The cortical localization of the lexicons. Positron emission tomography evidence.

Brain (1992), 115, 1769-1782

THE CORTICAL LOCALIZATION OF THE LEXICONS POSITRON EMISSION TOMOGRAPHY EVIDENCE

(From the 'MRC Cyclotron Unit, Hammersmith Hospital, London, the 2Neurology Department, Charing Cross Hospital, London, the -Psychology Department, Birkbeck College, University of London, the 4MRC Applied Psychology Unit, Cambridge, UK, the 5Radiology Department, University of Wisconsin Hospital, Madison, Wisconsin, USA and the ^Neurology Department, University of Essen, Germany)

SUMMARY Positron emission tomography was used to investigate changes in regional cerebral blood flow (rCBF) in neurologically normal subjects during word reading and word repetition. The blood flow in these conditions was compared with control conditions where subjects were presented with stimuli of comparable auditory and visual complexity to real words and said the same word on presentation of each stimulus. The control condition for word repetition (hearing spoken words presented backwards) resulted in bilateral activation of the superior temporal gyrus. Word repetition caused a significant increase in rCBF over this control condition in the left superior and middle temporal gyri. The control condition for word reading (seeing stimuli written in 'false fonts', i.e. non-existent letter-like forms) resulted in significant changes in rCBF bilaterally in the striate and extrastriate cortex. Word reading caused a significant increase in blood flow relative to this control in the posterior part of the left middle temporal gyrus. The implications of these results are discussed, and it is argued that they are consistent with localization of a lexicon for spoken word recognition in the middle part of the left superior and middle temporal gyri, and a lexicon for written word recognition in the posterior part of the left middle temporal gyrus.

INTRODUCTION

In 1874 Carl Wernicke described two aphasic patients with fluent, paraphasic speech and a prominent disorder in spoken word comprehension (Wernicke, 1874). Wernicke argued that the patients' difficulty was due to impaired auditory images of words; the disorder of speech production would follow from the fact that, in Wernicke's theory, auditory images mediate the production of speech from the centre for motor word images, located in Broca's area in the inferior frontal convolution. On autopsy for one of these patients, Wernicke subsequently demonstrated an infarction of the superior temporal gyrus in the left hemisphere. Wernicke argued that this area of the association cortex, located immediately posterior to the primary auditory cortex, was neurologically plausible as a cortical area specialized for world recognition: in more modern psychological terminology, a lexicon for spoken words. Correspondence to: David Howard, Psychology Department, Birkbeck College, Malet Street, London WC1E 7HX, UK. © Oxford University Press 1992

Downloaded from http://brain.oxfordjournals.org/ at Ryerson University on November 18, 2012

by DAVID HOWARD, 3 KARALYN PATTERSON, 4 RICHARD WISE, 1 2 W. DOUGLAS BROWN, 1 5 KARL FRISTON, 1 CORNELIUS WEILLER 1 6 and RICHARD FRACKOWIAK 1

1770

D. HOWARD AND OTHERS

The cortical location of a lexicon for recognition of written words has proved somewhat more elusive. Dejerine (1891) suggested that there was a centre for visual word images in the left angular gyrus. Wernicke (1906) was, however, sceptical of this suggestion, pointing out that the lesions of patients with damage in this area typically extended to the underlying white matter; thus their difficulty in reading might arise from a disconnection of association fibres rather than an impairment to a cortical centre. Despite Wernicke's criticisms, Dejerine's anatomical conjectures have been widely accepted. Dejerine (1892) argued that pure alexia would occur when the left angular gyrus was disconnected from the primary visual areas of both the left and right occipital lobes. Lesions associated with pure alexia typically involve the left occipital lobe and the posterior part of the corpus callosum, although any lesion or combination of lesions which results in the disconnection can apparently produce this pure disorder of reading (Geschwind, 1965; Greenblatt, 1973, 1983). Where damage is to the angular gyrus itself, both alexia and agraphia result (Nielson and Raney, 1938; Benson and Geschwind, 1969). According to Dejerine, this occurs because the centre for visual word images plays a role in written language production, in much the same way that, as Wernicke had suggested, the centre for auditory word images is involved in spoken language production. There has been much disagreement on whether understanding of written words involves, as a necessary step, access to the lexicon for spoken words. Wernicke (1874) argued that recognition of written words would require recoding into their phonological form, which could then access the centre for auditory word images (although he later modified this view; see De Bleser and Luzzatti, 1989). This became a standard view among some neurologists (e.g. Geschwind, 1965, 1979; Luria, 1970) and some psychologists (e.g. Rubenstein et ai, 1971). More recently, a number of arguments have been made for direct access from a reading lexicon to semantics. For instance, as Coltheart (1980) points out, normal people have no difficulty in retrieving the correct meanings of words with different spellings but identical pronunciations (e.g. pear, pair;


The correspondence that Wernicke noted between lesions in the left superior temporal gyrus and difficulties in spoken word recognition has proved remarkably durable. Over the ensuing decades a variety of post-mortem data broadly supported his original view, refining the localization of 'Wernicke's area' to the posterior half of the superior temporal gyrus (e.g. Nielson, 1936; Seldon, 1985). However, as Bogen and Bogen (1976) point out, there have been considerable differences in the postulated extent of Wernicke's area. Some researchers have confined the area to the posterior part of the superior temporal gyrus (e.g. Geschwind, 1969), while others have suggested that Wernicke's area extends over much of the posterior part of the superior and middle temporal gyri, the supramarginal gyrus and inferior portions of the parietal lobe (e.g. Marie and Foix, 1917). The posterior part of the superior temporal gyrus is the core common to most (but by no means all) accounts of the localization of Wernicke's area (Bogen and Bogen, 1976). More recently, studies correlating forms of aphasia with lesions as revealed by computerized tomography scan have shown that most patients with primary deficits of spoken word recognition have lesions which involve Wernicke's area in the left temporal lobe (e.g. Naeser and Hayward, 1978), although there are infrequent exceptions where Wernicke's area is destroyed without any marked disorder in spoken word comprehension (e.g. Basso etai, 1985).

