## Introduction

Developmental dyslexia is a neurocognitive disorder with a strong genetic component (see review by DeFries et al., 1997), and should not be confused with poor reading achievement which can occur for a variety of different reasons (Rutter and Yule, 1975). Definitions of dyslexia suffer greatly from problems inherent in behavioural observations, which change with age and general ability (Frith, 1997). The attempt to make definitions dependent on arbitrarily defined discrepancies between reading skill and general skills has proved unsatisfactory (Siegel, 1992; Lyon, 1995). While the hallmark of dyslexia is the pronounced difficulty in the acquisition of literacy, many dyslexics eventually achieve satisfactory scores on standard reading tests. Likewise, dyslexic readers in different languages experience quite different problems in reading and writing, since the orthography of a specific language has a profound impact on the manifestation of dyslexia in reading tests (Landerl et al., 1996). Subtle cognitive impairments in language processing outside reading and writing are apparent before school age (Scarborough, 1990), as well as in adulthood (e.g. Bruck, 1992; Elbro et al., 1994; Snowling et al., 1997). Behavioural criteria are therefore an inadequate basis for a unified definition of dyslexia. It is also necessary to identify the nature of the putative cognitive impairments and to investigate the physiological anomalies in the underlying neural systems.

Clinical signs of dyslexia that appear to be independent of age and ability include impairments in speech processing, in particular on tasks requiring manipulation and memory of segmental speech sounds (see Snowling and Nation, 1997). Similarities to specific language impairments indicate left hemisphere abnormalities. Difficulties in auditory temporal processing and temporal order judgement (Tallal, 1980; Tallal et al., 1993) and specific deficits in neuromotor tests (Kinsbourne et al., 1991) also point to left hemisphere impairments. Other signs of dyslexia concern low-level difficulties in the visual processing of fast moving stimuli (see Lovegrove, 1993). These difficulties implicate the magnocellular system bilaterally (Livingstone et al., 1991). Unstable binocular control has also been found to relate to reading difficulties (Cornelissen et al., 1992), and Slaghuis and colleagues suggested that visual and language processing deficits co-occur in dyslexia (Slaghuis et al., 1993).

A number of brain imaging studies have targeted some of the underlying processes that are assumed to be compromised in dyslexia, and have attempted to locate brain regions with abnormal activation in dyslexics. Flowers and colleagues were the first to test the hypothesis of abnormal left hemisphere activation in dyslexics during a language task (Flowers et al., 1991). They found that degree of reading impairment correlated with reduced activation in posterior temporoparietal cortex during an auditory rhyming task. Rumsey and colleagues (Rumsey et al., 1992), Paulesu and colleagues (Paulesu et al., 1996) and Shaywitz and colleagues (Shaywitz et al., 1998) also scanned dyslexics during rhyme judgements. The studies all show abnormal activation patterns in the left perisylvian and extrasylvian temporal cortices. Eden and colleagues targeted magnocellular system weakness, and scanned volunteers while they watched moving dot patterns (Eden et al., 1996). Demb and colleagues found a correlation between brain activation of components of this system and degree of reading difficulty in adult dyslexics (Demb et al., 1998). Lehmkuhle and colleagues showed abnormal evoked responses to low-spatial frequency targets in dyslexic children (Lehmkuhle et al., 1993). In all these studies dyslexics were found to differ at the physiological level even when task demands were simple. This strongly confirms the notion of dyslexia as a neurocognitive developmental disorder that must be distinguished from the variety of reading problems found in school-age populations.

The phonological processing hypothesis of dyslexia rests on robust behavioural evidence and claims to provide a direct link between the putative cognitive deficit and the specific problems of dyslexics in the acquisition of literacy (see review by Snowling, 1991). Literacy acquisition, especially in alphabetic writing systems, critically involves segmentation of speech into phonemes (Brady and Shankweiler, 1991). Impairment in this process leads to reading failure (Bryant and Bradley, 1978). Phonemes have to be related to graphemes, and grapheme–phoneme conversions have to be related to whole word sound. This involves both piecemeal or assembled phonology, and whole-word or addressed phonology (see Morton, 1980). In English, the mapping between these two processes is complex, because whole word sound cannot be reliably derived from piecemeal assembled sounds. The phonological deficit hypothesis of dyslexia proposes that all of these steps are problematic for dyslexics and cause reading skills to develop atypically. The phonological deficit hypothesis is amenable to be studied at the physiological level, since the neurological substrate of normal phonological processing has been investigated in a number of neuro-imaging studies. These studies have converged, reliably demonstrating that structures in the perisylvian and extrasylvian regions of the dominant language hemisphere are dedicated to phonological processes (Démonet et al., 1994). Tasks relying on segmented and assembled phonology elicit activations in the supramarginal gyrus/superior temporal gyrus (BA 40/22: Paulesu et al., 1993; Price et al., 1997b) and premotor cortex/Broca's area (BA 6/44: Paulesu et al., 1993; Démonet et al., 1994). In contrast, tasks that involve addressed phonology, picture, colour or letter-naming, for example, activate the left frontal operculum and the left posterior inferior temporal cortex (BA 37: Bookheimer et al., 1995; Price and Friston, 1997).

Abnormalities of activation in perisylvian and extrasylvian regions of the left hemisphere have been associated with impaired phonological processing in developmental dyslexia (Rumsey et al., 1992; Paulesu et al., 1996; Shaywitz et al., 1998). However, functional imaging methods for studying directly the processes involved in word recognition in dyslexic readers have not been used until recently. Salmelin and colleagues measured electrical activity (MEG) of the brains of Finnish-speaking developmental dyslexics and control readers while they silently read words (concrete and abstract) and pseudowords (Salmelin et al., 1996). In contrast to control readers, dyslexics failed to activate the left inferior temporo-occipital region during the first 200 ms of viewing the words. Even between 400–700 ms post-stimulus onset, the dyslexics' activation of the left temporal lobe was significantly less than that of the controls; however, the dyslexics activated the left inferior frontal lobe within this time window while the controls did not. Pseudowords elicited greater activation than words in the majority of the controls (in the left middle temporal lobe and central sulcus) and in half of the dyslexics (in the temporoparieto-occipital region). However, as no measure of reading accuracy was taken and participants gave no overt response to the stimuli, it is possible that some performance differences existed in the dyslexics' reading of the words and pseudowords compared with controls. It cannot be ruled out that the dyslexics' failure to activate the temporal language areas reflects a failure to decode the stimuli with sufficient accuracy to activate the semantic and naming areas of the brain.

In Study 1 we scanned compensated adult dyslexics while they read words and word-like pseudowords aloud. We presented one stimulus every 3 s, thus equating the rate of stimulus presentation for the two groups, and at the same time providing a sufficiently slow pace to allow accurate decoding by all individuals. Furthermore, we presented phonologically and orthographically simple stimuli, and produced very word-like pseudowords by only changing the internal consonants of the words (cassot for carrot). This was done to allow high reading accuracy in the dyslexic readers and thus to ensure that any differences in brain activation between dyslexics and controls would not merely reflect differences in performance levels. In line with the previous studies, we expected that the posterior temporal parietal regions of the left hemisphere would be less highly activated by dyslexic readers.

## Study 1: explicit reading of words and pseudowords

### Methods

#### Psychometric tests

Participants received a battery of standardized tests and experimental measures designed to assess literacy and literacy related skills. Verbal, performance and full-scale IQ were determined using short forms of the Wechsler Adult Intelligence Scale—Revised (WAIS-R). Reading and spelling ability were tested using the National Adult Reading Test (NART; Nelson, 1983) and the Wide Range Achievement Test—Revised (WRAT-R; Jastak and Wilkinson, 1984).

#### Word recognition tests

Specially constructed lists of words and pseudowords were presented on a laptop computer, and naming latency and accuracy were recorded via a voice-key to assess word recognition and grapheme–phoneme decoding ability. The stimuli consisted of two- and three-syllable words, avoiding irregular spelling patterns. Pseudowords were created from the words by maintaining the word envelope' whilst changing internal consonants. The word satin', for example, became samin'; and lemon became lexon'. More examples are provided in Appendix 1. Participants were asked to name the word as soon as they had recognized it. As soon as a verbal response had been detected, the stimulus disappeared. As a base line measure of response latency, we also used a dot stimulus.

