better, presumably because these items tend to be visually more distinctive (see below). Identification scores (H.J.A.’s drawings versus those of the Control 3). Comparisons of performance between H.J.A. and the control subject were analysed in separate ANOVAs for the 1985 and 1995 sets. For drawings produced in 1985, the data were analysed using an ANOVA with subject (H.J.A. versus the control), structure (SS versus SD) and name frequency (HF and LF) as factors. Significant main effects of subject [F(1,144) 5 6.86, P , 0.01] and structure [F(1,144) 5 47.39, P , 0.0001] were found. The main effect of frequency was not significant [F(1,144) 5 0.42]. The two-way interaction between subject 3 structure was reliable [F(1,144) 5 5.02, P , 0.03], as was the interaction between structure 3 Fig. 5 Histograms illustrating the proportion of H.J.A.’s drawings (1985 and 1995) and the control subject’s drawings from memory that were identified by raters as a function of category of the items (high name frequency, HF; low name frequency, LF). frequency [F(1,144) 5 4.90, P , 0.03]. The two-way interaction between subject 3 frequency and the three-way interaction were not reliable [subject 3 frequency F(1,144) 5 2.29, and subject 3 structure 3 frequency F(1,144) 5 1.8]. Post-hoc analysis (Student–Newman–Keuls) of the subject 3 structure interaction showed that while H.J.A. and Control 3 performed at essentially the same level for SD items, Control 3 performed significantly better than H.J.A. with SS items. In general, both H.J.A. and Control 3 performed better with SD than with SS items. Collapsing across subjects, post hoc analysis (Student–Newman–Keuls) of the structure 3 frequency interaction showed significantly better identification (at the 0.01 level) of SD than of SS items (irrespective of name frequency). In addition, SS LF items were identified better than SS HF items (at the 0.05 level). For drawings produced in 1995, there were significant main effects of subject [F(1,144) 5 17.50, P , 0.0001] and structure [F(1,144) 5 32.16, P , 0.0001]. The main effect of frequency was not significant [F(1,144) 5 0.45]. None of the two-way interactions were reliable [subject 3 structure: F(1,144) 5 1.58, structure 3 frequency: F(1,144) 5 3.05, subject 3 frequency: F(1,144) 5 1.92]. The three-way interaction was also not significant (subject 3 structure 3 frequency) [F(1,144) 5 0.85]. Stepwise regression analyses were conducted to assess in more detail the factors that affected the identification by raters of H.J.A.’s drawings from memory. They were conducted including the factors: name frequency, item familiarity, item complexity, image agreement (all taken from Snodgrass and Vanderwart, 1980), rated prototypicality and outline contour overlap within the category (using measures reported in Humphreys et al., 1988). For structurally dissimilar items (artefacts), the only reliable predictor of identification was Downloaded from http://brain.oxfordjournals.org/ at New York Medical College on August 22, 2014 with low name frequencies often have more distinctive perceptual properties (e.g. helicopter, elephant). Post hoc analyses (Student–Newman–Keuls) revealed that Control 3 performed significantly more poorly than Control 1. There was no difference in performance between Controls 1 and 2, or between Controls 2 and 3. (ii) Ratings: significant main effects were found for subject [F(2,216) 5 21.03, P , 0.0001] and structure [F(1,216) 5 18.16, P , 0.0001]. The main effect of frequency was not significant [F(1,216) 5 0.77]. None of the interactions were reliable (all F , 0.1). Post-hoc analyses (Student–Newman– Keuls) revealed that Controls 2 and 3 performed significantly more poorly than Control 1. There was no difference in performance between Controls 2 and 3. The analyses show that Control 3 performed more poorly in general than the other two control subjects. H.J.A.’s performance subsequently was compared with that of Control 3 (the worst control subject). 547 548 M. J. Riddoch et al. an overall positive frequency effect for structurally dissimilar items and a negative effect for structurally similar items. Rating scores (H.J.A.’s drawings versus those of control subject 3). Fig. 6 illustrates the difference in familiarity, in 1995 [F(1,36) 5 4.12, P , 0.05]. In 1985, none of the factors reliably predicted performance. In 1995, the more familiar items were more identifiable. For structurally similar items (natural kinds), contour overlap predicted performance in both 1985 and 1995 [F(1,36) 5 14.14 and 8.00, respectively, both P , 0.01]. Drawings of items with high contour overlap were more difficult to identify. With the variance due to contour overlap extracted, there was also an effect of name frequency, for 1995 only [F(2,35) 5 6.61, P , 0.01]. Items with high name frequencies were actually identified better than those with low name frequencies. Rating line drawings. Rating scores (H.J.A.’s 1985 versus 1995 data): the data were analysed using a mixed design (one within, two between) ANOVA, with time (1985 versus 1995) being the repeated measures factor, and structure (SS versus SD) and name frequency (HF and LF) being the between-items factors. Similar results were found with the rating of goodness of H.J.A.’s drawing as with the identification of the drawings. Significant main effects of structure [F(1,72) 5 94.45, P , 0.0001] and time [F(1,72) 5 20.05, P , 0.0001] occurred. The main effect of frequency was not significant [F(1,72) 5 0.03]. The two-way interaction between structure and frequency was reliable [F(1,72) 5 7.07, P , 0.01], but neither the other two-way interactions [F(1,72) 5 1.26 and F(1,72) 5 0.2 for structure 3 time and frequency 3 time, respectively] nor the three-way interaction [F(1,72) 5 0.06] were significant. The data are presented in Fig. 6. The structure 3 frequency interaction again reflects Downloaded from http://brain.oxfordjournals.org/ at New York Medical College on August 22, 2014 Fig. 6 Histograms illustrating the rating scores for H.J.A.’s drawings from memory relative to those of the control subject as a function of time (1985 and 1995), structure (SS versus SD) and name frequency (HF versus LF). Scores of 0 and 7 reflect low and high estimates, respectively, of the pictured representation. ratings per category of picture for the two sets of drawings produced by H.J.A. and the control subject’s drawings. Comparisons of performance between H.J.A. and the control subject were analysed in separate ANOVAs for the 1985 and 1995 set. For drawings produced in 1985, the data were analysed using an ANOVA, with subject (H.J.A. versus the control), structure (SS versus SD) and name frequency (HF and LF) as the factors. A significant main effect of structure [F(1,144) 5 74.81, P , 0.0001] was found. The main effects of subject [F(1,144) 5 0.38] and frequency [F(1,144) 5 0.97] were not significant. The two-way interaction between subject and structure was reliable [F(1,144) 5 14.06, P , 0.0003], but neither the other two-way interactions [F(1,144) 5 0.62 and F(1,144) 5 2.12 for subject 3 frequency and structure 3 frequency, respectively] nor the three-way interaction (subject 3 structure 3 frequency) [F(1,144) 5 2.86] were significant. A post hoc analysis (Student–Newman–Keuls) of the subject 3 structure interaction showed that H.J.A.’s drawings of SD items were rated as better representations than Control 3’s drawing of the same items; however, the control’s drawings of SS items received higher ratings than those for the same items drawn by H.J.A. (both P , 0.05). For both Control 3 and H.J.A., SD items received higher ratings than SS items (P , 0.01). For drawings produced in 1995, significant main effects of subject [F(1,144) 5 8.52, P , 0.004] and structure [F(1,144) 5 49.76, P , 0.0001] were found. The main effect of frequency was not significant [F(1,144) 5 0.35]. The twoway interaction between subject and structure