LOCALIZATION OF THE LEXICONS

1771


but cf Van Orden, 1987). Secondly, patients with acquired 'deep dyslexia' make semantic errors in reading aloud single words (e.g. for cat they say dog). These patients have clearly accessed (some aspects of) the meaning of the written word, where its phonological form is apparently unavailable (Coltheart et al., 1980). Geschwind (1979) provides a very precise hypothetical account of the cortical areas involved in word repetition and reading aloud. He suggests that in word repetition, the heard word activates the primary auditory cortex and then lexical representations in Wernicke's area. This sends activation, via the arcuate fasciculus, to Broca's area which, in turn, activates motor programming in the motor cortex. Word reading, on the other hand, activates the primary visual (striate) cortex, and then visual word forms in the region of the angular gyrus, involving both posterior superior temporal regions and the inferior portion of the parietal lobe. This, in turn, activates Wernicke's area, which transmits a representation to Broca's area, which is converted into an articulatory program in the motor cortex. Thus, in Geschwind's formulation, exactly the same areas are involved in word repetition and reading aloud, with the exceptions of (i) the primary auditory cortex which is involved only in word repetition, (ii) the striate cortex and the angular gyrus which are involved only in word reading. A theorist holding the view that written words can be recognized without recourse to their spoken forms would add one further difference between areas involved in the two tasks: Wernicke's area need not be activated in reading words aloud. Current forms of functional brain imaging permit a more direct attack upon the localization of speech and reading lexicons. The results of several important positron emission tomography (PET) activation studies by Petersen et al. (1988, 1990) suggest rather different locations for these lexical centres, especially for visual word forms, from the ones hypothesized by Wernicke, Dejerine or Geschwind. With regard to the reading lexicon, Petersen et al. (1988, 1990) place this in the left medial extrastriate cortex, thus essentially arguing that visual word forms are more to do with vision than with language. In the Petersen et al. (1990) study, subjects passively viewed four different types of experimental stimuli, to be compared with the baseline of a simple fixation cross. In one condition subjects saw 'false fonts', i.e. strings of letter-like forms of comparable visual complexity to real letters. In the second condition subjects saw random consonant strings (e.g. NLPFZ); in the third, stimuli were pronounceable pseudo-words (e.g. TWEAL), and, in the fourth, real words (e.g. BOARD). Relative to the fixation cross, all four conditions produced significant increases in activation in both the left and right lateral extrastriate cortex. When compared with presentation of false fonts and random letter strings, both real words and pseudo-words caused greater activation in the left medial extrastriate cortex (and, with reference to fig. 2A,B in Petersen et al., 1990, in several other places as well). The authors suggested that both real words and pseudo-words, because they conform to the orthographic constraints of English, activate the reading lexicon, which is located in the left medial extrastriate cortex. In support of this conclusion, Petersen et al. (1990) pointed out that left-hemisphere extrastriate lesions are often found in patients with pure alexia. However, as noted above, the classical account attributes pure alexia to a disconnection of visual sensory cortex from (intact) visual lexical representations. Moreover, recent experimental work with some pure alexic patients, for whom residual letter-by-letter reading is possible, suggests

1772



that their lexical representations for written words may be relatively intact (Bub et al., 1989; Coslett and Saffran, 1989; Schacter et al., 1990; Howard, 1991). The PET studies by Petersen et al. (1988, 1989, 1990) have concentrated more on locating visual than spoken word forms; however, one condition of the set employed by Petersen et al. (1988) compared passive hearing of spoken words, whilst fixating a cross-hair, to the fixation alone. Significant activation for the former minus the latter was found in several regions, including (i) the left and right primary auditory cortex, which Petersen et al. attribute to sensory processing, and (ii) the left temporoparietal cortex. Whilst acknowledging that more work is needed on this issue, Petersen et al. (1988, 1990) tentatively suggested that phonological processing is located in the left temporoparietal cortex. In support of this, Petersen et al. (1989) showed activation in this area during rhyme judgements on written words, a task which requires the generation of phonology. The present study, like the research of Petersen et al., used measurements of regional cerebral blood flow (rCBF) by PET in normal subjects, to investigate the location of cortical substrates for spoken and visual word recognition. The main intention was to shed some light on the apparent discrepancy between the regions classically associated with these lexical systems (Wernicke, 1874; Dejerine, 1891; Geschwind, 1979) and the regions identified by the PET studies of Petersen et al. (1988, 1990). A second intention was to clarify the extent to which written-word processing would activate the spoken-word lexicon. There were four experimental conditions in this PET study: two employed visual presentation and two used auditory presentation; within each modality, one condition employed real-word stimuli and the other used stimuli designed to have sensory characteristics of equivalent complexity to real words, without having any of the phonological, orthographic or semantic values associated with real words. Detailed aspects of the design of these conditions appear in Methods, but a basic description of each condition and the psychological processes that it should involve is as follows. (i) Word reading. A sequence of real words was presented on a computer screen, and the subject was asked to read each word in the sequence aloud. This task involves visual processing of complex patterns, access to the reading lexicon, retrieval of the corresponding word for spoken output and the motor aspects of speech production. There is also reason to believe that, although the task of reading a word aloud does not require access to its meaning, semantic representations will be automatically activated when subjects process real words in any task (e.g. Lupker, 1985; MacLeod, 1991; Neely, 1991). (ii) Word repetition. A sequence of real words in spoken form was presented to the subject who was asked to repeat each word after hearing it. This task involves auditory processing of complex sounds, access to the spoken-word lexicon, and then as in word reading, the further necessary stages for speech production. Furthermore, and again as with word reading, a semantic representation corresponding to the spoken word may be automatically activated. (iii) See and say. This is the control condition for word reading. Stimuli consisted of character strings where each character was a meaningless letter-like squiggle with roughly equivalent visual complexity to a real letter. These items are like the false fonts used by Petersen et al. (1990). On presentation of each string in this condition, the


