Category-specificity and modality-specificity in semantic memory

Neuropsycholu(/ta,

Vol. 27, No. 2, pp. 193-200,

1989

0028%3932/89 Ifi3.@0+0.00 ('1989Pergamon Press plc

Printed IDGreat Britain.

CATEGORY-SPECIFICITY AND MODALITY-SPECIFICITY SEMANTIC MEMORY MARTHA

J. FARAH,*

KATHERINE

M.

~~~GRAHAM *Carnegie-Mellon

HAMMOND,*

ZIYAH

IN

MEHTA~

RATCLIFF~

University,

fHarmarville

5000 Forbes Avenue, Pittsburgh, Pennsylvania 15213, U.S.A tOxford University, Oxford, U.K. Rehabilitation Center and University of Pittsburgh, U.S.A.

(Receiaed

17 November

1987: accepted

16 March 1988)

Abstract-Studies of agnosia have revealed two apparently orthogonal dimensions along which knowledge may break down. In some cases, knowledge of specific categories (such as living things) seems lost, regardless of the modality being tested. In other cases, knowledge in specific modalities (such as vision) seems lost, regardless of the category of stimuli being tested. These different sets of phenomena suggest different organizations for knowledge in the brain, the first by category and the second by modality. Unfortunately, possible confoundings between category, modality, and difficulty level in the previous studies prevent us from drawing strong conclusions from these data. The present study was aimed at assessing the nature of the breakdown in the semantic memory of a prosopagnosic patient, by orthogonally varying category and modality, while assessing diticulty level. The findings do not implicate a simple categorical or modality-dependent organization of his knowledge, but rather an organization in which both category and modality play a role.

ONE VIEW of the organization of knowledge in semantic memory is that it is fundamentally categorical, such that knowledge about objects in a given category will be represented together without regard to the modality through which that knowledge was gained. Information may be retained about the modality of origin of the knowledge, but modality is not an organizing factor. This view is consistent with the traditional theoretical approach to semantic memory taken by cognitive psychologists [3,11]. A very different view of semantic memory is that the organization of knowledge is fundamentally modality-specific, and that although one can form associations between knowledge of objects across modalities, the fundamental organization of knowledge is modality-specific. This view is consistent with that of certain modularity theorists in cognitive psychology [2], as well as with classical connectionist thinking in neurology [ 151. As descriptions of information-processing systems in the abstract, it would undoubtedly be difficult to tell these two types of organization apart. This is because associations linking representations across the major organizing subdivisions (category or modality) could in principle mimic the effects of the opposite organization. For example, in a semantic memory system organized by modality, cross-modality associations between modality-specific object representations are, in effect, equivalent to multimodal object representations. It thus becomes difficult to determine the psychological reality of organizations for semantic memory, or even what psychological reality would mean given the potential functional equivalence of category-specific and modality-specific knowledge systems. However, in the 193

194

MARTHA J. FARAH, KATHERINEM. HAMMOND,ZIYAH MENTA and GKAHAM RATC’LIFF

context of theories of brain function, the distinction between category-specific and modalityspecific organizations seems testable and, hence, more meaningful. In particular, we can ask whether the brain systems that subserve semantic memory subdivide, neurophysiologically, by category or by modality. To the extent that the effects of brain damage are selective, WC can assume that they select among biologically-defined subsystems of the brain (e.g. neuroanatomic, neurochemical, etc.). Therefore, to the extent that brain damage has selective effects upon components of semantic memory, this implies that these components are “biologically real”. Studies of the various forms of agnosia appear to address the issue of biologically real organizations for semantic memory. Traditionally, the agnosias were regarded as modality-specific disorders. Associative visual object agnosia, for example, involves adequate visual perception of stimuli, but an impairment in the recognition of visually presented stimuli which is not attributable to general intellectual deficits [4]. Why should an impairment in visual recognition processes be relevant to answering questions about semantic memory? With the criterion of relatively intact perception, associative agnosia is by definition not merely a disorder of perception, but rather of visual knowledge and its use. In TEUBER’S classic terms [27], associative agnosia is perception “stripped of its meaning”, a phrasing that highlights the role of knowledge and semantics in the agnosic deficit. Furthermore, most cases of visual agnosia have been found to have impaired visual mental imagery. This also implies that the disorder is not directly related to stimulus processing, and therefore reflects a loss of knowledge rather than of perceptual abilities [14]. This characterization of associative visual object agnosia itnplies that knowledge is organized in the brain by modality. The tidiness of this conclusion is spoiled by the observation that, even among the modality-specific visual agnosias, there is an indication of category specificity: prosopagnosic patients suffer from a visual agnosia that is most pronounced for faces, animals, and certain other categories of stimuli including foods, plants, and makes of automobile, leaving other categories relatively unaffected [S. 9, 19,23,24,25]. The stimuli that are most difficult for these patients can be roughly grouped under the category “living things”.