the semantic deficit hypothesis: perceptual parsing …caram/pdfs/1982_caramazza_berndt.pdf ·...

BRAIN AND LANGUAGE 15, 161-189 (1982)

The Semantic Deficit Hypothesis: Perceptual Parsing and Object Classification by Aphasic Patients

ALFONSO CARAMAZZA AND RITA SLOAN BERNDT

The Johns Hopkins Universit~~

AND

HIRAM H. BROWNELL

Aphasia Research Center, Boston Veterans’ Administration Hospital

Many aphasic patients are impaired in their ability to provide or to recognize the names of objects, but little is known about the processing deficits that underlie these difficulties. In this report, a model of object naming/name recognition is proposed, and a prediction is tested concerning one possible functional locus of impairment in name-recognition and object-naming disorders. A subgroup of aphasic patients is found to be impaired in the ability to perform perceptual similarity judgments for pairs of stimulus objects, and to be unable to classify the objects into one of two lexical categories. It is concluded that the classification disorder suffered by these patients results from an impairment at the level of the semantically guided perceptual parsing of objects.

The relationship between words and the objects in the world that they signify is complex, undoubtedly involving considerably more than a series of one-to-one associations between words and objects. Between the sight- ing of an object and the articulation of its name, or between the perception of a word and the selection from alternatives of the object it represents, the cognitive system must perform considerable sifting and weighting of information, categorization of events, and matching of different sets of abstract conceptual elements. Although some progress has been made

This research was supported by NIH Grant 14099 to the Johns Hopkins University. We would like to thank the staff members of the Department of Audiology and Speech Pa- thology, Fort Howard VA Medical Center, and the Department of Hearing and Speech, The Good Samaritan Hospital, for their cooperation in this research project. We are especially grateful to Michael Giordano for the time and care that he devoted to performing the data analyses reported here. Address correspondence to: Alfonso Caramazza, De- partment of Psychology, The Johns Hopkins University, Baltimore, MD 21218.

161

0093-934X/82/010161-29$02.00/0 Copyright 0 1982 by Academic Press. Inc.

All rights of reproduction in any form reserved.

162 CARAMAZZA, BERNDT, AND BROWNELL

by cognitive psychologists in describing aspects of this system, there is no detailed model available of the processing mechanisms that are necessary for the naming of objects or the identification of objects from their names. This lack of a psychological model has left the neuropsy- chologist with little structure to guide research on the functional locus of impairments that seem to involve this system.

Many aphasic patients have difficulty with the relationship between objects and their names. In keeping with the traditional emphasis on expressive disorders in aphasia research, the largest share of attention has been directed toward aphasic deficits affecting the ability to name objects. One type of aphasic patient-the anemic aphasic-is defined by a problem in naming objects that is disproportionate to other aphasic symptoms. The primary explanation for the naming deficit of these patients is that they are unable to “retrieve” or to “find” the correct word to match a particular object (Goodglass & Geschwind, 1976; Weigel- Crump & Koenigsnecht, 1973). Although it is not explicitly stated, an underlying assumption of this view is that the structure of the patient’s “mental dictionary” is undisturbed; the problem is one of access to the lexicon. This “retrieval/arousal deficit” hypothesis is supported by several common observations: a patient who cannot name an object on one occasion may be able to do so on another (Goodglass & Geschwind, 1976); a patient who is totally unable to name an object may easily identify the correct name when it is presented to him (Geschwind, 1967). These two widely reported observations suggest that anemic patients have difficulties retrieving lexical information (i.e., words) from an intact mental dictionary.

An alternative hypothesis concerning deficits in comprehension as well as in production of object names is that they result from a disruption of the mental dictionary (Caramazza & Berndt, 1978). That is, the semantic representations contained in the mental dictionary are disrupted, making it difficult to retrieve the word associated with an object or event. One important source of evidence in support of this latter hypothesis is the clinical observation that anemic aphasics often produce semantic paraphasias in spontaneous speech (Geschwind, 1967). The production of semantic paraphasias suggests an impairment of the semantic organization of the mental dictionary since the word that is produced will necessarily violate some of the semantic aspects of the word that is intended. Further evidence that naming deficits involve semantic disruption has been provided by the finding that patients who produced semantic paraphasias also committed semantically based errors in an object selection task that included semantic distracters (Gainotti, 1976).

A major difference between these two hypotheses is that the retrieval- based argument is applicable only to situations in which the patient must produce a name, while the semantic deficit hypothesis predicts that patients should manifest problems in a wide range of tasks that require

SEMANTIC DEFICIT HYPOTHESIS 163

semantic analysis. The emphasis of recent research has thus shifted somewhat from confrontation naming to other semantically based skills. Often, however, an effort is made to link performance in semantic tasks with the ability to name objects.

Support for the hypothesis that the semantic component of the lexicon can be disrupted independently of other language functions has been reported in studies of patients with focal lesions to the left hemisphere (Grober, Perecman, Kellar, & Brown, 1980; Grossman, 1978; Lhermitte, Derouesne, & Lecours, 1971; Zurif, Caramazza, Myerson, & Galvin, 1974). These studies were directed at documenting semantic impairments rather than at discovering the underlying cause of naming impairments.

Lhermitte and co-workers asked two groups of patients to sort words on the basis of semantic relatedness. They found that patients with posterior involvement (Wernicke’s aphasics) would violate semantic re- strictions by inappropriately grouping words of clearly different meaning-a so-called “broadening” of the semantic field. Importantly, the authors report that this deficit in semantic analysis appeared to be independent of “general intelligence.” In a somewhat similar study, Gross- man (1978) asked left-anterior-damaged and left-posterior-damaged patients to name as many members as they could of two superordinate categories (e.g., “furniture”). Grossman found that the posterior patients inappropriately listed items that were not members of the designated categories, demonstrating again a “broadening” of the semantic organization of the lexicon. More generally, Grossman (1978), Grober et al. (1980), and Buhr (1980) have shown that patients with posterior damage tend to have more difficulties with low-typical than with high-typical members of a category (Rosch, 1975).

Zurif et al. (1974) asked left-anterior- and left-posterior-damaged patients to decide which two of three words were most closely related in meaning. The word triads were generated from a semantically restricted set of 12 words (6 human and 6 animal terms). Multidimensional scaling and cluster analysis of the relatedness judgments produced by the posterior-damaged aphasics revealed markedly disorganized solutions and, by inference, a severe lexical-semantic impairment.

Two studies that are directly relevant to the issue of naming disturbances have been reported by Goodglass and Baker (1976) and by Cough- lan and Warrington (1978). Goodglass and Baker assessed the ability of a “low-comprehension” and a “high-comprehension” group of aphasics to recognize words associated with a pictured object. The low-comprehension patients (presumably posterior damaged) failed to recognize functional associates and functional contexts of the objects, especially those objects they were unable to name. On the basis of these results, Good- glass and Baker argue that the low-comprehension patients have a qual-


itatively different form of semantic organization. That is, these patients manifest a disruption of their semantic network that is related to their naming ability.