was reliable [F(1,144) 5 5.47, P , 0.02], but neither the other two-way interactions [F(1,144) 5 1.24 and F(1,144) 5 1.5 for subject 3 frequency and structure 3 frequency, respectively] nor the three-way interaction (subject 3 structure 3 frequency) [F(1,144) 5 2.1] were significant. A post hoc analysis (Student–Newman–Keuls) of the subject 3 structure interaction showed that drawings of SD items (whether by H.J.A. or Control 3) received higher ratings than drawings of SS items, with the difference between H.J.A. and the control being most pronounced on SS items (P , 0.01). For regression analyses, similar stepwise regression analyses were performed to those conducted with the identification data. For SD items, the only reliable predictor of H.J.A.’s drawing performance was complexity, for 1995 only [F(1,36) 5 7.9, P , 0.01]. More complex items were rated less highly. For SS items, the best predictor was contour overlap, in both 1985 and 1995 [F(1,36) 5 9.43 and F(1,37) 5 12.22, both P , 0.01]. In 1995, there was also a reliable, independent effect of prototypicality [F(2,36) 5 8.9, P , 0.01]. The more prototypical items were given higher ratings. Stored visual knowledge in visual agnosia 549 Discussion Fig. 7 Deterioration of visual respresentations over time? In the top section of the figure are two drawings from 1985 (celery to the left, an owl to the right). In the bottom section, are representations of the same two items drawn in 1995. deteriorated with age. Against this, H.J.A.’s copying remained accurate (see Fig. 3). We also know of no evidence that drawing skills decrease with age. (b) The deterioration occurs as a result of a specific degradation of H.J.A.’s visual memories which may be particularly susceptible to the effects of time in the absence of any reinforcing input (see Fig. 7 for examples of H.J.A.’s drawings from memory of celery and an owl in 1985 and 1995, respectively). We will return to discuss this proposition in more detail in the next section. (iv) H.J.A.’s drawings of SS items were worse than those of the poorest control, whilst his drawings of SD items differed little (in fact his 1985 drawings of SD items were rated as better than those of the control!). There does not seem to be a general deficit in drawing skill here but rather in depicting the features of objects from categories where objects have overlapping perceptual structures (SS items). Each of the experimental subjects (H.J.A. and the control subject) produced a subset of the drawings that the raters found difficult to identify. For H.J.A., three of the drawings from the 1985 set were not identified by any of the 20 raters (potato, beetle and lips); and in the later set of drawings (those performed in 1995), the numbers of drawings raters failed to recognize increased to 10 (lemon, lion, pepper, sheep, beetle, celery, gorilla, mushroom, strawberry and saltcellar) (all SS items apart from saltcellar). A number of the control subject’s drawings were not identified by any of Downloaded from http://brain.oxfordjournals.org/ at New York Medical College on August 22, 2014 The data on drawing from memory are essentially the same, irrespective of whether scores are based on the number of items that could be identified or on ratings of the quality of drawings. The main findings were: (i) Drawings of SS items received lower rating scores and were more difficult to identify than drawings of SD items; this effect was particularly marked for H.J.A. This is probably due to the fact that SS items (unlike SD items) often share a high degree of similarity with other category exemplars, which may make them more difficult to depict pictorially. Indeed, contour overlap across items was the best predictor of H.J.A.’s drawing of SS items. (ii) The overall identification scores for SS categories tended to be higher for LF than for HF items, while for H.J.A. the reverse occurred for SD items (higher HF than LF ratings). For H.J.A.’s drawings of SD items, identification scores also tended to be higher for more familiar items, as revealed in the regression analysis. Since familiarity and frequency are correlated (see Humphreys et al., 1988), it is difficult to assess which of these factors is crucial. For the controls, there tended to be negative effects of frequency for both SS and SD items, suggesting that the presence of distinctive perceptual features was more important than frequency or familiarity. For H.J.A., however, it appears that frequent encounters with SD objects, perhaps including touching as well as naming, are beneficial for drawing from memory. Note that SD items will tend to be touched more than SS items; the result is consistent with visual memories being reinforced via the somatosensory system. The apparent negative effect of name frequency on the drawings of SS items is misleading. In the regression analysis for H.J.A., contour overlap accounted for most variance (for both identification and rating measures), at both time intervals. H.J.A. was less good at generating drawings for items with high perceptual overlap relative to their neighbours. When the variance attributable to contour overlap was extracted, there was a positive effect of name frequency on drawing performance (as with SD items). We again attribute this to increased contact with these items. [The same argument can be applied to the positive effect of prototypicality on the drawing of SS items, 1995, for the rating data.] The negative frequency effect in the overall analysis is likely because SS items with high name frequencies tend also to have perceptual structures that overlap with their perceptual neighbours. (iii) H.J.A.’s drawings deteriorated significantly when he was tested in 1995 relative to 1985 (whether items were SS or SD). We can put forward at least two reasons for this. (a) The observed deterioration is simply a function of age in general. If this were the case, we might expect H.J.A.’s performance to have also deteriorated on other tests that he performed at the differing time intervals. As we have reported earlier, however, this is not the case, although it could be argued that drawing from memory requires more manual skill than the other tasks, and that it is this skill that has 550 M. J. Riddoch et al. the raters (fox, onion, orange, pipe and doorknob). There were also subsets of items produced by both the experimental and control subject that all the raters were able to identify. In H.J.A.’s 1985 set, 10 drawings (all SD items) were identified by all 20 raters (aeroplane, cup, leg, nut, ring, shoe, television, kettle, nail and necklace); in the 1995 set, the number of drawings identified by all 20 raters dropped to five (aeroplane, axe, pipe, shoe and television) (all SD items). Nineteen of the control subject’s drawings were identified by all the raters. These consisted of six SS items (lion, sheep, elephant, owl, bee and strawberry) and 13 SD items (clock, leg, broom, kite, ruler, scissors, stool, toaster, saltcellar, kettle, helicopter, TV and airplane). H.J.A. was given 76 sheets of paper, each sheet bearing the name of a single item at the top. The names corresponded to those used in Experiment 9 (drawing from memory). H.J.A. was asked to provide a definition for each of the words without using a dictionary. We were interested to discover whether H.J.A. was able to provide both visual and functional information about the different objects, and he specifically was requested to provide both forms of information in the instructions for the task. The instructions asked him to include the following information if possible (the first three points cover visual information about the object, the second three points functional and associative information): (i) what the object looks like; (ii) the typical colour of the object (if it has one); (iii) the size of the object; (iv) what the object may be used for; (v) where the object may be found; and (vi) items