* The category-specificity of this deficit has led some authors to conclude that faces and other living things are represented by a separate part of visual memory from other stimuli, that is, that the modality-specific visual memory is in turn subdivided by category [5,7,20, 291. Other authors have proposed that the difference between living things and other visual stimuli is a difference in degree of difficulty of recognition 16, 12, 181. Living things seem generally more complex and visually confusable, one for another, than nonliving things, and we generally require more particular inter-individual discriminations when we test recognition of living things than nonliving things. It is therefore argued that a mildly agnostic patient will make disproportionate errors with these stimuli and that, in effect, prosopagnosia is nothing more than a mild object agnosia. Striking evidence of category-specific agnosia was presented by WARRINGTON and SHAL.LICE [28]. They described the performance of four post-encephalitic patients on a variety of tasks testing knowledge of living and nonliving stimuli. These tasks included the naming of visually-presented stimuli, miming the use of visually-presented stimuli, and * For present purposes, this label can be considered a convenient short-hand for the set of stimuli which prosopagnosics typically find difficult to recognize. However, the label is not arbitrary: most of the members of this set are, in fact, living things, and one could imagine that exceptions to this rule (such as makes of automobile) are recognized as ifthey were living things, that is, using the same specific visual memory mechanisms that evolved for recognizing different species of animals.

DISSOCIATIONS

IN SEMANTIC

MEMORY

195

defining the meanings of spoken words. When the stimuli, visual or verbal, represented living thmgs, the subjects did significantly worse than when the stimuli represented nonliving things. At first glance, these data seem to imply that knowledge is organized by category rather than modality, as knowledge of a particular category, living things, was disproportionately impaired in ail modalities. Warrington and Shaiiice point out that there is another explanation of their findings that also deserves consideration. Knowledge of living things, even when tested verbally, involves modality-specific visual knowledge, whereas knowledge of nonliving things does so to a far smaller degree. For example, an adequate definition of a “vacuum cleaner” would be “a machine that uses suction to pick up dirt and dust, often used around the house”, a definition which contains no visual information. In contrast, an adequate definition of a “deer” would be “a forest animal with antlers and cloven feet, a bit bigger than a large dog”, a definition which is primarily visual. Therefore a patient with deficient visual knowledge would perform better at defining nonliving things. Similarly, it might be speculated that such a patient would be more likely to recognize a visually-presented object when the identity of the object is associated with a specific function, e.g. a vacuum cleaner, than when the identity depends almost exclusively on appearance: in the former case, visual recognition per se could be supplemented by inferences based on whether the object looks usable for a specific function. For example, the hose and nozzle of a vacuum cleaner would help to confirm that it is a suction cleaning device. In recent years, reports of other types of category-specific dissociations in cognitive tasks have appeared, including impairments for categories as specific as fruits and vegetables [ 161, body parts [13], and objects typically found indoors (7301. An important difference between these reports and the ones just summarized is that in these reports the deficit appears to be restricted to naming operations. Therefore, these reports are not directly relevant to the issue of the organization of semantic memory, but rather of the l.exicon or some aspect of its interface with semantics. To summarize, classical associative visual agnosia provides an example of the modaiityspecific breakdown of knowledge. Prosopagnosia, a type of associative visual agnosia, presents us with an apparent within-modality category-specific breakdown: specifically, greater impairment in the visual recognition of living things than of nonliving things. However, it has been suggested that prosopagnosia may differ only in degree rather than in kind from associative visual object agnosia, reflecting the greater difficulty, on average, of recognizing living things. If this is true, then prosopagnosia does not constitute evidence for the categorical organization of visual knowledge. The patients of Warrington and Shaiiice also present us with a partially selective impairment of