Dissociations in Processing Past Tense Morphology - Centre for

0 downloads 0 Views 2MB Size Report
English verbs form their past tense by adding the regular affix /-d/, a process that does not affect the form of the stem. This past tense marker is realized as [d], [t], ...
Dissociations in Processing Past Tense Morphology: Neuropathology and Behavioral Studies Lorraine K. Tyler1, Paul deMornay-Davies1, Rebekah Anokhina1, Catherine Longworth1, Billi Randall1, and William D. Marslen-Wilson2

Abstract & Neuropsychological research showing that the regular (‘‘jump–jumped’’) and irregular (‘‘drive/drove’’) past tense inflectional morphology can dissociate following brain damage has been important in testing claims about the cognitive and neural status of linguistic rules. These dissociations have been interpreted as evidence for two different computational systems—a rule-based system underlying the processing of regulars and the irregulars being individually listed in the mental lexicon. In contrast, connectionist accounts claim that these dissociations can be modeled within a single system. Combining behavioral data from patients with detailed information about their neuropathology can, in principle, provide strong constraints on accounts of the past tense. In this study, we tested five nonfluent aphasic patients, all of

INTRODUCTION English inflectional morphology has become the focus for a set of crucial issues about mental processes and representations. Linguists, psycholinguists, and connectionist modelers have used the inflectional system as a means of testing claims about the cognitive status of linguistic rules. The question driving this research is whether there is evidence for the existence of internally represented rules, expressed as symbols combined by a syntax, or whether distributed network representations, which require neither symbols nor syntax, are sufficient to account for the English inflectional system. Inflectional verb morphology in English consists of both regular and irregular forms. The vast majority of English verbs form their past tense by adding the regular affix /-d/, a process that does not affect the form of the stem. This past tense marker is realized as [d], [t], or [ed] depending on the final segment of the stem, as in ‘‘agree/agreed,’’ ‘‘jump/jumped,’’ ‘‘/hate/hated/.’’ All except about 160 past tense forms are created in this way. The exceptional 160, which are among the most frequent in the language, take some other, usually idiosyn1 University of Cambridge, 2MRC Cognition and Brain Sciences Unit, Cambridge

D 2002 Massachusetts Institute of Technology

whom had extensive left hemisphere (LH) damage involving the left inferior frontal gyrus and underlying structures, and four patients with semantic deficits following herpes simplex encephalitis (HSE) who had extensive damage to the inferior temporal cortex. These patients were tested in experiments probing past tense processing. In a large priming study, the nonfluent patients showed no priming for the regular past tense but significant priming for the irregulars (whereas controls show priming for both). In contrast, the HSE patients showed significantly impaired performance for the irregulars in an elicitation task. These patterns of behavioral data and neuropathology suggest that two separable but interdependent systems underlie processing of the regular and irregular past tense. &

cratic, form of past tense inflection. A few involve no stem change at all (e.g., ‘‘hit/hit’’), but most involve changes to their stem, either to the vowel (e.g., ‘‘sing/ sang’’; ‘‘drive/drove’’), the final consonant (e.g., ‘‘make/ made’’) or to both (e.g., ‘‘keep/kept’’). These changes are morphologically triggered and are not phonologically predictable. The key question is whether these linguistic differences reflect differences in the way in which the two types of past tense are mentally represented. Psychological studies comparing them consistently show different patterns of results for regular and irregular forms, suggesting distinct types of underlying mental representation. For example, studies using repetition priming tasks, where a test word is preceded by a related prime word, have generally found significant priming between regularly inflected past tense forms and their stems, but reduced priming (or none at all) for irregular past tense forms (Kempley & Morton, 1982; Stanners, Neiser, Hernon, & Hall, 1979). In a recent series of priming studies, Marslen-Wilson, Hare, and Older (1993, 1995) have extended earlier research that showed differential priming effects for regular and irregular past tense forms. This pattern of differential effects is consistent with studies using different paradigms, such as the elicitation Journal of Cognitive Neuroscience 14:1, pp. 79–94