#### Experimental tests

Participants also received a range of tasks thought to tap aspects of phonological processing. These included a Spoonerism task (to assess phonological segmentation and manipulation ability; after Perin, 1983), a measure of phonemic fluency (to assess efficiency of phonemic lexical access), a test of accuracy of immediate recall of either short or long words (to assess auditory verbal short-term memory for words; after Baddeley et al., 1975), the WAIS digit span test (a measure of verbal short-term memory for digits), and a digit-naming task (to measure naming speed; after Spring and Davis, 1988). See Appendix 1 for details.

The scanning period involved three experimental conditions: two activation conditions (reading aloud words and pseudowords) and a passive (eyes closed) rest condition. Eyes closed was selected as a baseline condition against which to compare the activations associated with reading words and pseudowords. The rest condition has been commonly used in our laboratory, so it will be possible to interpret the present data in terms of previous neuroimaging and neurophysiological findings that have attempted to localize the functional components involved in reading. The three experimental conditions were presented in a fixed counter-balanced design (ABCCBAABCCBA); each condition was presented four times. This study followed a factorial design with reading ability (dyslexic versus non-dyslexic) as the between-group factor and with task (reading words, reading pseudowords and rest) as the within-group factor.

The activation conditions required the reading of either familiar words (bi-syllabic, regular, concrete nouns, e.g. valley, body, carrot) or unfamiliar but word-like pseudowords, e.g. vassey, bofy, cappot). As in the pre-test of word-recognition ability, the pseudowords were created from the words by changing the internal consonants. Stimuli used during scanning had not been presented during the pre-test. The administration differed also from the pre-test in that each stimulus was presented for 1 s and the inter-stimulus interval was 2 s. Presentation of stimuli was via the SuperLab program (version 1.68: Cedrus Corporation) run on an Apple Power Macintosh computer. Stimuli were presented in lower-case Times, 48 point font size (black letters on a white background) at the rate of one every 3 s. A block of 22 words or pseudowords was presented per scan in the centre of a computer monitor; this monitor was suspended above the scanning bed at a comfortable distance from the subject, orthogonal to their line of vision. Participants' verbal responses (reading of words and pseudowords) were recorded.

#### PET scanning procedure

Regional cerebral blood flow (rCBF) was measured using a CTI Siemens Ecat HR + PET scanner (CTI Inc., Knoxville, Tenn., USA). Data were acquired with inter-detector collimating septa retracted (Townsend et al., 1991). Twelve consecutive scans were obtained for each subject following a transmission scan, implemented with an exposed 68Ge/68Ga source, to correct for attenuation. Each activation scan involved the intravenous administration of 5 mSv of H215O at the constant rate of 10 ml/min. From the start of each scan background radiation was measured for 30 s, after which the infusion was given. The activation task was started ~30 s later, 10 s prior to the onset of the acquisition period. Acquisition lasted for 90 s. After correcting for background activity, the true counts accumulated during this period were taken as an index of cerebral blood flow (Fox and Mintun, 1989). This protocol conforms with guidelines established by ARSAC UK and was approved by the Medical Ethics Committee of the Institute of Neurology.

#### Data analysis

Head movements which occurred during the course of the PET scan were corrected by realigning the time series with the first scan. The reconstructed transaxial PET images were spatially normalized for brain size and shape (Friston et al., 1995a) so that the spatially normalized images consisted of 2 × 2 × 4 mm voxels. These images were subsequently smoothed in three directions with a Gaussian filter (full-width at half-maximal) of 16 × 16 × 16 mm. This smoothing had the effect of increasing signal-to-noise ratio and allowing for normal between-subject variation in gyral anatomy. Task-related differences in rCBF were examined using statistical parametric mapping (SPM96) software (Friston et al., 1995b) on a SPARC1 workstation (Sun Microsystems Inc., Surrey, UK). Analyses were performed in MATLAB version 4.2c (MathsWorks Inc., Natick, Mass., USA). Data were analysed using subject-specific repeated measures ANCOVA with images scaled to a grand mean of 50. Statistical parametric maps of adjusted mean rCBF were produced for each condition/group along with adjusted error variance data for each. To protect against the reporting of false positive results, activations are only reported where they reach a significance level of P < 0.001 (Bailey et al., 1991).

The linear contrast, identified significant activations for (i) words – rest, (ii) pseudowords – rest, (iii) words – pseudowords and (iv) pseudowords – words (in both the dyslexic group and the normal group), and the interactions between each of these contrasts and subject group. To distinguish increased activation for words from decreased activation for pseudowords (and vice versa), we only report differences between words and pseudowords when one or other or both conditions resulted in significant activation relative to rest at a threshold of P < 0.01 (see Table 6).

### Results

#### Psychometric measures

The dyslexic readers were impaired on the Spoonerism task: they made more errors than the controls [F(1, 10) = 7.2; P = 0.02] and took over twice as long to complete the task [F(1, 10) = 9.4; P = 0.01]. They also took significantly longer than controls to complete the digit-naming task [F(1, 10) = 25.4; P < 0.001]. The difference in phonemic fluency was in the predicted direction, but not significant (P = 0.1). The dyslexic participants also demonstrated impairments in auditory short-term memory for word recall [F(1, 10) = 11.9; P < 0.01] and in digit span [F(1, 10) = 13.0; P < 0.01].

### Discussion

The two groups of readers proved to be well matched in terms of age, educational attainment and intelligence. While possessing reading skills that are average in terms of general population norms, the dyslexics were still significantly impaired relative to their peers on literacy tasks. They were also impaired on tasks tapping phonological ability in auditory presentation. This confirms previous studies indicating a persistent phonological deficit in developmental dyslexia even in highly able and compensated adults. The nature of this deficit can be explored through the functional imaging results.

The functional imaging results replicated some of the findings reported by Salmelin and colleagues (Salmelin et al., 1996) and Rumsey and colleagues (Rumsey et al., 1997). As in these previous studies, similar brain systems were activated for words and pseudowords. This suggests that similar strategies were used for reading words and pseudowords. Both normally skilled and dyslexic readers showed activations in the classical language regions and in regions associated with the sensory aspects of reading: visual activations in the occipital lobe, receptive auditory processing in the superior temporal lobe and articulation in precentral cortices (Price et al., 1994, 1996a; Bookheimer et al., 1995; Rumsey et al., 1997). However, there were also significant group differences. While both samples relied on a common distributed language system for word recognition, dyslexics showed an anomalous activation pattern as regards its various components. Again, as in the two previous studies, they showed less activation in the posterior parts, but more in the anterior parts. Our experimental design was successful in ensuring that the two groups did not differ in decoding accuracy during PET scanning. Thus, unlike the previous studies, differences in activation are less likely to be due to quantitative differences in performance level, but can be safely attributed to qualitative differences in reading processes.

The areas where dyslexics showed reduced activation relative to normals were the left posterior inferior temporal cortex, the left and midline cerebellum, the left thalamus and the medial extrastriate cortex. All but the medial extrastriate cortex have previously been found to be activated for reading, picture naming, letter naming and colour naming (Price and Friston, 1997), suggesting that they encompass part of a modality-independent lexical retrieval system. The medial extrastriate region is activated for reading aloud relative to reading silently (Price et al., 1997a). It is an area that Petersen and colleagues attributed to the word-form system (Petersen et al., 1990), but this conclusion has been questioned by the finding that it also responds to non-orthographic stimuli (Indefrey et al., 1997). One possible hypothesis is that reading aloud relative to reading silently normally alters the way that subjects attend to the visual features in the word (Price et al., 1997). For instance, activation in the medial extrastriate cortex is enhanced when subjects attend to the global structure of a visual stimulus (Fink et al., 1996). The hypothesis that the dyslexics are focussing less on the global properties of the word is consistent with evidence that they were more reliant on sublexical processing. This evidence was indicated by the finding that dyslexics showed increased activation in an area of the premotor cortex (BA 44/6) that has been associated with articulatory coding (Démonet et al., 1992). Hagoort and colleagues have argued recently that assembling articulatory programmes from sublexical segments greatly increases the computational load on articulation relative to familiar articulatory routines (Hagoort et al., 1999). The increased activation for the dyslexics in a region associated with articulation is therefore consistent with increased demands on sublexical processing. If our hypothesis is correct—that dyslexics are increasing the demands on articulatory processes to compensate for reduced efficiency in lexical retrieval—then a clear prediction arises: dyslexics should not show an increase in activation of Broca's area when reading aloud is not required. Our second study tests this prediction. In addition, the second study uses false font stimuli as the control rather than rest, which should reduce differences in visual processing between reading and control conditions.