knowledge about living things, and unlike prosopagnosic patients, these patients appear to be impaired in ail modalities. Warrington and Shaiiice point out that the underlying dissociation could nevertheless be modality-specific, because verbally expressed knowledge of living things weights visual characteristics much more heavily than is the case with nonliving things. Therefore, the question of categorical versus modality-specific organization of semantic memory in the brain remains unanswered. A helpful step towards answering this question would be to test visual and nonvisual semantic memory for living and nonliving things, in an experimental design that has the following two features: (1) visual and nonvisual knowledge wouid be tested for both living and nonliving categories, to eliminate the possible confounding, pointed out by Warrington and Shallice, of visual knowledge with living things and nonvisual knowledge with nonliving

196

MARTHA J. FARAH, KATHERINE M. HARIMONU,ZIYAH MEHTA and GRAHA~~ RATCLIFF

things, and (2) the patient’s performance on each type of question would be scaled against normal performance, to eliminate the possibility of differential difficulty of certain question types creating an artifactual “selective deficit”. This is what we have done, in our testing of a prosopagnosic patient. METHODS Subjects The brain-damaged subject, LH, is a 36-yr old minister currently working towards a second Masters degree. When he was 18 yrs old, he sustained a severe closed head injury in an automobile accident. Brain damage from the accident and subsequent surgery involved both temporo-occipital regions and the right inferior frontal lobe, as demonstrated by CT scan, neurological examination and surgical records. Details of LH’s medical history are published elsewhere 121, 221. LH made a remarkable recovery from his accident, eventually returning to college and going on IO earn a Masters degree. When tested on the Wechsler Adult Intelligence Scale 7 yr after his accident, hts verbal IQ was 132 and his performance IQ was 93. His memory quotient on the Wechsler Memory Scale was 121. He had no detectable language or motor skill deficits. His spatial localization of visual stimuli was normal, and his elementary visual capabilities were basically intact: acuity was 20150 in the left eye and 20,‘70 in the right, with blindness in the peripheral visual field, particularly in the upper left and lower right quadrants of the visual field. Despite his general intellectual and elementary visual capabilities, LH vvas and still is profoundly impaired in visual recognition. He is unable to recognize live or photographed faces. On the ALBERT et al. [l] test of “Famous Faces”, LH was able to recognize only 4 out of30 faces, despite his prolonged scrutiny ofeach one. Normal subjects score about 75% correct on this test. When shown a group photograph which included his face, he was unable to pick himself out. He cannot reliably recognize his wife and children if they do not wear some distinctive article ofelothing. By his own report. LH is also extremely impaired at recognizing animals, plants and some foods. This is consistent with his performance of recognizing only 8 out of 26 line drawings of animals, compared to 24 out of 32 line drawings of common objects drawn by the same artist. Visual recognition of words in reading is remarkably intact. Although we have not tested this ability formally, he is clearly not dyslexic, Reading is a constant requirement in his professional life. and he IS able to meet this requirement without any special aids. In the task described below, LH’s performance is compared with a control group of twelve age- and educationmatched men. The average age of the men was 35 yr (range 33 38 yr) and all had Masters degrees. All subjects, including LH, were paid for their participation.

MATERIALS

AND PROCEDURE

The 48 living and 48 nonliving items used by WARRINCTON and SHALLICE [28] formed the basis for our test. For each of these 96 items, we created four “yes/no’‘-format semantic memory questions: a question about visual appearance whose correct answer is “yes”, a question about visual appearance whose correct answer is “no”, a question about a nonvisual property of the object whose correct answer is “yes”, and a question about a nonvisual property of the object whose correct answer is “no”. This yielded 384 questions. The following are examples of each type of question: (I ) Licing- Visual Are the hind legs of a kangaroo Do ducks have long ears? (no)

larger than the front legs? (yes)

(2) Living-Nonvisual Are roses given on Valentine’s day? (yes) Is peacock served in French restaurants? (no) (3) Nonliring- Visual Is a canoe widest in the center? (yes) Does a guitar have a square-shaped opening

(for the sound to resonate)?

(no)

(4 J Nonlioing-Nonoisual Were wheelbarrows invented before 1920” (yes) Is a decanter thrown into the fireplace for good luck? (no) Questions were printed in twelve subsections of 32 questions each, with the constraint that each subsection have equal numbers of questions from the eight types of questions described above, and none of the 96 items be repeated within a subsection of 32. Within each subsection, the order of questions was random. Subjects answered the questions by circling a “y” or “n” next to each question to indicate “yes” or “no”.