task used by Bybee and Modor (1983). They asked subjects to produce past tense forms for nonwords that were similar to real verbs (these are termed ‘‘nonce’’ forms), and found that they were unlikely to produce irregular past tense forms except when the novel item closely resembled an existing irregular verb. For example, subjects produced ‘‘splung’’ as the past tense of ‘‘spling.’’ The nonce form ‘‘spling’’ is very similar to the irregular verb ‘‘spring,’’ which is one of about 12 irregular verbs sharing similar phonological alternations (as in ‘‘spring/sprung,’’ ‘‘fling/flung,’’ etc). In contrast, production of regularly inflected forms is argued not to be affected by phonological similarity, consistent with the application of a symbolic default rule (Prasada & Pinker, 1993). These behavioral differences between irregular and regular forms have been interpreted in two very different ways; in terms of either linguistic rule-based systems or of distributed connectionist models. According to current rule-based accounts, there are two distinct mechanisms underlying regular and irregular past tense forms, with irregulars being individually listed in the mental lexicon in a pattern-associating network (Prasada & Pinker, 1993). In contrast, the regular past tense is generated by the application of a default rule, which is applied without reference to the phonology of the verb stem involved. Connectionist models take a very different approach, claiming that dissociations between regular and irregular past tense forms can be captured within a single mechanism connectionist system ( Joanisse & Seidenberg, 1999; Marchman, 1993; Plunkett & Marchman, 1991). Regular and irregular verbs differ, according to these models, in terms of type/token frequency and phonological predictability. The relationship between stem and past tense form for the regular verbs is more phonologically predictable but irregulars are more frequent. Neuropsychological research showing that the regular and irregular past tense can dissociate following brain damage has been important in this debate. Nevertheless, the demonstration of a dissociation—where some patients have more problems with the regulars and other patients have more problems with the irregulars—does not, in and of itself, provide conclusive evidence for either a single or dual system account. Behavioral data alone can be modeled both in terms of two separate mechanisms and in terms of a single system, since some connectionist models are capable of demonstrating double dissociations (e.g., Joanisse & Seidenberg, 1999; Plaut & Shallice, 1993). However, neuropsychological data in the context of knowledge about a patient’s neuropathology does have the potential to provide strong constraints on accounts of the representation of the past tense. If deficits in processing either the regular or irregular past tense are associated with damage to different neural pathways, this undermines the case for a single system at the neural level. It is 80

Journal of Cognitive Neuroscience

important, therefore, to combine clear dissociations in processing the past tense with an analysis of the neuropathologies these dissociations are linked to. Early research with patients showing morphological deficits suggested that these were associated with damage to Broca’s area. However, this work was limited in that distinctions were rarely drawn between the inflectional and derivational morphology (but see Miceli & Caramazza, 1988) or between regularly and irregularly inflected words. More recently, Ullman et al. (1997) reported a study in which patients with various types of brain damage [Alzheimer’s disease (AD), Huntington’s disease (HD), Parkinson’s disease (PD), and aphasic patients who had damage to either posterior or anterior brain regions] were tested for their ability to produce past tense forms. The main task was a sentence completion paradigm in which subjects were asked to produce past tense verbs in a sentence context. They contrasted regular, irregular, and novel verbs that took the regular past tense form (e.g., ‘‘spuffed,’’ ‘‘prapped’’). They found that AD patients and aphasic patients with damage to the temporal or parietal cortex had more difficulty in generating the past tense forms of irregular verbs, while patients with damage to the frontal cortex (aphasic patients as well as PD and HD patients) had problems generating the past tense of regulars. However, the robustness of these effects have been questioned, given that some of the patients could not perform the completion task and were tested instead on a different task in which they had to read the verbs aloud, that few items were included in the analyses, and that the two sets of past tense items were not always fully matched on frequency (Joanisse & Seidenberg, 1999). Another limitation of the Ullman et al. report is that the neuropathology of some of the patients is inferred on the basis of their behavioral deficit rather than established by means of neuroanatomical evidence. The AD patients, for example, are assumed to have temporal lobe damage given their semantic deficits, even though the extent and location of neural damage in AD can be highly variable. Nonetheless, the general pattern observed by Ullman et al. (1997) is supported by some recent neuropsychological work of our own, contrasting patients with semantic deficits and agrammatic aphasic patients. The two agrammatic patients, DE and JG, had suffered extensive left hemisphere (LH) frontal damage including Broca’s area (Marslen-Wilson & Tyler, 1997, 1998), whereas ES, a patient suffering from semantic dementia, had bilateral damage to the temporal lobe with very little frontal involvement. Both agrammatic patients had typical nonfluent speech, combined with difficulties in interpreting inflected words in which a stem is combined with an affix (e.g., ‘‘smile-ing,’’ ‘‘smile-s’’; Tyler, 1992). We expected these patients to have problems with regular past tense forms as well, since these also involve combining a stem and an affix. The important Volume 14, Number 1