## Study 2: implicit reading of words and pseudowords

### Participants

The same selection criteria were used as in the previous study. Six dyslexic and six control readers were recruited from the same universities and were again matched for age, educational level and IQ. The mean age of the dyslexic group was 24.5 years (± 6.1), whereas the mean age of the control group was 24.5 years (± 2.9). Participants were tested with the same psychometric test battery as in Study 1, and gave informed consent before testing began.

### Methods

During each PET scan participants performed a feature detection task. This involved detecting the presence or absence of ascenders (graphic features which go above the midline of the word, e.g. b, l, t) within visually presented words, pseudowords and false font strings. The words and pseudowords were a subset of those employed in Study 1; the false fonts were created by substituting letters in the real words with non-letters matched for size and presence or absence of ascenders. Participants were familiarized with the task requirements prior to scanning and a short practice session was provided for each type of stimulus. It was emphasized that this was not a reading task but a feature detection task, and participants were told not to bother reading the stimuli. If participants experienced uncertainty with either the task demands or with the detection of ascenders, the practice trials were repeated.

The three stimulus conditions were presented in a counter-balanced order (ABCCBAABCCBA) but the requirements of the task remained constant across stimuli: participants pressed one key of a response box with their right hand index finger if one or more ascender was present and another key with their right middle finger if no ascenders were present. Rate and mode of stimulus presentation were as in Study 1, with each stimulus displayed for 1 s with a 2 s inter-stimulus interval. Reaction time and accuracy of response were recorded. rCBF was measured and analysed as reported above.

### Results

#### Feature detection task performance during PET activation

Although the mean response latencies for the dyslexics (623 ms) were longer than for the controls (587 ms), this difference was non-significant (P = 0.3). No significant between-group difference was found for accuracy of detection (P = 0.9).

### Discussion

The psychological profiles of the dyslexic and control participants were highly similar to those from Study 1. They confirmed the persistence of reading and spelling difficulties in spite of performance levels that are in the average range for the total population. Thus, these students were compensated in the sense of showing greatly improved performance on literary tasks compared with their earlier difficulties, but they were not cured'. In particular, they showed significant impairments on orally presented phonological processing tasks.

The presence of words and word-like stimuli in the visual field during a feature detection task elicited widespread activation of the classical language areas, including the superior, middle and posterior temporal regions, the superior, medial and inferior frontal regions and the precentral gyri, in both the controls and the dyslexics. These activations indicate that the stimuli were processed even though the participants were not required to read them. The results from the controls replicate the results from the original feature detection study of Price and colleagues (Price et al., 1996b) and are consistent with findings from a study of the passive viewing of real words (Price et al., 1994). While the two groups shared a network of common activation, the normal readers demonstrated greater levels of activation in the temporoparietal regions of the left hemisphere. No differences between subject groups were detected in the cerebellum, thalamus or medial extrastriate cortex, as seen in Study 1.

#### Areas activated in common during implicit and explicit reading

The comparison of the two studies shows that reduced activation relative to controls in the left inferior temporal region (BA 37) and bilateral cerebellum was a consistent finding. The combined analysis also revealed that the left frontal operculum surrounding the most anterior part of the insula was less active in the dyslexic readers in both studies, although this was not detected as significant in the individual analyses reported above. A previous experiment has found that the left frontal operculum and the left posterior inferior temporal cortex are activated together during modality-independent naming, along with activation of the left thalamus, cerebellum and left posterior fusiform gyrus (Price and Friston, 1997). These regions have therefore been associated with lexical phonological retrieval. Reduced activation in these regions for the dyslexics relative to the controls is thus consistent with a deficit in retrieving lexical phonology. In the first study, the dyslexics showed increased activity in the left premotor cortex (Broca's area), some 2 cm more lateral to the opercular region showing reduced activation. When explicit reading was not required, activation in the left premotor cortex was not greater for dyslexic than control readers. This was predicted from the hypothesis that the increase in left premotor activation reflects compensatory effort by dyslexics in order to decode print successfully for explicit reading (see Discussion for Study 1; for a similar account see Shaywitz et al., 1998).

## General discussion

The results from the explicit and implicit reading studies were highly consistent. The dyslexic and normal reader samples were very highly comparable across studies. Both groups of dyslexic readers showed impairments in literacy as well as in phonological tasks, despite a high degree of compensation. This pattern of results is similar to that seen in a large number of studies on developmental dyslexia (see review by Snowling and Nation, 1997). It does not rule out other deficits, for instance deficits in visual motion detection, which was not tested here (e.g. Eden et al., 1996). The present findings confirm the persistence of subtle language processing difficulties even in those dyslexic individuals who attain high standards of literacy, compatible with tertiary education.

In Study 1 participants were required to read stimuli aloud, in Study 2 they were required to perform a task that did not explicitly require reading. In both cases, the level of performance was similar in the two groups. Irrespective of task requirements, both dyslexics and controls activated a common system of language areas when faced with print. The most important finding was that in both studies the dyslexics showed reduced activation in the left posterior inferior temporal lobe (BA 37) with corresponding reduction of activation in the left frontal operculum—two regions associated with modality independent phonological retrieval (Price and Friston, 1997). The reduced activation in regions that subserve phonological processing relates to the comprehensive phonological processing impairments displayed by the dyslexic participants during behavioural testing. However, while the behavioural tasks do not point to any specific aspects of phonological processing as being affected more than others, the neuro-imaging studies indicate a selective impairment in regions associated with lexical phonological retrieval. Evidence supporting this claim is presented below.

What function has been ascribed to the left posterior inferior temporal gyrus where the dyslexics consistently showed reduced activation? This region has been shown previously to be active in normal readers during reading (Price et al., 1996a), performance of a visual phonological discrimination task (Rumsey et al., 1997), and picture, letter and colour naming (Price and Friston, 1997). In the latter study, activation in BA 37 was always accompanied by activation in the frontal operculum, a region that showed reduced activation in the dyslexics relative to controls in the combined analysis. The studies on normal subjects indicate that left BA 37 (and the left frontal operculum) are critical either in the specification or retrieval of phonological information irrespective of modality of access. The connections between the posterior inferior temporal region and Wernicke's area are normally strong (Di Virgilio and Clarke, 1997). In view of these connections and this region's role in the mediation of modality-independent naming, it has been suggested that BA 37 may be Wernicke's Wortschatz' (Luders et al., 1991). The term Wortschatz', or thesaurus, implies an access facility to names of words and can be seen as a precursor of the more modern concept of the internal lexicon, and in particular, Morton's output logogen system (Morton, 1980). In support of this hypothesis, neuropathological data indicate that lesions in the posterior inferior temporal area lead to anomia in the absence of impaired object recognition (Foundas et al., 1998).

What role does Wernicke's Wortschatz play in the reading process? If these regions are critically involved in lexical retrieval, they may have a special role in the lexical retrieval process during reading. This process is particularly crucial in English given the ambiguity of grapheme-to-phoneme mappings. A model of how the ambiguity may be resolved is provided by Berent and Perfetti's two-cycles process (Berent and Perfetti, 1995): essentially, a consonant skeleton is derived first, in bottom-up mode from grapheme to phoneme. This can be done since consonants have relatively reliable letter-to-sound mappings. Vowels, on the other hand, do not, and need to be resolved in context. Hence top-down lexical processes are needed to derive the correct vowel sound in a second processing cycle. Thus, the reduced activation of the lexical components of the language processing system in dyslexics could be a consequence of a general slowness of the recognition process. This would mean that the mapping between sublexical and lexical processes is delayed and hence relatively more effort appears to be demanded of sublexical processes, as indicated by the increased activation during reading aloud in a region associated with articulatory planning.

The finding of significantly reduced activation in the left posterior inferior temporal region (BA 37) in dyslexics is validated by its replication across two different studies with different subjects and different reading tasks. It is also consistent with previously reported anomalies in the activation of posterior inferior temporal region of dyslexics during reading. The MEG study by Salmelin and colleagues, for example, reported a failure of impaired readers to activate this region during the first 200 ms after the presentation of a word, while control readers showed strong posterior inferior temporal activations (Salmelin et al., 1996). Likewise, Rumsey and colleagues observed significant deactivations in the middle and inferior temporal gyri of dyslexics, but not of controls, during reading relative to fixation (Rumsey et al., 1997).

