Psychonomic Bulletin & Review 1999, 6 (1),5-27

Deficits in lexical and semantic processing: Implications for models ofnormal language

JENNIFER R. SHELTON and ALFONSO CARAMAZZA Harvard University, Cambridge, Massachusetts

The investigation of language processing following brain damage may be used to constrain models of normal language processing, Wereview the literature on semantic and lexical processing deficits, focusing on issues of representation ofsemantic knowledge and the mechanisms of lexical access, The results broadly support a componential organization of lexical knowledge-the semantic component is independent of phonological and orthographic form knowledge, and the latter are independent of each other. Furthermore, the results do not support the hypothesis that word meaning is organized into modality-specific subcomponents. Wealso discuss converging evidence from functional imaging studies in relation to neuropsychological results.

The investigation of cognitive deficits in brain-damaged that has benefited greatly from neuropsychological inves­ individuals can provide powerful constraints for theories tigations is lexical processing. of normal cognition. Brain damage does not lead to an un­ A basic assumption shared by almost all current theo­ differentiated loss of cognitive abilities but to richly struc­ ries of the lexicon is that it is composed of several dis­ tured patterns ofimpaired and spared performance. Thus, tinct components. This assumption supports the expec­ for example, a patient might show severe difficulties in tation that brain damage could result in the selective orally producing the names ofpictures but show no com­ disruption of each ofthe components that constitutes the parable difficulties in writing them. Or a patient might lexical system. The ways in which a cognitive system can show great difficulty in producing nouns but not verbs be damaged depends on how it is organized in the brain. and adjectives. Furthermore, in most cases impaired per­ Functionally independent lexical components can be dam­ formance deviates from normal performance in instruc­ aged selectively only if they are neuroanatomically sep­ tive ways. For example, in a picture naming task a patient arable.' But, within the constraints imposed by the neu­ might produce phonological distortions of the target re­ roanatomical organization ofthe lexical system and the sponse (e.g., chair ~ "share, chail"), suggesting the pos­ vagaries of damage to different regions of the brain (as sibility that the correct lexical node has been accessed but imposed, e.g., by the organization ofthe vascular system that subsequent processes are damaged. Or the patient in the brain in cases of cerebrovascular accidents), the might produce well-formed, semantically related re­ patterns of lexical impairments should reflect the func­ sponses (e.g., chair ~ "table, not that but close, some­ tional organization and computational structure of the thing you sit on"), suggesting that motor programming lexical system (Caramazza, 1986; Shallice, 1988). This and articulatory processes are intact but that lexical se­ basic expectation has received ample support in the neuro­ lection mechanisms are malfunctioning. The analysis of psychological literature and has encouraged detailed in­ the dissociation and association of symptoms and their vestigations aimed at uncovering the processing struc­ related error patterns severely constrains possible inter­ ture ofthe lexical system. pretations of the locus of functional impairment in the In this paper, we review recent results in the area of normal language production system. The logic of infer­ lexical processing by aphasic patients and show that the ences from impaired performance to normal cognition is data provide crucial constraints for models of normal straightforward: We prefer a theory of normal cognition language processing. Because the cognitive neuropsycho­ over alternative theories if that theory can explain both logical literature has become quite extensive, we focus normal performance and the various ways in which a sys­ on just two issues: the structure ofthe semantic compo­ tem breaks down in conditions ofbrain damage. One area nent and the organization oflexical form subsystems. We begin by considering some basic assumptions about the structure ofthe lexicon. This review was supported in part by NIH Grant NS2220 I. We thank Randi Martin, Michael Cortese, Matt Lambon Ralph. and two anony­ OVERVIEW OF THE LEXICAL SYSTEM mous reviewers for their helpful suggestions on an earlier version ofthe manuscript. Correspondence should be addressed to J. R. Shelton or Consider the lexical system sketched in Figure 1. This A. Caramazza, Cognitive Neuropsychology Laboratory. Department of diagram presents a set of interconnected, independent Psychology. Harvard University. 33 Kirkland St., Cambridge, MA 02138 (e-mail: [email protected] or [email protected]. components and is representative of a number of differ­ edu). ent theories of the functional architecture of the lexical

5 Copyright 1999 Psychonomic Society, Inc. 6 SHELTON AND CARAMAZZA

Phonological Input Orthographic Input Lexicon Lexicon

Semantic System

Phonological Output Orthographic Output Lexicon Lexicon

Figure 1. A general overview ofthe lexical system.

system (e.g., Caramazza, 1986; Dell, 1988; Levelt, 1989; version procedure (see papers in Patterson, Marshall, & Morton, 1981; Shallice, 1988). There are several impor­ Coltheart, 1985). In some of these patients, often labeled tant points to be made about this diagram. First, input "surface dyslexics," the impairment is thought to affect and output components are represented independently. the orthographic input lexicon--c-that is, the word forms There have been numerous debates concerning this issue that are stored in this processing component (see also (see, e.g., Allport, 1984; Howard & Franklin, 1988; Mon­ Howard & Franklin, 1987). Since these patients do not sell, 1987), which do not need to be reviewed here other have problems comprehending words from auditory input, than to say that to date the results that have been cited in the phonological input lexicon is presumed to be undam­ support ofseparate input and output processing compo­ aged and thus separate from the impaired orthographic nents can usually be accommodated by models specify­ input lexicon. ing single processing components (but see Nickels & A third aspect ofthe lexical system represented in Fig­ Howard, 1994). These components are presented sepa­ ure 1 is a separation oflexical semantics from lexical in­ rately here for ease of exposition since areas relevant to put and output processing (for simplicity, they are referred input and output processing will be reviewed separately. to as "semantic processing" and "lexical processing," re­ Each lexical form component is assumed to contain spectively). There are numerous examples ofwhy these the information relevant to a word's representation for components should be considered separate processing sub­ either recognition or production, separately for phono­ systems. For example, the above description of(one form logical and orthographic processing. Later in the paper, of) surface dyslexia refers to patients who can success­ arguments for a separation between phonological and or­ fully process information semantically, at least from thographic output components are reviewed. For input, phonological input, but who cannot process information there are theoretical and empirical reasons for separating in the orthographic input lexicon. There are also exam­ orthographic and phonological lexical representations: ples of patients who can process information in the For example, there must be a mechanism by which irreg­ phonological input lexicon but cannot use this informa­ ularly spelled words are recognized. That is, "stored"? tion to successfully access semantic information (see, e.g., word forms must exist that correspond to those words, such Ellis, 1984; Franklin, Howard, & Patterson, 1994; Kohn as yacht, soldier, or island, which cannot be pronounced & Friedman, 1986). As noted, there are many reports of using only a simple phonological conversion strategy. patients who have problems processing specific semantic Moreover, some brain-damaged patients read regularly information such as that associated with certain seman­ spelled words and pronounceable pseudowords (e.g., tic categories, and those deficits are not associated with plim, garper, burndle) at a significantly higher rate than selective problems of input or output processing. Also irregular words, presumably using a phonological con- discussed below are patients who have problems with LEXICAL AND SEMANTIC DEFICITS 7 orthographic and/or phonologic production that result is organized within this semantic system can be under­ from postsemantic processing deficits and not deficits to stood in part by examining the patterns of performance the semantic system. ofpatients who have deficits to semantic knowledge. Two The simplified lexical system presented in Figure 1 areas that have received considerable attention concern does not include the representation of syntactic infor­ the issue ofmodality-specific organization ofknowledge mation. It is assumed that syntactic information is rep­ and the issue ofcategory-specific knowledge deficits. resented separately from semantic and lexical form in­ formation. There are numerous examples of patients Separate Visual and Verbal Semantic Systems? who can process correctly the meaning of individual A popular view in cognitive psychology and neuro­ nouns and verbs in a sentence but cannot understand the psychology is that different modalities of semantic in­ meaning ofthe sentence as a whole (see, e.g., Caramazza formation are represented independently, and there are a & Zurif, 1976; Schwartz, Saffran, & Marin, 1980; for re­ number ofmodels of semantic memory that distinguish views, see Berndt, 1991; Caramazza & Berndt, 1978). between separate representations for "visual" and "ver­ Thus, a patient may be able to tell what girl, boy, andpush bal" knowledge (e.g., Allport, 1985; A. R. Damasio, 1990; mean but will be unable to select the appropriate picture Paivio, 1971; Shallice, 1988; Warrington & Shallice, to match the sentence "The girl pushes the boy," espe­ 1984). This position has been disputed, however, and cer­ cially when the choice involves selecting between a pic­ tainly does not represent the consensus in the cognitive ture of a girl pushing a boy or a boy pushing a girl. We literature (see Seymour, 1979, for a review). Snodgrass are not concerned here with this aspect ofsyntactic pro­ (1984), for example, has suggested that the evidence cessing (for a review, see Berndt, Mitchum, & Haendiges, from normal subjects does not strongly favor one view 1996). However, the role ofgrammatical features ofwords over the other. (e.g., grammatical class, gender, etc.) in lexical access is Inthe neuropsychological literature, many researchers discussed later in the paper. have adopted a multiple-modality view ofsemantic mem­ There is also neuroanatomical support for the idea that ory. For example, Shallice (1987, 1988, 1993) presented different aspects oflanguage processing are independent several lines of evidence from brain-damaged subjects of one another. Since the late l800s it has been known that he interpreted as supporting the existence of inde­ that damage to different areas ofthe brain could result in pendent, modality-specific representations ofknowledge. different patterns ofbehavior in language processing (for A schematic representation of this position is presented review, see McCarthy & Warrington, 1990). Recent ad­ in Figure 2. As can be seen, visual knowledge and ver­ vances in functional neuroimaging techniques have al­ bal knowledge are represented independently of each lowed researchers to explore in greater detail the neuro­ other and the bidirectional arrows between the two rep­ anatomical organization of language processing, with resentations indicate that the two types of knowledge results that are beginning to converge with the estab­ interact with each other. Furthermore, structural descrip­ lished neurobehavioral observations. Thus, both behav­ tions of objects directly access visual semantic knowl­ iorally and neurologically there is evidence for consid­ edge, whereas verbal semantic knowledge about objects erable structure and organization in the lexical system. is mediated by access to visual semantic knowledge. Even this simple overview provides quite a bit of in­ Likewise, verbal input descriptions (phonological or or­ formation from cognitive neuropsychology that constrains thographic lexical forms) directly access verbal seman­ models ofnormal processing: Components can be dam­ tic knowledge, and access ofvisual semantic knowledge aged independently ofone another, requiring certain as­ about an item from verbal input is mediated by access to sumptions regarding the organization and representation verbal semantic knowledge.' Shall ice and others (e.g., of knowledge of lexical and semantic information. We Beauvois, 1982; Warrington & Shallice, 1979; see Saf­ turn now to a consideration ofseveral aspects oflanguage fran & Schwartz, 1994, for a review) have presented sev­ processing that have been extensively studied within cog­ erallines ofevidence in favor ofthe multiple-semantics nitive neuropsychology, specifically focusing on issues position, including the performance ofpatients with "op­ ofsemantic and lexical output processing and represen­ tic aphasia" (and other "modality-specific" anomias) tation. We review several different interpretations and and the performance ofpatients with semantic memory discuss the evidence in favor ofcertain positions over oth­ impairments. ers, ultimately arriving at what we feel is the most plau­ Modality-specific anomia refers to a disorder charac­ sible interpretation ofthe collective body ofevidence. terized by naming difficulties restricted to stimuli pre­ sented in one modality (naming from other input modal­ SEMANTIC KNOWLEDGE: ities is unimpaired). Such disorders are distinguished from UNITARY VERSUS MULTIPLE FORMATS agnosia (disorders in recognizing objects shown in one modality) by the fact that anomie patients can access some The evidence from cognitive neuropsychology shows semantic information in the impaired input modality that the semantic component functions autonomously (Shallice, 1987, 1988). For example, patients with "tac­ from other lexical components; it also helps reveal the tile aphasia" cannot name from tactile input but can mimic internal structure ofthe semantic system. How knowledge the use of the unnamed object when blindfolded (e.g., 8 SHELTON AND CARAMAZZA

house Ihausl

STRUCTURAL ORTHOGRAPHIC PHONOLOGIC perceptllnl processillg DESCRIPTIONS DESCRIPTIONS DESCRIPTIONS

VISUAL SEMANTICS VERBAL SEMANTICS

5eruQulic procrssillg

visual information associated verbal information associated with house with house

Figure 2. An example of a model postulating separate modality-specific semantic systems.