1773

TABLE I. PSYCHOLOGICAL PROCESSES INVOLVED IN THE FOUR EXPERIMENTAL TASKS See and say

Word reading

Hear and say

Word repetition

Primary visual processes Complex pattern recognition

Primary visual processes Complex pattern recognition Visual input lexicon Semantic access Word retrieval Word production Articulatory programming

Primary auditory processes Complex pattern recognition

Primary auditory processes Complex pattern recognition Auditory input lexicon Semantic access Word retrieval Word production Articulatory programming

Word production Articulatory programming

Word production Articulatory programming

various processes. Subtracting rCBF for 'hear and say' from that for 'see and say' should identify the cortical areas involved in primary visual processing of complex, letter-like stimuli. Likewise, subtracting 'see and say' from 'hear and say' should find the cortical areas involved in primary, pre-lexical processing of complex auditory stimuli. Ideally, subtracting 'see and say' from word reading would isolate the visual word-form lexicon, and subtracting 'hear and say' from word repetition should locate the auditory wordform lexicon. However, as noted above, there are in fact two additional ways in which the real word conditions may differ from their control conditions: the words have meaning, which may be automatically activated even though the tasks do not require word comprehension; and the word conditions require retrieval of a different word pronunciation for each word in the sequence, whereas the same word ('crime') is uttered in response to every item in the control conditions. The design does, however, provide a potential solution to this problem: since the retrieval of pronunciations and meanings


subject was required to say 'crime', a word chosen to match the means of the word-sets used in this experiment on the variables of frequency, length and imageability. Although the word 'crime' clearly has a strong semantic representation, multiple repetition of a single word rapidly results in semantic satiation (e.g. Smith, 1984). Thus semantic processing should not make a significant contribution to this task. The see-and-say task was designed to recruit the processes of visual pattern analysis required in word reading. Furthermore, although this control condition does not require retrieval of a different word pronunciation on every trial as in the word-reading condition, the subject does speak in response to every control string; thus the cortical areas involved in word production and articulatory programming should be activated by see-and-say as well as by word reading. (iv) Hear and say. This is the control condition for word repetition. A taped sequence of real spoken words was re-recorded in reverse, yielding auditory stimuli with the same frequency spectra and auditory complexity as the stimuli in the word-repetition condition, just as in the 'see and say' condition, subjects were required to say 'crime' each time a hear-and-say auditory stimulus was presented. All four conditions involve spoken word production at the same mean rate (40 wpm), and therefore necessarily involved auditory input of the speaker's own voice. These processes which are constant across conditions should be irrelevant to comparisons of rCBF between conditions. The processes which we argue are recruited by these tasks are summarized in Table 1. The comparison methodology of PET should permit estimates of localization of the

1774


are likely to be involved to the same extent in both word reading and word repetition, differences between the two word conditions themselves should help to single out the visual and auditory input lexicons. In summary, we predicted that we would (i) identify two activation foci uniquely involved in word reading and word repetition (relative to their respective baselines)— these would represent the visual and auditory input lexicons, respectively; (ii) identify a third focus common to both tasks if automatic semantic processing causes significant changes in blood flow in the real word tasks; (iii) find activation in Wernicke's area during word reading only if Geschwind's thesis about obligatory phonological coding in reading is correct.

Each subject participated in six rCBF measurements made serially in a single session lasting a total of about 90 min. Two separate scans were made for each of the real word conditions (i.e. word reading, and word repetition), and a single scan for each of the control conditions (i.e. 'see and say' and 'hear and say'). The order of the six scans was determined by a Latin Square Design such that each condition occurred in each serial position equally often. In every one of the four conditions, the presentation rate (interval from onset of stimulus N to onset of stimulus N +1) averaged 1500 ms but varied randomly between 1200 ms and 1800 ms. The purpose of this procedure is related to the nature of the control conditions, where subjects were saying the same word to every stimulus. The slight variation in inter-stimulus interval was designed to ensure that subjects would say 'crime' in response to the arrival of the stimulus rather than establishing a rhythm of uttering the response at regular intervals. Since the inter-stimulus intervals were sampled with equal probability from the range of 1200 ms to 1800 ms, the average interval was 1500 ms, i.e. 40 items (spoken words, written words, auditory backwards words or visual non-letter strings) per minute. Visual stimuli were presented on the VDU of an Apple Mac Plus computer suspended over the PETscanning couch. Presentation was controlled by the PsychLab program (Bub and Gum, 1988), with each word or control item displayed for a fixed duration of 1000 ms followed by a blank screen varying in duration from 200 ms to 800 ms. Words were in lower-case print. Stimuli for the 'see and say' condition were generated by substituting letters in the real words with a false font of 26 non-letters specially created for this purpose; each ascender, descender and ordinary-sized letter was substituted by a false letter of the same overall shape. This procedure guaranteed that the stimuli presented in false fonts matched those in the real word sets in length, overall shape and the relative frequency of occurrence of various characters in the set. In the 'hear and say' and 'word repetition' conditions, the word lists were tape-recorded by a male speaker with the same presentation rate as in the visual conditions. The stimuli for 'hear and say' were generated simply by re-recording the word tape-recordings in reverse, which ensured that presentation rates, stimulus durations and frequency spectra were matched for the real-word and control conditions. Stimuli were presented binaurally through earphones at approximately 80 dB. Other than listening to ensure that subjects were repeating or reading or saying 'crime' aloud appropriately in each condition, the experimenter did not record subjects' responses, as the subjects all had normal speech and reading abilities and the items in the two word conditions were all familiar words. There were two word sets, matched for word frequency (from Kucera and Francis, 1967), number of letters, number of syllables and rated imageability (from the MRC Psycholinguistic Database; Coltheart, 1981). Of the words used in the study 82% were regular in their spelling-to-sound correspondences. Use of the different lists was counterbalanced across subjects and conditions. Subjects There were 12 adult subjects (seven male and five female), aged 18-70 yrs. All subjects were strongly right-handed as assessed by the Edinburgh handedness inventory (Oldfield. 1971). All had English as their first language, no history of neurological disorder and good literacy levels. All subjects gave their informed consent to the procedure.