issue was whether they would show similar problems with the irregular past tense, where the morphological relation between stem and past tense form does not involve the same type of combinatorial operation. These irregular forms do not have any internal morphological structure, and must be accessed (and generated) as whole forms.1 We tested the patients in a spoken word priming task using lexical decision, so as to avoid tasks that involve either reading or speech production, which typically pose problems for agrammatic patients. The patients showed significant priming for the irregular past tense forms (‘‘gave–give’’) and for word-pairs that were semantically related (such as ‘‘cello–violin’’), but they showed no priming for regularly inflected pairs (‘‘jumped–jump’’). The absence of priming for the regulars contrasted with the data from control subjects who showed robust priming for both regulars and irregulars. In contrast, the patient suffering from semantic dementia showed priming for the regulars but not for the irregulars nor the semantically related pairs. These double dissociations seem to provide strong evidence for functional and neurological distinctions between the underlying systems supporting the representation and processing of regular and irregular forms. They also led to the preliminary proposal (MarslenWilson & Tyler, 1998) that these dissociations reflect distinct processing routes to lexical representations, with an inferior temporal route for access of stored whole phonological forms (such as simple nouns and irregular past tenses), and a second route, involving the posterior frontal cortex, for complex forms such as regular past tenses, where access to the stem requires processes of phonological parsing that separate the inflectional affix from its stem. Present Research We had two main aims in carrying out the present set of studies. The first was to determine whether the double dissociation we reported in the earlier paper (MarslenWilson & Tyler, 1997) would hold up for larger groups of subjects, with different etiologies, and using a wider range of experimental tasks. The second was to determine whether the dissociations that we had reported earlier between the regulars and irregulars were related to specific patterns of neuropathology. In particular, we expected the left temporal lobe damage to result in deficits for the irregular past tense and damage to the posterior frontal cortex to be associated with impairments of the regular past tense. By relating neuropathological data for each patient to their behavioral pattern in the past tense studies, we should be able to place strong constraints on current accounts of the underlying mechanisms involved in processing the regular and irregular past tense.

In our original study, we reported data from two agrammatics and one semantic dementia patient, with the agrammatics showing no priming for the regulars and the semantic dementia patient showing the opposite pattern of no priming for the irregulars.2 For the present research we developed a new priming study, similar to the original experiment but including a new, larger, and better-matched set of items. We tested five agrammatic patients (two of the original patients and three others), all with middle cerebral artery infarcts implicating the left inferior frontal structures. In order to probe the other side of the dissociation—that problems in processing the irregular past tense are associated with semantic deficits due to temporal lobe lesions—we selected four patients with temporal lobe damage following herpes simplex encephalitis (HSE), all of whom had severe semantic disorders. None of the HSE patients could carry out the priming task, due to difficulties with lexical decision. Instead we developed a version of the past tense elicitation paradigm, allowing us to link this study more directly to earlier research using this task to probe the access and representation of regular and irregular past tense forms (Ullman et al., 1997; Prasada & Pinker, 1993; Bybee & Modor, 1987). Patients: Behavioral and Neuropathological Data Nonfluent Aphasic Patients The five nonfluent patients in this group all presented with a language profile characteristic of Broca’s aphasia and all had left inferior frontal damage as a result of a middle cerebral artery infarct (MCA). Details of their scores on a sample of different language tasks tests used to assess language deficits in aphasic patients are presented in Table 1. We also include data on the Ravens CPM test in this table. Nonfluent Broca’s aphasics typically have more difficulty in repeating sentences compared to single real words. Consistent with this, our group of five nonfluent patients are more accurate at repeating single words (Test 1) than nonwords (Test 2) or sentences (Test 3). Tests 4 and 5 evaluate single word (4) and nonword (5) reading and show the typical nonfluent pattern of poorer performance at reading nonwords compared to real words. Test 6 uses a sentence–picture matching paradigm to investigate sentence processing abilities in our patients. Subjects hear a sentence that they compare against a set of three pictures displayed in front of them. The sentences are all semantically ‘‘reversible’’ in that either person mentioned can potentially perform the action (e.g., ‘‘The girl pulls the boy’’), and the verb used in each sentence had a preference to take a noun phrase argument (as determined by pretests). Half of the sentences were passive and half were active. The two ‘‘foil’’ pictures in the three-picture set either contained a lexical distractor (e.g., a picture of a girl Tyler et al.