Beauvois, Saillant, Meininger, & Lhermitte, 1978). Pa­ (e.g., 31/32 = 97% judging functional similarity). Cos­ tient R.G. (Beauvois et aI., 1978) was impaired at nam­ lett and Saffran interpreted this pattern of performance ing objects presented in the tactile modality (62.5% cor­ as indicating that E.M. could access unimpaired seman­ rectj> but virtually unimpaired at naming from either the tic information from the impaired input modality. visual or the auditory modality (96% and 98.8%, respec­ Other patients have been described who were signifi­ tively). R.G. also demonstrated appropriate handling and cantly more impaired in one input modality over another use ofobjects he could not name from the tactile modal­ (Warrington, 1975) but who were not classified as ity, indicating that he could understand what the object "modality-specific" aphasics. One of Warrington's pa­ represented (i.e., he could access the "correct" semantics tients, E.M., (a different case from that described above), from tactile input). was more impaired in making semantic decisions from There are also reports ofpatients with impairments to verbal (70/120 = 58%) than from visual input (93/120 = other input modalities. Patients have been described who 78%), whereas the other patient, A.B., showed the oppo­ demonstrated impaired naming from visual input but site pattern (verbal input: 80/120 = 75%, visual input: unimpaired naming from auditory and tactile input (see, 75/120 = 63%). In another case (Schwartz, Marin, & Saf­ e.g., Coslett & Saffran, 1989, 1992; Hillis & Caramazza, fran, 1979), w.L.P. could correctly demonstrate the use 1995a; Lhermitte & Beauvois, 1973; Riddoch & Hum­ of objects (100%) but was unable to identify the words phreys, 1987); others have been described who demon­ associated with those objects (89/140 = 64%, where strated impaired naming from auditory input but not vi­ chance = 50%). sual or tactile input (Denes & Semenza, 1975). For Some investigators have interpreted the patterns of example, E.M. was described as a case of"optic aphasia" dissociations found in these patients as providing sup­ (Coslett & Saffran, 1992). E.M. was impaired at naming port for a multiple-semantics view ofknowledge repre­ objects presented visually (6/28 = 21%) and significantly sentation (e.g., Beauvois et aI., 1978; Coslett & Saffran, better at naming the same objects presented for touch 1989, 1992; Shallice, 1988). They argued that the ob­ (17/25 = 68%) and when naming the item to description served dissociations result from the patients' failure to (19/28 = 68%). E.M. provided appropriate gestures for access an intact verbal semantic system from an equally visually presented objects (100% correct) and performed intact modality-specific semantic system (visual or tac­ well on semantic classification tasks with visual input tile). Thus, in the case ofoptic aphasia, the visual seman- LEXICAL AND SEMANTIC DEFICITS 9

tic system was assumed to be intact since the patients were better miming than naming ofobjects from visual input supposedly able to access semantic information from vi­ (33/44 = 75%). However, 1.B. was impaired at answer­ sual input (as indicated, e.g., by the ability to mime the ing detailed semantic questions about visually presented action associated with objects that could not be named). objects (177/284 = 62%), indicating that he could not Given this assumption, the optic aphasic's failure to name fully access semantics from visual input. Riddoch and is argued to result from an inability to transmit informa­ Humphreys suggested that 1.B.had impaired visual input tion from the visual-semantic system to the verbal se­ processing that resulted in poor access of semantic in­ mantic system, which is required for naming. This rea­ formation from this input modality. This pattern ofresults soning has been applied to all sensory modality-specific does not require distinguishing between visual and ver­ anomias/aphasias (e.g., Allport, 1985, proposed a model bal semantics. ofsemantics that represents semantic information sepa­ More recently, Hillis and Caramazza (1995a) described rately for each sensory modality). a patient (D.H.Y.) with "optic aphasia" whose pattern of A second source ofevidence cited in favor ofseparate performance was similar to previously reported cases. visual and verbal semantic systems comes from patients D.H.Y. could not name from visual input (2/47 = 4%) who show priming in one modality but not another. War­ but could name from other input modalities (e.g., tactile rington and Shallice (1979) described a patient, A.R., who input: 44/47 = 94%), and she demonstrated access to se­ could read words significantly better when they were pre­ mantic knowledge from visual input by performing very ceded by an auditory prompt than when the same words well on a number of semantic processing tasks, includ­ were preceded by pictures. For example, A.R. was sig­ ing those tasks previously used to demonstrate "intact" nificantly better at reading pyramid when he heard the semantic knowledge in a case ofoptic aphasia (E.M.; Cos­ word "Egypt" prior to presentation ofthe target (49/80 = lett & Saffran, 1992). However, the appropriateness of 61%) than when he saw a picture ofa pyramid (36/80 = these semantic processing tasks for assessing the integrity 45%). Furthermore, A.R.'s picture naming improved only ofthe semantic system was questioned since tasks such slightly when the target was preceded by an auditory as miming or classification into semantic categories (an­ prompt. Given the lack ofpriming from pictures to words imal vs. plants) do not necessarily require detailed se­ and from words to pictures, Warrington and Shallice (1979) mantic analysis or an appreciation of the subtle differ­ argued that A.R. could not transmit information from vi­ ences between semantically related items. When D.H.Y. sual semantics to verbal semantics. was tested on semantic processing tasks that required Taken together, the reported cases would seem to pre­ appreciation of detailed semantic information, she per­ sent strong evidence in favor ofseparate visual and ver­ formed much more poorly (50%-88% correct) and dis­ bal semantic systems (and presumably other sensory input played strong evidence that she was not able to get suf­ systems) that can be selectively affected following brain ficient semantic information from the pictures to allow damage. However, these interpretations and explanations her to perform the tasks normally. For example, on tasks have not gone unchallenged, and there are several grounds that required D.H.Y. to sort animals by function (e.g., "Is on which researchers have objected to the distinction be­ it edible?"), answer detailed questions about pictured tween a visual semantic system and a verbal semantic sys­ items, or pick the two most related pictures from three tem, or, more generally, modality-specific semantic sys­ related items (e.g., lightswitch, lightbulb, traffic light), tems (e.g., Caramazza, Hillis, Rapp, & Romani, 1990; D.H.Y. performed well below the normal range. How­ Hillis, Rapp, Romani, & Caramazza, 1990; Humphreys ever, D.H.Y. performed normally on these same tasks & Riddoch, 1988; Riddoch & Humphreys, 1987; Rid­ when the semantic information was probed in nonvisual doch, Humphreys, Coltheart, & Funnell, 1988). modalities. Much like 1.B.(Riddoch & Humphreys, 1987), Riddoch et al. (1988) questioned whether the diffi­ this case clearly showed that access to semantics from the culty levels ofthe naming and comprehension tasks used impaired input modality was not normal. Taken together, in the studies of modality-specific anomias were equiv­ 1.B. and D.H.Y.'s patterns of performance raise the pos­ alent across the different modalities, since naming from sibility that at least in some cases, the tasks used to probe pictures and naming tactilely from real objects could semantic knowledge were not sensitive enough to pick up very well be ofunequal difficulty given that more infor­ subtle semantic deficits (or complete access to this knowl­ mation might be available from real objects than from edge) since the tasks could have been performed at a high pictures. Other aspects ofthe tasks used in different modal­ level ofaccuracy using incomplete semantic information. ities could also have made them of unequal difficulty. Objections have also been raised against studies that For example, miming and forced-choice tasks may be ac­ compared performance between semantic knowledge complished successfully on the basis ofpartial semantic tasks with pictures versus verbal stimuli on the grounds information in pictures and real objects. Riddoch and that there is information in the pictures that could allow Humphreys (1987) described 1.B., a patient with "optic patients to answer questions by relying on structural cues aphasia" who was impaired at naming objects in the vi­ found in pictures but not in verbal input (Caramazza et aI., sual modality (20/44 = 45%) but who could name those 1990; Riddoch et aI., 1988).6For example, there is a great objects better from the tactile modality (33/44 = 75%). deal ofinformation in a picture ofa cat that would allow Like other reported cases of optic aphasia, 1.B.also showed answers to questions about that object, but there is noth- 10 SHELTON AND CARAMAZZA

ing in the phonological or orthographic form ofthe name Several researchers have argued that the modality­ of that object that would give clues about its semantic specific aphasia cases could be easily interpreted within content. The fact that a picture ofa cat represents seman­ a single amodal semantic system (e.g., Caramazza et aI., tic information such as eyes, a mouth, and four legs could 1990; Hillis & Caramazza, 1995a; Riddoch et aI., 1988). be enough to allow a patient to discern that the object rep­ Forexample,Caramazza et al. (1990)presented a modality­ resented a living thing and was a nonhuman mammal (be­ neutral model of semantic memory, the "organized uni­ cause of the presence of four legs). Further, that infor­ tary content hypothesis" (OUCH), and demonstrated that mation could lead patients to deduce that the animal would it could account for the patterns of performance that orig­ have fur and would be capable ofself-initiated movement. inally motivated a multimodal semantic memory theory. All this information would be available through the un­ A schematic representation of this type of amodal se­ derstanding ofcertain features represented in the picture mantic memory model is shown in Figure 3. even without a complete understanding ofthe object. Two assumptions in this model are relevant here. The There is also empirical evidence that is highly prob­ first is the assumption of "privileged accessibility." This lematic for the modality-specific semantics theory. assumption captures the fact that there is an asymmetry Hillis et al. (1990) reported the performance ofa patient, in the relationships that words and objects have to se­ K.E., who showed similar levels of performance (36%­ mantic information: Words but not pictured objects have 47% correct) across a wide variety oftasks differing in an arbitrary relationship to their meanings (e.g., dog and terms of input and output requirements. Since virtually log are lexically very similar but have radically different all ofK.E.'s errors (60%-96%) across all tasks were se­ meanings). The relationship between pictures (or per­ mantic substitutions (tiger ~ "lion"), the pattern ofper­ ceptual features in general) and semantics is not arbitrary formance suggested a deficit at the semantic level. And, (e.g., a chair has a flattish surface that is used for sitting). since the level and type of difficulty was comparable Thus, perceptual features can be interpreted semanti­ across almost all tasks and modalities, K.E.'s performance cally with the consequence that the surface properties of was interpreted as indicating that the damage involved a pictures or objects provide much more semantic infor­ single semantic system shared in all tasks. mation than do words.

house Ihaus I

STRUCTURAL ORTHOGRAPHIC PHONOLOGIC ptrceplllni processillg DESCRIPTIONS DESCRIPTIONS DESCRIPTIONS

SEMANTIC SYSTEM selllnll/ic processing

all relevanl semantic information associated with house including visual and verbal information

Figure 3. An example of a model postulating a single, amodal semantic system. LEXICAL AND SEMANTIC DEFICITS 11

A second assumption of the model-the assumption formation such that the patient is then able to read the of "privileged relationships"-is that there is a special word. Patients do not completely lose all semantic infor­ relationship between types of semantic information as­ mation following brain damage, and thus there is always sociated with an object such that certain semantic fea­ some information available to them. If the tasks are de­ tures (e.g., chairs have a flattish surface) are more strongly signed so that providing additional information can im­ linked to certain semantic features of that object (e.g., prove performance and/or make use of the information chairs are used for sitting) than to others (e.g., chairs are that they still retain, the model would predict different found in houses). And, it could be argued that the action levels ofperformance in different modalities depending associated with an object may have a privileged rela­ on the amount ofinformation provided by the input (see tionship with certain perceptual features of that object Caramazza et aI., 1990; Hillis et aI., 1990; Rapp, Hillis, (e.g., a handle). As a consequence, miming in response & Caramazza, 1993, for detailed discussion). to an object or a picture could have been readily accom­ A distinction between visual and verbal semantics is plished despite the fact that naming ofthat object is im­ only as helpful as our understanding ofwhat is meant by paired, since naming ofthe object requires more informa­ "visual" and "verbal" semantics. Unfortunately, the tion than that required to support miming the action of multiple-semantics hypotheses have remained rather vague the object. And, obviously, there is no information in the on the precise meaning of these terms, with the conse­ surface structure ofa word, unlike the case ofthe picture, quence that it has been difficult to generate the predic­ to support miming. This second assumption applies not tions and interpretations ofperformance that are needed only to form-function relationships but also to any fea­ to further our understanding of "modality-specific" se­ tures of objects that are highly intercorrelated (see also mantic information. Consider the role that modality might McRae, de Sa, & Seidenberg, 1997, for a similar proposal). play in the acquisition of concepts. The function of an That is, certain properties ofobjects (e.g., has a mouth) are object is experienced visually (or perhaps in other sen­ likely to be highly related to other properties ofthat object sory modalities), yet this knowledge is assumed to be (e.g., is animate, has eyes, ingests food, produces sound). represented in the "verbal" semantic system. So, for ex­ This model predicts that "optic aphasic" patients should ample, we learn the function ofa fork by watching some­ show impaired performance on tasks requiring detailed one use this item (and presumably through sensory feed­ semantic knowledge, since they presumably have diffi­ back of ourselves using the item as well), which would culty accessing a complete semantic representation (in a suggest that function should be a property ofthe "visual" single semantic system) from visual input. Getting to se­ semantic system. Or, conversely, without having seen (or mantic information is not impossible from the impaired heard, or touched) a porcupine, we can read that this an­ input modality, as shown by the fact that these patients imal has long quills extending out of its body that are can perform some semantic processing tasks even when sharp and pointy. Although we haven't experienced this items are presented to the impaired modality (e.g., D.H.Y. information through sensory modalities, it is considered and lB. performed well on the same or similar tasks used sensory information, and thus should be stored in the to establish "unimpaired" semantics). But, the semantic corresponding sensory semantic systems. If this is the representation that is accessed is not complete and there­ case, we are left with trying to explain how information fore patients perform poorly with tasks that require the that is experienced through one modality in acquisition appreciation ofdetailed semantic information (or, at least, is stored in the appropriate modality-specific semantic that was the case for D.H.Y. and IB., who were tested on system based on the "modality" of the information and more extensive, detailed assessments of semantic pro­ not the modality ofthe input. cessing). Thus, previously reported cases of modality­ In contrast to the hypothesis that the data require the specific aphasias that have been cited in favor ofmodality­ existence of separate visual and verbal semantic sys­ specific semantic systems can be accommodated without tems, the unitary semantics hypothesis-OUCH-re­ difficulty within a unitary semantics model. viewed here can provide a principled explanation for all A model ofa single amodal semantic system can also the reported cases (see also Riddoch et aI., 1988) by mak­ accommodate priming effects and modality-specific com­ ing plausible assumptions about the structure of access prehension deficits in much the same way as it explains to semantic information and by assuming that the orga­ modality-specific aphasias. For example, the patient de­ nization ofthat information is based on the meanings of scribed earlier, A.R., was much better at reading a word concepts. Moreover, this account does not need special when it was preceded by an auditorily presented related assumptions regarding learning mechanisms that would word than when the word was preceded by a picture of require information to be "placed" in a specific system the concept represented by the word. Assuming that A.R. depending on, we guess, the nature of that information has partial damage to a semantic system and that similar (regardless ofthe modality through which that informa­ semantic information is activated when A.R. sees a word tion is acquired). Therefore, the neuropsychological lit­ or sees a picture ofthe concept represented by the word, erature does not support a multiple-semantics position. little additional semantic information is provided by the Also, results from a neuroimaging study with PET presence ofthe picture. However, priming with a seman­ that directly compared semantic processing ofwords and tically related word can activate additional semantic in- pictures, support the idea ofa common semantic system 12 SHELTON AND CARAMAZZA