DISSOCIATIONS

IN SEMANTIC

197

MEMORY

RESULTS In the course of conducting the experiment, fourteen out of the 384 questions were discovered to be erroneous, in that they did not have determinate “yes” or “no” answers (e.g. “Is a wallet bigger than a checkbook?“; most wallets are smaller than checkbooks but some are bigger). Therefore, the results to be presented will represent performance on the remaining 370 questions, of which 95 were from the Living- Visual group, 93 from the Liuingh’onrisual group, 89 from the Nonliving- Visual group, and 93 from the Nonliving-Nonvisual group. The percentage of correct answers in each of the four types of questions for each subject is shown in Table 1, along with the means and standard deviations for the control subjects. Figure 1 shows the performance of LH compared to the performance of the control group as a whole. Note that neither LH nor the control subjects are at floor or ceiling on any of the groups of questions. It is clear that the difference between LH and the normal controls is describable neither in terms of a simple effect of living vs nonliving categories, nor as a simple effect of visual vs nonvisual modalities. Rather, LH performs like the average control subject on Nonhing-Nonvisual questions, performs within the range of the normal control subjects on Living-Nonvisual and Nonliving-Visual questions, and performs far below the normal control subjects on the Living- Visual questions. Table 1. Performance (in percent correct) of the twelve normal control subjects and LH on the four types of questions. Means and standard deviations for the control subjects are also shown, in italics Normal subjects

1

Living visual

Living nonvisual

Nonliving visual

Nonliving nonvisual

2 3 4 5 6 7 8 9 10 11 12 mean SD

83.2 80.0 73.1 85.3 83.2 81.1 82.1 85.3 83.2 70.5 71.6 85.3 80.4 (5.4)

93.6 85.0 88.2 90.3 89.3 90.4 94.6 91.4 94.6 19.6 80.7 83.9 88.5 (5.2)

89.9 84.3 88.8 92.1 91.0 91.0 86.5 87.6 92.1 83.1 76.4 93.3 88.0 (4.9)

92.5 91.4 88.2 95.1 95.7 93.5 95.1 92.5 91.4 86.0 86.0 90.3 91.6 (3.5)

LH

63.2

84.8

80.9

91.4

To quantify the relation between LH and the control subjects for each of these questiontypes, we can consider the t value associated with the difference between LH’s score and the average normal score. For the Living- Visual questions, LH is indeed impaired, t(l1) = 3.06, P < 0.02. In contrast, for the Living-Nonvisual, Nonliving- Visual and Nonliving-Nonvisual questions, LH is not significantly different from the normal subjects, t( 11) = 0.72, 1.39, and 0.06, respectively, P> 0.1 in all cases. The foregoing tests tell us whether or not, within each group of questions, LH is impaired relative to normal subjects. These statistics do not measure the extent to which LH’s pattern of performance, across question-types, is abnormal. In particular, they do not tell us whether LH’s performance on the Living-Visual

198

MAKTHA J. FARAH, KATHERINE M. HAMMOND, Zw.4~

100

Living-Nonvisual

Living-Visual

1

MCHTA and GRAHAM RAKY.I~F

NonlivIng-Nonvisual

Nonliving-Visual

l-

l-

-I-

80

T

-l-

-r

Condition

I.

Mean performance (in percent correct) of twelve normal control subjects (“NC”) and LH (“LH”) on the four types of questions. Error bars show one standard deviation from the mean in each

FIG.

direction.

questions is unusually bad relative to his performance on other types of questions, given the normal within subjecf variation in scores on different question-types. To test whether LH is disproportionately impaired on &viny-Visual questions compared to his own scores on other question-types, we adopted the following conservative test. We compared the diserence between LH’s score on the Liuing- Visud questions and his next worst score (which was on the Nonliving- Visual questions) to the difference between each normal subject’s worst and best score. In a t-test between LH’s difference score and the 12 normal difference scores, LH’s difference score was significantly larger, t( 11) = 2.33, P< 0.05.