Coughlan and Warrington (1978) concentrated primarily on the types of errors patients make in an object-naming and in a naming-to-description task. They tested patients with unilateral damage to the left hemisphere, as well as two control groups of patients. A substantial proportion of the errors produced by the aphasic group (44 and 29% in the object- and description-naming tasks, respectively) could be considered to result from a failure to carry out adequate semantic analyses. These were errors that involved the production of semantically related words, phrases that indicated only partial knowledge of the object or word, or unrelated words. These authors failed to find any strong correlation between site of lesion and phoneme discrimination, ruling out the possibility that the lexical-semantic processing impairments (of the temporal-lobe- damaged patients) resulted from a perceptual processing deficit, as suggested by Luria (1970).

The research reviewed here suggests that brain damage can result in selective disruption of the semantic organization of the lexicon. In addition, several of the studies allow the conclusion that such semantic disruption can be related to the patient’s impaired ability to name objects. Stronger conclusions with regard to the functional locus of impairment of the various levels of lexical-semantic processing cannot be made on the basis of these studies, and as noted above a major reason for this is the lack of a well-developed psychological model of naming and/or lexical processing that, although still at a preliminary stage, is sufficiently detailed to allow predictions about performance in a range of tasks in addition to object naming.

Caramazza and Berndt (1978) have presented a skeletal representation of the processes involved in object naming and in word/object matching (choosing which object represents a presented word). We have identified at least four “stages” in each of the two processes. In the case of object naming these include a perceptual analysis of the presented object; a modality-specific analysis which consists of a semantically constrained parsing of the perceptual input; a modality-independent semantic description which allows the selection of an appropriate lexical form; and the execution of the phonological information specified for the selected lexical item. A parallel but reverse sequence of processes is assumed to be involved in a word-object matching task. This skeletal model serves as the basis for the further elaboration of the processes of lexical access and semantic analysis to be developed here.

We begin by assuming that a name will be assigned to an object if the object satisfies some set of criteria. Some of the properties could be functional attributes, such as the use to which the object may be put.


The meaning of a word (the concept) will thus consist of a list of perceptual, functional and other abstract properties that can be used jointly to define the members of the category that can be labeled by the word (Labov, 1973; Miller & Johnson-Laird, 1976; but see Fodor, Garrett, Walker, & Parkes, 1980). For example, the concept CUP may be defined by the following list of properties: “artifact, upwardly concave, may have handle, height about equal to top diameter, used as a container from which to drink hot liquids.” If an object has these properties, it can be labeled cup. ’

It may be noted that the list of properties given for the concept CUP includes a feature (“may have a handle”) that need not necessarily be present for inclusion of the object in the category cup. Other features in the list presented are necessary attributes of the concept (e.g., “concave upward”), and their absence from an object constitutes sufficient grounds for the exclusion of the object from the category.

Finally, it may be noted that no single feature is sufficient to uniquely determine the category membership of an object (see Hersh & Cara- mazza, 1976; Labov, 1973, for discussion). In the present case, concave upward must be supplemented by the proper diameter-to-height ratio and may be supplemented by the presence of a handle to distinguish cups from bowls.

This genera1 view of the naming/name recognition process, and of the componential structure of word meaning, can be further elaborated to include the kind of perceptual parsing of an object that must serve as input to a recognition and naming algorithm. The assumption can be made that the perceptual parsing of an object includes as part of its output those features that are represented as part of the concept. In this view, at some point in the parsing process relatively abstract values are assigned to parts of the stimulus-values that correspond to the componential structure of a category concept. We can imagine an exclusively “bottom-up” parsing which recovers from an object a set of perceptual features that are then matched to the mental representation of the category. However, this exclusively “bottom-up” processing view fails to specify how the critical step from strictly perceptual to conceptual properties is accomplished. In fact, there is no a priori justification for assuming that a “bottom-up” processing system will produce as output the unique parsing that is needed for a particular comparison. Consid- erations such as these have led some investigators to assume that the perceptual parsing process has a “top-down” component (Neisser, 1967;

’ We have chosen to cast our model in terms of a componential model of semantics because this approach is at present the most fully developed theoretically and because it seems best suited for analyses involving perceptually based semantic elements. It would be possible to describe the results of the present research in terms of, for example, a prototype model of meaning without altering the general conclusions.


1976; Palmer, 1975). Specifically, it seems reasonable to consider the parsing process as guided by potential feature matches that are represented in semantic memory.

With regard to the example of cup, the handle and diameter-to-height ratio of the object must be parsed as important subunits of the total stimulus, as opposed to any one of an indefinite number of other possible subgroupings of its parts. However, it seems unreasonable to assume that the abstract properties under consideration emerge spontaneously by the application of an exclusively “bottom-up” parsing process. A more reasonable explanation is that after preliminary low-level analyses an active top-down component guides the parser to search the stimulus for the presence of “semantically” interpretable components.

In light of the above remarks we can elaborate our model to include the following sequence of events:

1. Low-level perceptual analysis restricts the initial search space for an internal description of the presented object.

2. A parser uses semantic level information to produce “semantically interpreted” components that serve as inputs to a classification algorithm; the semantically interpreted components are modality specific.

3. The classification algorithm assigns category membership by determining whether an object has the criteria1 values that define a category; the category is represented in a modality-independent format.

4. The selected category maps onto a particular lexical item that specifies phonological and syntactic information.

5. The phonological representation selected is articulated. The retrieval/arousal deficit hypothesis presumably specifies a break-

down at stages 4 and/or 5, while the semantic deficit hypothesis locates the problem at stage 2, where semantic category representations are postulated to be at least partly disrupted. Note that a partial disruption of the feature list associated with a concept will result in a variety of disorders such as those described in the literature that involve word processing alone (Goodglass & Baker, 1976) or processing of words as well as of visual information (Whitehouse, Caramazza, & Zurif, 1978).

This description of the naming model has been developed primarily on logical grounds. However, there is neuropsychological evidence that can be interpreted as support for various aspects of our model. Probably least controversial is the claim that there is a level of semantic representation that is modality independent (Point 3). Neuropsychological support for this claim is reported by Goodglass, Barton, and Kaplan (1968), who found that in a large group of aphasic patients, naming defects were mostly independent of the four modaiities tested. In addition, Spreen, Benton, and Van Allen (1966) found that 16 of 21 aphasics tested did not show a disassociation between tactile and visual naming. Goodglass et al. (1968) conclude on the basis of their results that a “modality nonspecific” process mediates naming.


There is equally compelling evidence that there is a level of representation in the naming process at which analysis is carried out in a modality specific format (Point 2). An example is the existence of modality-specific anomias most often discussed in terms of the disconnection syndromes (Geschwind, 1965). The mental dictionary of these patients appears to be unimpaired; modality-specific processes are disrupted prior to the selection of a lexical item.

The model we have described is explicit enough to allow the construc- tion of experimental tasks that are directed at uncovering the functional site of breakdown in aphasic patients. Consider in this regard the line drawings of cup-like stimuli shown in Fig. 1.