shared by verbal and visual inputs (Vandenberghe, Price, Farah, Hammond, Mehta, & Ratcliff, 1989; Farah, Me­ Wise, Josephs, & Frackowiak, 1996). Vandenberghe Mullen, & Meyer, 1991; Farah, Meyer, & McMullen, 1996; et al. presented subjects with item triplets, either words Hart & Gordon, 1992; Hillis & Caramazza, 1991; Laia­ or pictures, and they were required to match the stimuli cona, Barbarotto, & Capitani, 1993; Laiacona, Capitani, for (I) meaning (verbal-associative semantic task), & Barbarotto, 1997; Laws, Evans, Hodges, & McCarthy, (2) for "real-life" size (visual semantic task), or (3) for 1995; McCarthy & Warrington, 1988; Pietrini et al., 1988; actual presented size (baseline condition). The authors Powell & Davidoff, 1995; Sartori & Job, 1988; Sheridan & measured the brain activity associated with the different Humphreys, 1993; Silveri & Gainotti, 1988; Sirigu, Du­ processing tasks. They found activation in overlapping hamel, & Poncet, 1991; Warrington & McCarthy, 1983, brain regions during both semantic processing tasks, 1987). However, several studies also demonstrated that specifically in the left superior occipital gyrus, middle patients may show deficits that are not restricted to just and inferior temporal cortex, and the inferior frontal the category of living things (Silveri & Gainotti, 1988; gyrus. They did find areas of activation specific to pro­ Warrington & Shallice, 1984). For example, Silveri and cessing ofpictures and words, but the activation was not Gainotti (1988) reported a patient who showed a deficit specific to the type ofsemantic processing required. Thus, for living things, foods, and musical instruments. the brain areas that were selectively activated for pictures The existence of patients who present with semantic and for words are those dedicated to recognizing pictures deficits for some categories ofknowledge but not others and words, respectively, whereas the areas that were ac­ could indicate that conceptual knowledge is organized tivated during semantic processing ofpictures and words categorically. This possibility has never really been en­ were shared and overlapping. These results, then, pro­ tertained seriously, however, in part because the reported vide at least initial neuroanatomical evidence for a uni­ category-specific deficits have been rather complex and tary semantic system representing both visual and non­ seemed to cut across the living/nonliving divide; for ex­ visual aspects of meaning. ample, the early cases described by Warrington and Shal­ Thus far we have argued that neuropsychological data lice showed difficulty with living things but also food concerning modality-specific semantic systems do not and musical instruments. Given the diverse nature ofcat­ support separate visual and verbal semantic systems. A egory deficits, Warrington and her colleagues (Warring­ second set of results that addresses the issue of the in­ ton & McCarthy, 1983, 1987; Warrington & Shallice, ternal structure of the semantic system comes from pa­ 1984) sought to account for their occurrence by proposing tients who demonstrate category-specific semantic def­ that conceptual knowledge is organized into modality­ icits. We turn now to a review ofthese results. specific components. In this theory, the principal distinc­ tion is between visual and functional properties." A sche­ Category-Specific Semantic Deficits: matic representation depicting the "sensory/functional" The Visual-Functional Knowledge Distinction theory ofsemantic organization is presented in Figure 4.9 Warrington and Shallice (1984) presented the first In this case, the modality ofinformation refers to the type well-documented cases of category-specific semantic of information associated with the object regardless of deficits resulting from brain injury (herpes simplex en­ the input modality or the nature ofthe response required. cephalitis).? Two of these patients (J.B.R. and S.B.Y.) According to this theory, sensory and functional prop­ showed deficits in comprehending and naming living erties are differentially important in identifying mem­ things such as animals, fruits, and vegetables in the con­ bers ofthe living and nonliving categories, with sensory text of relatively spared ability to comprehend and pro­ information being much more important for distinguish­ duce the names of nonliving things such as tools, furni­ ing among living things and functional information being ture, and kitchen utensils. For example, S.B.Y. identified much more important for distinguishing among nonliv­ or defined 36/48 (75%) ofnonliving things but no living ing things. Therefore, damage to sensory properties should things, and IB.R. identified or defined 45/48 (94%) of result in greater impairments to living things, since this nonliving things and only 2/48 (4%) of living things. information is more important in identifying members Further investigation ofIB.R.'s deficit revealed that the of this category. This explanation also supposedly ac­ distinction between spared and impaired categories was counts for the somewhat strange association ofcategories more fine-grained than a living/nonliving one. J.B.R. that have been reported to be damaged together, such as showed selective problems in comprehending and nam­ animals, foods, and musical instruments, since presum­ ing musical instruments, gemstones, metals, fabrics, and ably the items in these categories may all be more de­ foods in addition to animals, fruits, and vegetables. There­ pendent on sensory than functional properties for distin­ fore, the patient's deficits did not appear to respect a cat­ guishing among them. egorical principle such as living/nonliving. The theory also predicts that patients should be differ­ A number ofstudies since then have reported patients entially impaired with visual/perceptual knowledge within showing dissociations in processing between living and the impaired categories. Several researchers have claimed nonliving things (Basso, Capitani, & Laiacona, 1988; Cara­ to have confirmed this prediction (Basso et al., 1988; De­ mazza & Shelton, 1998; H. Damasio, Grabowski, Tranel, Renzi & Lucchelli, 1994; Farah et al., 1989; Sartori & Hichwa, & Damasio, 1996; DeRenzi & Lucchelli, 1994; Job, 1988; Silveri & Gainotti, 1988). For example, Sil- LEXICAL AND SEMANTIC DEFICITS 13

dog Idawgl

STRUcrURAL ORTHOGRAPHIC PHONOLOGIC ptrctplll.'proctssill8 DESCRIPTIONS DESCRIPTIONS DESCRIPTIONS

PERCEPTUAL INFORMATION NONPERCEI'TUAL INFORMATION

visual, audilory,lactile, functional, encyclopedic, st",.,,'ic proctssi"8 olfactory information associative information associated wilh "dog" associated with "dog"

SEMANTIC SYSTEM

Figure 4. An example of a model postulating separate knowledge-specific semantic systems based on sensory features versus nonsensory features. veri and Gainotti described a patient (L.A.) who showed this, Farah and McClelland determined that the ratio of a category X modality interaction effect: The patient was visual information to functional information for living more impaired with living items (11/54 = 20%) than with things and nonliving things is 7.7: 1 and 1.4:1, respec­ nonliving items (22/28 = 79%) and was more impaired tively. When Farah and McClelland damaged the visual with visual (1/11 = 9%) than with nonperceptual infor­ knowledge network, they found a category-specific def­ mation associated with living items (8/14 = 58%). Thus, icit for living things. This result rests, of course, on the the patient's deficit was disproportionately greater for a validity ofthe initial intuition by Warrington and her col­ category ofitems and a certain type ofinformation asso­ leagues that living things (or certain categories ofitems) ciated with those items. rely more heavily on visual information to distinguish Converging arguments in favor of this account came among them, and on the validity of Farah and McClel­ from a connectionist model developed by Farah and Me­ land's rating data. We return to this point below. Clelland (1991). In their model, visual and functional Although the sensory/functional theory has become properties are represented in two independent but inter­ the received view of category-specific deficits, various connected networks, and the ratio ofvisual to functional observations have undermined the usefulness and valid­ knowledge for living things is larger than that for non­ ity of this explanation. First, several researchers ques­ living things. The actual ratio of visual to functional tioned the validity of category-specific deficits in general, knowledge for living and nonliving categories was de­ suggesting that the results may have been artifacts offac­ termined by having subjects rate dictionary definitions tors such as familiarity, frequency, and/or visual complex­ of items in terms of the number of visual or functional ity (e.g., Funnell & Sheridan, 1992; Gaffan & Heywood, properties associated with each definition. Subjects were 1993; Stewart, Parkin, & Hunkin, 1992). For example, given a list ofdefinitions for living and nonliving items Stewart et al. described a patient who showed a category­ and told to circle either information that referred to the specific effect for animals (30/55 = 55%) when tested visual appearance of the item or information that re­ with materials used by other investigators, but the effect ferred to "what the item does" or "what it is for." From disappeared when they controlled for the factors of famil- 14 SHELTON AND CARAMAZZA iarity and visual complexity (animate, 15/36 = 42%; in­ for naming fruits and vegetables (67/107 = 63%), but animate, 13/36 = 36%). Moreover, Stewart et al. demon­ not animals or even other food items (262/269 = 97%) strated that normal control subjects were slower to respond (see also Farah & Wal1ace, 1992). Caramazza and Shel­ to visual information than to functional information as­ ton (1998) reported a patient who was impaired for ani­ sociated with animals and found that visual information mals (16/47 = 34%) but not any other living things (100%) for animals was less familiar than functional information. (or other categories such as musical instruments [8/10 = These findings raise several important concerns. They 80%] or food items [100%], which have been found to be suggest that category-specific deficits may be artifacts associated with living things), and Hil1is and Caramazza ofuncontrolled factors rather than true effects. They also (1991) reported a patient with selective sparing of animals suggest that the category X modality interactions re­ (72%-92% correct) but not fruits and vegetables (7%­ ported previously may wel1have been effects ofcategory 17% correct). These patterns ofperformance are not read­ variation in familiarity rather than true interactions. And, ily explicable by appeal to the sensory/functional theory; the results suggest that the influence of familiarity may according to this theory, damage to visual properties be responsible for some ofthe associations ofcategories should result in damage to those items relying most heav­ previously reported to be damaged together such as gem­ ily on those properties-presumably the categories ofan­ stones, fabrics, and musical instruments, categories one imals, fruits, and vegetables. These reports clearly pre­ would imagine would not be highly familiar to the aver­ sent a chal1enge to the theory. age person. Thus, those categories may not necessarily be A second set of results also chal1enges the sensory/ damaged together because ofsome underlying semantic functional theory. Several reports have demonstrated principle, but because they are all unfamiliar. quite convincingly that, when familiarity ofan item and/ The fact that familiarity strongly influences perfor­ or familiarity ofthe knowledge associated with that item mance could indicate that in some cases the putative are controlled, patients who show a deficit for living things category-specific semantic deficits are not dissociations do not show the predicted category X modality inter­ between categories ofknowledge but rather dissociations action (Barbarotto, Capitani, Spinnler, & Trivelli, 1995; between familiar and less familiar concepts, irrespective Caramazza & Shelton, 1998; Funnel1 & De Mornay of category. However, more recent reports of patients Davies, 1997; Laiacona et al., 1993; Laiacona et al., 1997; with category-specific deficits have control1ed for these Sheridan & Humphreys, 1993). For example, Caramazza factors and the patients stil1 showed category-specific and Shelton demonstrated that their patient (E.W.), who knowledge deficits (e.g., Caramazza & Shelton, 1998; had a clear deficit for animals only, was equally impaired Farah et al., 1996; Forde, Francis, Riddoch, Rumiati, & on visual (74/100 = 74%) and functional knowledge Humphreys, 1997; Gainotti & Silveri, 1996; Hart & Gor­ (115/150 = 77%) ofthe items in this semantic category. don, 1992; Laiacona et al., 1993; Laiacona et al., 1997; Moreover, E. W.'s performance was nearly perfect on Lambon-Ralph, Howard, Nightingale, & El1is, 1998; equal1y difficult questions concerning visual and func­ Moss, Tyler, Durrant-Peatfield, & Bunn, 1998; Samson, tional information for fruit and vegetable items. These Pil1on, & De Wilde, 1998; Sheridan & Humphreys, 1993). results strongly contradict the predictions from the sen­ Furthermore, there are reports ofpatients who show se­ sory/functional theory. lective damage to nonliving items-the easier or more And finally, although Farah and McClelland (1991) familiar category-suggesting that some ofthe category provided putative empirical support for the initial intu­ effects must be true effects-or at least not simply the ition that sensory properties are more important for result ofdifferences in the familiarity ofthe items within defining living things than nonperceptual properties, a category (e.g., Hil1is & Caramazza, 1991; Lambon­ problems with this study may undermine its usefulness Ralph et al., 1998; Sacchett & Humphreys, 1992; War­ for addressing the issues of how knowledge can frac­ rington & McCarthy, 1983, 1987). However, the lesson to tionate to result in category-specific deficits. The results be learned from the critiques by Stewart et al. and Fun­ obtained by these authors may simply reflect an artifact nel1 and Sheridan (1992) is an important one: Familiar­ of the instructions given to subjects for identifying sen­ ity has a strong influence on performance and may have sory and nonperceptual features of objects in the living accounted for some of the strange categorical associa­ and nonliving categories. Specifically, the instructions tions and for the category X modality interactions that given to subjects may have biased them against identify­ have been reported for some studies ofcategory-specific ing nonperceptual features for living things. In the attempt deficits. Although this conclusion does not directly un­ to determine the amount offunctional information asso­ dermine the sensory/functional theory, it raises doubt ciated with living and nonliving categories, Farah and about the validity of various results that have been cited McClelland asked subjects to underline information show­ in its support. 10 ing "what the item does or what it is for" (p. 342). These However, a number of empirical results are problem­ instructions could have led subjects to exclude a large atic for the sensory/functional theory. First, there are re­ amount ofnonperceptual information that is included in ports ofdissociations ofperformance that are not easily definitions ofliving things, such as "lives in Africa," "car­ explained by the theory. For example, Hart, Berndt, and nivore,""grows underground," "flies," and so on. However, Caramazza (1985) reported a patient who was impaired this information is crucial to the definition ofliving things LEXICAL AND SEMANTIC DEFICITS 15 and is the type ofinformation that has been included in property "being capable of self-initiated motion"), and the "functional" set ofinformation on which patients have members of superordinate categories have many proper­ been tested (e.g., Farah et aI., 1989). Caramazza and ties in common. If brain damage can selectively affect Shelton (1998) replicated this experiment but changed specific areas of semantic space (because of how this the instructions so that subjects were asked to underline knowledge is distributed in the brain), category-specific either all sensory properties or all nonsensory properties deficits can arise. That is, damage to any area populated in the definitions used by Farah and McClelland. They by properties that are important for the category of liv­ found that the ratios ofsensory to nonsensory properties ing things would result in a category-specific deficit. Al­ for living and nonliving things were 2.9:2.5 and 2.2:2.3, though these types oftheories can easily account for the respectively. Thus, there appears to be little empirical ev­ emergence of category-specific deficits, they fail to pro­ idence for the initial intuition that living things rely heav­ vide an explanation as to why category-specific deficits ily on sensory properties for determining their meaning respect a tripartite distinction ofanimals, plants, and arti­ (see also McRae et a!., 1997). facts and as to why the majority ofcategory-specific def­ Two other explanations have been offered for the ex­ icits demonstrate a deficit to animals or living things in istence ofcategory-specific deficits. One explanation is general as opposed to a deficit for artifacts. that category-specific deficits reflect the fact that con­ A recent proposal that attempts to provide an expla­ cepts within a category have many correlated properties nation for these two facts is that the semantic system may in common (see, e.g., Caramazza et aI., 1990; McRae have even more internal structure than previously thought et a!., 1997; Tyler & Moss, 1997). For example, the model and that some semantic category deficits can be inter­ reviewedin the last section, OUCH, can accommodate the preted as truly categoricaldeficits. On this account, knowl­ category-specific results by assuming that certain prop­ edge is organized in broad domains that reflect evolu­ erties of objects are strongly intercorrelated (e.g., the tionarily salient distinctions ofconceptual knowledge­ property "having eyes" is strongly correlated with the "domain-specific knowledge hypothesis" (Caramazza &

dog Idawgl

STRUCTURAL ORTHOGRAPHIC PHONOLOGIC DESCRIPTIONS DESCRIPTIONS DESCRIPTIONS percept".1 processing