81

82

Journal of Cognitive Neuroscience

Volume 14, Number 1

Single real word repetition (PALPA) Nonword repetition (PALPA) Sentence repetition (Spreen & Benton, 1969) Single word reading PALPA Nonword reading(PALPA) Same–different word minimal pairs (PALPA) Syntax: spoken sentence–picture matchb Spoken word–picture match (Moss et al., 1998) Ravens CPM 36/36 correct

Initial: 100%, Final: 83% RR errors = 32%, Lexical errors = 0% 100% correct

64% imageability effect 4.2% correct

18% correct

43% correct

93% correct

Not done

Initial: 86%, Final: 80% RR errors = 41%, Lexical errors = 2% 96% correct

50% imageability effect 0% correct

13% correct

37% correct

76% correct

JG

34/36 correct

Initial: 84%, Final: 67% RR errors = 20%, Lexical errors = 0% 100% correct

90% correct; No imageability effect 29% correct

14% correct

60% correct

97% correct

CH

25/36 correct

Initial: 86%, Final: 92% RR errors = 29%, Lexical errors = 8% 88% correct

77% correct, mild imageability effect 42% correct

36% correct

57% correct

80% correct

MR

32/36 correct

Initial: 100%, Final: 100% RR errors = 32%, Lexical errors = 6% 100% correct

96% correct

100% correct

41% correct

73% correct

99% correct

AS

Syntax: spoken sentence–picture matcha Picture naming (Bunn et al., 1998) Spoken word–picture matching (Moss et al., 1998) Within-category word–picture matchb Picture sorting (Moss & Tyler, 1998) Naming to description (Moss & Tyler, 2000) Pyramids and palm trees (Howard & Patterson, 1992) Property verification (Moss & Tyler, 2000)

2

80% correct

58% correct

43% correct

75% correct

65% correct

25% correct

83% correct

77% correct

82% correct 81% correct

77% correct

9% reverse role errors, 0 lexical errors 42% correct

0 reverse role errors, 6% lexical errors 73% correct 75% correct

33/36 correct

35/36 correct

RC

83% correct

Not done

45% correct

74% correct

88% correct

80% correct

3% reverse role errors, 0 lexical errors 59% correct

30/36 correct

JH

76% correct

77% correct

13% correct

81% correct

84% correct

68% correct

9% reverse role errors, 0 lexical errors 27% correct

36/36 correct

WL

92–100% correct

98–99% correct

80–100% correct

90–100% correct

100% correct

90–100% correct

94% correct

99% correct

29–35 correct

Norms

29–35 correct

99% correct

99% correct

98% correct

95% correct

99% correct

No norms available

No norms available

99%

Normsa

b

Unpublished data from CSL. The aim of this experiment was to compare performance on living and nonliving things in a task that required no verbal response. We presented subjects with a four-picture array, consisting of a target, a within-category distractor, and two across-domain distractors, which were matched for familiarity. We included two categories within the living things domains (animals and fruit) and two categories from the nonliving domain (tools and vehicles). The pictures were taken from our set of colored photographs that was extensively normed for name agreement and many other variables. Details of the picture set are given in Bunn et al. (1998).