METHODS


1775

RESULTS

Sensory processing of complex auditory and visual stimuli When rCBF in the 'see and say' condition is compared with that found in 'hear and say', there is highly significant activation, with z-scores of more than eight, over several planes from 8 mm below to 20 mm above the AC-PC line in both cerebral hemispheres extending over the whole of the extrastriate cortex; there is significant, but less marked activation in striate cortex in both hemispheres {see Fig. 1A). Clearly visual presentation of letter-like strings provokes extensive visual processing bilaterally in the striate and extrastriate cortex. The comparison of 'hear and say' with 'see and say' shows a highly significant increase in the auditory condition of blood flow involving most of the superior temporal gyrus bilaterally; this includes both the primary auditory cortex (Heschl's gyrus) and the auditory association cortex {see Fig. 1B). The peak significance is located at Talairach and Tournoux (1988) x,y,z coordinates 50,-30,8 mm (peak z = 7.78, P < 0.00001) in the left hemisphere, and at 52,-18,8 mm (z = 8.07, P < 0.00001) in the right hemisphere. Peak activation therefore occurs in the auditory association cortex slightly posterior to the primary auditory cortex. Visual word processing When 'word reading' is compared with 'see and say', there is a small unilateral area of significant increase in cerebral blood flow in the left posterior middle temporal gyrus (peakz = 3.18, P < 0.001, at - 5 0 , - 4 8 , + 8 mm). In a comparison between two similar


Procedure Scans were obtained with a PET scanner (CTI model 931-08/12 Knoxville, USA) whose physical characteristics have been described (Spinks et al., 1988). Following reconstruction, the images had a transaxial resolution of 8.5 mm. The reconstructed images contained 128x 128 volume elements (voxels), each corresponding to 2.05x2.05x2.25 mm, following bi-linear interpolation to 43 slices. A measured attenuation correction was used. Subjects inhaled C15O2 at a concentration of 6 m Bq/ml and a flow rate of 500 ml/min through a standard oxygen face mask for a period of 2 min. parametric rCBF images were derived from dynamic PET scans collected for a period of 3.5 min starting 0.5 min prior to C'5O2 delivery, according to a protocol described by Lammertsma et al. (1990). Images were analysed using statistical parametric mapping (Friston and Frackowiak, 1991). Calculations and image matrix manipulations were performed by statistical parametric mapping (SPM; MRC Cyclotron Unit, London, UK) in PRO MATLAB (Mathworks Inc., New York, USA). The intercommissural line was identified directly from the primary (PET) image and the volume transformed into a standard stereotactic space (Friston et al., 1989, 1991a). In this space 1 voxel represents 2 x 2 x 4 mm in the atlas of Talairach and Tournoux (1988). Stereotactic normalization allows voxel by voxel pooling of data and correlation of function with anatomy. In order to account for variability in functional anatomy each image was smoothed with a Gaussian filter 10 pixels (20 mm) wide. Neurophysiological activation of rCBF was assessed with the appropriate linear contrast (weighting of the six condition means) using the t statistic following ANCOVA with whole brain activity as covariate (Friston et al., 1990). This analysis was performed for all voxels in parallel and the resulting set of/ values constitute the t statistical parametric map [SPM(0] for each comparison. The subsets of voxels exceeding a threshold of P = 0.001 were displayed as volume images in three orthogonal projections. This threshold was chosen as empirical studies have shown it to protect from false positives (Bailey et al., 1991); while the significance level adopted may appear to be conservative, it is chosen to allow for the multiple comparisons involved in the analysis. Indeed this significance level is close to that obtained with a Bonferroni corrected threshold of P = 0.05.

1776

D. HOWARD ANfl O T H E R S

conditions, Petersen et al. (1990) located the major area activated by words but not false fonts in the left medial extrastriate cortex. In our study, although examination of the blood flow changes suggests that there may have been a small trend towards a change in left medial extrastriate activation, this did not approach statistical significance. There is also a small area of activation in the left anterior cingulate cortex (peak z = 3.46, P < 0.001, at - 1 6 , - 4 , + 3 6 mm; see Fig. lc). In view of recent demonstrations of important decreases in rCBF during activation tasks (e.g. Friston et al., 1991c; Frithetal., 1991), we tested for statistically significant decreases in rCBF in 'word reading' when compared with 'see and say'. No significant changes were found in this analysis in the left hemisphere. There were, however, large areas of significant decrease in blood flow in the right hemisphere. One area of decrease involved the inferior and middle frontal gyri (peak at 44,12,36 mm, z = 3.95, P < 0.001); there was an area of decrease near the junction between the middle occipital gyrus and the temporal lobe (peak at 40,-74,4 mm, z = 3.23, P < 0.001), another in the extrastriate cortex (peak at 22,-68,36 mm, z = 3.69, P < 0.001) another in the supramarginal gyrus (peak at 48,—46,36 mm, z = 3.78, P < 0.001) and another in the medial frontal cortex (peak at 10,38,36 mm, z = 3.75, P < 0.001). Auditory word processing When 'word repetition' is compared with 'hear and say', there is a unilateral increase in blood flow for the word condition in the left superior and middle temporal gyri (see Fig. ID). The peak change is at - 4 8 , - 3 8 , - 4 mm (z = 4.22, P < 0.00001) at the junction between the superior and middle temporal gyri immediately beneath the primary