ANIMALS

-sernantic information -sernanlic information -semantic information associated with all associated with all associated with all semantic processillg animals lanl life, including nonliving things r.ruits and vegetables

SEMANTIC SYSTEM

Figure 5. An example of a model postulating a single semantic system organized according to semantic category. 16 SHELTON AND CARAMAZZA

Shelton, 1998). A schematic representation ofthis type category-specific deficits (Gainotti, Silveri, Daniele, & ofmodel is presented in Figure 5. This model is based on Giustolisi, 1995) and by examining neuroimaging stud­ the assumption that evolutionary pressures have led to ies measuring brain activity using PET (H. Damasio et al., specific adaptations-highly specialized neural mecha­ 1996; Martin, Wiggs, Ungerleider, & Haxby, 1996;Perani nisms-for recognizing animals because oftheir roles as et al., 1995). However, not all cases with a deficit to predators and prey. Similar assumptions could be made these categories have damage to left temporal areas; there for the identification ofconspecifics, plant life, and per­ have also been reports of category-specific deficits for haps even for artifacts (see Hauser, 1997, for a discus­ animals/living things when the damage primarily involved sion concerning the potential value for recognizing "tools" the right temporal lobe (Barbarotto et al., 1995; Laws as an evolutionarily important category). et al., 1995). And, there are also a few cases of deficits One expectation that follows from this model is that cat­ to living things resulting from damage to frontal and in­ egory deficits should be seen only for evolutionarily im­ ferior parietal areas (Caramazza & Shelton, 1998; Hillis portant categories: humans, animals, and plants (and arti­ & Caramazza, 1991; Laiacona et al., 1993). facts, either by contrast to the other categories or because Unfortunately, there is even less consistency concern­ this category has its own specialized neural mechanisms). ing the processing ofnonliving things. A selective defi­ That is, one should not find finer grained distinctions cit to nonliving things has been reported for patients with among category deficits (once familiarity is controlled), left frontal and parietal lesions (e.g., Sacchett & Hum­ and currently, there are no reports of patients who show phreys, 1992; Warrington & McCarthy, 1983, 1987) and true categorical deficits for very specific categories (e.g., lesions to the left temporal lobe and basal ganglia (Hillis just vehicles or just land animals or just fruit). Therefore, & Caramazza, 1991). Thus, damage to the left frontal, one testable prediction of this hypothesis is that specific parietal, and/or temporal lobes appears to result in defi­ categorical deficits should exist only for those categories cits to nonliving things. With regard to neuroimaging that can be shown to have important standing in evolution, studies, there is little agreement among them concern­ and we shouldnot find finer grainedcategory-specific def­ ing the processing ofnonliving things, although they all icits (see also Shelton, Fouch, & Caramazza, 1998, for dis­ found evidence ofactivation in the posterior middle tem­ cussionof potentiallyimportant categories in evolution that poral area. In addition, Martin et al.'s (1996) and Perani expand on this tripartite distinction). Moreover, this hy­ et al.'s (1995) studies found activation of the frontal pothesis accounts for the greater number of cases with def­ lobes for processing ofnonliving things. icits to animals/living things by assuming that those brain Although the results from lesion site and neuroimag­ areas dedicated for processing these concepts are highly ing studies are not conclusive regarding the relationship localized and thus more susceptible to circumscribed dam­ ofcertain brain areas to processing the different seman­ age. It follows, then, that there should always be a greater tic categories, evidence is beginning to emerge that dis­ number of cases with specific impairments to animals/liv­ tinct areas ofthe brain are differentially involved in the ing things, once important factors such as familiarity, age processing ofparticular semantic categories. At the very of acquisition, frequency, imageability, and visual com­ least, it appears that inferior areas of the temporal lobe plexity have been taken into account. are important in processing living things, whereas more This hypothesis was proposed as a way to account for posterior areas of the temporal lobe and frontoparietal the two factors that the other hypotheses could not ac­ regions are important in processing nonliving things. count for: the consistency in the categories for which dis­ sociations have been reported (animals, plant life, arti­ Unitary Versus Multiple Semantics: Conclusion facts) and the disparate number ofcases with deficits to The study of patients with semantic knowledge defi­ animals/living things. However, this account leaves open cits provides the opportunity to examine the internal or­ the question ofthe organization ofsemantic information ganization of semantic knowledge. The results to date within a specific domain. It is not unreasonable to suppose pose an interesting challenge for theories ofthe organi­ that something along the lines ofOUCH (or other corre­ zation of meaning: how to account for highly selective lational accounts) could provide the organizational basis deficits to specific domains ofknowledge. Some theories within a semantic domain. In the latter case, it might be of the organization of conceptual knowledge have pro­ possible to find fine-grained category-specific deficits posed a modality-based principle-such as the distinc­ that would reflect the latter organization. tion between perceptual and nonperceptual knowledge­ Neuroanatomical evidence suggests that different brain to explain category-specific semantic deficits (e.g., areas are involved in processing the different categories Allport, 1985; Paivio, 1971; Shallice, 1988). Alternatively, ofinformation (at least comparing animals/living things category- and modality-specific semantic deficits may with nonliving things), although there are conflicting arise because of principles such as those proposed by findings about the specific areas and structures that are OUCH (e.g., the assumption that members ofa category involved in different semantic domains. There is general share highly correlated semantic properties that can be agreement that the left temporal lobe is involved in the damaged selectively; Caramazza et al., 1990). In addi­ processing of animals/living things. This is supported tion, the most intriguing possibility is that conceptual by examining the lesion sites of the patients who show knowledge may be organized into broad domains defined LEXICAL AND SEMANTIC DEFICITS 17 by the role these domains may have played in the evolu­ erature has also typically supported a role ofphonology tion ofthe human brain. At the very least, the patterns of in reading comprehension (e.g., Perfetti & Bell, 1991; performance of patients with semantic deficits, and es­ Van Orden, Johnston, & Hale, 1988), although for present pecially those patients who have deficits to very specific purposes we will concentrate on the evidence concerning types ofknowledge, are highly problematic for modality­ the possible role of phonology in production (but see based theories ofthe organization ofconceptual know l­ note 14). The evidence we review shows that writing is edge and point toward a modality-neutral, unitary seman­ not dependent on phonological processes.'! tic system that has a great deal ofinternal structure. One type ofevidence that has been cited in support of the autonomy ofwriting processes from phonology is the PHONOLOGICAL AND ORTHOGRAPHIC dissociation of writing and speaking impairments (e.g., OUTPUT AND GRAMMATICAL Assai, Buttet, & Jolivet, 1981; Blanken, de Langen, Ditt­ CLASS EFFECTS mann, & Wallesch, 1989; Bub & Kertesz, 1982; Cara­ mazza, Berndt, & Basili, 1983; Ellis, Miller, & Sin, 1983; Disorders of language production is a well-studied Hier & Mohr, 1977; Kremin, 1987; Patterson & Shewell, area within cognitive neuropsychology, and the patterns 1987). However, these reports did not explore the deficits of dissociations that have been observed in this area of in enough detail to resolve issues concerning the influ­ research provide useful information for models of nor­ ence of both sublexical and lexical phonological pro­ mal production. However, because the literature in this cessing on writing. More recently, detailed evidence has area is so vast, we have chosen to focus on topics that re­ been brought to bear on this issue that has led neuropsy­ flect the areas in which our interests have been concen­ chologists to view output phonology and output orthog­ trated and that also reflect recent discoveries that have raphy as independent and dissociable processes (e.g., significant implications for models ofnormal processing. Hanley & McDonnell, 1997; Miceli, Benvegnu, Capasso, Specifically, we will review evidence on the relationship & Caramazza, 1997; Rapp, Benzing, & Caramazza, 1997; between phonological and orthographic production to re­ Shelton & Weinrich, 1997), although this viewpoint is veal the role ofgrammatical information in lexical access. not unanimously held (for a review, see Barry, 1994). Two issues have been addressed in investigations of The Independence ofPhonological and the independence between speaking and writing pro­ Orthographic Output Representations cesses: (I) the role ofsublexical phonological mediation Traditionally, writing has been viewed as fully depen­ and (2) the role of lexical phonological mediation. The dent on phonological processing, both in the neurologi­ distinction drawn between these two mechanisms ofsup­ cal literature (e.g., Geschwind, 1969; Head, 1926; Luria, port is that (I) does not rely on stored word forms and in­ 1966; Wernicke, 1886/1989) and in the cognitive litera­ stead relies on sublexical phonological-to-orthographic ture (e.g., Frith, 1980; Hotopf, 1980). The cognitive lit- conversion mechanisms whereas (2) relies on the sup-

SEMANTIC REPRESENTATIONS sublexical-phonological mediation

LEXICAL-PHONOLOGICAL LEXICAL-0RTIiOGRAPHIC REPRESENTATIONS REPRESENTATIONS lexical-phonological mediation

Figure 6. An example of a model postulating an obligatory relationship between phonological support and orthographic output. 18 SHELTON AND CARAMAZZA

SEMANTIC REPRESENTATIONS sublexical-phonological mediation

LEXICAL-PHONOLOGICAL LEXICAL-ORTIiOGRAPHIC REPRESENTATIONS REPRESENTATIONS lexical-phonological mediation

Figure 7. An example ora model postulating a nonobligatory relationship between phonological support and orthographic output.