The objects depicted in this figure vary on two dimensions: a continuous dimension of diameter-to-height ratio and a discrete dimension of presence/absence of a handle. It is clear that the narrower stimuli could be labeled “cups” while the wider ones could be labeled “bowls.” The particular label applied to any of the depicted objects depends on the specific diameter to height ratio, whether the object has or does not have a handle, and the functional use of the object (whether one is drinking coffee or eating cereal from it). Thus, for example, the fourth stimulus in the top row could be labeled “cup,” depending on how much importance one assigns to the presence of a handle in relation to the specific diameter-to-height ratio of the depicted object (Labov, 1973). Thus, proper naming of these objects requires normal parsing of the visual stimulus to recover the two important dimensions of diameter-to-height ratio and presence or absence of a handle, which are part of the semantic descriptions of the concepts cup and bowl. If we were to ask aphasic patients to name each of the line drawings in Fig. 1, both the semantic deficit and the retrieval/arousal deficit hypothesis predict poor performance. However, if we were to provide the patient with two categories (“Cup” and “bowl”) and ask them to decide between these two categories for each of the depicted objects, only the semantic deficit hypothesis predicts poor performance in this task, since providing the names of the objects circumvents the retrieval deficit.

Whitehouse et al. (1978) have tested these predictions. They compared the classification performance of a group of anemic and a group of

I 2 3 4 5 6 7

FIG. 1. Stimulus items with two components of variation.


Broca’s aphasics using objects similar to those in Fig. 1. The Broca patients did not present any marked difficulty choosing the proper label for the objects on the basis of the two perceptual dimensions we have described and a functional context cue (type of food or drink being poured into the object). The anemic aphasics, on the other hand, were clearly unable to integrate the two perceptual dimensions for the proper classification of objects despite the fact that the labels were provided as alternatives. In addition, they were unable to integrate successfully the functional context cue to determine the appropriate category of an object.

Whitehouse et al. (1978) and Caramazza and Berndt (1978; in press) have argued that this result is inconsistent with the retrieval/arousal deficit hypothesis. That is, the lexical categories were severely constrained, and the labels were provided for the patient so that the actual names for the objects did not need to be “aroused.” Nonetheless, the precise functional locus of breakdown is still not clear.

We have emphasized that the perceptual parsing of an object is guided by semantic considerations, such that the output of the perceptual parser consists of modality-specific, semantically interpreted features. If the semantic component is disrupted, it should affect the functioning of the parser in its production of semantically interpreted features. In this regard, it is interesting to consider what should happen when a subject is asked to judge the perceptual similarity between pairs of the cup-like objects presented in Fig. 1. It has been demonstrated repeatedly that similarity judgments reflect the componential structure of the stimulus set (e.g., Shepard, 1964; Torgerson, 1965). For example, in studies in which the stimulus set consisted of color patches or geometric shapes, multidimensional scaling analyses of the similarity data revealed that subjects used the dimensional (feature) structure of the stimuli to make similarity judgments. That is, the recovered dimensions are just the sa- lient perceptual attributes that characterize the stimulus set. It has also been shown that judgments of similarity of words or objects clearly reflect the categorical structure of the stimulus set, as well as underlying dimensions shared across categories (Caramazza, Hersh, & Torgerson, 1976; Degerman, 1970). Thus, for our cup-like stimuli we expect judgments of similarity to reflect the quantitative perceptual variation (e.g., diameter-to-height ratio) as well as the categorical differences among stimuli (e.g., cups or bowls).

The logic underlying this claim can best be clarified by considering once again the stimuli presented in Fig. 1. These stimuli vary in equal increments on the dimension of diameter-to-height ratio such that the lower ratio stimuli look like cups while the higher ratio stimuli look like bowls. This particular stimulus set was chosen because it has the property that two different concepts or categories are specified by different values


on the same dimension; because there is no well-defined criteria1 value on the diameter-to-height ratio dimension that can be used to distinguish between the two concepts; and because this stimulus set can have a perceptual attribute (handle) that is characteristically associated with one of the concepts (cup) but not the other (bow0 (Labov, 1973).

Specific predictions about how subjects will judge the similarity of pairs of the cup-like objects can be made on the basis of the description of the naming process that we have offered. Consider first the case in which the ratio value for a pair of stimuli clearly fall within the range of values of a particular concept-either both cups or bowls. Since values on this dimension are defining attributes of the categories, and since the ratio values are within the same category, the pair of stimuli will be judged to be similar. If, on the other hand, the two stimulus values are not clearly from the same category, then the overall judged similarity for the pair will be relatively low. The assumption underlying these predictions is that in parsing the visual stimuli the subject encodes the diameter-to-height ratio values. The similarity judgment is then based on the abstract representation of these attributes which are “identical” in one instance (the same category) and “different” in the other instance (different categories).

For patients who have difficulty classifying the objects as “cups” or “bowls,” however, the semantic deficit hypothesis predicts that the categorical information about the stimuli should be impaired. Thus, the similarity judgments of these patients should reflect disordered relations despite the fact that the only judgment required is one that involves a perceptual comparison, since the impaired semantic categorical information will not provide sufficient guidance for the parsing mechanism. Furthermore, a comparison of patients’ similarity judgments with their ability to classify or to label the stimuli should provide important information about the relationships among perceptual parsing, object classification, and name recognition.

To address these issues we tested the performance of a group of 20 left-hemisphere damaged, 10 right-hemisphere damaged, and 10 neurologically intact patients on a similarity-judgments task and a classification or labeling task. We did not use an a priori classification of the aphasic patients as in earlier work (Whitehouse et al., 1978), but used the data obtained from the experimental tasks themselves to determine empirically the subgrouping of patients. In this way we avoided the problem of masking potentially important differences that might be present within clinically defined aphasic types. That is, we are not looking simply for “anemic” deficits within a particular clinical group, but are probing for possibly subtle lexical-semantic deficits that might affect other types of aphasic patients as well.


Subjects

METHOD

The experimental group included 20 right-handed male aphasic patients, all victims of a single left-hemisphere cerebrovascular accident. The mean age of the group was 54 years (range: 35-65 years) and mean level of education was 12 years (range: 6-19 years). The majority of patients (14) were out-patients in speech therapy at the time of testing; the remainder were hospital in-patients undergoing intensive stroke rehabilitation therapy. Minimum time postonset was approximately 2 months, the maximum was 5 years, and the mean was 20 months.

Ten right-handed male patients who had suffered a right-hemisphere cerebrovascular accident were also tested. The mean age of this group was 58 years (range: 48-66 years), with an average education level of 9 years (range: 6-13 years). All patients were free from any clinical sign of language disturbances, and all but two were hospital in-patients undergoing stroke rehabilitation therapy.

Ten neurologically intact male patients were tested as controls. Mean age of this group was 59 years (range: 53-66 years), and average educational level was 10 years (range 5-15 years). Seven were in-patients and three out-patients undergoing rehabilitation therapy for various nonneurological complaints.

Patients in all three groups were free from clinical signs of diffuse cortical damage, including the pathology presumed to accompany chronic alcohol abuse.

Stimuli Similarity-judgment task. Line drawings of cup-like stimuli such as those presented in

Fig. 1 were used as stimuli. The stimuli varied on a dimension defined by the ratio of diameter to height. Two stimuli were prepared for each of the following seven diameter- to-height ratios: 0.8, 1.2, 1.6, 2.0, 2.4, 2.8, 3.2. For each ratio value pair, one stimulus was represented with a handle and the other without a handle to produce the 14 stimuli shown in Fig. 1. Each of these 14 stimuli was paired with each other stimulus to produce a total of 91 pairs of stimuli. Each pair was reproduced as a slide and projected onto a screen.