Fie. 1. Statistical parametric maps of significant changes in rCBF shown in a horizontal section through the brain 4 mm above the AC-PC line for each of the four comparisons. To facilitate comparison of the regions involved in the different tasks, this figure displays regions where the change is significant at P < 0.01 or better. The greatest changes in cerebral blood flow are shown in white; in each section the areas which have a significant increase in activation are those whose colour is further towards white than the small square shown at the bottom left-hand corner of the section. A, 'see and say' compared with 'hear and say'; significant activation can be seen centred bilaterally in the extrastriate cortex, but also including the striate cortex, B, 'hear and say' compared with 'see and say'; activation can be seen bilaterally involving the primary auditory cortex and the middle part of the superior temporal gyrus. c, 'word reading' compared with 'see and say'; there is significant activation in the posterior part of the left middle temporal gyrus, and a small focus of activation in the right anterior cingulate. D, 'word repetition' compared with 'hear and say'; there is significant activation centred in the middle part of the left superior temporal gyrus.


1777

auditory cortex in Heschl's gyrus, i.e. close to but slightly anterior to the classical location of Wernicke's area. There is a second, much smaller, although significant, change in blood flow at 10,-80,12 mm (peak z = 3.19, P < 0.001) at the junction between the striate cortex and the cuneus in the right hemisphere. There were no significant decreases in rCBF in 'word repetition' compared with 'hear and say' in the left hemisphere. There were, however, decreases in the right hemisphere involving the middle frontal gyrus (peak at 38,48,4 mm, z = 3.76, P < 0.001), the superior and middle temporal gyri (peak at 52,-56,16 mm, z = 3.44, P < 0.001) and the inferior part of the parietal lobe (peak at 26, -52,40 mm, z = 3.13, P < 0.001).

The results, and issues raised by them, will be discussed in the following order: (i) principle findings regarding the localization of areas for spoken and written word recognition; (ii) additional regions of significant activation observed in this study; (iii) separation of regions for auditory/visual word recognition from those involved in word comprehension or production; (iv) the question of whether written word processing activates representations of spoken words. The results from the auditory conditions of this PET study agree rather precisely with the tenets of classical neuropsychology. Processing of complex auditory stimuli yielded bilateral activation of extensive areas of both the primary and secondary auditory cortex. The condition where the complex auditory stimuli also happened to be words produced a significant increase in blood flow localized to Wernicke's area in the left superior temporal gyrus. The activation extended to include much of the middle part of the left middle temporal gyrus but, as noted by Bogen and Bogen (1976), many classical accounts of the localization of Wernicke's area incorporate much of the middle temporal gyrus as well as the posterior superior temporal gyrus. The results from the visual conditions also correspond reasonably well, though perhaps not quite so precisely, to the predictions of classical neuropsychology. There is, of course, no surprise in the results from the visual baseline task: seeing false fonts produced extensive bilateral activation in both the striate and extrastriate cortex. Furthermore, this outcome, which presumably reflects visual processing of complex stimuli, confirms results previously published by Petersen etal. (1990). When the complex visual stimuli were words, peak activation was observed in the left posterior middle temporal gyrus; this is on the margin of, but not exactly coincident with, the classical localization of visual word forms, namely the angular gyrus (Dejerine, 1891; Nielson and Raney, 1938; Nielson, 1939). We would, however, argue that there is support from lesion data for locating visual word forms in the left temporal lobe. The acquired disorder of reading known as 'surface dyslexia' is usually interpreted as a deficit in the lexicon for written word recognition (Marshall and Newcombe, 1973; Patterson et al., 1985; Behrmann and Bub, 1992); and patients with surface dyslexia typically have lesions or atrophy in the left-temporal areas (Vanier and Caplan, 1985; Hodges et al., 1992). The proposal of a visual word-form system in the left temporal cortex differs markedly, however, from that suggested by a previous PET study (Petersen et al., 1990). In their comparison between viewing real words and false fonts, Petersen etal. (1990) observed


DISCUSSION

1778



a significant increase in rCBF centred in the left medial extrastriate cortex. In the present study, this location was the site of a weak trend towards an increase in rCBF which did not approach statistical significance. A less stringent criterion for significance would of course boost this (and other) trends; but (i) it remains the case that, in this study, the left middle temporal gyrus revealed the greatest activation when subjects were reading words, and (ii) in the earlier PET study, there was also some heterogeneity of activation associated with processing of written words, possibly including the left-temporal region identified here (see Petersen et al., 1990; fig. 2B). The medial extrastriate location is the only one discussed by Petersen et al. (1990). There are several procedural differences between the two PET studies and, although none of these appears substantial or critical, each or indeed all might have contributed to the discrepant outcomes, (i) Petersen et al. (1990) presented words in upper case at a rate of 1 item per second with an exposure duration of 150 ms; here, lower-case words were presented at an average (but slightly varying) rate of 1 item per 1.5 s with an exposure duration of 1000 ms. (ii) The stimulus words in the Petersen et al. (1990) experiment were all names of animals and objects, thus highly concrete; in the current study, words were all nouns but varied widely in concreteness. (iii) In the earlier study, subjects viewed stimuli silently; in this experiment, each word was read aloud. Existing results do not suggest which, if any, of these procedural differences may be responsible for the discrepant findings. We are conducting further studies to examine the impact of some of these variables on rCBF during written word recognition. In both word reading and word repetition there were extensive areas of significant decrease in rCBF in the right hemisphere relative to the control conditions. This finding complements the significant increases in the two tasks localized in the left hemisphere, demonstrating that, in linguistic tasks, the left hemisphere increases in rCBF are accompanied by extensive decreases in the right hemisphere. In each of the two modality-specific word conditions (as compared with its control condition), there was one additional significant region of increased blood flow. For the auditory case, this was a location in the right occipital cortex, at the junction between the striate cortex and the cuneus. We had no grounds for predicting this area of activation and have no useful interpretation of it. For the visual case, a significant increase in rCBF was observed in the left anterior cingulate cortex. Although we did not specifically predict this finding either, it does have a plausible interpretation. Various authors have suggested that the anterior cingulate may be involved in attention and selection for action (e.g. Petersen et al., 1988); it is therefore possible that the activation in this area, when subjects were reading, reflects a process of word selection for output. Although it might seem that deliberate word selection should be equally implicated in word reading and repetition, cognitive psychological experiments suggest that word reading is an attentiondemanding task (Herdman and Dobbs, 1989) but that word repetition is a rather automatic task requiring little attention (McLeod and Posner, 1984). Given that the principal regions activated by the two word conditions were both located in the left temporal lobe, how confident can one be that these are distinct foci? The peak of the area activated for reading words was 12 mm higher than, and 10 mm posterior to, the peak of activation for repeating words. The absolute distance between the two peaks was 16 mm, substantially greater than the resolution of statistical parametric mapping which is approximately 11 mm. (This resolution is estimated empirically from