port ofstored word forms. Two models contrasting these there is still the issue ofthe role oflexical phonology in positions are shown in Figures 6 and 7. The dashed lines writing. Some models ofwriting assign an obligatory role in Figure 7 represent connections that may be established to stored phonological word forms in writing (Figure 6). but are not obligatory. However, recent evidence suggests that this role is not Evidence that sublexical phonological mediation does necessary and that successful writing can take place in­ not play an obligatory role in writing initially came from dependently ofall phonological support (Figure 7). the performance ofpatients who could write words suc­ In the case just described, E.A. (Shelton & Weinrich, cessfully despite their inability to write nonwords (e.g., 1997) had significant impairments in verbal production, Shall ice, 1981). However, since the ability to write non­ and these were shown to result from damage to phono­ words is typically not completely abolished in these pa­ logical output processing rather than semantic process­ tients, there could be some support ofsublexical phonol­ ing, articulation, and/or motor programming problems. ogy in successful writing (e.g., Barry, 1994; Nolan & E.A. demonstrated excellent written picture naming Caramazza, 1983). A recent case reported by Shelton and (228/265 = .86) but severely impaired oral picture nam­ Weinrich (1997) countered this criticism. Their patient, ing (97/265 = .37). The fact that he could write the name E.A., was able to successfully write nouns with approx­ ofthe picture he could not name orally indicates that he imately 85% accuracy and yet could not write a single could successfully process the picture semantically; the nonword (0/75). In fact, E.A. correctly produced only fact that he could successfully repeat the name ofthe pic­ 2/187 (.01) individual target graphemes when writing ture he could not spontaneously generate in oral produc­ nonwords. Given these results, it is difficult to argue that tion indicates that he did not have problems with articu­ E.A.'s good performance in writing words reflects partial lation or motor programming; this conclusion is further support from sublexical phonology-to-orthography in­ supported by the fact that E.A. almost never made formation since his performance suggested that he had phonological errors in naming.l? Thus, damage to verbal none of this information available to him. Other recent production must be at the level of lexical output pro­ cases have obtained similar results: Successful writing cessing, indicating that he could not use stored lexical­ can occur even when there is minimum support from phonological word forms in a normal manner. Despite sublexical phonology-to-orthography information (e.g., this, however, his ability to write the names of pictures Hanley & McDonnell, 1997; Miceli et al., 1997; Rapp was excellent. And the fact that sublexical phonology-to­ et al., 1997). orthography processing was completely damaged in E.A. Thus, there is now clear evidence to support the idea leads to the conclusion that his ability to write was inde­ that sublexical phonology-to-orthography conversion pendent of his ability to use either lexical or sublexical mechanisms do not have an obligatory role in writing, but phonology-to-orthography information. Thus, ortho- LEXICAL AND SEMANTIC DEFICITS 19 graphic and phonological (output) lexical forms are stored given the target picture peppers, W.M.A. wrote tomato and used separately. and said "artichoke." Crucially, the inconsistencies were Hanley and McDonnell (1997) presented a patient, lexical. That is, the patients would either make semantic P.S., with severely restricted ability to use sublexical errors in only one modality ofoutput or would make dif­ phonology-to-orthography conversion mechanisms. PS. ferent semantic errors in the two modalities. The presence was at chance (19/35 = .54) in deciding whether or not of semantic errors, in the absence of a semantic knowl­ the names of two pictures rhymed, but could accurately edge deficit, indicates that the patients' deficits were in write the names of both pictures. This suggests that he accessing output lexical forms and not in accessing the could not access the phonological representations of phonological and/or orthographic information associated words he spelled correctly. Another test ofhis ability to with the output. 13 The production ofdifferent lexical re­ use phonological information to support spelling was a sponses in the two modalities implies that the written re­ naming-to-definition task for homophones. PS. was given sponse could not have been determined by its phonolog­ a definition that fit one ofthe homophones (e.g., "seven ical content, for otherwise the same lexical response ofthem in a week") and was asked to write the word be­ would have been produced. These results are incompat­ ing defined (e.g., days) or to write the word that sounded ible with a theory oflexical organization that states that the same as the word being defined (e.g., daze). When he phonological word forms have an obligatory role in writ­ was asked to spell the homophone that matched the def­ ten word production. inition, he succeeded on all but one trial (19/20 = .95). The studies we have reviewed investigated the issue of When he was given both the definition and the spoken the independence between phonological and orthographic word form (Ideiz I) and asked to write the word that output using a variety of techniques with different pa­ sounded the same as the one being defined, he very often tients, and the results converge in support ofthe conclu­ correctly wrote the nondefined homophone (17/20 = sion that writing can be carried out independently of .85). However, when he was only given the definition and speaking. The implication to be drawn from these results was required to write the non defined homophone, his is not that phonology and orthography do not interact in performance dropped considerably (8/20 = .40). Since the normal course of writing but that phonology is not he could spell the nondefined homophone when given obligatory for successful writing.U-'> We can use this the definition and the spoken word form, his poor per­ conclusion as the basis for addressing more general ques­ formance cannot be attributed either to problems in pro­ tions concerning mechanisms involved in lexical access­ ducing the alternative word or to problems spelling that specifically, questions about the role of factors, such as specific word form. IfPS. was using phonological word grammatical class, in writing and speaking. forms to support spelling, he should have been equally capable of spelling both forms of the homophone since Grammatical Class Effects the identical phonological form would be used to address An extremely robust and well-studied finding with orthography. Instead, PS.'s failure to write homophone brain-damaged patients is the influence of grammatical pairs in conjunction with his chance-level performance class on production. There are numerous reports demon­ on rhyme judgments ofwords he wrote successfully sug­ strating that some aphasic patients produce nouns much gests that he was only able to access orthography directly better than either main verbs or function words (e.g., from semantics without relying on support from stored Berndt, Mitchum, Haendiges, & Sandson, 1997; Cara­ phonological word forms (or sublexical phonological mazza & Hillis, 1991; A. R. Damasio & Tranel, 1993; information). Kohn, Lorch, & Pearson, 1989; McCarthy & Warring­ Another way in which this issue has been addressed has ton, 1985; Miceli, Silveri, Villa, & Cararnazza, 1984; been through the use of a double naming task in which Myerson & Goodglass, 1972). There are also reports of patients are asked to both speak out and write the name patients who produce verbs better than nouns (e.g., Bax­ ofpictured objects. Researchers have presented patients ter & Warrington, 1985; Berndt et al., 1997; Breedin & (WM.A., Miceli et aI., 1997; PW, Rapp et aI., 1997) Martin, 1996; Caramazza & Hillis, 1991; A. R. Damasio whose naming difficulties could be localized at the level & Tranel, 1993; De Renzi & di Pellegrino, 1995; Mar­ ofphonological output and who were unable to use sub­ shall, Chiat, Robson, & Pring, 1996; Miceli, Silveri, No­ lexical phoneme-to-grapheme conversion mechanisms centini, & Caramazza, 1988; Zingeser & Berndt, 1988) to write (as demonstrated by their inability to spell non­ and patients who produce open-class words (nouns and words). The patients were asked to name pictures orally verbs) much better than closed-class words (e.g., An­ and in writing in the same trial (i.e., the patient was asked dreewsky & Seron, 1975; Caramazza, Berndt, & Hart, to orally name a picture and then immediately write the 1981; Gardner & Zurif, 1975; and papers in Coltheart, name or vice versa). Both WM.A. and PW demonstrated Patterson, & Marshall, 1980). Better production offunc­ inconsistencies in responding between the two modali­ tion words as compared with open-class words was also ties, even though the responses were produced immedi­ described in patients with semantic dementia (e.g., ately one after the other. For example, given the target Breedin & Saffran, 1997; see also Schwartz et al., 1979) picture pillow, P W said "pillow" and wrote "bed" And, and jargon aphasia (Buckingham, 1991). 20 SHELTON AND CARAMAZZA

These patterns ofdissociations allow a number ofplau­ verb, to watch), the same pattern of contrasts was ob­ sible hypotheses about the organization oflexical knowl­ tained, and in fact for H.W the difference in performance edge, but for the purposes of this review the focus will was more pronounced; he orally produced 46% of the be on the contrasting patterns of performance between verbs and 88% of the nouns but wrote 96% and 98%, re­ noun and verb production. The fact that nouns are often spectively. These findings suggest that the impairments found to be processed better than verbs following brain are not the result of particular difficulties in producing damage could merely reflect an abstractness effect-verbs specific phonological or orthographic forms, but rather are typically more abstract than nouns, and many patients a difficulty associated with access to lexical forms. show more problems processing abstract than concrete Error patterns for the 2 patients also differed across nouns (e.g., McCarthy & Warrington, 1985). However, output modalities. H.W.and S.lD. made semantic errors the existence of patients who perform better with verbs in only one modality ofoutput (the more impaired modal­ than nouns effectively eliminates this explanation as a ity for each patient). 16 Both patients' normal performance general explanation ofall selective verb impairments (see, on tests ofcomprehension eliminated a lexical-semantic e.g., Berndt et aI., 1997, for discussion ofthis argument). locus for the naming impairment and instead localized Furthermore, the grammatical class effect is obtained the deficit at the level of selection of phonological and even when the concreteness of test items is controlled orthographic output forms. Moreover, a semantic deficit (Caramazza & Hillis, 1991). for a specific grammatical class was ruled out given the In some patients, the differential performance in pro­ normal performance by each patient in the intact produc­ cessing nouns and verbs is most likely the result ofa def­ tion modality (writing for H.W. and speaking for S.lD.). icit at the lexical-semantic level (i.e., in understanding These results have important implications for theories the meaning of either the nouns or verbs, e.g., Berndt oflexical access in production. The fact that verbs were et aI., 1997; McCarthy & Warrington, 1985; Miceli et aI., selectively impaired in different modalities in each pa­ 1988). For these patients, it is not clear whether the selec­ tient indicates that the deficit was one ofactivating a spe­ tive deficit in processing one class ofwords-say, nouns­ cific grammatical category within a modality-specific is due to their grammatical status-being a noun-s-or their lexical output component, rather than a modality­ semantic status-being an object or thing. However, there independent lexical output component. Thus, the impli­ are reports of patients whose selective deficit in pro­ cation is that lexical output forms are represented inde­ cessing nouns or verbs is not due to damage at the seman­ pendently by modality (this also was supported by the tic level and, thus, allows the inference that the basis for evidence reviewed in the previous section). Furthermore, the differential performance is due to the grammatical sta­ the results invite the hypothesis that each ofthe modality­ tus ofthe words. specific lexical output forms is independently specified There is evidence to suggest that the dissociation in for grammatical class. I? In other words, we have a tripar­ performance between different grammatical classes can tite distinction between semantic information, lexical be a result ofdamage to output lexical processing and not form information (phonological and orthographic), and damage to the semantic system itself(e.g., Caramazza & grammatical information, with each type ofinformation Hillis, 1991; Hillis & Caramazza, 1995b; Rapp & Cara­ independently specified. The distinction between output mazza, 1997). Patients H.Wand SJ.D., described by Cara­ forms and grammatical information could be realized in mazza and Hillis (1991), are typical ofthe cases who have (at least) two ways. Modality-specific lexical output forms been shown to produce differential patterns of perfor­ may be organized according to grammatical class. On mance with nouns and verbs in speaking and writing. this view, words ofdifferent grammatical classes would One patient (H.W) had significantly greater problems be represented in different areas of the brain.If Or, the with spoken (reading and picture naming; 240/356 = connections between modality-specific output forms and 67%) than with written (writing to dictation and picture syntactic nodes representing grammatical class could naming; 100%) output. The other patient (S.lD.) had be damaged. On this view, a syntactic node is shared by significant problems with written output (293/356 = the two output forms. Damage to the connections from a 82%) and no problems with spoken output (352/356 = modality-specific output lexicon would result in a gram­ 99%). Interestingly, when the patients' performance was matical class deficit for that modality. To date, there are examined in spoken and written output for different no data to differentiate between these two possibilities. 19 grammatical classes, H.W. performed significantly worse Additional evidence regarding the independence of in spoken production of verbs (22%) than nouns (56%) lexical output forms and syntactic information comes and equally well with both grammatical classes in writ­ from studies ofpatients who can retrieve grammatical in­ ten production (99%). S.J.D., on the other hand, per­ formation about words they are unable to produce (e.g., formed significantly worse in written production ofverbs Badecker, Miozzo, & Zanuttini, 1995; Henaff Gonon, (70%) than nouns (99%) and equally well with both gram­ Bruckert, & Michel, 1989; Miozzo & Caramazza, 1997). matical classes in spoken production (97% and 99%, re­ For example, Miozzo and colleagues studied an Italian spectively). And, importantly, when these 2 patients were patient, Dante, who showed severe difficulties retrieving tested on homonym production (e.g., noun, the watch; phonological output but who seemed to have normal un- LEXICAL AND SEMANTIC DEFICITS 21 derstanding oflanguage. This patient was shown to recall nent and measured ERPs for these items during a lexical correctly gender information (97%) (a syntactic feature decision task. They found that ERPs to verbs were larger ofthe language) associated with words he was unable to at central brain sites close to motor cortices as compared produce or even guess their initial (53%) or final sounds with ERPs to nouns, which were larger at occipital sites (47%). Dante could also accurately report the appropriate over visual cortices, thus suggesting that nouns and verbs auxiliary form of verbs he could not produce (69/70 = have distinct cortical topographies. Ofcourse, it is pos­ 99%). This pattern of results suggests that syntactic in­ sible that the differences in processing arose because of formation is represented independently oflexical output the differences in the visual components between the two information (i.e., phonological or orthographic word types ofwords, rather than because ofa difference in pro­ form informationj.s? cessing the different grammatical word classes." Neuroanatomical evidence from (I) studies investigat­ The results from neuroanatomical/neuroimaging stud­ ing lesion sites in patients with selective deficits in pro­ ies ofnoun and verb processing using different techniques cessing either nouns or verbs (see Gainotti et aI., 1995, for converge on a general principle: Verb processing tends to review), (2) neuroimaging studies (Fiez, Raichle, Balota, be associated with left frontal areas and noun processing Tallal, & Petersen, 1996; Martin, Haxby, Lalonde, Wiggs, tends to be associated with left temporal lobe areas. While & Ungerleider, 1995; Petersen, Fox, Posner, Mintun, & there are exceptions to this rule, one thing is clear: There Raichle, 1989; Wise et al., 1991), and (3) evoked poten­ are grammatical word class differences in processing that tial (EP) studies (Dehaene, 1995; Pulvermuller, Preissl, are also established neuroanatomically. Lutzenberger, & Birbaumer, 1997) support the idea that nouns and verbs are represented separately in the brain, Deficits ofLexical Access: Implications for Our although none ofthese reports distinguishes between se­ Understanding ofNormal Language Processing mantic and lexical form processing of nouns and verbs. Detailed investigations of phonological and ortho­ Gainotti et al. (1995) reviewed the reported cases ofpa­ graphic production disorders have led to intriguing find­ tients with selective deficit to either nouns or verbs and ings regarding the independence ofphonological and or­ found that selective deficits to nouns are associated with thographic output and the role ofgrammatical information lesions in the left temporal lobe and posterior associa­ in accessing those phonological and orthographic forms. tion areas. Selective deficits to verbs are generally asso­ Although there has been tremendous resistance to the ciated with lesions in the left anterosuperior portions of notion that phonological and orthographic output could the temporal lobe and inferoposterior parts of the left be represented and processed independently, recent re­ frontal lobe (although there are exceptions, such as, e.g., ports strongly support the view that these two types of H.W., discussed above, who had problems orally pro­ knowledge are independent and can be accessed au­ ducing verbs and whose lesion involved only the left pari­ tonomously. That is, the careful documentation and ex­ etal area). Nevertheless, there is a general trend ofselec­ amination of the role of phonology in writing in recent tive problems with verbs being associated with frontal case reports strongly supports the position of lexical­ areas, and selective problems with nouns being associated orthographic autonomy. with temporal (and perhaps some temporal-occipital) areas. In addition, recent reports provide convincing evidence Neuroimaging studies using PET also support this that favors the idea that grammatical information can be general rule. For example, Martin et al. (1995) investigated accessed and processed at the lexical output level for both word generation involving color words and action words. phonological and orthographic forms, and recent case They found that both types of words activated left pre­ reports also support the idea that there is independence frontal dorsolateral and left posterior parietal areas but between the representation and processing of syntactic that color words activated bilateral fusiform gyri more and semantic information. Taken together, the findings strongly in the left, and action words activated left pos­ reviewed in this section support a theory oflexical orga­ terior and superior temporal gyri and left inferior frontal nization in which semantic, syntactic, and lexical form areas. Basically, their results indicate that temporo­ information are represented autonomously both func­ frontal lobe areas are involved with action word genera­ tionally and neuroanatomically. tion and the posterior temporal areas are involved with One important implication from these findings con­ color word generation. Other imaging studies have found cerns the structure ofspeech production models. The re­ similar results, demonstrating that prefrontal and temporo­ sults we have reviewed allow a very strong inference about parietal areas are activated in verb generation tasks (Fiez the structure oflexical access: Semantic information in­ et aI., 1996; Petersen et al., 1989; Wise et al., 1991). dependently activates phonological and orthographic And finally, event related potentials (ERP) studies have lexical forms, which are specified for grammatical class. also provided evidence ofa neuroanatomical distinction This conclusion is supported by two observations: (I) the between nouns and verbs. Pulvermuller et al. (1997; see fact that words ofone grammatical class can be damaged also Dehaene, 1995) investigated processing of nouns selectively in only one output modality, and (2) the fact whose meanings have a strong visual feature component that the patients who show these dissociations make se­ and verbs whose meanings have a strong motor compo- mantic errors in only one output modality. These two facts 22 SHELTON AND CARAMAZZA