Could the reduced activation of the area identified as Wernicke's Wortschatz in dyslexics point to a specific impairment in lexical processing? Although there is much evidence for difficulties in sublexical phonological processing (seen in the present study, for instance, in the Spoonerism task), there is also evidence for difficulties in lexical processing (seen in the present study, for instance, in poorer verbal fluency). The notion of a lexical impairment is also supported by existing behavioural data. Word finding difficulties are frequently seen in dyslexic children (Snowling et al., 1988), particularly on tasks requiring rapid naming (Denckla and Rudel, 1976; Bowers and Swanson, 1991; Wolf, 1991). Such a finding was reported recently by Swan and Goswami, who found picture naming deficits in dyslexics and in garden variety' poor readers relative to chronological and reading-age matched controls (Swan and Goswami, 1997). Unlike the garden variety poor readers, the dyslexics were often unable to name pictures that they could recognize; this would point to a specific difficulty in retrieving the phonological codes that correspond to the items to be named. The present findings are therefore consistent with dyslexia involving a core deficit in accessing phonological word forms.

Whether this deficit is itself a primary source of the failure of the system, or whether system failure, in terms of weak connectivity between otherwise well-functioning components, causes name retrieval to be slow, remains an open question. An answer would require a demonstration that dyslexics can activate BA 37 normally in the context of a different neural system. For instance, Paulesu and colleagues demonstrated that developmental dyslexics were able to activate Broca's area during rhyming and the left temporoparietal cortex during a short term memory task, but in contrast to normals, these areas were not activated in concert (Paulesu et al., 1996). Paulesu and colleagues concluded that Broca's area and the left temporoparietal cortex were functioning normally but in a disconnected fashion, perhaps due to weak connectivity between anterior and posterior regions. In other words, no single component of the system may be malfunctioning, but the system as a whole is malfunctioning because the components do not work in concert. A similar account might underlie the abnormal activations we observe in the present study, but further experiments are required.

Finally, a somewhat disappointing aspect of our studies was that no striking differences were shown between reading words and pseudowords. Any differences shown were simply those of greater activation of the same areas for the pseudowords compared with the words. This was true for both normal and dyslexic readers and may possibly reflect the high degree of lexical credibility' of the pseudowords used. It is plausible that the reader would engage in a lexical search for an existing entry. Interestingly, previous PET studies have also indicated few differences in the reading of words and word-like stimuli in normal readers (Price et al., 1996b) and dyslexics (Rumsey et al., 1997) where different types of non-words were used. The systematic manipulation of similarity between words and non-words remains a task for the future.

In summary, despite the high level of compensation that dyslexics can reach in reading and writing skills, their difficulties persist at the cognitive level. This is reflected in the dyslexics' patterns of brain activation and in their performance on the more taxing phonological tasks at the behavioural level. Our findings, which pinpoint the deficit to the neural system underlying lexical retrieval, demonstrate the utility of functional neuroimaging studies of developmental disorders when the behavioural manifestations of the disorder have ceased to be clinically abnormal.

## Appendix 1

In this test four blocks each of 20 words or pseudowords were presented in an ABBA order; the first block comprised pseudowords. The words were highly concrete (>520 on the Kucera–Francis database) and highly familiar (> 450 on the Kucera–Francis database) nouns. They were also as regular as possible in terms of grapheme–phoneme conversion. Within each block half of the stimuli were two syllables in length and half had three syllables. Words with strong onsets, such as plosives, were favoured while sibilants and fricatives were avoided. Pseudowords were created from these words by maintaining the onset and coda while changing internal consonants. Examples of the words and corresponding pseudowords include: carrot, table, garden, pencil, vinegar, corridor, microscope, monument; cassot, tagle, garlen, pectil, vipelar, collipor, mibrostope and mofukent.

Words and pseudowords were presented individually with the SuperLab program. Stimuli were presented in black capitals (Times, 48 point size font) in the centre of the computer screen, against a white background. A six letter word such as rabbit' measured ~7 cm. Each word was displayed for a maximum of 3 s. Participants were asked to read each word/pseudoword as soon as it appeared on the screen. Once the subject had responded and the latency had been recorded via a voice key, the word disappeared; there was a 1-s interval before the next stimulus was presented.

### Spoonerisms

The Spoonerism task involves the segmentation and manipulation of the constituent sounds of words (Perin, 1983). The 24 words used in the Spoonerism task were two syllable highly familiar concrete words selected on the basis that they had clear syllable divisions (e.g. BAS-KET) and no consonant clusters in their onsets. No real words were produced by recombining the onsets and rhymes of the pairs of words (see below):

Prior to the test, the concept of Spoonerisms was introduced and defined in the following way: A Spoonerism is where you have two words and you swap over the beginning sounds of each word to make two new words or pseudowords. For example, the words RED/PEN become PED/REN. Can you hear what I've done? I've taken the /r/ sound from RED and the /p/ sound from PEN and swapped them around to produce PED/REN'. A couple of practice trials followed this introduction as participants were asked What do the words DESK/HAT become?' If subjects responded correctly, HESK/DAT', they were told that this was correct and they were given the second practice trial, FISH/BOX'. Responses were scored as correct only when the word order was maintained in the response. In response to the words BASKET/LEMON, for example, the subject must respond with LASKET/BEMON'; the response BEMON/LASKET' would not be allowed. If subjects made errors during practice they were given feedback and a further trial. Word order errors were corrected during the test only the first time they occurred.

Time taken to complete the test was recorded with a stopwatch from just before the first pair was given until the last Spoonerism was completed. The time taken per pair was calculated by dividing this time by 12. During the task the stopwatch was stopped in the event of any interruptions from outside and during correction of the subject. In the event of hesitation or repeated attempts timing continued. One repeat by the experimenter was allowed per word pair if requested, but timing continued during this repetition. If the subject experienced real difficulty and discomfort the task was discontinued at the discretion of the experimenter.

#### Phonemic fluency

Participants were first given a practice trial of semantic fluency (how many animals can you think of in 30 seconds?'), then the phonemic fluency task was introduced. Participants were asked to think of as many words as possible beginning with a particular sound (list as many words as you can beginning with /s/; when I say begin' . . . Okay? . . . Begin' . . .'). On saying begin' the stopwatch was started; after 30 s the subject was stopped. The task was repeated for words beginning with /f/. The phonemic fluency score was calculated from the mean number of words generated across these two trials.

#### Auditory short-term memory for short and long words

Before testing, participants were shown the words to be used in the task. These were seven short words (worst, sum, yield, harm, bond, hate, twice) and seven long words (immediately, university, organization, individual, opportunity, association, representative), taken from Baddeley et al. (1975). Participants were asked to read the words aloud before they were removed from view. Two practice trials were then given in which, as in the experimental trials, participants heard lists of six of the words spoken at the rate of one word every second; as soon as the experimenter had spoken the last word from a list the participant was prompted to recall the words in the order of presentation. If a word could not be remembered participants were told to use the word something' to maintain the order of the surrounding words. Only words recalled in the appropriate position were scored as correct. Ten lists of short words and 10 lists of long words were presented in alternating order, and the number of words recalled correctly was summed across the two lengths of list.

#### Digit naming

In this task participants were asked to read aloud, as fast as possible, strings of 50 single syllable digits (the number 7 was excluded as the only digit with a two syllable name). Digits were presented in blocks of five (e.g. 68248 83542 99634), although participants were told to read each digit as a single number, i.e. the string 51368 should be read as five, one, three, six, eight'. The task was presented twice with different strings of 50 digits. The time taken to read each string was recorded with a stopwatch, starting from the experimenter saying the word go' of ready, steady, go', until the last digit had been said. Accuracy of reading the digits was monitored by the experimenter although it was not scored. When participants (rarely) made mistakes, they always self-corrected immediately; any repetitions were included in the timing. A mean score, in seconds, was obtained over the two trials.