DISCUSSION According to these results, LH has a selective deficit in semantic memory for visual information about living things. Thus, the deficit appears to have both modality-specificity and category-specificity. This conclusion can be drawn without reservations about the differential difficulty of retrieving visual or nonvisual information about living or nonliving things, because all comparisons have been made relative to the performance of normal subjects who were neither at floor nor at ceiling in their performance on these questions. This represents the first evidence, ofwhich we are aware, that speaks to the issue of whether prosopagnosia and visual object agnosia differ in degree or in kind from one another. According to the first of the hypotheses discussed earlier, the differential performance of prosopagnosic patients with living stimuli (primarily faces and animals) reflects an underlying impairment in a distinct visual subsystem specialized for this category of stimuli. According to the second of the hypotheses discussed earlier, this pattern of performance reflects the greater difficulty of a single, impaired visual-gnostic faculty with more complex and confusable stimuli. The critical data for distinguishing between these hypotheses is an

DISSOCIATIONS IN SEMANTIC MEMORY

199

independent measure of the difficulty of living and nonliving stimuli. The conclusion to be drawn from our data is that sheer difficulty, as measured by normal subjects’ performance, does not account for the difference between a prosopagnosic patient’s knowledge of the visual appearance of living and nonliving stimuli. Further work is needed to ascertain the generality of this result for other prosopagnosic patients and for other tasks. The present results do not conform neatly to either of the one-word characterizations of the organization of semantic memory in terms of “category” or “modality”. They imply that semantic memory is organized by both modality and category. The interpretation that we would tentatively suggest for these data is that the representation of visual appearance is not a unitary ability, and that the representation of visual stimuli that fall under the general category of “living things” may require particular capacities not needed for other categories of stimuli. This does not necessarily imply that the visual representation of living things depends upon an entirely distinct system from the visual representation of other stimuli. The present data are also consistent with a category-specific system that is used in conjunction with a more general system of visual representation for purposes of representing the appearances of living things. It is also worth noting that the category of stimuli handled by the hypothesized special system need not correspond literally to the English language category “living things”. It has already been noted that certain types of nonliving stimuli, such as specific makes of automobile, are generally not recognized by prosopagnosics. The true underlying nature of the distinction is probably formal rather than semantic. That is, it is probably based on differences in the aspects of shape that are typically important for recognizing living and nonliving stimuli, rather than differences in aliveness per se. Nevertheless, “living things”is a good approximation to the category ofstimuli most affected in prosopagnosia. Why would semantic memory be organized this way? More specifically, why would modality-specific visual knowledge be sub-divided categorically? Any answer to this question must at present be ad hoc, but we can suggest an answer which has some measure of independent support. Researchers in computational vision have been unsuccessful in finding a single system of representational primitives that is effective for all types of visual stimuli. This has led some authors to conclude that the problem of visual recognition requires more than one type of system of representation [17,26]. Psychologists who study face recognition have also noted that representational systems that work well for representing a wide range of other concrete three-dimensional objects are unsatisfactory for representing faces [lo]. These considerations provide a possible reason for the categorical dissociation within visual knowledge observed here. The brain has developed more than one system of visual representation in order to effectively recognize different types of stimuli, They are also consistent with the suggestion made earlier that the basis for the dissociation between living and nonliving things is formal rather than semantic in nature. What do the present results have to say about the other proposed organizations for semantic memory discussed earlier? We wish to point out that our results do not disconfirm these other hypotheses. On an empirical level, the pattern of dissociation that we have observed does not preclude the possible future observation of other patterns. On a theoretical level, the organization of semantic memory suggested by our data does not preclude other cross-cutting organizations. The implications of our data are narrower in scope, but nonetheless informative. In an experimental paradigm that avoids many of the methodological pitfalls of research in this area, we found evidence ofboth modality-specificity and category-specificity in the organization of memory for the appearances of living things.

200

MAKT‘HAJ. FARAH, KATHERINE M. HAMMOND, ZIYAH MEHTA and GRAHAM RATCLIFF

Acknowledgements-This research was supported by NIH grant NS23458. ONR Contract NOOl-l-86-0094, the Alfred P. Sloan Foundation, and an NIH program project grant to the Aphasia Research Center of the Boston University School of Medicine. We thank Saul Sternberg for his helpful comments on an earlier draft of this paper. We also thank David N. Levine and Ron Calvanio for introducing us to LH and for sharing many insights with us on the nature of LH’s neuropsychological condition. Last, we thank LH for his contribution to this research.