1779

ACKNOWLEDGEMENTS We are grateful to Dr Cathy Price and Dr Stuart Ramsey for their help in the analysis of the data used in this research. This work was partly supported by a grant from the McDonnell-Pew Foundation. Dr David Howard was supported


the variance of the field derivative; see Friston et al., 1991b, for a formal description.) In other words, a plausible account of the observed results is that these represent two distinct foci unique to word reading and word repetition. As noted earlier in the list of putative psychological processes recruited by the various experimental conditions, the two word conditions differed from their control conditions not only in activating representations for familiar, modality-specific word forms, but also in two other ways: they required the retrieval of, or computation of, a different pronunciation for every stimulus; and, whilst not explicitly requiring it, they might automatically have engaged the retrieval of, or computation of, a word's meaning. However, no single local area of cortex showed significant increases in rCBF in both the word reading and repetition conditions. With regard to phonological representations for output, this may mean that the control response (pronunciation of the same word for each stimulus) was in fact an adequate control for the word conditions. With regard to word meanings, there are two possible explanations for the failure to observe a difference between word and control conditions that was common to both modalities. One is simply that (passive, automatic) evocation of semantic representations, although taking place in localized regions of the cortex, is not a process which is reflected in rCBF changes. More plausibly, we suggest (along with many other researchers, e.g. Wernicke, 1874; Allport, 1985) that representations of word meanings are not well localized, but rather depend on activation of semantic features, and connections between them, distributed over large regions of the cerebral cortex. Finally, there is the question of whether readers translate written words into some form of speech-based code in order to recognize and comprehend them. This is a complex issue, debated at great length in the literature on cognitive psychological studies of reading (see, for example, Coltheart, 1980; Van Orden, 1987), and a full discussion of it is beyond the scope of this paper. The finding of a left-temporal locus for the writtenword condition does not rule out the possibility that some form of phonological code is crucially involved in comprehending written words. Our study does, however, address one specific version of this idea, namely the hypothesis, proposed by Geschwind (1979), that the processing of written words activates representations in the auditory word-form system. Since results from the auditory-word condition implicate Wernicke's area as the critical location for recognition of spoken words, the absence of activation in Wernicke's area for the written-word condition argues against Geschwind's proposal. In conclusion, this PET study provides specific hypotheses regarding cortical localization of the processes involved in auditory and visual lexical access in normal subjects. These hypotheses are that access to the lexicon for spoken words is a function of Wernicke's area in the left superior temporal gyrus and the middle part of the left temporal gyrus, and access to the lexicon for written words depends on an area about 16 mm away in the left posterior middle temporal gyrus. Both processes are apparently exclusively carried out in the left cerebral hemisphere. These findings accord rather well with lesion data from patients with specific disorders of spoken and written word recognition.

1780


by a Senior Research Fellowship from The Medical Research Council, and Dr Cornelius Weiller was Feodor-Lynen Fellow of the Alexander von Humboldt Stiftung, Bonn-Bad Godesborg, Germany. REFERENCES ALLPORT DA (1985) Distributed memory, modular subsystems and dysphasia. In: Current Perspectives in Dysphasia. Edited by S. Newman and R. Epstein. Edinburgh: Churchill Livingstone, pp. 32—60. BAILEY DL, JONES T, FRISTON KJ, COLEBATCH JG, FRACKOWIAK RSJ (1991) Physical validation of statistical

parametric mapping. Journal of Cerebral Blood Flow and Metabolism, 11, 2, S150. BASSO A, LECOURS AR, MORASCHINI S, VANIER M (1985) Anatomoclinical correlations of the aphasias

as defined through computerized tomography: exceptions. Brain and Language, 26, 201-229. BEHRMANN M, BUB D (1992) Surface dyslexia and dysgraphia: dual routes, single lexicon. Cognitive

FRISTON KJ, PASSINGHAM RE, NUTT JG, HEATHERJD, SAWLE GV, FRACKOWIAK RSJ (1989) Localisation

in PET images: direct fitting of the intercommissural (AC-PC) line. Journal of Cerebral Blood Flow

and Metabolism, 9, 690-695. FRISTON KJ, FRITH CD, LIDDLE PF, DOLAN RJ, LAMMERTSMA AA, FRACKOWIAK RSJ (1990) The

relationship between global and local changes in PET scans. Journal of Cerebral Blood Flow and Metabolism, 10, 458-466. FRISTON KJ, FRITH CD, LIDDLE PF, FRACKOWIAK RSJ (1991a) Plastic transformation of PET images.