Table 1

Subject characteristics: reading and phonological abilities

Variable Dyslexics (n = 6) Controls (n = 6) P
Mean SD Mean SD
n.s. = non-significant. *P < 0.05; **P < 0.01; ***P < 0.001; latency based on correct responses only.
Age (years)  23.0  5.2  23.2  2.9 n.s.
IQ tests (WAIS-R)
Full-scale  107.7  10.2 112.8  5.8 n.s.
Verbal  106.0  7.9 114.0  7.2 n.s.
Performance  108.5  14.7 108.3 10.3 n.s.
Standardized literacy tests
Reading NART (errors; maximum = 50)  28.3  8.5  12.0  2.6 ***
Reading WRAT (standardized scores)  101.5  13.9 114.2  3.7
Spelling WRAT (standardized scores)  82.3  17.6 108.2  8.7 **
Accuracy (maximum = 40):
words  38.8  1.2  40.0
pseudowords  29.7  6.1  38.8  1.3 **
Latency (ms):
words  817.6 199.4 558.5 45.0 **
pseudowords 1429.0 251.1 727.0 56.2 ***
Phonological measures
Spoonerisms (errors; maximun = 12)  5.2  4.2  0.5  0.8
Spoonerisms (s/pair)  19.4  8.0  7.7  3.9 **
Phonemic fluency (words in 30 s)  18.0  5.1  24.2  7.3 n.s.
Digit naming: 50 digits (s)  20.2  3.7  12.2  1.1 **
Auditory STM (mean correct; maximum = 6)  2.2  0.91  4.0  0.90 **
WAIS digit span (maximum = 19)  7.7  1.9  11.2  1.5 **
Variable Dyslexics (n = 6) Controls (n = 6) P
Mean SD Mean SD
n.s. = non-significant. *P < 0.05; **P < 0.01; ***P < 0.001; latency based on correct responses only.
Age (years)  23.0  5.2  23.2  2.9 n.s.
IQ tests (WAIS-R)
Full-scale  107.7  10.2 112.8  5.8 n.s.
Verbal  106.0  7.9 114.0  7.2 n.s.
Performance  108.5  14.7 108.3 10.3 n.s.
Standardized literacy tests
Reading NART (errors; maximum = 50)  28.3  8.5  12.0  2.6 ***
Reading WRAT (standardized scores)  101.5  13.9 114.2  3.7
Spelling WRAT (standardized scores)  82.3  17.6 108.2  8.7 **
Accuracy (maximum = 40):
words  38.8  1.2  40.0
pseudowords  29.7  6.1  38.8  1.3 **
Latency (ms):
words  817.6 199.4 558.5 45.0 **
pseudowords 1429.0 251.1 727.0 56.2 ***
Phonological measures
Spoonerisms (errors; maximun = 12)  5.2  4.2  0.5  0.8
Spoonerisms (s/pair)  19.4  8.0  7.7  3.9 **
Phonemic fluency (words in 30 s)  18.0  5.1  24.2  7.3 n.s.
Digit naming: 50 digits (s)  20.2  3.7  12.2  1.1 **
Auditory STM (mean correct; maximum = 6)  2.2  0.91  4.0  0.90 **
WAIS digit span (maximum = 19)  7.7  1.9  11.2  1.5 **
Table 2

Behavioural measures during PET scanning: stimuli read correctly

Stimuli Dyslexics Controls P
Mean SD Mean SD
n.s. = non-significant.
Words (maximum = 88) 85.2  3.8 87.0 1.1 n.s.
Pseudowords (maximum = 88) 73.3 10.3 83.0 4.5 n.s.
Stimuli Dyslexics Controls P
Mean SD Mean SD
n.s. = non-significant.
Words (maximum = 88) 85.2  3.8 87.0 1.1 n.s.
Pseudowords (maximum = 88) 73.3 10.3 83.0 4.5 n.s.
Table 3

Common areas activated by dyslexics and controls

Foci of activation Words – rest Pseudowords – rest
x y z Z score x y z Z score
This table shows focal maxima and Z scores for activated brain regions. The x, y and z coordinates refer to the three orthogonal dimensions of Talairach and Tournoux's atlas (Talairach and Tournoux, 1988). The x coordinate indicates regions (millimetres) to the left (–) and right (+) of the anterior commissure; y indicates mm anterior (+) and posterior (–) to the commissure; z to regions superior (+) and inferior (–) to this reference point. The Z scores indicate the magnitude of rCBF change at each location, all significant at P < 0.001. n.s. = non-significant.
Cerebellum –20 –60 –18 6.4 –16 –58 –34 3.6
–34 –40 n.s. –38 –34 3.5
Bilateral fusiform/medial extrastriate –22 –94 –10 8.1 –22 –94 –10 8.8
–28 –60 –20 7.0 –28 –60 –20 9.1
–34 –72 –20 6.5 –32 –74 –10 6.5
–40 –56 –26 6.0 –38 –60 –24 8.7
28 –98 –8 7.5 34 –88 –8 7.1
38 –62 –20 3.1 42 –64 –20 4.7
Medial extrastriate –6 –74 3.7 –6 –78 10 6.7
–10 –90 5.0 –8 –86 –10 4.7
12 –68 –20 6.8 –68 –18 7.8
Insula –30 3.0 –34 14 3.9
Thalamus –16 –14 4.6 –18 –10 5.4
–24 –6 16 3.3 –22 –4 12 4.2
Inferior parietal    n.s. 44 –58 56 3.3
Superior temporal gyri –68 –42 12 3.4 –68 –42 12 5.5
–70 –26 4.0 –70 –26 5.3
–52 12 3.2 –52 14 4.2
50 –24 3.9 56 –24 4.6
64 –4 3.8 68 –4 4.8
64 –36 12 5.6 62 –36 10 5.1
Bilateral precentral gyri –52 –6 46 4.4 –48 –8 44 4.8
–54 –4 30 4.9 –54 –4 30 5.6
54 –8 44 6.9 54 –8 42 6.6
66 –2 24 4.3 66 22 4.4
Frontotemporal 50 10 4.8 50 10 6.1
62 22 –4 n.s. 60 26 –8 3.6
Ventromedial frontal (BA 11) –26 26 –32 5.9 –26 26 –32 6.1
Inferior frontal gyrus (BA 47) 22 30 –28 4.6 22 30 –28 4.5
Foci of activation Words – rest Pseudowords – rest
x y z Z score x y z Z score
This table shows focal maxima and Z scores for activated brain regions. The x, y and z coordinates refer to the three orthogonal dimensions of Talairach and Tournoux's atlas (Talairach and Tournoux, 1988). The x coordinate indicates regions (millimetres) to the left (–) and right (+) of the anterior commissure; y indicates mm anterior (+) and posterior (–) to the commissure; z to regions superior (+) and inferior (–) to this reference point. The Z scores indicate the magnitude of rCBF change at each location, all significant at P < 0.001. n.s. = non-significant.
Cerebellum –20 –60 –18 6.4 –16 –58 –34 3.6
–34 –40 n.s. –38 –34 3.5
Bilateral fusiform/medial extrastriate –22 –94 –10 8.1 –22 –94 –10 8.8
–28 –60 –20 7.0 –28 –60 –20 9.1
–34 –72 –20 6.5 –32 –74 –10 6.5
–40 –56 –26 6.0 –38 –60 –24 8.7
28 –98 –8 7.5 34 –88 –8 7.1
38 –62 –20 3.1 42 –64 –20 4.7
Medial extrastriate –6 –74 3.7 –6 –78 10 6.7
–10 –90 5.0 –8 –86 –10 4.7
12 –68 –20 6.8 –68 –18 7.8
Insula –30 3.0 –34 14 3.9
Thalamus –16 –14 4.6 –18 –10 5.4
–24 –6 16 3.3 –22 –4 12 4.2
Inferior parietal    n.s. 44 –58 56 3.3
Superior temporal gyri –68 –42 12 3.4 –68 –42 12 5.5
–70 –26 4.0 –70 –26 5.3
–52 12 3.2 –52 14 4.2
50 –24 3.9 56 –24 4.6
64 –4 3.8 68 –4 4.8
64 –36 12 5.6 62 –36 10 5.1
Bilateral precentral gyri –52 –6 46 4.4 –48 –8 44 4.8
–54 –4 30 4.9 –54 –4 30 5.6
54 –8 44 6.9 54 –8 42 6.6
66 –2 24 4.3 66 22 4.4
Frontotemporal 50 10 4.8 50 10 6.1
62 22 –4 n.s. 60 26 –8 3.6
Ventromedial frontal (BA 11) –26 26 –32 5.9 –26 26 –32 6.1
Inferior frontal gyrus (BA 47) 22 30 –28 4.6 22 30 –28 4.5
Table 4

Differences in brain areas activated, between dyslexics and controls, irrespective of word type