together imply that the damage must concern a post­ mechanisms. The selective deficits in performance re­ semantic stage ofprocessing (semantics is intact in com­ sulting from damage restricted to specific brain areas prehension and in one output modality) but it must be a converge with results from neuroimaging studies. In­ stage ofprocessing prior to segmental selection since er­ creasingly, it seems that a highly organized language pro­ rors involved lexical and not segmental substitutions. The cessing system corresponds to highly organized and spe­ only possible stage left is the stage where lexical forms cialized neural processing mechanisms. are selected. And since the damage is selective for an out­ put modality, the impairment must concern a modality­ REFERENCES specific lexical form. ALLPORT, D. A. (1984). Speech production and comprehension: One The evidence from neuropsychology is problematic lexicon or two? In W. Prinz & A. F. Sanders (Eds.), Cognition and for theories oflexical access (e.g., Bock & Levelt, 1994; motor processes (pp. 209-228). Berlin: Springer. Dell, 1988, 1990; Dell, Schwartz, Martin, Saffran, & ALLPORT, D. A. (1985). Distributed memory, modular subsystems and Gagnon, 1997; Garrett, 1980; Roelofs, 1992) that distin­ dysphasia. In S. K. Newman & R. Epstein (Eds.), Current perspec­ tives in dysphasia (pp. 207-244). Edinburgh: Churchill Livingstone. guish between a level ofrepresentation where modality­ ANDREEWSKY, E., & SERON, X. (1975). Implicit processing of gram­ neutral lexical forms associated with syntactic features matical rules in a classical case ofagrammatism. Cortex, 11,379-390. are specified (lemma level) and a level where modality­ ASSAL, G., BUTTET,J., & JOUVET, R. (1981). Dissociations in aphasia: specific lexical nodes are specified (lexeme level).22 For A case report. Brain & Language, 13,223-240. on these theories, damage at the level of lemma, where BADECKER, W., MIOZZO, M., & ZANUTTINI, R. (1995). The two-stage model of lexical retrieval: Evidence from a case ofanomia with se­ grammatical class is specified, should result in qualita­ lective preservation ofgrammatical gender. Cognition, 57, 193-216. tively similar impairments in speaking and writing. And BARBAROTTO, R., CAPITANI, E., SPINNLER, H., & TRIVELU" C. (1995). damage at the level ofmodality-specific forms should not Slowly progressive semantic impairment with category specificity. result in selective impairments for words of one gram­ Neurocase, 1, 107-119. BARRY, C. (1994). Spelling routes (or roots or rutes). In G. D. A. Brown matical class. Thus, these theories fail to account for the & N. C. Ellis (Eds.), Handbook ofspelling: Theory. process and in­ patterns of selective impairments for grammatical class tervention (pp. 27-49). New York: Wiley. by modality (see Caramazza, 1997, for discussion). BASSO, A., CAPITANI, E., & LAIACONA, M. (1988). Progressive language impairment without dementia: A case with isolated category specific semantic defect. Journal ofNeurology, Neurosurgery. & Psychiatry, CONCLUSIONS 51, 1201-1207. BAXTER, D. M., & WARRINGTON, E. K. (1985). Category-specific This review has examined cognitive neuropsycholog­ phonological dysgraphia. Neuropsychologia, 23, 653-666. ical results specifically in relation to issues of semantic BEAUVOIS, M.-F. (1982). Optic aphasia: A process of interaction be­ and lexical processing and demonstrated that these re­ tween vision and language. Philosophical Transactions ofthe Royal Society ofLondon: Series B, 298, 35-47. sults place important constraints on models of normal BEAUVOIS, M.-F., SAILLANT, B., MEININGER, v.. & LHERMITTE, F. language processing. The structure of semantic knowl­ (1978). Bilateral tactile aphasia: A tacto-verbal dysfunction. Brain, edge has long been of interest in cognitive psychology, 101,381-401. and the results from patients who present with semantic BERNDT, R. S. (1991). Sentence processing in aphasia. In M. T Sarno processing deficits have provided us with important clues (Ed.), Acquiredaphasia (2nd ed., pp. 223-270). New York: Academic Press. regarding the nature ofthis structure. It would seem that BERNDT, R. S., MITCHUM, C. c.,& HAENDIGES, A. N. (1996). Compre­ neuropsychological results are best interpreted within a hension ofreversible sentences in agrammatism: A meta-analysis. Cog­ model of semantics that is highly structured and poten­ nition, 58, 289-308. tially organized according to several biologically salient BERNDT, R. S., MITCHUM, C. c., HAENDIGES, A. N., & SANDSON, J. (1997). Verb retrieval in aphasia: I. Characterizing single word im­ categories. pairments. Brain & Language, 56,68-106. The mechanisms involved in lexical access have also BLANKEN, G., DE LANGEN, E. G., DITTMANN, J., & WALLESCH, C. W. been ofconsiderable interest to cognitive scientists. The (1989). Implications of preserved written language abilities for the results from patients who demonstrate differential ef­ functional basis ofspeech automatisms (recurring utterances): A sin­ fects of grammatical class in production and especially gle case study. Cognitive Neuropsychology, 6, 211-249. BOCK,J. K, & LEVELT, W. J. M. (1994). Language production: Gram­ the comparison in production between writing and speak­ matical encoding. In M. A. Gernsbacher (Ed.), Handbook of psycho­ ing have provided us with intriguing possibilities con­ linguistics (pp. 945-984). San Diego: Academic Press. cerning the mechanisms involved in production. These BREEDIN, S. D. & MARTIN, R. C. (1996). Patterns of verb impairment data place severe constraints on current models ofspeech in aphasia: An analysis of four cases. Cognitive Neuropsychology, 13,51-91. production and ultimately require a rethinking ofthe num­ BREEDIN, S. D., & SAFFRAN, E. M. (1997). Sentence processing in the ber and type of levels of representation needed in lan­ face 0/ semantic loss: A case study. Unpublished manuscript. guage production models. BUB, D., & KERTESZ, A. (1982). Evidence of lexicographic processing Patient data not only provide us with a unique oppor­ in a patient with preserved written over oral single word naming. tunity to study the effect ofhighly selective factors on per­ Brain, 105, 697-717. BUCKINGHAM, H. W. (1991). Explanations for the concept ofapraxia of formance but also contribute to our understanding ofthe speech. In M. 1. Sarno (Ed.), Acquired aphasia (2nd ed., pp. 271­ relationship between language processing and brain 312). San Diego: Academic Press. LEXICAL AND SEMANTIC DEFICITS 23

CARAMAZZA, A. (1986). On drawing inferences about the structure of DERENZI, E., & DI PELLEGRINO, G. (1995). Sparing of verbs and pre­ normal cognitive systems from the analysis of patterns of impaired served but ineffectual reading in a patient with impaired word pro­ performance: The case for single patient studies. Brain & Cognition, duction. Cortex, 31, 619-636. 5,41-66. ELLIS, A. W. (1984). Introduction to Bramwell's (1897) case of word CARAMAZZA, A. (1997). How many levels of processing are there in lex­ meaning deafness. Cognitive Neuropsychology, 1,245-258. ical access? Cognitive Neuropsychology, 14, 177-208. ELLIS, A. w., MILLER, D., & SIN, G. (1983). Wernicke's aphasia and CARAMAZZA, A., & BERNDT, R. S. (1978). Semantic and syntactic pro­ normal language processing: A case study in cognitive neuropsy­ cesses in aphasia: A review of the literature. Psychological Bulletin, chology. Cognition, IS, 111-144. 85,898-918. FARAH, M. J., HAMMOND, K. M., MEHTA, Z., & RATCLIFF, G. (1989). CARAMAZZA, A., BERNDT, R. S., & BASILI, A. G. (1983). The selective Category-specificity and modality-specificity in semantic memory. impairment ofphonological processing: A case study. Brain & Lan­ Neuropsychologia, 27, 193-200. guage, 18, 128-174. FARAH, M. J., & MCCLELLAND, J. L. (1991). A computational model of CARAMAZZA, A., BERNDT, R. S., & HART, J. (1981). "Agrammatic" semantic memory impairment: Modality specificity and emergent reading. In F.1. Pirozzolo & M. C. Wittrock (Eds.), Neuropsycholog­ category specificity. Journal ofExperimental Psychology: General, ical and cognitive processing in reading (pp. 297-317). New York: 120,339-357. Academic Press. FARAH, M. J., McMULLEN, P.A., & MEYER, M. M. (1991). Can recog­ CARAMAZZA, A., & HILLIS, A. E. (1991). Lexical organization of nouns nition ofliving things be selectively impaired? Neuropsychologia, 29, and verbs in the brain. Nature, 349, 788-790. 185-193. CARAMAZZA, A., HILLIS, A. E., RAPP, B., & ROMANI, C. (1990). Multi­ FARAH, M. J., MEYER, M. M., & McMULLEN, P.A. (1996). The living/ ple semantics or multiple confusions? Cognitive Neuropsychology, nonliving dissociation is not an artifact: Giving an a priori implausible 7,161-190. hypothesis a strong test. Cognitive Neuropsychology, 13, 137-154. CARAMAZZA, A., & MIOZZO, M. (1997). The relation between syntactic FARAH, M. J., & WALLACE, M. A. (1992). Semantically-bounded and phonological knowledge in lexical access: Evidence from the anomia: Implications for the neural implementation of naming. Neuro­ "tip-of-the-tongue" phenomenon. Cognition, 64, 309-343. psychologia, 30, 609-621. CARAMAZZA, A., & MIOZZO, M. (1998). More is not always better: A FIEZ,J. A., RAICHLE, M. E., BALOTA, D. A., TALLAL, P., & PETERSEN, response to Roelofs, Meyer, and Levelt. Cognition, 89, 231-241. S. E. (1996). PET activation of posterior temporal regions during audi­ CARAMAZZA, A., & SHELTON, J. R. (1998). Domain specific knowledge tory word presentation and verb generation. Cerebral Cortex, 6, 1-10. systems in the brain: The animate-inanimate distinction. Journal of FORDE, E. M. E., FRANCIS, D., RIDDOCH, M. J., RUMIATI, R. I., & Cognitive Neuroscience, 10,1-34. HUMPHREYS, G. W. (1997). On the links between visual knowledge CARAMAZZA, A., & ZURIF, E. B. (1976). Dissociation of algorithmic and and naming: A single case study of a patient with a category-specific heuristic processes in language comprehension. Brain & Language, impairment for living things. Cognitive Neuropsychology, 14, 403­ 3, 572-582. 458. COLTHEART, M., & COLTHEART, V. (1997). Reading comprehension is FRANKLIN, S., HOWARD, D., & PATTERSON, K. (1994). Abstract word not exclusively reliant upon phonological representation. Cognitive meaning deafness. Cognitive Neuropsychology, II, 1-34. Neuropsychology, 14, 167-176. FRITH, U.(1980). Unexpected spelling problems. In U.Frith (Ed.), Cog­ COLTHEART, M., CURTIS, B., ATKINS, P., & HALLER, M. (1993). Mod­ nitive processes in spelling (pp. 495-515). London: Academic Press. els of reading aloud: Dual-route and parallel-distributed-processing FUNNELL, E., & DE MORNAYDAVIES, P. (1997). JBR: A reassessment approaches. Psychological Review, 100,589-608. of concept familiarity and a category-specific disorder for living COLTHEART, M., PATTERSON, K., & MARSHALL, J. (1980). Deep dys­ things. Neurocase, 9,135-153. lexia (2nd ed.). London: Routledge & Kegan Paul. FUNNELL, E., & SHERIDAN, J. S. (1992). Categories of knowledge? Un­ COSLETT, H. B., & SAFFRAN, E. M. (1989). Preserved object recognition familiar aspects of living and nonliving things. Cognitive Neuropsy­ and reading comprehension in optic aphasia. Brain, 1I2, 1091-1110. chology, 9,135-153. COSLETT, H. B., & SAFFRAN, E. M. (1992). Optic aphasia and the right GAFFAN. D., & HEYWOOD, C. A. (1993). A spurious category-specific hemisphere: A replication and extension. Brain & Language, 43, visual agnosia for living things in normal humans and nonhuman pri­ 148-161. mates. Journal ofCognitive Neurosciences,S, 118-128. DAMASIO, A. R. (1990). Category-related recognition defects as a clue GAINOTTI, G., & SILVERI, M. C. (1996). Cognitive and anatomical locus to the neural substrates of knowledge. Trends in the Neurosciences, of lesion in a patient with a category-specific semantic impairment 13,95-98. for living beings. Cognitive Neuropsychology, 13, 357-389. DAMASIO, A. R., & ThANEL, D.(1993). Nouns and verbs are retrievedwith GAINOTTI, G., SILVERI, M. c., DANIELE, A., & GIUSTOLISI, L. (1995). differently distributed neural systems. Proceedings ofthe National Neuroanatomical correlates of category-specific semantic disorders: Academy ofSciences ofthe United States ofAmerica, 90, 4957-4960. A critical survey. In R. A. McCarthy (Ed.), Semantic knowledge and DAMASIO, H., GRABOWSKI, T. J., TRANEL, D., HICHWA, R. D., & DAMA­ semantic representations. Memory (Vol. 3, Issues 3 & 4, pp. 247­ SIO, A. R. (1996). A neural basis for lexical retrieval. Nature, 380, 264). Hove, England: Erlbaum, Taylor & Francis. 499-505. GARDNER, H., & ZURIF, E. (1975). Bee, but not be: Oral reading of sin­ DEHAENE, S. (1995). Electrophysiologicalevidence for category-specific gle words in aphasia and alexia. Neuropsychologia, 27, 193-200. word processing in the normal human brain. NeuroReport, 6, 2153­ GARRETT, M. F. (1980). Levels of processing in sentence production. In 2157. B. Butterworth (Ed.), Language production (pp. 177-220). New York: DELL, G. S. (1988). The retrieval of phonological forms in production: Academic Press. Test of predictions from a connectionist model. Journal ofMemory GESCHWIND, N. (1969). Problems in the anatomical understanding of &Language, 27, 124-142. aphasia. In A. L. Benton (Ed.), Contributions ofclinical neuropsy­ DELL, G. S. (1990). Effects offrequency and vocabulary type on phono­ chology (p. 107). Chicago: University of Chicago Press. logical speech errors. Language & Cognitive Processes, 4, 313-349. HANLEY, J. R., & McDONNELL, V. (1997). Are reading and spelling DELL, G. S., SCHWARTZ, M. F., MARTIN, N., SAFFRAN., E. M., & phonologically mediated" Evidence from a patient with a speech pro­ GAGNON, D. A. (1997). Lexical access in aphasic and nonaphasic duction impairment. Cognitive Neuropsychology, 14, 3-34. speakers. Psychological Review, 104, 801-838. HART, J., BERNDT, R. S., & CARAMAZZA, A. C. (1985). Category-specific DENES, G., & SEMENZA, C. (1975). Auditory modality-specific anomia: naming deficit following cerebral infarction. Nature, 316, 439-440. Evidence from a case of pure word deafness. Cortex, 11,401-411. HART, J., & GORDON, B. (1992). Neural subsystems for object knowl­ DERENZI, E., & LUCCHELLI, F.(1994). Are semantic systems separately edge. Nature, 359, 60-64. represented in the brain? The case ofliving category impairment. Cor­ HAUSER, M. (1997). Artifactual kinds and functional design features: tex, 30, 3-25. What a primate understands without language. Cognition, 64, 285-308. 24 SHELTON AND CARAMAZZA