that may be associated with the object. The same procedure was followed twice, in 1985 and in 1995. A similar method of scoring was used for the definitions as had been used for the drawing from memory. Raters were first asked if they could identify the item from the written definition, and then (having been told the correct name of the item) to rate the accuracy of the description on a 1–7 scale. Different raters were used in Experiment 10 relative to Experiment 9. Finally, a further analysis was then performed on the definitions, which were analysed for the number of attributes given (according to the list above) and for the number of words. This was conducted independently by three judges. When producing written definitions, H.J.A. made a number of writing errors, a fact observed in other agnosic patients (Adler, 1944, 1950; Kertesz, 1979; Sparr et al., 1991). These writing errors were also assessed at the two time intervals. Ten undergraduate students (aged between 18 and 23 years) acted as raters (five male, five female). SS HF SS LF SD HF SD LF 1985 definitions 1995 definitions 0.90 0.93 0.93 0.87 0.92 0.87 0.92 0.86 description of an object written by H.J.A. using different random orderings of the total corpus of descriptions. If H.J.A. named any item within the text of an individual definition, the name was blanked out when the item subsequently was scored. Seventy-six descriptions were written in 1985 and 74 in 1995 (two items had to be omitted from the 1995 corpus as H.J.A. misread the words to be described; KETTLE was misread as RATTLE, and DOORKNOB as DOORKNOCKER). The missing data slots were filled with the mean score of that particular category for the subsequent analysis. NUT and PEPPER were also not included, as the definition of a seed was given for NUT rather than a definition of a tool, and the definition of a spice was given for PEPPER rather than the description of a fruit. Procedure Each booklet was prefaced with a set of instructions asking raters to read through each definition carefully and to try to identify what was being described. In the second part of the experiment, raters were asked to reread the definitions and to rate each for its adequateness (given the object name) on a scale of 1 (poor) to 7 (excellent). Five raters were assigned to each booklet, three raters working through the booklet from front to back and two from back to front. Results Naming the items from the definition. The proportions of raters able to identify the items successfully from the written definition are presented in Table 1. In each condition, performance was near ceiling. The data were analysed using a three-factor analysis of variance (one within factor, and two between factors). The within-items factor was time (1985 versus 1995), and the between-items factors were structure (SS versus SD) and frequency (HF versus LF). None of the effects were significant (all F , 1.0). Rating the definitions for adequacy. The mean ratings of subjects for items in each condition are presented in Table 2. Again, the scores in each condition were near ceiling. Although the data were analysed using the same design as above, no factors were significant (all F , 1.0). Materials Two 150-page booklets were constructed for this study. Each booklet contained 150 A4 sheets each with a single verbal Number of attributes: totals. The total number of attributes in the 1985 corpus of definitions was 557 (the total Downloaded from http://brain.oxfordjournals.org/ at New York Medical College on August 22, 2014 Experiment 10: written definitions from memory Method Table 1 Proportions of correct identifications of written definitions as a function of condition Stored visual knowledge in visual agnosia Table 2 Mean rating scores for written definitions as a function of condition SS HF SS LF SD HF SD LF 1985 Definitions 1995 Definitions 5.51 5.79 5.78 5.74 5.56 5.69 5.80 5.75 Table 3 Mean number of attributes in H.J.A.’s definitions as a function of the category of the items and their name frequency SS HF SS LF SD HF SD LF 1995 Visual Verbal Visual Verbal 3.2 3.8 2.7 2.8 4.2 4.8 3.7 4.1 2.4 3.5 2.7 2.7 5.4 5.4 4.7 4.7 words used in the descriptions was 6484), and for the 1995 set of definitions it was 569 (the total number of words used was 6946). The total numbers of visual attributes went down over time (from 237 in 1985 to 203 in 1995). On the other hand, the total numbers of verbal attributes increased (from 320 in 1985 to 366 in 1995). A detailed breakdown of the mean number of attributes per structural and name frequency category is given in Table 3. The data were analysed using a four-factor analysis of variance. The main factors were category (SS versus SD), name frequency (HF versus LF), time (1985 versus 1995) and attribute type (visual versus verbal). A significant effect was shown for all the main factors: [for category, F(1,72) 5 11.99, P , 0.0009; for frequency, F(1,72) 5 3.98, P , 0.05; for time F(1,72) 5 88.32, P , 0.0001; for attribute type, F(1,72) 5 4.11, P , 0.05]. Only the attribute type 3 time two-way interaction was significant [F(1,72) 5 19.48, P , 0.0001]. All the remaining interactions were not significant. The attribute type 3 time interaction occurred because the overall number of visual attributes given by H.J.A. decreased while the overall number of verbal attributes increased in the 1985 relative to the 1995 definitions (e.g. contrast the definitions for a CAT given in 1985 and 1995 shown in Fig. 8). As indicated in Table 3, the decrease in visual attributes was particularly associated with the SS HF items (from 3.2 to 2.4 on average). There was no change in the number of visual attributes recorded for SD items (a mean of 2.75 in 1985 and 2.70 in 1995). The drop in visual attributes for SS LF items was relatively small (3.8 to 3.5). Artefacts (i.e. SD items) are not usually described in terms of visual characteristics but more usually in terms of function; in contrast, living things typically are described in terms of their physical characteristics (Farah and McClelland, 1991). If H.J.A.’s visual knowledge is degrading over time due to impaired perceptual input, then we would expect such items to be affected differentially as we have indeed shown. One possibility is that SS items, being more similar within their categories, show stronger effects of degraded perceptual input than SD items. The larger effect with similar items having high name frequencies may be because these items tend to have more typical shapes within their categories. Without updating from normal perceptual input, the visual properties that differentiate these items appear to be degraded. Another possibility is that visual attributes are weighted more strongly in the representation of living things relative to the representations of artefacts (this could hold even if living things were not more structurally similar). Degradation of visual knowledge may then affect living things more. Number of attributes: regression analyses. The number of visual and verbal attributes listed by H.J.A. were assessed in a regression analysis using the same factors as those used in the analysis of his drawings (Experiment 9). For SS items, the retrieval of visual attributes was accounted for by contour overlap [F(1,75) 5 11.98, P , 0.001]. Year exerted a negative effect, though this was only marginally significant [F(2,73) 5 3.0, P , 0.1]. For SD items, recall of visual attributes was not accounted for by any factor. Report of the verbal attributes of SS items was affected most strongly by year [F(1,75) 5 13.55, P , 0.001], then by name agreement [F(2,73) 5 14.66, P , 0.001], and by prototypicality [F(3,72) 5 11.55, P , 0.001]. Report of the verbal attributes of SD items was affected by year [F(1,75) 5 7.83, P , 0.01], then by familiarity [F(2,73) 5 7.24, P , 0.01], and by prototypicality [F(3,72) 5 6.40, P , 0.001]. For SS items, visual attributes were poorly reported