Foci of activation Controls > Dyslexics Dyslexics > Controls
x y z Z score x y z Z score
Stereotactic conventions as described in Table 3. Z scores significant at P < 0.001.
Medial extrastriate/lingual gyrus –2 –80 3.03
–80 –18 4.06
16 –100 –8 3.93
12 –104 3.67
Cerebellum –28 –86 –26 3.80
–16 –54 –28 3.02
–10 –58 –42 3.38
–30 –48 3.00
Basal temporal (BA 37) –50 –48 –12 3.64
–54 –68 –14 3.58
–48 –56 –6 3.13
Caudate/thalamus –14 –20 3.15
–16 –22 18 3.03
–18 –26 2.79
Premotor (BA 6/44)     –64 26 3.09
Foci of activation Controls > Dyslexics Dyslexics > Controls
x y z Z score x y z Z score
Stereotactic conventions as described in Table 3. Z scores significant at P < 0.001.
Medial extrastriate/lingual gyrus –2 –80 3.03
–80 –18 4.06
16 –100 –8 3.93
12 –104 3.67
Cerebellum –28 –86 –26 3.80
–16 –54 –28 3.02
–10 –58 –42 3.38
–30 –48 3.00
Basal temporal (BA 37) –50 –48 –12 3.64
–54 –68 –14 3.58
–48 –56 –6 3.13
Caudate/thalamus –14 –20 3.15
–16 –22 18 3.03
–18 –26 2.79
Premotor (BA 6/44)     –64 26 3.09
Table 5

Subject characteristics: reading and phonological abilities

Variable Dyslexics (n = 6) Controls (n = 6) P
Mean SD Mean SD
Stereotactic conventions as described in Table 3. Z scores significant at P < 0.001; n.s. = non-significant. *P < 0.05; **P < 0.01;***P < 0.001; latency based on correct responses only.
Age (years)  24.5  6.1  24.5  2.9 n.s.
IQ tests (WAIS-R)
Full-scale  119.2  6.1 122.2  13.9 n.s.
Verbal  118.0  3.3 124.0  17.3 n.s.
Performance  115.5  11.6 114.0  10.2 n.s.
Standardized literacy tests
Reading NART (errors; maximum = 50)  23.3  7.1  8.7  5.1 **
Reading WRAT (standardized scores)  103.8  8.6 117.7  4.6 **
Spelling WRAT (standardized scores)  89.8  12.5 117.7  6.0 ***
Accuracy (maximum = 40):
words  39.5  0.8  40.0 n.s.
pseudowords  34.8  3.7  39.7  0.5 **
Latency (ms):
words  725.9  67.0 573.0  36.0 ***
pseudowords 1405.0 195.0 817.1 154.0 ***
Phonological measures
Spoonerisms (errors; maximum = 12)  5.2  4.2  0.5  0.8 **
Spoonerisms (s/pair)  13.2  3.8  4.8  1.7 ***
Phonemic fluency (words in 30 s)  20.3  3.4  27.0  5.4
Digit naming: 50 digits (s)  19.8  2.6  15.4  3.9
Auditory STM (mean correct; maximum = 6)  2.2  0.8  3.5  0.5 **
WAIS digit span (maximum = 19)  7.7  1.9  11.2  1.5
Variable Dyslexics (n = 6) Controls (n = 6) P
Mean SD Mean SD
Stereotactic conventions as described in Table 3. Z scores significant at P < 0.001; n.s. = non-significant. *P < 0.05; **P < 0.01;***P < 0.001; latency based on correct responses only.
Age (years)  24.5  6.1  24.5  2.9 n.s.
IQ tests (WAIS-R)
Full-scale  119.2  6.1 122.2  13.9 n.s.
Verbal  118.0  3.3 124.0  17.3 n.s.
Performance  115.5  11.6 114.0  10.2 n.s.
Standardized literacy tests
Reading NART (errors; maximum = 50)  23.3  7.1  8.7  5.1 **
Reading WRAT (standardized scores)  103.8  8.6 117.7  4.6 **
Spelling WRAT (standardized scores)  89.8  12.5 117.7  6.0 ***
Accuracy (maximum = 40):
words  39.5  0.8  40.0 n.s.
pseudowords  34.8  3.7  39.7  0.5 **
Latency (ms):
words  725.9  67.0 573.0  36.0 ***
pseudowords 1405.0 195.0 817.1 154.0 ***
Phonological measures
Spoonerisms (errors; maximum = 12)  5.2  4.2  0.5  0.8 **
Spoonerisms (s/pair)  13.2  3.8  4.8  1.7 ***
Phonemic fluency (words in 30 s)  20.3  3.4  27.0  5.4
Digit naming: 50 digits (s)  19.8  2.6  15.4  3.9
Auditory STM (mean correct; maximum = 6)  2.2  0.8  3.5  0.5 **
WAIS digit span (maximum = 19)  7.7  1.9  11.2  1.5
Table 6

Common areas activated by dyslexics and controls

Foci of activation Words – false font Pseudowords – false font Words – pseudowords Pseudowords – words
x y z Z score x y z Z score x y z Z score x y z Z score
Stereotactic conventions as described in Table 3
Cerebellum –64 –20 4.9 –68 –22 4.4
–2 –78 –20 3.8
Left posterior basal temporal –42 –48 –24 3.1 –40 –50 –22 5.1     –42 –54 –20 3.0
–46 –32 –26 3.6     –34 –38 –16 3.1
Left middle temporal (BA 20)     –48 –30 –12 3.3
Left superior temporal (BA 22) –60 –40 16 3.7 –66 –50 16 4.1
–32 –32 10 3.6 –56 4.0
Thalamus 12  –6 14 3.6    n.s. 14 –10 14 3.7
–20 3.5
–8 –44 3.1
Bilateral precentral (BA 4/6) 58 44 3.2    n.s. 58 44 3.5
Cingulate      –6 24  38 3.8
Left inferior/middle frontal (BA 44/45/46) –54 32 10 4.2 –50 30 12 5.0
–58 24 32 3.5 –56 18 26 4.1
–50 30 3.4 –50 34 28 3.1
–48 26 4.0     –48 26 3.0
Left superior frontal (BA 8)     –50 48 3.2
Foci of activation Words – false font Pseudowords – false font Words – pseudowords Pseudowords – words
x y z Z score x y z Z score x y z Z score x y z Z score
Stereotactic conventions as described in Table 3
Cerebellum –64 –20 4.9 –68 –22 4.4
–2 –78 –20 3.8
Left posterior basal temporal –42 –48 –24 3.1 –40 –50 –22 5.1     –42 –54 –20 3.0
–46 –32 –26 3.6     –34 –38 –16 3.1
Left middle temporal (BA 20)     –48 –30 –12 3.3
Left superior temporal (BA 22) –60 –40 16 3.7 –66 –50 16 4.1
–32 –32 10 3.6 –56 4.0
Thalamus 12  –6 14 3.6    n.s. 14 –10 14 3.7
–20 3.5
–8 –44 3.1
Bilateral precentral (BA 4/6) 58 44 3.2    n.s. 58 44 3.5
Cingulate      –6 24  38 3.8
Left inferior/middle frontal (BA 44/45/46) –54 32 10 4.2 –50 30 12 5.0
–58 24 32 3.5 –56 18 26 4.1
–50 30 3.4 –50 34 28 3.1
–48 26 4.0     –48 26 3.0
Left superior frontal (BA 8)     –50 48 3.2
Table 7

Reduced activation for dyslexics during implicit reading, irrespective of word type

Foci of activation Interaction Main effects
x y z Z score Controls Dyslexics
Stereotactic conventions as described in Table 3. Z scores significant at P < 0.001; n.s. = non-significant.
Left posterior basal temporal (BA 37) bordering cerebellum –42 –50  –4 3.8 3.5 n.s.
–42 –62 –14 3.1 4.2 n.s.
–44 –76 –18 3.3 3.1 n.s.
Left inferior temporal (BA 20) –46 –36 –24 3.1 5.0 n.s.
Left middle temporal (BA 21) –48 –48 3.3 3.5 n.s.
Left inferior parietal (BA 40/7) –40 –50 46 3.6 3.2 n.s.
Foci of activation Interaction Main effects
x y z Z score Controls Dyslexics
Stereotactic conventions as described in Table 3. Z scores significant at P < 0.001; n.s. = non-significant.
Left posterior basal temporal (BA 37) bordering cerebellum –42 –50  –4 3.8 3.5 n.s.
–42 –62 –14 3.1 4.2 n.s.
–44 –76 –18 3.3 3.1 n.s.
Left inferior temporal (BA 20) –46 –36 –24 3.1 5.0 n.s.
Left middle temporal (BA 21) –48 –48 3.3 3.5 n.s.
Left inferior parietal (BA 40/7) –40 –50 46 3.6 3.2 n.s.
Table 8