Labeling task. The 14 line drawings used in the similarity judgment task were used in the labeling task. In addition, line drawings of a coffee pot shown in a position appropriate for pouring coffee and a box of cereal shown in a position appropriate for pouring cereal were prepared to test the effects of functional context on classification. The cup-like stimuli and the two context pictures could be combined to produce composite pictures showing a cup-like stimulus and either the box of cereal or the coffee pot pouring its contents into the cup-like object. These stimuli were drawn individually on 8 x IO-in. sheets of white paper.

Procedure Similariry-judgment tusk. Patients were required to judge the perceptual similarity be-

tween all possible pairs of the cup-like drawings, which were projected by individual pairs onto a screen. Patients’ judgments were conveyed by means of a 7-point scale, with “1” representing “very similar” and “7” representing “very different.” A vertically oriented scale with the numbers 1 through 7 and the words “similar” and “different” at the appropriate endpoints was kept in the patient’s view; his task was to point to the number on the scale that represented his judgment for each pair.

An extensive training procedure was employed to ensure that patients understood the nature of the task. Training material included a set of seven pairs of geometric forms and a set of eight pairs of animal terms. The geometric forms differed on three dimensions: size, shape, and color. The initial segment of the training was essentially a demonstration of the scale using pairs of geometric forms, beginning with the most similar items, pro-


gressing to the most dissimilar. The dimensions of variation were pointed out to the patient if he did not appear to appreciate them spontaneously. It was emphasized that the scale values did not represent any absolute “right” or “wrong” answers, but were only a subjective judgment that might differ somewhat from person to person. After the demonstration, the geometric stimuli were presented randomly a second time for the patient’s judgments. If responses were not reasonable (e.g., were perseverative or clear reversals of the scale), the demonstration was repeated. If the patient completed this set satisfactorily, he was presented with pairs of animal names (typed on a card and read aloud). Patients were not provided with a demonstration for this set of items, but were required to produce their own judgments, with appropriate feedback and encouragement. The experimental task was administered only when the interviewer was confident that the patient understood the nature of the task and could respond appropriately. Successful completion of the training was interpreted as a demonstration of two important points. First, the patient was assumed capable of performing a similarity judgments task. Second, he demonstrated an intact ability to perform a low-level perceptual analysis of visually presented stimuli.

The slides containing the cup-like stimuli were presented to each patient in one of six predetermined quasi-random orders. Each slide was allowed to remain in view until the patient produced a response. Twenty pairs of stimuli (a different random selection for each patient) were rated twice to provide an index of reliability. In most cases, all 91 judgments and the reliability items were completed during one testing session. When it was necessary to complete the task in more than one session, the training was repeated briefly and limited cross-session reliability was obtained.

Labeling rusk. The labeling task was administered after the similarity judgments had been obtained, but not in the same testing session. Patients were told that they would be shown pictures that they had seen before, each of which could be called either a “cup” or a “bowl.” It was emphasized that for some of the items it was difficult to decide which name should apply, but that they should attempt to label each item with one of the two names. No other label (e.g., “dish, ” “glass”) was entertained. Each of the 14 stimuli was presented to the patient six times; in half of these trials (the “bowl” context) the container was shown with a picture of a cereal box, held so that the cereal appeared to be pouring into the container. In the other 42 trials, no context was provided. However. to avoid the implication that “no-context” should be interpreted as “no-cereal context.” four trials were randomly interspersed in which the container was shown with a picture of a coffee pot held so that coffee appeared to be pouring into the container. These items were not scored. One complete replication of the set included a random presentation of each item, each item with cereal context and the coffee context items. The set was repeated three times, in a different order. to each patient. The printed words “cup” and “bowl” were kept in view, and the patients were occasionally reminded aurally of the two alternatives. This task was most often completed in two sessions, and a composite profile across the three replications was constructed.

RESULTS AND DISCUSSION

Judgments of Similarity

The reliability of patients’ similarity judgments, computed on a random set of 20 stimulus pairs rated twice, were calculated separately for each clinical group. The mean correlation between these repeated judgments was .67 for the left-hemisphere-damaged patients, .51 for the right-hemisphere-damaged patients, and .71 for the control group. Since a different random sample of items was repeated for each subject, direct compar- isons of individual patients’ reliability scores cannot easily be made.


Nonetheless, since some individual patients did not exhibit a high degree of reliability in their judgments, particular attention was paid to assessing the contribution of the reliability of patients’ judgments to the outcome of the analyses that follow.

The patients tested in this study were initially grouped on the basis of whether they had suffered left- or right-hemisphere damage, or were free of neurological impairment. It is obvious that this gross grouping will mask important differences within the three groups, particularly within the left-hemisphere-damaged group. It is well established that patients with damage to the left-hemisphere can vary considerably in their ability to process language. Similarly, the right-hemisphere-damaged patients can differ in their ability to perform visual-perceptual analyses (e.g., Benton, 1979). It is important, therefore, that we establish the extent to which patients within each group are homogeneous in their performance of a task. Should there be significant variation in performance within groups, the data obtained for homogeneous subgroups can be analyzed separately.

The approach we have adopted to separate patients into homogeneous subgroups does not rely on the clinical profiles of the patients. Although grouping by clinical profiles can be extremely useful, it can also be misleading. That is, the theoretical basis for the clinical classification is unlikely to be sufficiently detailed to allow the fine differentiations in performance typically sought in experimental investigations. Thus, in- discriminately averaging patients’ performance in clinically determined groups can result in severe distortions of individual patient abilities. A reasonable alternative is to examine patients’ performance on the experimental tasks to determine the extent of individual differences and to base the subgrouping of patients on the experimental data. In this way, the subgroups that emerge will reflect the extent to which the task taps processes and structures that may be differentially available to the patient group tested.

In the present case, we will separate patients on the basis of the internal metric that they used to judge the similarity of the cup-like stimuli. The justification for using this criterion has already been discussed in the introduction. The basic notion is that the similarity judgments reflect the complex process of perceptual parsing and semantic analysis. Disruptions at different levels of the categorization process will presumably result in the application of different similarity metrics. The identification of patients’ use of a particular similarity metric will serve as the empirical justification for separating patients into homogeneous subgroups.

An empirical procedure for separating subjects into subgroups utilizes an inverse principal components analysis (Tucker & Messick, 1963). This analysis takes as input a stimulus pairs x subjects matrix and treats


subjects as variables in a principal components factor analysis. The results of the analysis show which subjects treat the experimental task in similar fashion; those subjects who cluster together in the principal components space (i.e., those with similar factor loadings) may reasonably be grouped together with respect to task performance. The first component (or dimension) accounts for the most variance and, generally, serves to identify major subject groups. More specifically, subjects who load highly on dimension 1 have performed the task similarly and, by consensus, “correctly,” while other subjects have performed discrepantly.