for items high in contour overlap, and fewer attributes tended to be reported across time. Report of verbal attributes for both SS and SD items improved over time, and it was better for more typical items. High name agreement (for SS items) and high prototypicality (for SD items) also improved recall of verbal attributes. Number of words. The number of words in each corpus of definitions was calculated. The total number of words produced in 1985 and 1995 was 6484 and 6946, respectively. The mean number of words per category of definition is given in Table 4. Significant mean effects were found for structure [F(1,72) 5 6.4, P , 0.01] and time [F(1,72) 5 12.3, P , 0.0008]. The main effect of frequency was not significant [F(1,72) 5 0.002]. None of the interactions were found to be reliable [F(1,72) 5 0.31, F(1,72) 5 1.5 and F(1,72) 5 0.03 for structure 3 frequency, time 3 structure and time 3 frequency, respectively; F(1,72) 5 0.45 for structure 3 frequency 3 time]. The 1995 set of definitions overall contained more words than the 1985 set, with more words being used to describe SD than SS items (particularly in 1985). Downloaded from http://brain.oxfordjournals.org/ at New York Medical College on August 22, 2014 1985 551 552 M. J. Riddoch et al. Table 4 Mean number of words in both the 1985 and 1995 sets of definitions as a function of the category of the items and their name frequency SS HF SS LF SD HF SD LF 1985 1995 93.0 93.2 77.9 77.2 97.6 103.0 94.3 90.4 the total corpus of words, a number of letters were used interchangeably: ‘p’s and ‘b’s (n 5 28), ‘g’s and ‘c’s (n 5 15), ‘c’s and ‘s’s (n 5 14), ‘t’s and ‘l’s (n 5 14), and ‘i’s and ‘e’s (n 5 12). In addition, there was a tendency for some letters to be replaced by others (‘u’ was replaced by ‘i’ or ‘o’, n 5 24; ‘k’ was replaced by ‘c’, n 5 7; and ‘w’ was replaced by ‘v’, n 5 5). Discussion Nature of written errors. For the 1985 set of definitions, H.J.A. made writing errors on 103 words (representing 1.6% of the total number of words used); for the 1995 set, he made 133 errors on words (representing 1.9% of the total words used). Errors could be classified as either substitutions (69.9 and 73.7% of the 1985 and 1995 definitions, respectively), omissions (18.4 and 18.0% of the 1985 and 1995 definitions, respectively) or additions (16.7 and 8.3% of the 1985 and 1995 definitions, respectively). Although there was a small numerical increase in spelling errors, this only represented 0.3% of the total word corpus and was not significant (χ2 , 1.0). There was no evidence that writing errors were tied to the positions of letters in words, rather they tended to be tied to particular letters; for instance, in When H.J.A.’s written definitions were scored in terms of their adequacy and whether items could be named from them, no effect of time was found. This contrasts with the effects of time shown in the quality of H.J.A.’s drawings. H.J.A.’s good ability still to provide nameable definitions goes against an account of deterioration in his drawings due to general ageing effects. Furthermore, H.J.A. used more words in the 1995 set of definitions than in the 1985 set, and the number of verbal attributes listed increased. There is no evidence of a general decrease in his semantic knowledge about objects. Nevertheless, there was a deterioration in the number of listed visual attributes in his definitions, and this was particularly the case for SS items (natural kinds) with high frequency names. We note again that the frequency itself was not the likely Downloaded from http://brain.oxfordjournals.org/ at New York Medical College on August 22, 2014 Fig. 8 Examples of H.J.A.’s written definitions of a cat in 1985 (left) and 1995 (right). Stored visual knowledge in visual agnosia neuropsychological literature to examine patients’ long-term visual knowledge for stimuli. The tests were also given to 12 undergraduate students (aged between 19 and 21 years). Experiment 11: memory for animals The Animal Ears Task (see Kosslyn et al., 1985) In this test, subjects were instructed to produce a mental image of an animal, having been given its name (e.g. poodle, cat, etc.), and then to determine whether the ears protruded above the skull, or flopped down by the side of the head. H.J.A. scored at ceiling (20/20), and scored above the mean score of the control subjects (16.9/20, standard deviation 5 1.8). The Animal Legs Task (see Kosslyn et al., 1985) Subjects were again instructed to produce a mental image of a named animal (e.g. crocodile, giraffe, etc.), and then to determine whether its legs were generally short or long in proportion to its body. H.J.A. scored 19/20, which was at a similar level to the control subjects (mean score 5 19.3/20, standard deviation 5 0.7). The Animal Size Test (Paivio, 1975; Kosslyn et al., 1985) The names of 28 animals were presented verbally in pairs (e.g. ostrich–stork, blackbird–robin, etc.), and the subjects were asked to name the larger animal of the pair. H.J.A. scored at ceiling (14/14) and at a higher level than the mean score of the control subjects (12.4/14, standard deviation 5 2.1). Perceptual Differences Test (De Renzi and Lucchelli, 1994; Forde et al., 1997) Ten animal pairs were presented, and subjects were asked to identify the perceptual features which allow the two to be distinguished. H.J.A. again scored at ceiling (10/10; the mean control score 5 9.5/10, standard deviation 5 0.5). Experiment 12: Memory for artefacts High versus Wide Test (see Kosslyn et al., 1985) The names of 20 artefacts were presented in turn. The task was to determine whether vertical dimensions were greater than horizontal or vice versa (e.g. frying pan, traffic lights, etc.). H.J.A. scored 19/20 (control mean 5 18.5/20, standard deviation 5 1.4). Probing visual knowledge Sharp versus Rounded Test (see Policardi et al., 1996) To test H.J.A.’s visual knowledge further in 1997, he was given sets of tests that have been used in the Subjects were given the names of 20 items and were asked to indicate whether the end of the item was predominantly Downloaded from http://brain.oxfordjournals.org/ at New York Medical College on August 22, 2014 causal variable, and any effects are more likely to be due to high frequency items having more average shapes (having higher levels of contour overlap within their categories). In the regression analyses, contour overlap accounted for most of the variance of SS items along with the deleterious effects of time. This reduction in the number of visual attributes listed supports the evidence from drawing from memory, suggesting that there has been some change in H.J.A.’s longterm memory for the visual properties of objects. The present data indicate that the effects tend to be more serious for natural objects, belonging to categories with structurally similar examples. This is consistent with the idea that, when perceptual input cannot be used to update memory, there is some loss of knowledge for the visual properties of objects in such categories. We elaborate on why natural kinds might suffer most in the General discussion. Like other patients with visual agnosia (e.g. those reported by Adler, 1944, 1950; Kertesz, 1979; Sparr et al., 1991), H.J.A. makes some writing errors. However, he does not superimpose letters, produce multiple loops or produce multiple strokes (cf. Adler, 1944, 1950; Kertesz, 1979; Sparr et al., 1991). Nevertheless, he does make substitution errors along with some omissions and additions. These errors are not due to neglect, since they are not position-related, and they are not generated randomly, since they tend to be more frequent with some letters than others. It may be that substitutions occur between visually similar items and reflect a monitoring failure when visual input is impoverished. A detailed analysis of H.J.A.’s spelling errors is not of primary concern in this paper. More important is to note that the