a

9

8

7

6

5

4

3

Ravens CPM

1

JBR

Table 2. Background Tests on the Four HSE Patients

b

Norms taken from PALPA (Kay, Lesser, & Coltheart, 1992) or collected in our lab on groups of healthy subjects aged between 60–73 years, with 10–12 subjects per group. Unpublished data from CSL.

a

9

8

7

6

5

4

3

2

1

DE

Table 1. Background Language Tests on Nonfluent Patients

painting a boy) or a reverse role distractor. In a picture that serves as a reverse role distractor, the agent of the action becomes the recipient of the action (e.g., The boy pulls the girl). Typically, nonfluent patients make more reverse role errors and few lexical errors on this kind of test (Black, Nickels, & Byng, 1991). As Table 1 shows, the performance of our patients was consistent with this pattern; all patients made a large number of reverse role (RR) errors. Finally, in keeping with most nonfluents, the patients were very accurate on a spoken word–picture matching test (Test 8) designed to probe for semantic deficits. Thus, the overall pattern that emerges from this sample of tests is that our patients all show a similar pattern of intact lexical semantics and impaired syntax (for further details on the language abilities of some of these patients see also Ostrin & Tyler, 1995; Tyler, 1985, 1992). (1) DE has been described extensively elsewhere (e.g., Tyler, 1985, 1992; Tyler & Ostrin, 1994; Tyler, Moss, & Jennings, 1995). He is a right-handed man who had a CVA in 1970 following a motor-scooter accident. His speech output is slow, effortful, and hesitant. He was classified as an ‘‘agrammatic’’ aphasic on the BDAE and in addition, he is a deep dyslexic (Patterson & Marcel, 1977). In a variety of on-line studies, he typically shows significant semantic priming effects and problems with syntax (Tyler, 1992). A CT scan carried out in 1978 revealed a large LH lesion affecting the middle and posterior parts of the inferior frontal lobe and most of the superior and middle temporal lobe. There was no damage to the left inferior temporal lobe or the parietal lobe and no abnormality in the right hemisphere. A magnetic resonance imaging (MRI) scan taken in 1996 confirmed this, showing extensive frontal (including Broca’s area) and temporal damage, with sparing of the inferior temporal cortex and the right hemisphere (see Figures 1 and 2). (2) JG has also been described in detail elsewhere (Ostrin & Tyler, 1995; Tyler, 1992, Tyler et al., 1995). A right-handed man, he suffered a CVA in 1980 when he was 55 years old. He has a dense right-sided hemiplegia. JG is an agrammatic aphasic, with reading impairments typical of deep dyslexia. His spontaneous speech is slow, hesitant, and consists of short, simple phrases. Just like DE, in a wide range of studies probing his on-line language comprehension, he shows significant semantic priming effects but severe impairments in various aspects of syntactic processing (Ostrin & Tyler, 1995; Tyler, 1992). JG had aneurysm clips inserted, and neither MR nor CT scans are available for him. A PET study carried out in 1992, measuring cerebral blood flow, showed that he had a large perisylvian lesion involving most of the region supplied by the left MCA, including the left temporo-parietal cortex and the left inferior and middle frontal cortices with relative sparing of the inferior temporal cortex (Price et al., 1998). Figure 3a shows