HEAD, H. (1926). Aphasia and kindred disorders ofspeech. London: Neural correlates of category-specific knowledge. Nature, 379, Cambridge University Press. 649-652. HENAFF GONON, M., BRUCKERT, R., & MICHEL, F. (1989). Lexicaliza­ MCCARTHY, R A., & WARRINGTON, E. K. (1985). Category specificity tion in an anomie patient. Neuropsychologia, 27, 391-407. in an agrammatic patient: The relative impairment of verb retrieval HIER, D. B., & MOHR, J. P. (1977). Incongruous oral and written nam­ and comprehension. Neuropsychologia, 23,709-723. ing: Evidence for a subdivision ofthe syndrome Wernicke's aphasia. MCCARTHY, R A., & WARRINGTON, E K. (1988). Evidence for modality­ Brain & Language, 4, 115-126. specific meaning systems in the brain. Nature, 334, 428-430. HILLIS, A. E., & CARAMAZZA, A. (1991). Category-specific naming and MCCARTHY, R. A., & WARRINGTON, E K. (1990). Cognitive neuropsy­ comprehension impairment: A double dissociation. Brain, II4, 2081­ chology: A clinical introduction. San Diego: Academic Press. 2094. McRAE, K., DE SA, V. R., & SEIDENBERG, M. S. (1997). On the nature HILLIS, A. E., & CARAMAZZA, A. (1995a). Cognitive and neural mech­ and scope of featural representations for word meaning. Journal of anisms underlying visual and semantic processing: Implications from Experimental Psychology: General, 126, 99-130. "optic aphasia." Journal ofCognitive Neuroscience, 7,457-478. MICELI, G., BENVEGNU, B., CAPASSO, R, & CARAMAZZA, A. (1997). HILLIS, A. E., & CARAMAZZA, A. (I995b). Representation ofgrammat­ The independence of phonological and orthographic forms: Evidence ical categories of words in the brain. Journal ofCognitive Neuro­ from aphasia. Cognitive Neuropsychology, 14, 35-70. science, 7, 396-407. MICELI, G., SILVERI, M. C; NOCENTINI, U., & CARAMAZZA, A. (1988). HILLIS, A. E., RAPp, B., ROMANI, C., & CARAMAZZA, A. (1990). Selec­ Patterns of dissociation in comprehension and production of nouns tive impairments of semantics in lexical processing. Cognitive Neu­ and verbs. Aphasiology, 2, 351-358. ropsychology,7,191-243. MICELI, G., SILVERI, M. c, VILLA, G., & CARAMAZZA, A. (1984). On HOTOPF, N. (1980). Slips of the pen. In U. Frith (Ed.), Cognitive pro­ the basis for agrammatics difficulty in producing main verbs. Cortex, cesses in spelling (pp. 287-309). London: Academic Press. 20,207-220. HOWARD, D., & FRANKLIN, S. (1987). Three ways for understanding MIOZZO, M., & CARAMAZZA, A. (1997). On knowing the auxiliary ofa written words and their use in two contrasting cases of surface dys­ verb that cannot be named: Evidence for the independence ofgram­ lexia. In D. A. Allport, D. MacKay, W Prinz, & E. Scheerer (Eds.), matical and phonological aspects of lexical knowledge. Journal 0/ Language perception and production: Common processes in listen­ Cognitive Neuroscience, 9,160-166. ing, speaking, reading and writing (pp. 340-366). London: Acade­ MONSELL, S. (1987). On the relation between input and output pathways mic Press. for speech. In D. A. Allport, D. G. MacKay, W Prinz, & E. Scheerer HOWARD, D., & FRANKLIN, S. (1988). Missing the meaning? A cogni­ (Eds.), Language perception and production relationships between tive neuropsychological study ofthe processing ofwords by an apha­ speaking, reading and writing (pp. 273-311). London: Erlbaum. sic patient. Cambridge, MA: MIT Press. MORTON, J. (1981). The status ofinformation processing models oflan­ HUMPHREYS, G. W, & RIDDOCH, M. J. (1988). On the case for multiple guage. Philosophical transactions ofthe Royal Society ofLondon, semantic systems: A reply to Shallice. Cognitive Neuropsychology, 295,387-396. 5,143-150. Moss, H. E, TYLER, L. K., DURRANT-PEATFIELD, M., & BUNN, E. M. KOHN, S. E., & FRIEDMAN, R. (1986). Word meaning deafness: A (1998). Two eyes of a see-through: Impaired and intact semantic phonological-semantic dissociation. Cognitive Neuropsychology, 3, knowledge in a case ofselective deficit for living things. Neurocase, 291-308. 4,291-310. KOHN, S. E., LORCH, M. P, & PEARSON, D. M. (1989). Verb finding in MYERSON, R., & GOODGLASS, H. (1972). Transformational grammars aphasia. Cortex, 25, 57-69. ofthree agrammatic patients. Language & Speech, 15,40-50. KREMIN, H. (1987). Is there more than ah-oh-oh? Alternative strategies NICKELS, L., & HOWARD, D. (1994). A frequent occurrence? Factors af­ for writing and repeating lexically. In M. Coltheart, G. Sartor, & fecting the production of semantic errors in aphasic naming. Cogni­ R. Job (Eds.), The cognitive neuropsychology oflanguage (pp. 295­ tive Neuropsychology, II, 289-320. 336). London: Erlbaum. NOLAN, K. A., & CARAMAZZA, A. (1983). An analysis of writing in a LAIACONA, M., BARBAROTTO, R, & CAPITANI, E (1993). Perceptual and case ofdeep dyslexia. Brain & Language, 20, 305-328. associative knowledge in category specific impairment of semantic PAIVIO, A. (1971). Imagery and verbal processes. London: Holt, Rine­ memory: A study oftwo cases. Cortex, 29, 727-740. hart & Winston. LAIACONA, M., CAPITANI, E., & BARBAROTTO, R (1997). Semantic cat­ PATTERSON, K. E., MARSHALL, J. c.,& COLTHEART, M. (1985). Surface egory dissociations: A longitudinal study of two cases. Cortex, 33, dyslexia: Neuropsychological and cognitive studies ofphonological 441-461. reading. London: Erlbaum. LAMBON-RALPH, M. A., HOWARD, D., NIGHTINGALE, G., & ELLIS, A. PATTERSON, K. E., SEIDENBERG, M. S., & MCCLELLAND, J. L. (1989). (1998). Are living and nonliving category-specific deficits causally Connections and disconnections: Acquired dyslexia in a computa­ linked to impaired perceptual or associative knowledge? Evidence tional model ofreading processes. In R. G. M. Morris (Ed.), Parallel from a category-specific double dissociation. Neurocase, 4, 311-338. distributed processing: Implications for psychology and neuro­ LAWS, K. R., EVANS, J. J., HODGES, J. R., MCCARTHY, R A. (1995). science (pp. 131-181). Oxford: Oxford University Press. Naming without knowing and appearance without awareness: Evi­ PATTERSON, K. E., & SHEWELL, C. (1987). Speak and spell: Dissociations dence for constructive processes in semantic memory? Memory, 3, and word class effects. In M. Coltheart, G. Sartori, & R. Job (Eds.), 409-433. The cognitive neuropsychology oflanguage (pp. 273-294). London: LEVELT, W. J. M. (1989). Speaking: From intention to articulation. Erlbaum. Cambridge, MA: MIT Press. PERANI, D., CAPPA, S. E, BETTINARDI, v., BRESSI, S., GORNO-TEMPINI, M., LHERMITTE, E, & BEAUVOIS, M.-F. (1973). A visual-speech discon­ MATARRESE, M., & FAZIO, F. (1995). Different neural systems for nexion syndrome: Report ofa case with optic-aphasia, agnosic alexia the recognition of animals and man-made tools. Neuro Report; 6, and colour agnosia. Brain, 96, 695-714. 1637-1641. LURIA, A. R. (1966). Higher corticalfunctions in man. New York: Basic PERFETTI, D., & BELL, L. (1991). Phonemic activation during the first Books. forty milliseconds of word identification: Evidence from backward MARSHALL, J., CHlAT, S. ROBSON, J., & PRING, T.(1996). Calling a salad masked priming. Journal ofMemory & Language, 30, 473-485. a federation: An investigation of semantic jargon. Part 2-Verbs.Jour­ PETERSEN, S., Fox, P., POSNER, M., MINTUN, M., & RAICHLE, M. nal ofNeurolinguistics, 9, 251-260. (1989). Positron emission tomography studies ofthe processing of sin­ MARTIN, A., HAXBY, J. Y, LALONDE, F. M., WIGGS, C. L., & UNGER­ gle words. Journal ofCognitive Neuroscience, I, 153-170. LEIDER, L. G. (1995). Discrete cortical regions associated with knowl­ PIETRINI, Y, NERTEMPI, P., VAGLIA, A., REVELLO, M. G., PINNA, Y, & edge of color and knowledge ofaction. Science, 270, 102-105. FERRO-MILONE, F. (1988). Recovery from herpes simplex encephali­ MARTIN, A., WIGGS, C. L., UNGERLEIDER, L. G., & HAXBY, J. V. (1996). tis: Selective impairment ofspecific semantic categories with neuro- LEXICAL AND SEMANTIC DEFICITS 25