Reduced activation for dyslexics, explicit and implicit reading combined

Foci of activation Interaction Main effects
x y z Z score Controls Dyslexics
Stereotactic conventions as described in Table 3. Z scores significant at P < 0.001; n.s. = non-significant.
Bilateral cerebellum –18 –70 –28 3.0 5.7 n.s.
10 –76 –18 3.3 5.8 n.s.
–48 –48 –26 3.2 7.3 4.3
–48 –60 –24 3.0 6.0 n.s.
Left inferior/middle temporal (BA 21/37) –42 –60 –12 3.2 6.0 3.3
–42 –48  –6 4.0 5.9 3.0
Left frontal operculum/anterior insula –40 22 3.1 5.1 3.4
Foci of activation Interaction Main effects
x y z Z score Controls Dyslexics
Stereotactic conventions as described in Table 3. Z scores significant at P < 0.001; n.s. = non-significant.
Bilateral cerebellum –18 –70 –28 3.0 5.7 n.s.
10 –76 –18 3.3 5.8 n.s.
–48 –48 –26 3.2 7.3 4.3
–48 –60 –24 3.0 6.0 n.s.
Left inferior/middle temporal (BA 21/37) –42 –60 –12 3.2 6.0 3.3
–42 –48  –6 4.0 5.9 3.0
Left frontal operculum/anterior insula –40 22 3.1 5.1 3.4
Word 1: Word 2: Spoonerisms:
Button Turtle Tutton/Burtle
Doctor Window Woctor/Dindow
Fabric Pocket Pabric/Focket
Motor Tiger Totor/Miger
Novel Table Tovel/Nable
Paper Satin Saper/Patin
Rabbit Sofa Sabbit/Rofa
Razor Medal Mazor/Redal
Ribbon Silver Sibbon/Rilver
Word 1: Word 2: Spoonerisms:
Button Turtle Tutton/Burtle
Doctor Window Woctor/Dindow
Fabric Pocket Pabric/Focket
Motor Tiger Totor/Miger
Novel Table Tovel/Nable
Paper Satin Saper/Patin
Rabbit Sofa Sabbit/Rofa
Razor Medal Mazor/Redal
Ribbon Silver Sibbon/Rilver
Fig. 1

Networks of rCBF activation common to the dyslexics and controls reading aloud words and pseudowords. Activated systems include the bilateral cerebellum/fusiform, precentral gyri, superior temporal gyri, inferior parietal and frontal regions. Significant voxels are rendered onto lateral and medial surfaces of idealized brains corresponding to the sterotactic space defined by Talairach and Tournoux (1988). As in Fig. 2, areas of significant activation are shown in colour.

Networks of rCBF activation common to the dyslexics and controls reading aloud words and pseudowords. Activated systems include the bilateral cerebellum/fusiform, precentral gyri, superior temporal gyri, inferior parietal and frontal regions. Significant voxels are rendered onto lateral and medial surfaces of idealized brains corresponding to the sterotactic space defined by Talairach and Tournoux (1988). As in Fig. 2, areas of significant activation are shown in colour.

Fig. 2

Regions of reduced rCBF activation in the dyslexics relative to the controls across Studies 1 and 2, irrespective of word type. The dyslexics consistently activated the left inferior/middle temporal region, the left frontal operculum and the cerebellum to a lesser degree than the controls. As in Fig. 1, significant voxels are rendered onto lateral and medial surfaces.

Regions of reduced rCBF activation in the dyslexics relative to the controls across Studies 1 and 2, irrespective of word type. The dyslexics consistently activated the left inferior/middle temporal region, the left frontal operculum and the cerebellum to a lesser degree than the controls. As in Fig. 1, significant voxels are rendered onto lateral and medial surfaces.

The authors are grateful to Alison Gallagher for advice on stimulus materials and tests, and to Eraldo Paulesu and Jean-Francois Démonet for their vital contribution to the design of the experiments. This study was carried out under Biomed contract No. BMH4-CT96-0274. Additional funding was provided by the MRC, the Wellcome Trust and the Gatsby Foundation.

## References

Baddeley AD, Thompson N, Buchanan M. Word length and the structure of short-term memory.
J Verb Learn Verb Behav

1975
;
14
:
575
–89.
Bailey DL, Jones T, Friston KJ, Colebatch JG, Frackowiak RSJ. Physical validation of statistical parametric mapping.
J Cereb Blood Flow Metab

1991
;
11
:
S150
.
Berent I, Perfetti CA. A rose is a REEZ: the two-cycles model of phonology assembly in reading English.
Psychol Rev

1995
;
102
:
146
–84.
Bookheimer SY, Zeffiro TA, Blaxton T, Gaillard W, Theodore W. Regional cerebral blood flow during object naming and word reading.
Hum Brain Mapp

1995
;
3
:
93
–106.
Bruck M. Persistence of dyslexics' phonological awareness deficits.
Dev Psychol

1992
;
28
:
874
–86.
DeFries JC, Alarcon M, Olson RK. Genetic aetiologies of reading and spelling deficits: developmental differences. In: Hulme C, Snowling M, editors. Dyslexia: biology, cognition and intervention. London: Whurr; 1997. p. 20–37.
Demb JB, Boynton GM, Heeger DJ. Brain activity in visual cortex predicts individual differences in reading performance.

1997
;
94
:
13363
–6.
Démonet J-F, Chollet F, Ramsay S, Cardebat D, Nespoulous J-L, Wise R, et al. The anatomy of phonological and semantic processing in normal subjects.
Brain

1992
;
115
:
1753
–1768.
Denckla MB, Rudel RG. Rapid automatized' naming (R.A.N.): dyslexia differentiated from other learning disabilities.
Neuropsychologia

1976
;
14
:
471
–9.
Di Virgilio G, Clarke S. Direct interhemispheric visual input to human speech areas.
Hum Brain Mapp

1997
;
5
:
347
–54.
Eden GF, VanMeter JW, Rumsey JM, Maisog JM, Woods RP, Zeffiro TA. Abnormal processing of visual motion in dyslexia revealed by functional brain imaging [see comments].
Nature

1996
;
382
:
66
–9. Comment in: Nature 1996; 382: 19–20.
Elbro C, Nielsen I, Petersen DK. Dyslexia in adults: evidence for deficits in non-word reading and in the phonological representation of lexical items.
Ann Dyslexia

1994
;
44
:
205
–26.
Flowers DL, Wood FB, Naylor CE. Regional cerebral blood flow correlates of language processes in reading disability.
Arch Neurol

1991
;
48
;
637
–43
Fiez JA, Raichle ME, Balota DA, Tallal P, Petersen SE. PET activation of posterior temporal regions during auditory word presentation and verb generation.
Cereb Cortex

1996
;
6
:
1
–10.
Foundas A, Daniels SK, Vasterling JJ. Anomia: case studies with lesion localisation.
Neurocase

1998
;
4
:
35
–43.
Fox PT, Mintun MA. Non-invasive functional brain mapping by change-distribution analysis of averaged PET images of H215O tissue activity.
J Nucl Med

1989
;
30
:
141
–9.
Frackowiak RS, Friston KJ. Functional neuroanatomy of the human brain: positron emission tomography—a new neuroanatomical technique. [Review].
J Anat

1994
;
184
:
211
–25.
Friston KJ, Frith CD, Liddle PF, Frackowiak RSJ. Comparing functional (PET) images: the assessment of significant change.
J Cereb Blood Flow Metab

1991
;
11
:
690
–9.
Friston KJ, Ashburner J, Poline J-B, Frith CD, Heather JD, Frackowiak RS. Spatial realignment and normalisation of images.
Hum Brain Mapp

1995
;
2
:
165
–89.
Friston KJ, Holmes AP, Worsley KJ, Poline J-B, Frith CD, Frackowiak RSJ. Statistical parametric maps in functional imaging. A general linear approach.
Hum Brain Mapp

1995
;
3
:
189
–210.
Frith CD, Kapur N, Friston KJ, Liddle PF, Frackowiak RSJ. Regional cerebral activity associated with the incidental processing of pseudo-words.
Hum Brain Mapp

1995
;
2
:
153
–60.
Frith U, Landerl K, Frith CD. Dyslexia and verbal fluency: more evidence for a phonological deficit.
Dyslexia: Int J Res Pract

1995
;
1
:
2
–11.
Frith U. Brain, mind and behaviour in dyslexia. In: Hulme C, Snowling M, editors. Dyslexia: biology, cognition and intervention. London: Whurr; 1997: p. 1–19.
Galaburda AM, Menard MT, Rosen GD. Evidence for aberrant auditory anatomy in developmental dyslexia.