REFERENCES amnesia of patients with 1. ALBERT, M. S., BUTTERS. N. and LEVIN, J. Temporal gradients in the retrograde alcoholic Korsakoff’s disease. Brain 36, 21 l&216, 1979. 2. ALLPORT, D. A. Patterns and actions: cognitive mechanisms are content-specific. In Cognitirr Psychology .l’nf Directions, G. CLAXTON (Editor). Routledge & Kegan Paul, London, 1980. 3. ANDERSON, J. R. The .4rchitecture o/Coynirion. Harvard University Press, Cambridge, MA, 1983. 4. BAUER, R. M. and RUBENS, A. B. Agnosia. In Clinical Neuropsychology, (2nd Edition), K. M. Hy~t.hl~lu’ and E VALENSTE~N(Editors). Oxford University Press, New York, 1985. 5. BENTON, A. L. and VAN ALLEN, M. W. Prosopagnosia and facial discrimination. J. neural. Sci. IS, I67 171, 1972. 6. BEYN. E. S. and KNYAZEVA, G. R. The problem of prosopagnosia. J. Neural. Neurosury. Psychirctr. 25,154 158, 1962. I. BOUAMER,J. Die Prosop-Agnosie. Archs Psychiutr. Z. Neuro. 179, 6 54, 1947. 8. BOKNSTEIN,B. Prosopagnosia. In Problems ofDynamic~ Neurolo~g):. L. HALPEKN (Editor). Hadassah Medical Organization, Jerusalem, 1963. 9. BOKNSTEIN,B., STROKA, H. and MUNITZ. H. Prosopagnosia with animal face agnosia. Cortex 5, 164 169, 1969. 10. BRUCE, V. and YOUNG, A. Understanding face recognition. Br. J. Psycho/. 77, 305- 327, 1986. 11. COLLINS, A. M. and QUILLIAN, M. R. Retrieval time from semantic memory. .I. Verb. Learn. Vrrh. Behar. 8, 24&248, 1969. 12. DAMASIO, A. R., DAMASIO, H. and VAE~ HOESEN, G. W. Prosopagnosia: anatomic basis and behavioral mechanisms. Neurology 32, 331-341, 1982. 13. DENNIS, M. Dissociated naming and locating of body parts after left anterior temporal lobe resection: an experimental case study. Brain Lang. 3, 147-163, 1976. 14. FARAH, M. J. The neurological basis ofmental imagery: a componential analysis. Cognition 18,245~ 272. 1984. 15. GESCHWIND, N. Disconnexion syndromes in animals and man. Part Il. Brain 88, 585 645. 1965. 16. HART, J., BERNDT. R. S. and CARAMAZZA, A. Nature, Lond. 316, 439~440, 1985. Cognition 18, 65-96, 1985. 17. HOFFMAN, D. D. and RICHARDS, W. The parts of recognition. 18. HUMPHREYS, G. W. and RIDDOCH, M. J. To See but Not to See: A Cu.se Study of Visuul ACgnosiu. Erlbaum Associates, Hillsdale, NJ, 1987. 19. JOSSMAN, P. Zur psychopathologic des optisch-agnostichen storungen. Monatsschrifi fir Psychiatric und Neurologie 72, 81-149, 1929. 20. KAY. M. C. and LEVIN, H. S. Prosopagnosia. Am. J. Opthamol. 94, 75 80, 1982. 21. LEVIXE, D., CALVANIO, R. and WOLF, E. Disorders of visual behavior following bilateral posterior cerebral lesions. Psycho/. Res. 41, 217 234, 1980. 22. LEVINE, D. N., WARA(.H, J. and FARAH, M. J. Two visual systems in mental imagery: dissociation of”What”and “Where” in imagery disorders due to bilateral posterior cerebral lesions. Neurology 35, lot@ 1018, 1985. 23. LHERMITTE,J., CHAIN, F., ESTOUKELLE,R., DUCARNE. B. and PILLON. B. Etude anatomo-clinique d’un cas de prosopagnosie. Revue Neurologiqur 126, 329-346. 1972. 24. LHERMITTE,F. and PILLON, B. La prosopagnosie. Role de l’hemisphere droit dana la perception and visuelle. Rewe Neurologique 131, 791- 812, 1975. 25. MACRAE, D. and TROLLE, E. The defect of function in visual agnosia. Brain 79, 94 110, 1956. In Visual Cognition, S. PINKFR (Editor). MIT Press. Cambridge, 26. PINKER. S. Visual cognition: an introduction. 1985. 27. TEUBER, H. L. Alteration of perception and memory in man. In Andysis of’5ehariorul Change, Id. WLISKRAXTZ (Editor). Harper & Row, New York, 1968. 28. WARRINGTON, E. K. and SHALLICE, T. Category specific semantic impairments. Brcrin 107, X29- 854, 1984. a clinical, psychological, and anatomical study of 29. WHITEIXY, A. M. and WAKRINGTON. E. K. Prosopagnoaia: three patients. J. Neural., Neurosurg. Psychiatr.40, 395 403, 1977. 30. YAMADORI, A. and ALBERT, M. L. Word category aphasia. Corfev 9, 112-l 15. 1973.