Journal of Computer Assisted Tomography, 15, 634 — 639. FRISTON KJ, FRITH CD, LIDDLE PF, FRACKOWIAK RSJ (1991 A>) Comparing functional (PET) images: the

assessment of significant change. Journal of Cerebral Blood Flow and Metabolism, 11, 690 — 699. FRISTON KJ, FRITH CD, LIDDLE PF, FRACKOWIAK RSJ (1991c) Investigating a network model of word

generation with positron emission tomography. Proceedings of the Royal Society of London, B, 244, 101-106. FRITH CD, FRISTON KJ, LIDDLE PF, FRACKOWIAK RSJ (1991) A PET study of word finding.

Neuropsychologia, 29, 1137-1148. GESCHWIND N (1965) Disconnexion syndromes in animals and man. Part 1. Brain, 88, 237-294. GESCHWIND N (1969) Problems in the anatomical understanding of the aphasias. In: Contributions to clinical neuropsychology. Edited by A. L. Benton. Chicago: Aldine, pp. 107—128.


Neuropsychology, 9, 209-251. BENSON DF, GESCHWIND N (1969) The alexias. In: Handbook of Clinical Neurology, Volume 4. Edited by P. J. Vinken and G. W. Bruyn. Amsterdam: North-Holland, pp. 112-140. BOGEN JE, BOGEN GM (1976) Wernicke's region—where is it? Annals of the New York Academy of Sciences, 280, 834-843. BUB D, GUM T (1988) PsychLab Experimental Software. Montreal: McGill University. BUB DN, BLACK S, HOWELL J (1989) Word recognition and orthographic context effects in a letter-byletter reader. Brain and Language, 36, 357-376. COLTHEART M (1980) Reading, phonological recoding, and deep dyslexia. In: Deep Dyslexia. Edited by M. Coltheart, K. E. Patterson and J. C. Marshall. London: Routledge and Kegan Paul, pp. 197-226. COLTHEART M (1981) The MRC psycholinguistic database. Quarterly Journal of Experimental Psychology, 33A, 497-505. COLTHEART M, PATTERSON KE, MARSHALL JC (1980) Deep Dyslexia. Edited by M. Coltheart, K. E. Patterson and J. C. Marshall. London: Routledge and Kegan Paul. COSLETT HB, SAFFRAN EM (1989) Evidence for preserved reading in 'pure alexia'. Brain, 112, 327-359. DE BLESER R, LUZZATTI C (1989) Models of reading and writing and their disorders in classical German aphasiology. Cognitive Neuropsychology, 6, 501—513. DEJERINE J (1891) Sur un cas de ce'cite' verbale avec agraphie, suivi d'autopsie. Comptes Rendus Hebdomadaires des Seances et Memoires de la Societe de Biologie, Ninth series, 3, 197 — 201. DEJERINE J (1892) Contribution a I'etude anatomo-pathologique et clinique des differentes varie'tes de cecite verbale. Comptes Rendus Hebdomadaires des Seances et Memoires de la Societe de Biologie, Ninth series, 4, 6 1 - 9 0 . FRISTON KJ, FRACKOWIAK RSJ (1991) Imaging functional anatomy. In: Brain Work and Mental Activity: Quantitative Studies with Radioactive Tracers. Edited by N. A. Lassen, D. H. Ingvar, M. E. Raichle and L. Friberg. Copenhagen: Munksgaard, pp. 267-277.


1781

LAMMERTSMA AA, CUNNINGHAM VJ, DEIBER MP, HEATHER JD, BLOOMFIELD BM, NUTT J et al. (1990)

Combination of dynamic and integral methods for generating reproducible functional CBF images. Journal of Cerebral Blood Flow and Metabolism, 10, 675 — 686. LUPKER SJ (1985) Relatedness effects in word and picture naming: parallels, differences, and structural implications. In: Progress in the Psychology of Language, Volume 1. Edited by A. W. Ellis. London: Lawrence Erlbaum, pp. 109—142. LURIA AR (1970) Traumatic Aphasia: Its Syndrome, Psychology and Treatment. Translated by D. Bowden. The Hague: Mouton. MACLEOD CM (1991) Half a century of research on the Stroop effect: an integrative review. Psychological Bulletin, 109, 163-203. MCLEOD P, POSNER MI (1984) Privileged loops from percept to act. In: Attention and Performance X: Control of Language Processes. Edited by H. Bouma and D. G. Bouwhuis. London: Lawrence Erlbaum, pp. 55-66. MARIE P, FOIX C (1917) Les aphasies de guerre. Revue Neurologique, i, 53 — 87. MARSHALL JC, NEWCOMBE F (1973) Patterns of paralexia: a psycholinguistic approach. Journal of Psycholinguistic Research, 2, 175 — 199. NAESER MA, HAYWARD RW (1978) Lesion localization in aphasia with cranial computed tomography and the Boston Diagnostic Aphasia Exam. Neurology, New York, 28, 545—551. NEELY JH (1991) Semantic priming effects in visual word recognition: a selective review of current findings and theories. In: Basic Processes in Reading: Visual Word Recognition. Edited by D. Besner and G. W. Humphreys. Hillsdale, NJ: Lawrence Erlbaum, pp. 264—336. NIELSON JM (1936) Agnosia, Apraxia, Aphasia: Their Value in Cerebral Localization. New York: P. B. Hoeber. NIELSON JM (1939) The unsolved problems in aphasia: II. Alexia resulting from a temporal lesion. Bulletin of the Los Angeles Neurological Society, 4, 168—183. NIELSON JM, RANEY RB (1938) Symptoms following surgical removal of the major (left) angular gyrus. Bulletin of the Los Angeles Neurological Society, 3, 42—46. OLDFIELD RC (1971) The assessment and analysis of handedness: the Edinburgh Inventory. Neuropsychologia, 9, 97—113. PATTERSON KE, MARSHALL JC, COLTHEART M (1985) Surface Dyslexia: Neuropsychological and Cognitive Studies of Phonological Reading. Edited by K. E. Patterson, J. C. Marshall and M. Coltheart. London: Lawrence Erlbaum. PETERSEN SE, FOX PT, POSNER MI, MINTUN M, RAICHLE ME (1988) Positron emission tomographic