The results of this analysis carried out for all patients together is shown in Fig. 2. Especially for those patients with left-hemisphere lesions, there is a relatively clear separation on dimension 1, which accounted for 45% of the total subject variance. One group (31 patients) loaded highly on dimension 1 (loadings > 0.0) and a second group (8 patients) had lower loadings (~0.0) on this dimension. A single subject, forming the third group, was well separated from the other patients. The second dimension, accounting for 7.5% of the total variance, did not produce clear separation among patients.

As can be seen in Fig. 2, most of the patients in the two control groups, and 14 of the left-hemisphere patients, appeared to be using the same internal metric in making the similarity judgments. Importantly, six left-hemisphere aphasics appeared to base their judgments on factors other than those employed by the majority of patients.

Separate inverse principal components analyses performed on the similarity judgments produced by the left-hemisphere-damaged, right-hemisphere-damaged, and neurologically intact patients supported the group-

e C

‘! L :R ! -

‘k .R

R .c

RL .L.L

E :f I

.R c LR ’

L. .L R’ ‘.c

.L . R .L

C l L .L ‘*L l

t

h .L

‘L

FIG. 2. Inverse principal-components analysis of similarity judgments for all patients (L = left-hemisphere-damaged patients: R = right-hemisphere-damaged patients; C = control patients).

174 CARAMAZZA. BERNDT. AND BROWNELL

.V.O’F.

0;R.

l B.L.

l R.L.

Ad? l l C.E

GiW. .E.W.

?.E. I

3iD. c.T l l W.R

i;K. C;R.

l F.S.

3.6 l E.D.

FIG. 3. Inverse principal components analysis of similarity judgments for left-hemisphere patients (position of individual patients marked by their initials).

ings obtained in the overall analysis. The results of the analysis carried out separately for the left-hemisphere-damaged group are shown in Fig. 3. As can be seen, there is a clear separation of patients on dimension 1, which accounted for 50% of the variance. Subgroup 1 (14 patients) loaded highly on dimension 1 and fell together in a cluster; subgroup 2 (6 patients) had a low loading on dimension 1. The further separation of subgroup 2 on the basis of differential loading on dimension 2, which accounted for 9.5% of the variance, did not prove useful in subsequent analyses and will not be considered further.’ All subsequent analyses of the aphasic patients’ performance will be carried out separately for subgroups 1 and 2. It should be emphasized that the aphasic patients were divided into these groups on the basis of the principal components analysis before any other analyses were performed.

Comparison of the mean reliability coefficients for subgroups 1 and 2 indicate that patients in subgroup 1 were considerably more reliable as a group (r = .73) than patients in subgroup 2 (Y = .5 1). Within subgroup 2, however, there were three patients whose reliabilities were relatively high (>.65) and three others whose judgments were less reliable. In subsequent analyses, the judgments of these two subgroups of three patients were analyzed separately to ensure that the performance

* Separate multidimensional scaling analyses were carried out for subgroups 2a and Zb, consisting of three patients each. No appreciable differences were observed. The reason for a lack of difference may involve the relative insensitivity of the analyses to differences other than substantial ones. It should be noted that the composition of subgroups 2a and 2b does not match the division of patients within subgroup 2 that is made on the basis of the computed reliability of their judgments.

SEMANTIC DEFICIT HYPOTHESIS

of each subgroup reflected the profile that is discussed as characterizing group 2 as a whole. This method provides good evidence that the abnormal pattern produced by subgroup 2 is not merely a function of a lack of reliability in performance of the task.

Separate inverse principal components analyses were carried out for the right-hemisphere damaged and the neurologically intact patients, Eight patients in the right-hemisphere group loaded highly on dimension 1 (53% of the variance). The other two subjects were clearly separated from these eight patients and their data will not be considered further. Similarly, the neurologically intact group separated into a homogeneous subgroup of nine patients who loaded highly on dimension 1 (53% of the variance) and one patient who was clearly separated from the rest.’

The results of the inverse principal components analysis of the similarity data generated by the aphasic patients dictate the separation of this group into two subgroups for further analysis. Accordingly, the similarity data were collapsed across individual subjects within each of these two subgroups. The mean similarity ratings for the 91 pairs of stimuli were analyzed for each subgroup using the KYST-2 Multidi- mensional Scaling Program (Kruskal, Young, & Seery, 1977). Solutions of varying dimensionality were tried, but in both cases a two-dimensional solution proved best in terms of interpretability and closeness of fit of the solution to the data. Stress for the two-dimensional solution for subgroup 1 was .08, indicating a relatively close monotonic fit between the obtained configuration and the data (Kruskal, 1964). A less acceptable stress value was obtained for subgroup 2; for this group, stress was .15, indicating a poorer monotonic fit between the obtained configuration and the data.

The two-dimensional solution for subgroup I of the aphasic patients is shown in Fig. 4, in which the numbers represent stimulus items. It is apparent that dimensions 1 and 2 correspond to stimulus variation on the ratio of diameter-to-height and the presence or absence of a handle, respectively. The aphasic patients in this group were clearly able to assess independently the contribution of the two physical dimensions that characterized the stimulus set in determining the similarity between members of each pair of stimuli.

The two-dimensional solution for subgroup 2 is presented in Fig. 5. In this solution the only readily interpretable dimension is dimension 1, which corresponds to the presence/absence of a handle variation among stimuli. The clear separation among items on the basis of this feature demonstrates that patients in group 2 could perform a reliable perceptual analysis of the stimulus items and that they could perform the rating

’ There is no theoretically motivated explanation for the lack of total homogeneity within these two “control” groups. In addition, the very small number of nonconforming patients mitigates against further analysis of their judgments using multidimensional scaling techniques.


II

z 4

2 ’ . ? B . !

I

12 II

. .

‘P

? !3

? 14 .

FIG. 4. Two-dimensional multidimensional scaling solution for similarity judgments of subgroup 1 (N= 14) left-hemisphere-damaged patients (numbers label stimulus items, with stimuli 1-7 having handles and stimuli 8-14 without handles).

task. The categorically important dimension of diameter-to-height ratio is not clearly reflected in this solution. This result suggests that the aphasic patients in this group were not making adequate use of a definitionally important dimension of the stimuli in making their judgments of similarity.

In an attempt to evaluate the contribution of the low reliability scores for some of the patients in subgroup 2 to this aberrant scaling solution,

I I

. ‘2 14 . 13 .

IO .

I I .

0

FIG. 5. Two-dimensional multidimensional scaling solution for similarity judgments of subgroup 2 (N =6), left-hemisphere-damaged patients (numbers label stimulus items, with stimuli 1-7 having handles, and stimuli 8-14 without handles).


a separate multidimensional scaling analysis was carried out for the three patients with relatively high reliabilities. The solution obtained for this group of ‘reliable” group 2 patients differed from the normal (i.e., group 1) solution to the same extent as did the group 2 solution as a whole. That is, the similarity judgments of the three patients failed to reflect the dimensional information in the stimulus set. These results support the view that the deviant solution obtained for the similarity judgments produced by group 2 is not simply a reflection of the inability of this group of aphasic patients to perform the task consistently.