number of errors represented only a small proportion of the written word corpus, and the percentage of errors changed very little over time (from 1.6 to 1.9%). Again, there is no evidence here of a general deterioration in H.J.A.’s faculties. This lack of deterioration in spelling over time is of some interest to the argument concerning the updating of visual memories. It suggests either that (i) visual updating plays little role in the maintenance of graphemic production codes; (ii) the visual feedback H.J.A. achieves is sufficient to maintain memories for writing, although it seems less effective for memories of more complex objects; or (iii) in writing, visual feedback is supported by motor feedback, which helps H.J.A. maintain memories for graphemic production codes. We note here that Adler’s (1944) patient also showed little deterioration over time in her written spelling (Sparr et al., 1991). The visual memories held by this patient were not tested in sufficient detail to enable conclusions to be made about their relative maintenance compared with memories of graphemic production codes. For H.J.A., memories for graphemic production codes seem less affected than visual memories for objects. 553 554 M. J. Riddoch et al. sharp (or pointed), or whether it was rounded (e.g. toothpick, matchstick, etc.). H.J.A. scored 19/20 (control mean 5 19.0/20, standard deviation 5 0.9). writing errors. H.J.A. remains an intellectually bright man, with a good deal of semantic knowledge about objects. Our discussion concentrates on H.J.A.’s perceptual abilities and his long-term memory for objects. Object Thickness Test (see Policardi et al., 1996) Subjects were asked to decide whether named objects exceed 1.5 cm in thickness (e.g. coin versus bicycle pump). H.J.A. scored 20/20 (control mean 5 18.5/20, standard deviation 5 2.5). Ten object pairs were presented, and the task was to identify the perceptual features which allow the two to be distinguished (e.g. the differences between a needle and a pin). H.J.A. again scored at ceiling (10/10; mean control score 5 9/10, standard deviation 5 0.8). Discussion On a range of forced-choice tests designed to assess longterm visual knowledge of both animals and artefacts, H.J.A. performed extremely well and was clearly unimpaired even relative to young control subjects (and even when the controls performed below ceiling). Previous authors have taken a profile of performance such as this to suggest that both longterm visual memory and visual imagery is intact (Behrmann et al., 1992; Jankowiak et al., 1992; Servos et al., 1993; Servos and Goodale, 1995). However, the findings are at variance with the results from H.J.A.’s drawing from memory and his verbal definitions, both of which indicated some degradation in long-term visual knowledge of objects. We suggest that this contrast reflects the sensitivity of the tests. When given a forced-choice discrimination between alternative stimuli, H.J.A. may be able to use degenerate visual knowledge to perform at an adequate level. In contrast, in ‘free report’ tasks, such as drawing from memory and producing definitions, the degraded visual input may provide inadequate support leading to relatively impaired performance. The important point here is that many clinical tests of long-term visual knowledge (using forced-choice procedures) may not provide sensitive measures of more subtle memory loss. General discussion We have documented tests of H.J.A.’s visual perceptual abilities and long-term memory some 16 years following the lesion that led to his visual agnosia. The first result to note is that there was no evidence of any general deterioration over time. On some perceptual tests his performance improved (e.g. with real objects), his listing of verbal/functional properties of objects increased and there was no increase in H.J.A. generally showed a stable pattern of performance in his perceptual abilities over time. With line drawings, he remained poor at both naming and retrieving semantic (associative) information, he continued to be differentially affected by perceptual overlap and he again showed a trend (opposite to normal) to perform better with silhouettes than line drawings in an object decision task. These results demonstrate little change in H.J.A.’s underlying pathology. Nevertheless, there were positive changes in some abilities: he improved at naming real objects, he improved in his ability to match line drawings on the basis of functional characteristics (the Item Match Test) and he showed less of an effect of short exposure durations relative to when he was first tested (Riddoch and Humphreys, 1987a). In previous follow-up studies of agnosia, some improvement in the identification of line drawings has been noted (Adler, 1944, 1950; Davidoff and Wilson, 1985; Wilson and Davidoff, 1993). In H.J.A., this improvement was not accompanied by a change in his ability to encode and integrate contour information, when naming line drawings or when overlapping figures were used. This suggests that the improvement with real objects was due to better use of non-contour-based visual information, such as surface detail and depth cues. One of H.J.A.’s deficits in naming line drawings is due to his tendency to use internal edge detail to segment parts inappropriately (Humphreys and Riddoch, 1987b). The presence of additional depth cues in objects (particularly from stereo) may be helpful in this respect, preventing objects being parsed into inappropriate regions. Note that H.J.A. has good stereo depth (Humphreys and Riddoch, 1987b). In this respect, H.J.A.’s improvement with real objects can be considered to reflect the adoption of a compensatory visual process (e.g. an increased weighting in the use of 3D depth relative to 2D edge information in object segmentation and parsing), rather than the restitution of an impaired process (given his consistent deficits with line drawings). It will be important for future work to tease apart exactly which properties of 3D objects are useful here. In addition to the improvement with real objects, H.J.A. became better able to deal with briefly presented line drawings (Experiment 8) and in matching drawings on the basis of function (Experiment 2). This might be due to the development of faster visual processing over the time period since his lesion, though the stability of the deficit in the other perceptual tests argues against this. Instead we propose that, again as a compensatory process, H.J.A. has become better at coping with fragmentary visual information, maintaining such information in short-term visual memory, or guessing on the basis of such information. He can also Downloaded from http://brain.oxfordjournals.org/ at New York Medical College on August 22, 2014 Perceptual Differences Test (De Renzi and Lucchelli, 1994; Forde et al., 1997) Perceptual abilities Stored visual knowledge in visual agnosia use this information to match objects when functionally equivalent items are also visually similar. Long-term visual memory representations initially can remain quite robust in the face of a severe perceptual impairment and that the effects of impoverished perceptual up-dating only become apparent over a longer time period. This initial robustness, however, should not be taken to indicate that perceptual and memorial processes are functionally dissociated. In addition to this, tests can be shown to vary in their sensitivity. Even in the current tests, H.J.A. did extremely well when given forcedchoice questions that probed his stored visual knowledge about many perceptual properties of objects (their size, the shape of their parts, etc.) (see ‘Probing visual knowledge’), and frequently scored above the level achieved by younger, neurologically intact, university students. We suggest that this is because forced-choice questions constrain the decision space and enable a patient to function at a high level even if