cerebral blood flow obtained during this study (courtesy of Dr C. Price, Wellcome Department of Cognitive Neurology). (3) MR is a right-handed 65-year-old woman. She left school at 16 and worked as a receptionist. She suffered an MCA infarct in February 1986. Her language abilities in both comprehension and production are typical of an agrammatic aphasic (see Table 1). Her speech is nonfluent, she has problems with syntax but is relatively unimpaired on lexical semantic tests. Her single word repetition is considerably better than her sentence repetition. A CT scan in 1996 showed a large LH infarct involving the lateral aspect of the frontal lobe (Broca’s area, prefrontal region, and paraventricular and supraventricular areas), the parietal lobe (supramarginal gyrus, paraventricular and supraventricular areas), the insula, and auditory cortex. The left inferior temporal lobe was undamaged (see Figure 3b). The presence of aneurysm clips did not allow MR scanning. (4) CH is a right-handed 37-year-old woman. She left school at 16, moving to Hong Kong in 1981 to work as a secretary and later as an office manager. She had a left MCA infarct in 1997. Her spontaneous language is typical of Broca’s aphasia, consisting of single words combined into short two to three word phrases with no or little syntax. In addition, she has problems in comprehending syntax, making a large number of reverse role errors in a sentence–picture matching task (see Table 1), but she is unimpaired on tests of lexical semantics. A CT scan in 1997 showed a hypodense lesion over the left basal ganglia region. An MRI scan in 2000 revealed extensive damage to the left perisylvian region, extending into the left corpus striatum, with preservation of the temporal cortex (see Figures 1 and 2). (5) AS is a right-handed 57-year-old woman. She worked as a housing manager for a mental health unit up to her MCA aneurysm in August 1998. Typical of nonfluent aphasic patients, she makes many reverse role errors on a sentence–picture matching task, suggesting that she has problems with syntactic processing. In contrast, she shows no evidence of a semantic impairment. In addition, although she can repeat single words she has problems repeating sentences (see Table 1). Her spontaneous speech is nonfluent and consists of simple structures. A CT scan taken in September 1998 showed a hypodensity in the left superior temporal lobe and in Broca’s area and surrounding cortex (see Figure 3b). An MR scan was not possible because of the presence of aneurysm clips. Semantically Impaired Patients Four patients with a history of HSE participated in this study: JBR, RC, JH, and WL. The diagnosis of HSE was based on EEG, virological studies, and MR scans that Tyler et al.

83

JBR

RC

JH

WL

DE

CH

Figure 1. This figure shows the status of the left inferior frontal gyrus (BA 44) in the four HSE patients ( JBR, RC, JH, and WL) and two nonfluent patients (DE and CH), with the cross hair indicating the following coordinates on normalized structural MRI scans within 3 mm: JBR: 51.8, 10.3, 10.6; RC: 54.3, 10.3, 10.6; JH: 53.5, 10.3, 10.6; WL: 53.5, 10.3, 10.6; DE: 54.3, 10.3, 10.6; and CH: 51.8, 10, 10.6.

84

Journal of Cognitive Neuroscience

Volume 14, Number 1

JBR

RC

JH

WL

DE

CH

Figure 2. This figure shows the status of the left inferior temporal gyrus (BA 20) in the four HSE patients ( JBR, RC, JH, and WL) and two nonfluent patients (DE and CH), with the cross hair indicating the following coordinates on normalized structural MRI scans within 3 mm: JBR: 38.0, 3.4, 37.6; RC: 38.8, 4.3, 37.6; JH: 38.0, 4.3, 37.6; WL: 38.8, 4.3, 37.6; DE: 38.8, 4.32, 38.5; and CH: 38, 3.4, 37.6.

showed low attenuation in the temporal cortex. All of these patients have been tested extensively to determine the nature and extent of their language impairments. JBR (Bunn, Tyler, & Moss, 1997; Warrington &

Shallice, 1984) and RC have been reported previously (Moss, Tyler, Durrant-Peatfield, & Bunn, 1998) and have well-documented semantic deficits that are summarized in Table 2 (Tests 3–9). In contrast, they have Tyler et al.

85

JG Cerebral Blood Flow

Healthy MRI

AS

MR

Figure 3. (a) A resting PET scan for JG; ( b) CT scans for MR and AS with and without lesion tracing.

no significant syntactic difficulties, as shown by their performance on the sentence–picture matching task using reversible sentences described earlier (Table 2, Test 2). Unlike the nonfluent patients they make very few reverse role errors in this task. The other two patients, WL and JH, who have not yet been reported in the literature, are described below. They too have severe semantic deficits and minor syntactic problems, as shown in Table 2. 86

Journal of Cognitive Neuroscience

JH is a 35-year-old right-handed woman who suffered HSE in 1982 when she was 17 years old. She shows a dense amnesia (e.g., scoring on the fifth percentile for words and third percentile for faces on the Warrington Recognition Memory Test) and a significant semantic deficit (see Tests 3–9 in Table 2). Her syntax appears to be unimpaired as reflected by her fluent, grammatical speech and her normal performance on various syntactic tests. Volume 14, Number 1