1994
;
91
:
8010
–3.
Geschwind N, Galaburda AM. Cerebral lateralization: biological mechanisms, associations and pathology.
Arch Neurol

1985
;
42
:
428
–59.
Goswami U. Orthographic analogies and reading development.
Q J Exp Psychol

1988
;
40A
:
239
–68.
Hagman JO, Wood F, Buchsbaum MS, Tallal P, Flowers L, Katz W. Cerebral brain metabolism in adult dyslexic subjects assessed with positron emission tomography during performance of an auditory task.
Arch Neurol

1992
;
49
:
734
–9.
Hagoort P, Indefrey P, Brown C, Herzog H, Steinmetz H, Seitz RJ. The neural circuitry involved in the reading of German words and pseudowords: a PET study. J Cogn Neurosci. In press 1999.
Hynd GW, Semrud-Clikeman M, Lorys AR, Novey ES, Eliopulos D. Brain morphology in developmental dyslexia and attention deficit disorder/hyperactivity.
Arch Neurol

1990
;
47
:
919
–26.
Indefrey P, Kleinschmidt A, Merbolt K-D, Kruger G, Brown C, Hagoort P, et al. Equivalent responses to lexical and nonlexical visual stimuli in occipital cortex: a functional magnetic resonance imaging study.
Neuroimage

1997
;
5
:
78
–81.
Jastak S, Wilkinson GS. The Wide Range Achievement Test. Wilmington, DE: Jastak Associates Systems; 1984.
Landerl K, Wimmer H, Frith U. The impact of orthographic consistency on dyslexia. A German-English comparison.
Cognition

1997
;
63
:
315
–34.
Larsen J, Hoien T, Lundberg I, Odegaard H. MRI evaluation of the size and symmetry of the planum temporale in adolescents with developmental dyslexia.
Brain Lang

1990
;
39
:
289
–301.
Lehmkuhle S, Garzia RP, Turner L, Hash T, Baro JA. A defective visual pathway in children with reading disability [see comments].
N Engl J Med

1993
;
328
:
989
–96. Comment in: N Engl J Med 1993; 329: 579.
Livingstone MS, Rosen GD, Drislane FW, Galaburda AM. Physiological and anatomical evidence for a magnocellular defect in developmental dyslexia [published erratum appears in Proc Natl Acad Sci USA 1993; 90: 2556].

1991
;
88
:
7943
–7.
Luders H, Lesser RP, Hahn J, Dinner DS, Morris HH, Wyllie E, et al. Basal temporal language area.
Brain

1991
;
114
:
743
–54.
Lyon GR. Toward a definition of dyslexia.
Ann Dyslexia

1995
;
45
:
3
–27.
Macleod CM. Half a century of research on the Stroop effect: an integrative review. [Review].
Psychol Bull

1991
;
109
:
163
–203.
Morton J. The logogen model and orthographic structure. In: Frith U, editor. Cognitive processes in spelling. London: Academic Press; 1980. p. 117–33.
Paulesu E, Frith CD, Frackowiak RS. The neural correlates of the verbal component of working memory [see comments].
Nature

1993
;
362
:
342
–5. Comment in: Nature 1993; 363: 583–4.
Paulesu E, Frith U, Snowling M, Gallagher A, Morton J, Frackowiak RS, et al. Is developmental dyslexia a disconnection syndrome? Evidence from PET scanning.
Brain

1996
;
119
:
143
–57.
Perin D. Phonemic segmentation and spelling.
Br J Psychol

1983
;
74
:
129
–44.
Price CJ, Friston KJ. Cognitive conjunction: a new approach to brain activation experiments.
Neuroimage

1997
;
5
:
261
–70.
Price CJ, Wise RJ, Watson JD, Patterson K, Howard D, Frackowiak RS. Brain activity during reading: the effects of exposure duration and task.
Brain

1994
;
117
:
1255
–69.
Price CJ, Moore CJ, Frackowiak RS. The effect of varying stimulus rate and duration on brain activity during reading.
Neuroimage

1996
;
3
:
40
–52.
Price CJ, Wise RJ, Frackowiak RS. Demonstrating the implicit processing of visually presented words and pseudowords.
Cereb Cortex

1996
;
6
:
62
–70.
Price CJ, Moore CJ, Friston KJ. Subtractions, conjunctions and interactions in experimental design of activation studies.
Hum Brain Mapp

1997
;
5
:
264
–72.
Price CJ, Moore CJ, Humphreys GW, Wise RJS. Segregating semantic from phonological processes during reading.
J Cogn Neurosci

1997
;
6
:
727
–33.
Rack JP, Snowling MJ, Olson RK. The nonword reading deficit in developmental dyslexia—a review.

1992
;
1
:
28
–53.
Rumsey JM, Andreason P, Zametkin AJ, Aquino T, King AC, Hamburger SD, et al. Failure to activate the left temporoparietal cortex in dyslexia: an oxygen 15 positron emission tomographic study [published erratum appears in Arch Neurol 1994; 51: 243].
Arch Neurol

1992
;
49
:
527
–34.
Rumsey JM, Andreason P, Zametkin AJ, King AC, Hamburger SD, Aquino T, et al. Right frontotemporal activation by tonal memory in dyslexia, an H215O PET study.
Biol Psychiatry

1994
;
36
:
171
–80.
Rumsey JM, Nace K, Donohue B, Wise D, Maisog M, Andreason P. A positron emission tomographic study of impaired word recognition and phonological processing in dyslexic men.
Arch Neurol

1997
;
54
:
562
–73.
Salmelin R, Service E, Kiesila P, Uutela K, Salonen O. Impaired visual word processing in dyslexia revealed with magnetoencephalography.
Ann Neurol

1996
;
40
:
157
–62.
Shaywitz SE, Shaywitz BA, Pugh KR, Fulbright RK, Constable RT, Mencl WE, et al. Functional disruption in the organization of the brain for reading in dyslexia.

1998
;
95
:
2636
–41.
Siegel LS. An evaluation of the discrepancy definition of dyslexia.
J Learn Disabil

1992
;
25
:
618
–29.
Snowling MJ. Developmental reading disorders. [Review].
J Child Psychol Psychiatry

1991
;
32
:
49
–77.
Snowling MJ, Nation KA. Language, phonology and learning to read. In: Hulme C. Snowling M, editors. Dyslexia: biology, cognition and intervention. London: Whurr; 1997. p. 153–66.
Snowling M, Nation K, Moxham P, Gallagher A, Frith U. Phonological processing skills of dyslexic students in higher education: a preliminary report.

1997
;
1
:
31
–41.
Spring C, Davis JM. Relations of digit naming speed with three components of reading.
Appl Psycholinguist

1988
;
9
:
315
–34.
Swan D, Goswami U. Picture naming deficits in developmental dyslexia: the phonological representations hypothesis.
Brain Lang

1997
;
56
:
334
–53.
Talairach J, Tournoux P. Co-planar stereotaxic atlas of the human brain. Stuttgart: Thieme; 1988.
Townsend DW, Geissbuller A, Defrise M, Hoffman EJ, Spinks TJ, Bailey DL, et al. Fully three-dimensional reconstruction for a PET camera with retractable septa.
IEEE Trans Med Imaging

1991
;
10
:
505
–12.
Van Orden GC, Johnston JC, Hale BL. Word identification in reading proceeds from spelling to sound to meaning.
J Exp Psychol Learn Mem Cogn

1988
;
14
:
371
–86.
Zatorre RJ, Evans AC, Meyer E, Gjedde A. Lateralization of phonetic and pitch discrimination in speech processing.
Science

1992
;
256
:
846
–9.