there is an underlying impairment (for a similar argument, see Rapp and Caramazza, 1993). Such questions may be adequate to demonstrate problems in patients with more marked impairments in stored knowledge (for examples, see De Renzi and Lucchelli, 1994; Forde et al., 1997), but may fail to reveal more subtle degradation of visual memory. The conclusion that visual memory is intact in cases of apperceptive agnosia is perhaps premature (cf. Behrmann et al., 1992; Jankowiak et al., 1992; Servos et al., 1993; Servos and Goodale, 1995). Our evidence demonstrates that impairments are most apparent on relatively unconstrained tasks such as giving verbal definitions and drawing from memory, and with items that themselves have relatively unremarkable shapes within categories with perceptually similar examples (SS categories with high name frequencies). Previous tests of long-term visual memory in agnosic patients have not probed such items in sufficient depth to enable confident conclusions to be made. Clearly, it would also be interesting to examine the effects of time on stored visual knowledge in the adventitiously blind; as far as we know, such studies have not yet been performed. Our data are quite consistent with the notion that visual perception and longterm visual memory utilize common neural strata (e.g. in imagery tasks). Visual processing and category-specific deficits The present results also have implications for so-called category-specific deficits in neurological patients. Numerous studies have been reported in which patients have been shown to manifest category-specific deficits for the identification of natural objects relative to artefacts (e.g. Warrington and Shallice, 1984; Sheridan and Humphreys, 1993; De Renzi and Lucchelli, 1994; Forde et al., 1997; Caramazza and Shelton, 1998). Several accounts of these deficits have been offered, ranging from the effects being due to the familiarity or complexity of the depicted stimulus (Funnell and Sheridan, 1992; Stewart et al., 1992), to the loss of semantic knowledge for particular categories of items (e.g. Sheridan and Humphreys, 1993; Caramazza and Shelton, 1998). H.J.A.’s case is relevant to these arguments because he too shows Downloaded from http://brain.oxfordjournals.org/ at New York Medical College on August 22, 2014 Although H.J.A.’s basic perceptual abilities appeared to be reasonably stable, there was evidence for a deterioration in his long-term memory for the visual properties of objects. In particular, there was a worsening in his ability to draw objects from memory, and he tended to give fewer visual details in his definitions of items (Experiments 9 and 10). It cannot be argued that the shift in H.J.A.’s definitions reflected a general worsening of performance, since the number of verbal/ functional details increased (see above). However, it might be that H.J.A. simply chose not to include visual details in these definitions since defining objects from memory is a relatively poorly constrained task (even though we asked H.J.A. to provide visual as well as functional information, if he could). This last proposal would have to explain why he chose to omit visual details for natural (SS) items only (almost exactly the same number of visual details were produced for artefacts over time), and it cannot account for the deterioration in H.J.A.’s drawing from memory. In drawing from memory, the patient is constrained to produce visual details if possible. We propose that there is not simply a shift in strategy here, but rather some degradation in H.J.A.’s stored knowledge about objects over time. This is reflected in both his definitions and his drawing from memory. The evidence for a degradation in visual knowledge is consistent with the idea that perceptual and memorial processes do interact in vision. Thus perceptual processing of objects may lead to a consistent up-dating of visual memories over time, so that these memories remain tuned to the visual properties of objects in the world. Such consistent tuning of memorial processes to vision fits with connectionist accounts of cognition, in which recognition systems change adaptively to the inputs they receive over time (Ellis and Humphreys, 1998). When perceptual inputs are impaired, as in forms of apperceptive agnosia, visual memorial processes may gradually show some decline, with there being less fine tuning of the system to the visual properties of objects. In this case, some coupling can be observed between the perceptual processes that are impaired and the effect on particular memorial processes (Young et al., 1994). The decline in H.J.A.’s ability to define and draw objects from memory goes against our earlier report in which we note a dissociation between perceptual and memorial processes (Riddoch and Humphreys, 1987a), and also other cases where long-term visual knowledge has been reported as being intact in the face of severe agnosia (e.g. Behrmann et al., 1992; Jankowiak et al., 1992; Servos et al., 1993; Servos and Goodale, 1995). However, this contradiction may reflect factors such as the sensitivity of the tests and the time of testing post-lesion. We have shown here that, using the same tests of defining and drawing from memory, a decline can be shown over time. It may be that visual memory 555 556 M. J. Riddoch et al. artefacts. Tyler, Moss and colleagues (Tyler et al., 1996; Tyler and Moss, 1997) point out that natural objects tend to have many perceptual features in common that are associated with basic biological functions (e.g. legs, heads and bodies which are associated with walking, breathing, reproducing). In everyday learning, the perceptual features of objects will be associated with their function. For natural objects, common features will be reinforced in learning since the relationships between the features and functional properties hold across many objects. However, since these common features do not differentiate between the items within their categories, they are not useful for the identification and drawing from memory of individual items. For identification and drawing, the distinctive features that specify a particular object must be computed. These distinctive features, though, are not typically tied to functional information about natural items, and they hold for only a few objects; as a consequence, they will be reinforced only weakly during learning. Now, if we hold that visual–functional relationships are reinforced not only during initial learning, but also during the up-dating of memories, we might expect distinctive features of natural objects to be vulnerable to loss. Such a loss would make it difficult to recover natural objects from memory. In contrast, the distinctive features of individual artefacts are strongly linked to their functions and so these features may be reinforced during the visual up-dating process. Artefacts are thus less affected than natural objects when the up-dating process is disrupted. Irrespective of whether the impairment for natural objects is attributed just to their greater visual similarity, or to an effect linked to the distinctive features of these objects, our data suggest that visual up-dating has a differential impact on different object categories. Although we do not wish to over-generalize this argument to all cases of category-specific deficit, the present results point to the importance of visual factors in producing these deficits, at least for some patients. Brain mechanisms of perceptual learning Several recent studies using PET methodology have examined the neural structures involved in visual perceptual learning. For instance, an object in a degraded image which appears meaningless when seen for the first time can be easily recognized when an undegraded image of the object is presented first (when the recognition system is ‘primed’). Dolan, Frith and colleagues (Dolan et al., 1997; Frith and Dolan, 1997) found that changes