To characterize further the patients’ ability to use categorical information in their similarity judgments, we analyzed the similarity matrices for groups 1 and 2 using a hierarchical clustering program (Johnson, 1967). The solutions obtained are shown in Figs. 6 and 7 for subgroups 1 and 2, respectively. The hierarchical clustering solutions are shown superimposed on the two-dimensional solutions obtained through the application of the multidimensional scaling programs. The clustering program yields a pictographic model of the similarity ratings produced by the patients by generating a hierarchical representation of the judged “closeness” among the stimuli. Stimulus items within the smallest circles are judged most similar, with decreasing similarity represented by inclusion in larger circles. It is apparent that the patients in subgroup 1 clustered the stimuli by semantic category. That is, the clustering solutions reveal two clearly separable clusters that correspond to the categories cup and bowl. The clustering solution obtained for the patients in subgroup 2 shows a clustering determined by a physical feature (handle/no handle) of secondary importance to the categorical nature of the stimulus set. In other words, these patients failed to make use of the categorical value of the individual stimuli in judging the relatedness among them.

FIG. 6. Hierarchical clustering solution for similarity judgments of subgroup 1, left- hemisphere group. superimposed on multidimensional scaling solution.


FIG. 7. Hierarchical clustering solution for similarity judgments of subgroup 2, left- hemisphere group, superimposed on multidimensional scaling solution.

The major feature distinguishing the two groups of aphasics appears to be the importance they assign to the diameter-to-height ratio dimension and the corresponding categorical value of this dimension in determining whether a stimulus belongs to the cup or the bowl category. This conclusion is inferred from a consideration of the separate solutions obtained for the two groups of patients. In these solutions it is clear that patients in subgroup 2 do not adequately weight the diameter-to-height ratio dimension in making their similarity judgments.

This conclusion can be further validated by analyzing the similarity data for all aphasic patients together using the Individual Differences Scaling program, INDSCAL (Carroll & Chang, 1970). This program ana- lyzes individual subject differences by determining the pattern of weights that each subject seems to adopt for a multidimensional solution of common dimensionality. A matrix of the weights that were apparently adopted by the twenty patients in the left-hemisphere group is shown in Fig. 8, and the two-dimensional solution that resulted from the application of those weights appears in Fig. 9.

The first important point to note in these results is that the two-dimensional solution for the whole group of aphasic patients retains the normal structure of the two physical dimensions that define the stimulus set (variance accounted for by two-dimensional solution = 55%). How- ever, when we consider the subject weights it is clear that the six patients who comprise subgroup 2 assign low weights to dimension 1 (diameter- to-height ratio), while the 14 patients in subgroup 1 assign high weights

II

l.O-

7.5 -

5.0-

2.5 -

4.T.

V.?‘F ; K. A.P C.5.’

l C.B. E.W.* .G.W

?A. O.E. B;L. C.R.: l W.R.

Y.L. F. S..

E.D.* l C.T.

l R.C.

I I B.D.* , I 7 0.01 0.0 .25 .50 .75 1.0 1


M.B .

179

FIG. 8. Plot of individual left-hemisphere-damaged patients’ weights (from INDSCAL) for diameter-to-height ratio (I) and handle/no handle (1 I) dimensions.

to this dimension. Thus, the results of INDSCAL support the interpretation of the other scaling results that patients in subgroup 2 do not adequately weight the dimension of physical variation that is critical for the proper categorization of the stimuli as cups or bowls.

The performance of the other two groups of patients, the right-hemisphere-damaged and the neurologically intact controls, was similar to that of subgroup 1, as expected on the basis of the results of the inverse

6 4 ! . 1 3 .

l 2 . 5

I

12 .

‘P II 13 . 14

? .

?

FIG. 9. Two-dimensional INDSCAL multidimensional scaling solution of similarity judgments for all left-hemisphere-damaged patients.


FIG. 10. Hierarchical clustering solution for similarity judgments of right-hemisphere- damaged patients, superimposed on INDSCAL multidimensional scaling solution.

principal components analysis for the entire group of patients. That is, the aphasic patients in subgroup 1 and those in the two control groups all clustered together, loading highly on dimension 1 of the inverse principal-components analysis. The similarity matrices for the two control patient groups were also analyzed using multidimensional scaling and clustering procedures, and the results of these analyses are shown in Figs. 10 and 11 for the right-hemisphere-damaged and the neurologically

FIG. Il. Hierarchical clustering solution for similarity judgments of control group, superimposed on INDSCAL multidimensional scaling solution.


intact groups, respectively. The two-dimensional solutions for these two groups exhibited an acceptable monotonic fit with the data, as indicated by stress values of .12 for the right-hemisphere-damaged group and . 10 for the neurologically intact group. The major point to note in these results is that the two dimensions recovered by the patients reflect the physical dimensional structure of the stimuli. More importantly, the major weight is assigned to dimension 1, which is the more important dimension in defining the two categories CUP and bowl. It should be noted that a normal solution was obtained for the right-hemisphere-damaged patients despite the low reliability scores obtained for several of the patients in this group.

The analyses of the similarity data we have carried out for the different groups of patients reveal an interesting pattern of processing difficulties for a subgroup of the aphasic patients. The 14 patients of subgroup 1 performed normally on the judgment of similarity task, suggesting that they had an intact semantic system or, at least, that they could carry out semantically guided perceptual parsing of graphic inputs. The other subgroup of six aphasic patients showed an impairment in perceptual parsing. This impairment is not a strictly perceptual one, however. These patients could adequately use one perceptual feature, presence/absence of a handle, to judge the similarity among stimuli. Their perceptual parsing deficit appears to be one that is related to a deeper linguistic deficit. Namely, these patients could not make adequate use of semantic information to carry out a semantically interpreted parsing of the graphic inputs. If this interpretation of the deficit underlying the performance of patients in subgroup 2 is correct, then we should be able to predict a specific pattern of differential performance between this subgroup bf patients and patients in subgroup 1 in a categorization task.

In the introduction we argued that a minimally adequate model of naming or categorization must postulate a level of analysis at which the parsing of the graphic input is guided by semantic considerations. That is, there must be a level of perceptual representation that corresponds to the semantically relevant elements or features that define a concept. A disruption of the conceptual representation will, consequently, ad- versely affect the semantically interpreted parsing of an object. If patients in subgroup 2 have a conceptual representation deficit, then these patients should not produce normal semantically interpreted parsings of graphic inputs. Furthermore, if these patients are asked to categorize graphic inputs into semantic categories they should reveal a deficit comparable to the one that is demonstrable at the level of perceptual parsing. Stated in more concrete terms, patients in subgroup 2 should fail to make adequate use of the definitionally important dimension of diameter-to- height ratio in deciding whether a stimulus is a “cup” or a “bowl.” More generally, these patients should have difficulty integrating various


bits of information relevant to the definition of a concept in deciding the category of an object.