radiological correlation. Journal ofNeurology, Neurosurgery, & SILVERI, M. C, & GAINOTTI, G. (1988). Interaction between vision and Psychiatry, 51,1284-1293. language in category-specific impairment. CognitiveNeuropsychology, PLAUT, D. c., MCCLELLAND, J. L., SEIDENBERG, M. S., & PATTER­ 5,677-709. SON, K. (1996). Understanding normal and impaired word reading: SIRIGU, A., DUHAMEL, J. R., & PONCET, M. (1991). The role of sen so­ Computational principles in quasi-regular domains. Psychological rimotor experience in object recognition: A case of multi modal ag­ Review, 103, 56-115. nosia. Brain, 114,2555-2573. POWELL, J., & DAVIDOFF, J. (1995). Selective impairments of objects SNODGRASS, J. G. (1984). Concepts and their surface representations. knowledge in a case ofacquired cortical blindness. Memory, 3, 435­ Journal ofVerbalLearning & Verbal Behavior, 23, 3-22. 461. STEWART, E, PARKIN, A. J., & HUNKIN,N. M. (1992). Naming impair­ PULVERMULLER, E (1995). Agrammatism: Behavioral description and ments following recovery from herpes simplex encephalitis: neurobiological explanation. Journal ofCognitive Neuroscience, 7, Category-specific? Quarterly Journal ofExperimental Psychology, 165-181. 44A, 261-284. PuLVERMULLER, E, PREISSL, H., LUTZENBERGER, W., & BIRBAUMER, N. TYLER, L. K., & Moss, H. E. (1997). Functional properties ofconcepts: (1997). Nouns and verbs in the intact brain: Behavioral and electro­ Studies ofnormal and brain-damaged patients. Cognitive Neuropsy­ cortical evidence. Unpublished manuscript. chology, 14,426-486. RAPP, B. [C.], BENZING, L., & CARAMAZZA, A. (1997). The autonomy VANDENBERGHE, R., PRICE, C, WISE, R, JOSEPHS, 0., & FRACKOWIAK, oflexical orthography. Cognitive Neuropsychology, 14, 71-104. R. S. J. (1996). Functional anatomy ofa common semantic system for RAPP, B. [Cj, & CARAMAZZA, A. (1997). The modality-specific orga­ words and pictures. Nature, 383, 254-256. nization ofgrammatical categories: Evidence from impaired spoken VAN ORDEN,G. C, JANSEN OP DE HAAR, M. A., & BOSMAN, A. M. 1. and written sentence production. Brain & Language, 56, 248-286. (1997). Complex dynamic systems also predict dissociations, but RAPP, B. C, HILLIS, A. E., & CARAMAZZA, A. (1993). The role ofrep­ they do not reduce to autonomous components. Cognitive Neuropsy­ resentations in cognitive theory: More on multiple semantics and the chology, 14, 131-166. agnosias. Cognitive Neuropsychology, 10,235-249. VAN ORDEN,G. C, JOHNSTON, J. C, & HALE,B. L. (1988). Word identi­ RIDDOCH, M. J., & HUMPHREYS, G. W. (1987). Visual object processing fication in reading proceeds from spelling to sound to meaning. Jour­ in optic aphasia: A case ofsemantic access agnosia. Cognitive Neuro­ nal ofExperimental Psychology: Learning, Memory, & Cognition, psychology,4,131-185. 14,371-386. RIDDOCH, M. J., HUMPHREYS, G. W., COLTHEART, M., & FUNNELL, E. WARRINGTON, E. K. (1975). The selective impairment ofsemantic mem­ (1988). Semantic system or systems? Neuropsychological evidence ory. Quarterly Journal ofExperimental Psychology, 27, 635-657. re-examined. Cognitive Neuropsychology,S, 3-26. WARRINGTON, E. K., & MCCARTHY, R (1983). Category specific access ROELOFS, A. (1992). A spreading-activation theory of lemma retrieval dysphasia. Brain, 106, 859-878. in speaking. Cognition, 42, 107-142. WARRINGTON, E. K., & MCCARTHY, R (1987). Categories of knowl­ SACCHETT, c., & HUMPHREYS, G. W. (1992). Calling a squirrel a squir­ edge: Further fractionation and an attempted integration. Brain, 100, rel but a canoe a wigwam: A category-specific deficit for artefactual 1273-1296. objects and body parts. Cognitive Neuropsychology, 9, 73-86. WARRINGTON, E. K., & SHALLICE, 1. (1979). Semantic access dyslexia. SAFFRAN, E. M., & SCHWARTZ, M. E (1994). Of cabbages and things: Brain, 102, 43-63. Semantic memory from a neuropsychological perspective-A tutor­ WARRINGTON E. K., & SHALLlCE, 1. (1984). Category-specific seman­ ial review. In C. Umilta & M. Moscovitch (Eds.), Attention and Per­ tic impairments. Brain, 107, 829-853. formance xv Conscious and nonconscious information processing WERNICKE, C. (1989). Neurology: Recent contributions on aphasia. Cog­ (pp. 507-536). Cambridge, MA: MIT Press, Bradford Books. nitive Neuropsychology, 6, 547-569. (Original work published 1886) SAMSON, D., PILLON, A., & DE WILDE, V. (1998). Impaired knowledge WISE,R, CHOLLET, E, HADAR, U., FISTON, K., HOFFNER, E., & FRACK­ of visual and nonvisual attributes in a patient with a semantic im­ OWIAK, R. (1991). Distribution ofcortical neural networks involved pairment for living entities: A case ofa true category-specific deficit. in word comprehension and word retrieval. Brain, 114, 1803-1817. Neurocase, 4, 273-290. ZINGESER, L., & BERNDT, R. S. (1988). Grammatical class and context SARTORI, S., & JOB,R. (1988). The oyster with four legs: A neuropsycho­ effects in a case ofpure anomia: Implications for models oflanguage logical study on the interaction of visual and semantic information. processing. Cognitive Neuropsychology,S, 473-516. Cognitive Neuropsychology,S, 105-132. SCHWARTZ, M. E, MARIN,O. S. M., & SAFFRAN, E. M. (1979). Disso­ NOTES ciation oflanguage function in dementia: A case study. Brain & Lan­ guage, 7, 277-306. I. Functional components could also be distinguished neurocherni­ SCHWARTZ, M. E, SAFFRAN, E. M., & MARIN, O. S. M. (1980). The cally. However, since all the relevant evidence available to us-lesion word order problem in agrammatism: I. Comprehension. Brain & site, PET, and fMR1 data---concerns spatial segregation ofcomponents, Language, 10,249-262. we will be concerned only with neuroanatomical distinctions. SEYMOUR, P. H. K. (1979). Human visual cognition. London: Collier 2. Currently, there is ongoing debate concerning the need to separate Macmillan. input phonology and input orthography, with some models supporting SHALLICE, 1. (1981). Phonological agraphia and the lexical route in separate input pathways (e.g., Coltheart, Curtis, Atkins, & Haller, 1993) writing. Brain, 104,413-429. and other models supporting distributed representations that do not re­ SHALLlCE,1.(1987). Impairments ofsemantic processing: Multiple dis­ quire the representation of orthographic word forms (e.g., Plaut, Mc­ sociations. In M. Coltheart, G. Sartori, & R. Job (Eds.), The cognitive Clelland, Seidenberg, & Patterson, 1996). This issue is too complex to neuropsychology oflanguage (pp. 111-127). London: Erlbaum. review here. However, we will assume that input orthography and input SHALLlCE,1. (1988). From neuropsychology to mental structure. Cam­ phonology are represented separately since this assumption provides us bridge: Cambridge University Press. with a parallel configuration to our assumptions about output phonol­ SHALLlCE, 1. (1993). Multiple semantics: Whose confusions" Cogni­ ogy and orthography, discussed later in the paper (see also note 15 for tive Neuropsychology, 10, 251-262. further comment on the relationship between assumptions for input and SHELTON, J. R., FOUCH, E., & CARAMAZZA, A. (1998). The selective for output and how this may relate to the issue of separate [input and sparing ofbody part knowledge: A case study. Neurocase, 4,339-351. output] phonological and orthographic stores). SHELTON, J. R., & WEINRICH, M. (1997). Further evidence of a disso­ 3. Attempts have been made to use distributed network models to ciation between output phonological and orthographic lexicons: A simulate the performance ofpatients who present with surface dyslexia case study. Cognitive Neuropsychology, 14,105-130. (e.g., Patterson, Seidenberg, & McClelland, 1989; Plaut et al., 1996). SHERIDAN, J., & HUMPHREYS, G. W. (1993). A verbal-semantic category­ Contrary to the authors' claims, however. these attempts have not been specific recognition impairment. Cognitive Neuropsychology, 10, very successful in accounting for thefitll range of patient performance 143-184. (see, e.g., Coltheart et al., 1993). 26 SHELTON AND CARAMAZZA

4. The assumptions concerning routes of access to meaning are not vation could be used to support his written naming. Although we cannot common to all models of multiple-modality semantics but are specific conclusively state that E.A. was not activating some lexical phonology to the model described by Shall ice, as depicted in Figure 2 (see also in naming, there is some evidence to suggest that he was not getting Warrington & McCarthy, 1987). much, if any, phonological information that could then be used to sup­ 5. Unless otherwise noted, levels ofperformance for patients will al­ port successful writing. First, E.A. rarely made phonological errors in ways refer to proportion correct. naming, suggesting that his deficit was not one of postlexical phono­ 6. Riddoch and Humphreys (1987) questioned the magnitude ofthe logical output processing (i.e., there is no evidence that he successfully difference in performance between the different modalities ofinput for retrieved any ofthe phonological information), and he did not show any at least one ofthe modality-specific semantic memory deficit patients effects ofphonological variables on naming. Second, E.A. would often studied by Warrington (1975). When performance was collapsed across produce two different responses in the different modalities when at­ all the probe question tasks, rather than each probe question being ex­ tempting to name at the same time as he was performing the naming task amined individually, A.B. 's overall performance was 112/160 for picture (i.e., when writing a name he would say a different name). input and 109/160 correct for auditory input (E.M. showed a stronger 13. W.M.A. was also tested on a double naming task within the same difference in performance, however: 132/160 for picture input and modality. In this task, he always produced the same responses in the two 104/160 for auditory input). Riddoch and Humphreys pointed out that trials. Thus, different responses were produced only across different both patients performed equally well or better with auditory input than modalities of output. The contrast in results in the two tasks suggests with picture input when the question was one that could not be answered that the different responses produced in speaking and writing are not by information provided in the picture ("Is it English?"). All the other merely the result of independent attempts at naming but, rather, reflect probe questions used by Warrington could have been answered on the the fact that different modalities are implicated. basis ofinformation available in the picture. 14. Shelton and Weinrich (1997) developed a similar argument for 7. Although they presented 4 patients, only 2 ofthese patients were the role of phonology in reading comprehension. Their patient, E.A., studied extensively (J.B.R., S.B.Y.), and we will focus on them. The was shown to have little support from sublexical and lexical phonolog­ other 2 patients (K.B., I.N.G.) were very impaired and were tested on a ical information for input processing but demonstrated quite good read­ subset ofcomprehension tasks only. They did, however, show a similar ing comprehension. This pattern of performance does not support an pattern ofselective deficits for living things on these tasks. obligatory role ofphonology in reading comprehension (see also Colt­ 8. Although the terminology refers to "visual" and "functional" se­ heart & Coltheart, 1997; Hanley & McDonnell, 1997; but see Van mantic information, the terms are misleading since they have come to Orden, Jansen op de Haar, & Bosman, 1997, for commentary). be short-hand terms for "perceptual" and "nonperceptual" knowledge. 15. These data may also contribute to our understanding of input For example, the sound an animal makes is considered perceptual knowl­ phonology and orthography. As mentioned earlier (see note 2), there is edge and the country from which an animal comes is considered non­ ongoing debate as to whether or not separate word forms are needed to perceptual knowledge. Thus, type of knowledge does not refer to only represent input phonology and orthography or whether input processing visual or only functional knowledge, but more generally to perceptual can be represented in one distributed network. If one distributed net­ and nonperceptual knowledge. work were used to represent input processing, it would likely be the case 9. This model is a more "general" example ofthe sensory/functional that one distributed network would represent output processing (or even theory but is the "received" view in the literature. The model proposed that the same network would be used for input and output processing). by Warrington and colleagues distinguishes first between visual and Yet the data reported here support separate representations for output verbal semantics and then, within each system, perceptual and nonper­ phonology and output orthography. lt has yet to be demonstrated whether ceptual knowledge. or not a distributed network ofeither speaking or writing can handle the 10. It is important to note that even when categories have been equated dissociations in speaking and writing described here. And-more for familiarity, frequency, and visual complexity, there are more cases challenging-is the determination of whether or not one distributed net­ reported that have a deficit to animals/living things than those that have work could handle all the findings from input processing as well as out­ a deficit to nonanimals/nonliving things, a fact that becomes important put processing. in attempting to explain category-specific deficits (see below). However, 16. Patients who make semantic errors in only one modality of out­ some researchers have argued that other factors such as age of acquisi­ put might be argued to have a semantic deficit that reveals itself only in tion and imageability can influence performance with category-specific the impaired output modality if degraded semantic information could items and may account for the apparent discrepancy in the number of not strongly address degraded information for output in the impaired cases that present with deficits to animals/living things as compared modality. However, for this argument to work we would have to assume with nonanimals/nonliving things (e.g., Lambon-Ralph et al., 1998). that the impoverished semantic information is enough to activate the Lambon-Ralph et al. claimed that animate concepts are acquired at an correct output (even though the semantic information is degraded) in earlier age than nonanimate concepts and are more imageable, which the unimpaired output modality. This argument does not seem likely could lead to better performance with animate items. While the point is given the differential levels of performance between the two output well taken that other factors such as age ofacquisition and imageabil­ modalities. In at least some patients, performance in the unimpaired ity may influence performance and could lead to spurious identifica­ modality is perfect. This suggests that they can accurately access unim­ tion of category-specific deficits, this explanation cannot account for paired semantics, which allows them to produce the correct name on the greater number of cases with deficits to animals/living things. In ever\, trial. lt would be difficult to argue, for example, that a patient can fact, this explanation would predict just the opposite since earlier age of perform perfectly on an oral naming task because he/she is unimpaired acquisition and greater imageability would make animals the "easier" in speaking and still has a semantic impairment. category (although at the same time this category has been shown to be 17. Rapp and Caramazza ( 1997) reported a patient who demonstrated less familiar than nonanimate categories and claimed to be the "harder" impaired verbal production of open-c lass words and impaired written category). Thus, this "easier" category should not be expected to be production of closed-class words with spared noun production. Thus. identified as the "deficient" category in the majority ofcases. there is some evidence that nouns and verbs are not the only grammat­ II. Although we are suggesting that writing and speaking can occur ical class along which lexical output processing is organized. independently, we are not claiming that they must occur independently. 18. An account that represents grammatical class within each modality­ Evidence that supports a role ofphonology in writing is not contradic­ specific output lexicon (i.e., output lexicons are organized according to tory to our arguments here. We are only suggesting that writing and grammatical class) would have difficulty accounting for patients who speaking do not have to rely on each other; that is, phonology does not perform similarly with a specific grammatical class in both modalities have an obligatory role in successful writing. (e.g.. poor naming of verbs in both speaking and writing). 12. The data from E.A. do not allow us to rule out completely the 19. There is another way in which a modality-specific grammatical possibility that he was activating a partial phonological code when at­ class effect could arise. If we assume that objects and actions are repre­ tempting to orally name a picture and that this partial phonological acti- sented separately in the semantic system, damage to the connections LEXICAL AND SEMANTIC DEFICITS 27

from semantic information about nouns or verbs to the modality­ and phoneme decision, we would not expect the above patterns to ob­ specific output form could result in a modality-specific grammatical tain. And, one cannot account for Dante's performance with verb aux­ class effect (Caramazza & Miozzo, 1998). Although this distinction iliaries by using an explanation that depends on "ease ofthe task" since does not require us to postulate a tripartite distinction between seman­ the number of different auxiliaries used in Italian (two) is the same as tic, syntactic, and lexical output forms, there is other converging evi­ the number ofdifferent genders (two). dence to support this distinction, which is summarized below. 21. Pulvermuller (1995) reviewed evidence suggesting that content 20. It might seem that determining the initial phoneme or final pho­ words and function words were also processed and represented in neu­ neme is much more difficult than gender decision, even in those cases roan atomically distinct areas. where the phoneme decision is a forced-choice task (i.e., the patient only 22. Although Dell has not specifically stated that the "lemma" level has two choices for phoneme decision, the same as gender decision). of representation is modality neutral, he has stated that the lemma is However, Caramazza and Miozzo (1997; see also Miozzo & Caramazza, "nonphonological" (Dell et al., 1997). Presumably, this implies that the 1997) have demonstrated that correct gender decision is not dependent lemma level is modality neutral since the lemma is assumed to be an ab­ on correct phoneme decision and that phoneme decision is made with stract representation specified for grammatical information but clearly better-than-chance accuracy (by normal subjects). Thus, the lack ofre­ not for phonological (or, by extension, orthographic) information. This lationship between gender decision and phoneme decision is not be­ line of reasoning applies to other models of speech production as well cause subjects are very good with gender decision and very poor with (e.g., Bock & Levelt, 1994; Roelofs, 1992). phoneme decision, but rather that the two do not depend on each other (are not correlated), and a subject is frequently correct with phoneme decision but incorrect with gender decision. Ifit were just ease ofdcci­ (Manuscript received December 8, 1997; sion that was driving the lack of relationship between gender decision revision accepted for publication May 20, 1998.)