between the first (naive) and second (primed) state were associated with enhanced processing of inferior temporal regions involved in visual object processing (particularly the fusiform gyrus). Effects were lateralized differentially for objects and faces, suggesting that item-specific learning takes place in these regions. The data match results on learning-related tuning of temporal lobe activity in the monkey (Sakai and Miyashita, 1991, 1994; Tovee et al., 1996). In addition, Frith and Dolan (1997) reported enhanced activity in medial and lateral Downloaded from http://brain.oxfordjournals.org/ at New York Medical College on August 22, 2014 worse visual identification of natural objects relative to artefacts. This was first reported by Riddoch and Humphreys (1987a) and is confirmed here (Experiment 3), even with sets of natural objects matched for name frequency. In addition, we have replicated this result in a test with a larger set of items matched for complexity and familiarity: H.J.A. scored 16/44 (36%) correct for natural objects and 28/44 (64%) correct on artefacts using items from the matched sets of Rumiati et al. (1994); χ2(1) 5 5.5, P , 0.01. H.J.A.’s deficits for natural objects occur even though his semantic memory concerning verbal/functional knowledge about these items is very good (Experiment 10, Tables 1–3). It is difficult to attribute the problem to an impairment in semantic knowledge in his case. Nor is the effect due to differential familiarity, complexity or name frequency across the items used. In fact, the present report goes beyond the simple demonstration of a category-specific effect, since we present for the first time direct evidence for the impact of a visual factor (the degradation of stored visual knowledge) on the retrieval of information about natural objects relative to artefacts. The deterioration of visual knowledge, evident in H.J.A.’s reports of visual details in definitions, had a greater impact on (SS) natural objects than on (SD) artefacts (Experiment 10). This supports the argument that visual updating is more necessary for the maintenance of knowledge about natural objects than the maintenance of knowledge about artefacts. For example, one of the factors that can mediate the poor report of natural objects is that these items tend to be more visually similar within their category than artefacts (Humphreys et al., 1988). Due to the greater visual similarity within their categories, the visual identification of natural objects tends to suffer due to increased competition between items, relative to the identification of artefacts—a proposal consistent with data showing category differences in the speed of identification even in the normal population (Humphreys et al., 1988, 1997; Lloyd-Jones and Humphreys, 1997). The same proposal holds when the visual properties of individual items must be drawn from memory, when definitions are required or when items must be drawn from memory; again, increased competition will affect performance for natural kinds. In a patient whose stored visual knowledge has suffered because of impaired up-dating, performance on natural kinds may suffer most because visual representations of these items tend in any case to cluster. We suggest that visual similarity within the category is important here, rather than just the strong weighting of visual feature information in the representation of natural objects (Farah and McClelland, 1991). The similarity account provides a better explanation for why contour overlap was the best predictor of performance. Also the overall negative effect of name frequency for these items is otherwise difficult to understand. However, the effect can be accounted for on the basis of structural similarity; items with high name frequencies tend to have prototypical shapes within their category. In addition to the above, there can be contrasting roles for different forms of visual information for natural objects and Stored visual knowledge in visual agnosia Ellis R, Humphreys GW. Connectionist psychology. London: Psychology Press. In press 1998. Farah MJ. Visual agnosia: disorders of object recognition and what they tell us about normal vision. Cambridge (MA): MIT Press; 1990. Farah MJ, McClelland JL. A computational model of semantic memory impairment: modality specificity and emergent category specificity. [Review]. J Exp Psychol Gen 1991; 120: 339–57. Farah MJ, Levine DN, Calvanio R. A case study of mental imagery deficit. Brain Cogn 1988a; 8: 147–64. Farah MJ, Peronnet F, Gonon MA, Giard MH. Electrophysiological evidence for a shared representational medium for visual images and visual percepts. J Exp Psychol Gen 1988b; 117: 248–57. Finke RA, Pinker S, Farah MJ. Reinterpreting visual patterns in mental imagery. Cogn Sci 1989; 13: 51–78. Acknowledgements Forde EM, Francis D, Riddoch MJ, Rumiati RI, Humphreys GW. On the links between visual knowledge and naming: a single case study of a patient with a category-specific impairment for living things. Cogn Neuropsychol 1997; 14: 403–58. We wish to thank H.J.A. who has always been such a willing collaborator. This work was supported by the MRC (M.J.R. and G.W.H.) and the EU (G.W.H.). Frith C, Dolan RJ. Brain mechanisms associated with top-down processes in perception. [Review]. Philos Trans R Soc Lond B Biol Sci 1997; 352: 1221–30. Funnell E, Sheridan J. Categories of knowledge? Unfamiliar aspects of living and nonliving things. Cogn Neuropsychol 1992; 9: 135–53. References Adler A. Disintegration and restoration of optic recognition in visual agnosia. Arch Neurol Psychiatry 1944; 51: 243–59. Goldenberg G. Loss of visual imagery and loss of visual knowledge. Neuropsychologia 1992; 30: 1081–99. Adler A. Course and outcome of visual agnosia. J Nerv Ment Dis 1950; 111: 41–51. Goldenberg G. Is there a common substrate for visual recognition and visual imagery? Neurocase 1998; 4: 141–7. Behrmann M, Winocur G, Moscovitch M. Dissociation between mental imagery and object recognition in a brain-damaged patient [see comments]. Nature 1992; 359: 636–7. Comment in: Nature 1992; 359: 590–1. Boucart M, Humphreys GW. The computation of perceptual structure from collinearity and closure: normality and pathology. Neuropsychologia 1992; 30: 527–46. Caramazza A, Shelton JR. Domain-specific knowledge systems in the brain: the animate–inanimate distinction. [Review]. J Cogn Neurosci 1998; 10: 1–34. Chatterjee A, Southwood MH. Cortical blindness and visual imagery. Neurology 1995; 45: 2189–95. Davidoff J, Wilson B. A case of visual agnosia showing a disorder of pre-semantic visual classification. Cortex 1985; 21: 121–34. De Renzi E, Lucchelli F. Are semantic systems separately represented in the brain? The case of living category impairment. Cortex 1994; 30: 3–25. Decety J, Kawashima R, Gulyàs B, Roland PE. Preparation for reaching: a PET study of the participating structures in the human brain. Neuroreport 1992; 3: 761–4. Dolan RJ, Fink GR, Rolls E, Booth M, Holmes A, Frackowiak RS, et al. How the brain learns to see objects and faces in an impoverished context. Nature 1997; 389: 596–9. Goldenberg G, Artner C. Visual imagery and knowledge about the visual appearance of objects in patients with posterior cerebral artery lesions. Brain Cogn 1991; 15: 160–86. Goldenberg G, Podreka I, Steiner M, Willmes K, Suess E, Deecke L. Regional blood flow patterns in visual imagery. Neuropsychologia 1989; 27: 641–64. Goldenberg G, Müllbacher W, Nowak A. Imagery without perception—a case study of anosagnosia for cortical blindness. Neuropsychologia 1995; 33: 1373–82. Humphreys GW, Riddoch MJ. Routes to object constancy: implications from neurological impairments of object constancy. Q J Exp Psychol 1984; 36A: 385–415. Humphreys GW, Riddoch MJ. The fractionation of visual agnosia. In: Humphreys GW, Riddoch MJ, editors. Visual object processing: a cognitive neuropsychololgical approach. Hove (UK): Lawrence Erlbaum; 1987a. p. 