Object Class$cation

The patients’ object-classification performance will be presented separately for each homogeneous group as defined by performance on the similarity judgment tasks. Patients’ responses were collapsed within groups to obtain classification profiles as a function of diameter-to-height ratio, presence/absence of a handle, and functional context. Classification profiles for the four groups of patients are presented in Figs. 12-15. Figures 12 and 13 represent the performance of subgroups 1 and 2 of the aphasic patients, respectively, while Figs. 14 and 15 represent the right-hemisphere-damaged and the neurologically intact patients, respectively. In these figures, the vertical axis represents the percentage of times a stimulus was called “cup.” The horizontal axis represents the ordinal relationship of diameter-to-height ratio of the stimuli; the smaller numbers represent the narrower, more “cup-like” stimuli, the larger numbers represent the wider, more “bowl-like” stimuli (see Fig. 1). Four profiles are represented for each patient group: one for the stimuli with handles in the no-context condition, one for the stimuli with handles in the cereal context condition, one for the stimuli without handles in the

‘““V go-

0 HANDLES

l NO HANDLES

---CEREAL CONTEXT -NO CONTEXT

; 60-

5 % ?O-

z

=a 60-

;= 50-

ti

F zi 40-

Y w 30- a

20-

IO-

FIG. 12. Stimulus classification profiles for subgroup 1 (N= 14), left-hemisphere-damaged patients.

I I I \e -- L--* - I 2 3 4 5 6 7

STIMULI

100 0 HANDLES

9 l NO HANDLES ‘, ---CEREAL CONTEXT

\, -NO CONTEXT

w” 00 -

ln

6 70-

it

g 60-

5 -0 so-

w :: 40-

!G

g 30-

2 20-

IO-


\ \

I I I I *

I 2 3 4 5 6 7

STIMULI

183

FIG. 13. Stimulus classification profiles for subgroup 2 (N=6), left-hemisphere-damaged patients.

100 0 HANDLES . NO HANDLES

---CEREAL CONTEXT -NO CONTEXT

I 2 3 4 5 6 7

STIMULI

FIG. 14. Stimulus classification profiles for right-hemisphere-damaged group.


0 HANDLES

. NO HANDLES ---CEREAL CONTEXT

-NO CONTEXT

m 80-

z

i5 ?O-

2

2 60-

5 -” 50-

FIG. 15. Stimulus classification profiles for control group.

I 2 3 4 5 6 7

STIMULI

no-context condition and one for the stimuli without handles in the cereal context condition. Thus, for example, the top curve in Fig. 12 represents the no-context, handle stimuli; as can be seen the narrower stimuli are almost always labeled “cup” while the larger stimuli are more often labeled “bowl.”

The patterns of performance of the two subgroups of aphasic patients were clearly different. Patients in subgroup 1 made use of the definitionally important dimension of diameter-to-height ratio to classify stimuli into the cup and bowl categories. These patients performed an appropriate integration of the presence/absence of a handle feature and the functional context feature with the diameter-to-height dimension in determining the category of an object. Indeed, these patients’ performance is qualitatively identical to that of the right-hemisphere damaged (Fig. 14) and neurologically intact patients (Fig. 15). All three groups made effective use of the critical dimension of diameter-to-height ratio with appropriate modulation by the two secondary features of presencejab- sence of a handle and functional context in assigning an object to a category. That is, in all three cases there was a substantial increase in the number of the “bowl” responses as a function of increasing ratio of diameter-to-height, and a clear decrease in the number of “cup” responses as a function of the absence of a handle and the presence of the cereal context. In other words, as previously shown by Lavov (1973)


for normal subjects, there is a predictable interaction among the features that define the cup and bowl concepts in the application of these terms to an object.

The aphasic patients in subgroup 2 were clearly impaired in their ability to integrate the perceptual information characterizing each stimulus to determine its category membership. In particular, it is evident that patients in this group relied excessively on the presence/absence of handle feature to assign category membership. This result is predicted by these patients’ performance in the similarity judgments task; i.e., the only clearly interpretable dimension obtained in the multidimensional scaling solution was precisely that of presence/absence of a handle. Furthermore, it is quite clear that these patients did not adequately use the functional context information to modulate their interpretation of the perceptual properties of the stimuli in determining their category membership. In- terestingly, the performance of these patients in the labeling task was not unlike that of the anemic aphasics tested by Whitehouse et al. (1978). On the basis of the results obtained in the classification and similarity- judgment experiments for patients in subgroup 2, we suggest that these patients manifest a disorder of the semantic component of the lexicon.

An alternative interpretation of these results is that the six patients’ poor performance on both the scaling and the labeling tasks results from an impairment in processing information that is coded “relatively.” That is, it might be argued that these patients fail to process adequately dimensions with continuous variation. Although this interpretation is con- sistent with the results obtained here, it is purely ad hoc and bears no obvious relationship to other language or perceptual deficits that characterize the patient group in question. In contrast, the semantic deficit hypothesis has explanatory power that extends beyond the range of experimental data we have reported.

The conclusions that have been reached on the basis of group data can be substantiated at the individual patient level, with some reserva- tions. Inspection of individual subject profiles for group 2 patients reveals classification performance that was abnormal in some respect for all members of the group. Thus, the averaged profiles obtained for patients in subgroup 2 reflect the performance of each patient in the group relatively accurately.

Inspection of individual patient performance within subgroup 1, as well as within the two control groups, reveals something less than com- pletely homogeneous performance. That is, within these groups there were several patients whose classification profiles deviated in some way from the “normal” pattern that characterized the groups as a whole. There is no obvious explanation for these deviations except that they are not apparently related to an impairment at the level of the semantically guided perceptual parsing of objects. It may well be that the lexical


classification abilities of these patients have broken down at some other point in the processing chain or that idiosyncratic classification strategies were adopted by some patients.

To this point, we have not provided information on the clinical clas- sifications of the patients in the aphasic group. As noted above, we attempted to avoid a priori classification in the hope of uncovering lexical-semantic deficits that might escape notice in the clinic. To this end, we tested all available aphasic patients, some of whom do not present easily classifiable symptoms. Some generalizations can be made concerning the types of patients who fell into subgroups 1 and 2, however. Five of the six patients in group 2 presented signs of posterior pathology-fluent speech and/or moderately severe comprehension impairment. The remaining patient in this group has been classified as a Broca’s aphasic, but there is evidence that he suffers from a fairly large left- hemisphere lesion that extends posteriorly. The composition of subgroup 1, however, cannot be characterized even in general terms. That is, of the patients who could be classified on the basis of their symptoms, several presented classic signs of posterior pathology, while several others had presumed anterior damage. It appears, then, that the strongest statement that can be made at this time is that the type of semantically based deficit we have uncovered appears to be associated with some types of posterior pathology, but not with all posterior lesions.

CONCLUSION

In this report we have emphasized the close link between the semantic structure of words and visual-perceptual and categorical processes. In particular, we have tried to articulate a view in which processing units for natural objects at the perceptual and semantic levels are drawn from the same vocabulary (Miller & Johnson-Laird, 1976). It has been assumed that properties can be either relatively well defined (e.g., presence/absence of a handle) or fuzzy (diameter-to-height ratio) and that a conjoined set of such properties must be satisfied for the proper classification of an item into a category (see, for example, Labov, 1973). Further, we have argued that the process of extraction from a graphic input of the properties that are relevant to the classification of an object as a member of a category is not merely a bottom-up process but one that interacts with the semantic component of the lexicon in such a way that perceptual parsing is constrained by semantic considerations. The development of a model of naming that includes as a stage of processing the semantically guided perceptual parsing we have described leads to the nonobvious prediction of a correlation between performance in a classification task and in a similarity-judgment task. Specifically, the semantic deficit hypothesis we have developed predicts that aphasic patients who manifest classification deficits should also fail to use appropriately semantically


important perceptual information in a judgment-of-similarity task. The basis for this prediction is that if the classification defect is truly a reflection of a disorder of the semantic component of the lexicon, then the perceptual parsing stage should be disrupted since this stage depends critically on adequate semantic support for a proper analysis.