studies of the cortical anatomy of single-word processing. Nature, London, 331, 585-589. PETERSEN SE, Fox PT, POSNER MI, MINTUN M, RAICHLE ME (1989) Positron emission tomographic

studies of the processing of single words. Journal of Cognitive Neuroscience, 1, 153 — 170. PETERSEN SE, Fox PT, SNYDER AZ, RAICHLE ME (1990) Activation of extrastriate and frontal cortical areas by visual words and word-like stimuli. Science, 249, 1041 — 1044. RUBENSTEIN H, LEWIS SS, RUBENSTEIN MA (1971) Evidence for phonemic recoding in visual word recognition. Journal of Verbal Learning and Verbal Behavior, 10, 645—657.


GESCHWIND N (1979) Specializations of the human brain. In: The Brain. San Francisco: W. H. Freeman. GREENBLATT SH (1973) Alexia without agraphia or hemianopsia: anatomical analysis of an autopsied case. Brain, 96, 307-316. GREENBLATT SH (1983) Localization of lesions in alexia. In: Localization in Neuropsychology. Edited by A. Kertesz. New York: Academic Press, pp. 323—356. HERDMAN CM, DOBBS AR (1989) Attentional demands of visual word recognition. Journal of Experimental Psychology: Human Perception and Performance, 15, 124—132. HODGES JR, PATTERSON KE, OXBURY S, FUNNELL E (1992) Semantic dementia: fluent aphasia with temporal lobe atrophy. Brain. 115, 1783-1806. HOWARD D (1991) Letter-by-letter readers: evidence for parallel processing. In: Basic Processes in Reading: Visual Word Recognition. Edited by D. Besner and G. W. Humphreys. Hillsdale, NJ: Lawrence Erlbaum, pp. 34—76. KuCERA H, FRANCIS WN (1967) Computational Analysis of Present-Day American English. Providence, RI: Brown University Press.

1782


SCHACTER DL, RAPSCAK SZ, RUBENS AB, THARAN M, LAGUNA J (1990) Priming effects in a letter-by-

letter reader depend upon access to the word form system. Neuropsychologia, 28, 1079-1094. SELDON HL (1985) The anatomy of speech perception: human auditory cortex. In: Cerebral Cortex, Volume 4. Edited by A. Peters and E. G. Jones. New York: Plenum Press, pp. 273-327. SMITH LC (1984) Semantic satiation affects category membership decision time but not lexical priming. Memory and Cognition, 12, 483—488. SPINKS TJ, JONES T, GILARDI MC, HEATHER JD (1988) Physical performance of the latest generation

(Received February 11, 1992. Revised May 13, 1992. Accepted June 14, 1992)


of commercial positron scanner. IEEE Transactions on Nuclear Science, 35, 721 —725. TALAIRACH J, TOURNOUX P (1988) Co-Planar Stereotaxic Atlas of the Human Brain. Stuttgart: Thieme. VANIER M, CAPLAN D (1985) CT scan correlates of surface dyslexia. In: Surface Dyslexia: Neuropsychological and Cognitive Studies of Phonological Reading. Edited by K. E. Patterson, J. C. Marshall and M. Coltheart. London: Lawrence Erlbaum, pp. 511—525. VAN ORDEN GC (1987) A ROWS is a ROSE: spelling, sound, and meaning. Memory and Cognition, 15, 181-198. WERNICKE C (1874) Der aphasischer Symptomenkomplex: eine psychologische Studie auf anatomischer Basis. Breslau: Cohnand Weigert. Translated by G. H. Eggert (1977) in Wernicke's Works on Aphasia: A Sourcebook and Review. The Hague: Mouton, pp. 91 — 145. WERNICKE C (1906) Der aphasischer Symptomenkomplex. Die deutsche Klinik am Eingdnge des 20 Jahrhunderts, 6,487. Translated by G. H. Eggert (1977) in Wernicke's Works on Aphasia: A Sourcebook and Review. The Hague: Mouton, pp. 219-283.

Cortical and subcortical localization of response to pain in man using positron emission tomography.

Language localization with activation positron emission tomography scanning.

Positron emission tomography.

Temporoparietal cortex in aphasia. Evidence from positron emission tomography.

computed tomography: evidence from a large series of studies.

Positron emission tomography in the differential diagnosis of parkinsonism.

The precision of positron emission tomography: theory and measurement.

magnetic resonance imaging of the breast.

Lung cancer and schwannoma--the pitfalls of positron emission tomography.

Positron-emission computed tomography of the pancreas: a preliminary study.

The applications of positron emission tomography to oncology.

Positron emission tomography imaging of atherosclerosis.

Positron emission tomography imaging of coronary atherosclerosis.

The methodology of TSPO imaging with positron emission tomography.

Positron emission tomography applied in the study of pituitary adenomas.

positron emission tomography in neuroscience research.

Positron Emission Tomography (PET) in Oncology.

Fluorodeoxyglucose positron emission tomography in eosinophilic fasciitis.

Positron emission tomography radiotracers for imaging hypoxia.

Amyloid positron emission tomography and cognitive reserve.

Positron emission tomography in Shy-Drager syndrome.

Quantitative cardiac positron emission tomography: the time is coming!

Positron Emission Tomography in Breast Cancer.

Positron emission tomography in spinal infections.