281–306. Humphreys GW, Riddoch MJ, To see but not to see: a case study of visual agnosia. Hove (UK): Lawrence Erlbaum; 1987b. Humphreys GW, Riddoch MJ, Quinlan PT. Cascade processes in picture identification. Cogn Neuropsychol 1988; 5: 67–103. Humphreys GW, Riddoch MJ, Boucart M. The breakdown approach to visual perception: neuropsychological studies of object recognition. In: Humpreys GW, editor. Understanding vision: an interdisciplinary perspective. Oxford: Blackwell; 1992a. p. 104–25. Downloaded from http://brain.oxfordjournals.org/ at New York Medical College on August 22, 2014 parietal cortex that was specific to perceptual learning (the change between the naive and primed state). They suggested that the parietal activity could reflect the involvement of imagery in reconstructing degraded stimuli and in binding the parts of these stimuli together. H.J.A.’s lesion involves the inferior occipito-temporal regions, consistent with the site of item-specific learning in PET studies. We suggest that this brain area is intimately involved in the up-dating of our visual knowledge about specific objects in everyday life, a process that is impaired in H.J.A. On the other hand, there is no evidence of parietal damage in H.J.A.’s case. Parietal involvement in perceptual learning may be confined to instances when spatial reconstruction of stimuli is needed for task performance. Under these circumstances, a larger circuit may be recruited, involving parietal as well as inferior temporal lobe areas. 557 558 M. J. Riddoch et al. Humphreys GW, Riddoch MJ, Quinlan PT, Price CA, Donnelly N. Parallel pattern processing and visual agnosia. Can J Psychol 1992b; 46: 377–416. Humphreys GW, Troscianko T, Riddoch MJ, Boucart M, Donnelly N, Harding GFA. Covert processing in different visual recognition systems. In: Milner AD, Rugg MD, editors. The neuropsychology of consciousness. London: Academic Press; 1992c. p. 39–68. Humphreys GW, Donnelly N, Riddoch MJ. Expression is computed separately from facial identity and it is computed separately for moving and static faces: neuropsychological evidence. Neuropsychologia 1993; 31: 173–81. Rapp B, Caramazza A. On the distinction between deficits of access and deficits of storage; a question of theory. Cogn Neuropsychol 1993; 10: 113–41. Rensink RA, Enns JT. VScope and EMaker: micropsych software. 1992. Riddoch MJ, Humphreys GW. A case of integrative visual agnosia. Brain 1987a; 110: 1431–62. Riddoch MJ, Humphreys GW. Visual object processing in optic aphasia: a case of semantic access agnosia. Cogn Neuropsychol 1987b; 4: 131–85. Jankowiak J, Kinsbourne M, Shalev RS, Bachman DL. Preserved visual imagery and categorization in a case of associative visual agnosia. J Cogn Neurosci 1992; 4: 119–31. Riddoch MJ, Humphreys GW. BORB: the Birmingham Object Recognition Battery. Hove (UK): Lawrence Erlbaum; 1993. Kaplan E, Goodglass H, Weintraub S. Boston Naming Test: experimental edition. Boston: Boston University Aphasia Research Centre; 1982. Roland PE, Gulyás B. Visual imagery and visual representation [see comments]. [Review]. Trends Neurosci 1994; 17: 281–7. Comment in: Trends Neurosci 1994; 17: 287–94, Comment in: Trends Neurosci 1994; 17: S14–6. Kertesz A. Visual agnosia: the dual deficit of perception and recognition. Cortex 1979; 15: 403–19. Kosslyn SM. Image and brain: the resolution of the imagery debate. Cambridge (MA): MIT Press; 1994. Kosslyn SM, Holzman JD, Farah MJ, Gazzaniga MS. A computational analysis of mental image generation: evidence from functional dissociations in split-brain patients. J Exp Psychol Gen 1985; 114: 311–41. Kosslyn SM, Alpert NM, Thompson WL, Maljkovic V, Weise SB, Chabris CF, et al. Visual mental imagery activates topographically organized visual cortex: PET investigations. J Cogn Neurosci 1993; 5: 263–87. Lawson R, Humphreys GW. The effects of view in depth on the identification of line drawings and silhouettes of familiar objects: normality and pathology. Vis Cogn. In press 1998. Lissauer H. Ein fall von Seelenblindheit nebst einem Beitrage zur Theorie derselben. Arch Psychiat Nervkrankh 1890; 21: 222–70 Lloyd-Jones TJ, Humphreys GW. Perceptual differentiation as a source of category effects in object processing: evidence from naming and object decision. Mem Cogn 1997; 25: 18–35. McKenna P, Warrington EK. The graded naming test. Windsor: NFER-Nelson; 1983. Milner AD, Perrett DI, Johnston RS, Benson PJ, Jordan TR, Heeley DW, et al. Perception and action in ‘visual form agnosia’. Brain 1991; 114: 405–28. Paivio A. Perceptual comparisons through the mind’s eye. Mem Cogn 1975; 3: 635–74. Policardi E, Perani D, Zago S, Grassi F, Fazio F, Làdavas E. Failure to evoke visual images in a case of long-standing cortical blindness. Neurocase 1996; 2: 381–94. Poppelreuter W. Die psychischen Schädigungen durch Kopfschuss Rumiati RI, Humphreys GW, Riddoch MJ, Bateman A. Visual object agnosia without prosopagnosia or alexia: evidence for hierarchical theories of visual recognition. Vis Cogn 1994; 1: 181–225. Sakai K, Miyashita Y. Neural organization for the long-term memory of paired associates [see comments]. Nature 1991; 354: 152–5. Comment in: Nature 1991; 354: 108–9. Sakai K, Miyashita Y. Neuronal tuning to learned complex forms in vision. Neuroreport 1994; 5: 829–32. Servos P, Goodale MA. Preserved visual imagery in visual form agnosia. Neuropsychologia 1995; 33: 1383–94. Servos P, Goodale MA, Humphrey GK. The drawing of objects by a visual form agnosic: contribution of surface properties and memorial representations. Neuropsychologia 1993; 31: 251–9. Sheridan J, Humphreys GW. A verbal-semantic category-specific recognition impairment. Cogn Neuropsychol 1993; 10: 143–84. Snodgrass JG, Vanderwart MA. A standardized set of 260 pictures: norms for name agreement, image agreement, familiarity, and visual complexity. J Exp Psychol [Hum Learn] 1980; 6: 174–215. Sparr SA, Jay M, Drislane FW, Venna N. A historic case of visual agnosia revisited after 40 years. Brain 1991; 114: 789–800. Stewart F, Parkin AJ, Hunkin HN. Naming impairments following recovery from herpes simplex encephalitis: category-specific [see comments]. Q J Exp Psychol 1992; 44A: 261–84. Comment in: Q J Exp Psychol 1993; 46A: 489–509. Tovee MJ, Rolls ET, Ramachandran VS. Rapid visual learning in neurones of the primate temporal visual cortex. Neuroreport 1996; 7: 2757–60. Tyler LK, Moss HE. Functional properties of concepts: studies of normal and brain damaged patients. Cogn Neuropsychol 1997; 14: 511–45. Downloaded from http://brain.oxfordjournals.org/ at New York Medical College on August 22, 2014 Humphreys GW, Riddoch MJ, Price CJ. Top-down processes in object identification: evidence from experimental psychology, neuropsychology and functional anatomy. [Review]. Philos Trans R Soc Lond B Biol Sci 1997; 352: 1275–82. im Kriege 1914/16 mit besonderer Berücksichtigung der pathopsychologischen, pädagogischen, gewerblichen und sozialen Beziehungen: Vol. 1. Die Störungen der niederen und höheren Schleistungen durch Verletzungen des Okzipitalhirns. Leipzig: Voss; 1917. Stored visual knowledge in visual agnosia Tyler LK, Durrant-Peatfield M, Levy J, Voice K, Moss H. Distinctiveness and correlations in the structure of categories: behavioural data and a connectionist model. Brain Lang 1996; 55: 89–92. Warrington EK, Shallice T. Category specific semantic impairments. Brain 1984; 107: 829–54. Warrington EK, Taylor AM. Two categorical stages of object recognition. Perception 1978; 7: 695–705. 559 Wilson BA, Davidoff J. Partial recovery from visual object agnosia: a 10 year follow-up study. Cortex 1993; 29: 529–42. Young AW, Humphreys GW, Riddoch MJ, Hellawell DJ, de Haan EH. Recognition impairments and face imagery. Neuropsychologia 1994; 32: 693–702. Received April 27, 1998. Revised September 15, 1998. Accepted October 7, 1998 Downloaded from http://brain.oxfordjournals.org/ at New York Medical College on August 22, 2014