The results we have reported support the semantic deficit hypothesis of the naming defect. As predicted by that hypothesis, empirically obtained subgroupings of the aphasic subjects revealed that those patients who failed to weight adequately definitionally important perceptual dimensions in the similarity-judgment task also performed abnormally in the classification task. The basis for this prediction was a theoretical analysis of the naming process and a consideration of the available descriptions of aphasic disorders in naming and lexical-semantic processing. One important aspect of this hypothesis is the assumption that the underlying disorder in naming is at the level of the semantic component of the lexicon. This assumption is a very strong one and has clear implications not only for naming objects but also for lexical-semantic processing and for classification of objects. Stated differently, the hypothesis we have presented provides a motivated account for the apparent cooc- currence of several disturbances in aphasics with lesions to the posterior regions of the language zones as discussed in the introduction-naming defects, word-finding difficulties, semantic paraphasias, and object classification errors.

REFERENCES

Benton, A. 1979. Visuoperceptive, visuospatial and visuoconstructive disorders. In K. Heilman & E. Valenstein (Eds.), Clinical neuropsychology. New York: Oxford Univ. Press.

Buhr, R. D. Semantic categorization and naming deficits in aphasia. Manuscript submitted for publication, 1980.

Caramazza, A., & Berndt, R. S. 1978. Semantic and syntactic processes in aphasia: A review of the literature. Psychological Bulletin, 85 (4). 898-918.

Caramazza, A., & Berndt, R. S. A psycholinguistic assessment of adult aphasia. In S. Rosenberg (Ed.), Handbook of applied psycholinguistics, in press.

Caramazza, A., Hersh, H., & Torgerson, W. S. 1976. Subjective structures and operations in semantic memory. Journal of Verbal Learning and Verbal Behavior, 15, 103-I 17.

Carroll, .I. D., & Chang, J. J. 1970. Analysis of individual differences in multidimensional scaling via N-way generalization of “Eckart-Young” decomposition. Psychometrika, 35, 283-319.

Coughlan, A. K., & Warrington, E. 1978. Word comprehension and word-retrieval in patients with localized cerebral lesions. Brain, 101, 163-185.

Degerman, R. L. 1970. Multidimensional analysis of complex structure: Mixtures of class and quantitative variation. Psychometrika, 35, 475-491.

Fodor, J. A., Garrett, M. F., Walker, E. C.. & Parkes, C. H. 1980. Against definitions. Cognition, 8, 263-367.

Gainotti. G. 1976. The relationship between semantic impairment in comprehension and naming in aphasic patients. British Journal of Disorders of Communication, 11, 57-61.


Geschwind, N. 1965. Disconnection syndromes in animals and man. Bruin, 88, 237-294;

585-644. Geschwind, N. 1967. The varieties of naming errors. Cortex, 3, 97-112. Goodglass, H., & Baker, E. 1976. Semantic field, naming and auditory comprehension in

aphasia. Brain and Language, 3, 359-374. Goodglass, H., Barton, M. I., & Kaplan, E. F. 1968. Sensory modality and object-naming

in aphasia. Journal of Speech and Hearing Research, 11, 488-496. Goodglass, H., & Geschwind, N. 1976. Language disorders (aphasia). In E. C. Carterette

Jr M. P. Friedman (Eds.), Handbook of perception, Vol. 7: Speech and language. New York: Academic Press.

Goodglass, H., & Stuss, D. T. 1979. Naming to picture versus description in three aphasic subgroups. Cortex, 15, 199-211.

Grober, E., Perecman, E., Kellar, L., & Brown, J. 1980. Lexical knowledge in anterior and posterior aphasics. Brain and Language, 10, 318-330.

Grossman, M. 1978. The game of the name: An examination of linguistic reference after brain damage. Brain and Language, 6, 112-119.

Hersh, H., & Caramazza, A. 1976. A fuzzy set approach to modifiers and vagueness in natural language. Journal of Experimental Psychology: General, 14, 21 I-216.

Johnson, S. C. 1967. Hierarchical clustering schemes. Psychometrika, 32, 241-254. Kruskal, J. 1964. Multidimensional scaling by optimizing goodness of fit to a non-metric

hypothesis. Psychometrika, 29, l-27. Kruskal, J., Young, F. W., & Seery, J. B. 1977. How to use KYST-2, a very flexible

program to do multidimensional scaling and unfolding. Bell Laboratories Technical Report.

Labov, W. 1973. The boundaries of words and their meanings. In C.-J. N. Bailey & R. W. Shuy (Eds.), New ways of analyzing variation in English. Washington, D. C.: George- town Univ. Press.

Lhermitte, F., Derouesne, J., & Lecours, A. R. 1971. Contribution a I’etude des troubles semantiques dans I’aphasie. Revue Neurologique, 125, 81-101.

Luria, A. R. 1970. Traumatic aphasia. The Hague: Mouton. (English translation; originally published 1947.)

Miller, G. A., & Johnson-Laird, P. N. 1976. Language and perception. Cambridge, MA: The Belknap Press of Harvard Univ. Press.

Neisser, U. 1967. Cognitive psychology. New York: Appleton-Century-Crofts. Neisser, U. 1976. Cognition and reality: Principles and implications of cognitive psy-

chology. San Francisco: Freeman. Palmer, S. E. 1975. Visual perception and world knowledge: Notes on a model of sensory-

cognitive interaction. In D. A. Norman & D. E. Rumelhart (Eds.). Explorations of cognition. San Francisco: Freeman.

Pease, D. M., & Goodglass, H. 1978. The effects of cuing on picture naming in aphasia. Cortex, 14, 178-189.

Rosch, E. 1975. Cognitive representations of semantic categories. Journal of Experimental Psychology: General, 104, 192-233.

Shepard, R. N. 1964. Attention and the metric structure of the stimulus space. Journal of Mathematical Psychology, 1, 54-87.

Spreen, O., Benton, A. L., & Van Allen, M. W. 1966. Dissociation of visual and tactile naming in amnesic aphasia. Neurology, 16, 807-814.

Torgerson, W. S. 1965. Multidimensional scaling of similarity. Psychometrika, 38379-393. Tucker, L. R., & Messick, S. 1963. An individual differences model for multidimensional

scaling. Psychometrika, 28, 333-367. Weigel-Crump, C., & Koenigsknecht, R. A. 1973. Tapping the lexical store of the adult

aphasic: Analysis of the improvement mode in word retrieval skills. Cortex, 9,410-417.


Whitehouse, P., Caramazza, A., & Zurif, E. B. 1978. Naming in aphasia: Interacting effects of form and function. Brain and Language, 6, 63-74.

Zurif, E. B., Caramazza, A., Myerson, R., & Galvin, J. 1974. Semantic feature representations for normal and aphasic language. Brain and Language, 1, 167-187.

the semantic deficit hypothesis: perceptual parsing …caram/pdfs/1982_caramazza_berndt.pdf ·...

Documents