WL is a right-handed 68-year-old man who suffered HSE in 1998. He left school at 14 and joined the army, training as a radio operator. He left the army in 1972 and worked in civilian jobs until 1978 when he began training as a priest. He was ordained a vicar in 1980. His encephalitic incident left him with short-term amnesia (Rivermead Behavioural Memory Test standardized score 11, screening score 6) and a semantic deficit. His speech is fluent and he is cognitively relatively unimpaired (e.g., 100% correct on Ravens Progressive Matrices and a forward digit span of 6). All four patients had structural MR scans. In the case of three of the patients, JH, JBR, and RC, these scans were subsequently analyzed by means of voxel-based morphometry, a technique for analyzing changes in gray or white matter density in the brain (Ashburner & Frison, 2000). This analysis showed that the areas of greatest gray matter abnormality were in the lateral temporal cortex, temporal pole, and medial temporal cortex, with some degree of asymmetry across patients. Although all patients had damage to the left temporal lobe, two of them (JH and JBR) also had extensive right temporal damage (Gitelman, Ashburner, Friston, Tyler, & Price, 2000). Comparison of Lesion Sites All of the HSE patients and two of the agrammatic patients (DE and CH) had structural MRI scans. As a way of illustrating the commonalities and differences in the location of their lesions, we compare the six patients with respect to whether they have damage to two critical LH cortical regions that have been implicated in deficits to either regular or irregular past tense processing—Broca’s area (left inferior frontal gyrus; BA 44) and inferior temporal cortex (BA 22). We first compare the patients with respect to the left inferior frontal gyrus (BA 44) by locating the same region (to within 3 mm) on each patient’s scan. These coordinates are presented for different slices in Figure 1 and show that only DE and CH have significant damage in this region. In Figure 2, the coordinates show the location of the left inferior temporal gyrus in each patient (to within 3 mm). All four HSE patients have extensive damage to this region whereas this area remains outside DE and CH’s lesion. The remaining three nonfluent patients ( JG, AS, and MR), could not have MRI scans because of their aneurysm clips. Two of these patients (AS and MR) had CT scans and these are shown in Figure 3a and b. These patients have extensive LH damage primarily located around the left frontal cortex, involving Broca’s area and surrounding regions. The figure also shows a PET scan of JG, illustrating the damage caused by his MCA infarct, involving the left inferior and middle frontal cortex, with relative sparing of the left inferior temporal lobe. The dark areas in the scan indicate regions of

reduced or absent blood flow. A matched series of sections from an intact brain is provided as a guide to interpreting the PET images.

RESULTS AND DISCUSSION Experiment 1: Past Tense Priming Experiment We tested the five nonfluent patients in an intramodal lexical decision priming study, contrasting regular and irregular past tense forms. Prime words, such as the regularly inflected verb ‘‘jumped’’ or the irregularly inflected verb ‘‘sung,’’ were followed by either the verb stem (‘‘jump,’’ ‘‘sing’’) or an unrelated control word. In addition, we included a semantic priming condition in which the prime–target pairs were semantically but not phonologically or morphologically related (‘‘cello–violin’’) to determine whether we could replicate the link between preserved irregular and semantic priming. A fourth condition, in which the prime–target pairs were phonologically but not semantically or morphologically related (‘‘whisky–whisk’’), acted as a control for the effects of phonological overlap. Since the inflected-stem pairs share a great deal of phonological overlap, we need to determine whether phonological effects alone could produce the priming typically seen for morphologically related pairs. Controls Lexical decision errors on real words ranged from 1– 4%, and from 3–7% on nonwords. Each subject’s data was cleaned individually. For each subject, we replaced outliers by the cutoff value (mean ± 2 SDs), and excluded from the analysis any item in which there was not a lexical decision RT in both the test and control conditions. The mean RTs in each condition are given in Table 3. The data from each subject were combined, and the following group ANOVAs were carried out. In the overall ANOVA there was a main effect of priming [F1(1,6) = 160.995, p < .001; F2(1,123) = 179.47, p < .001], and a significant priming by condition interaction [F1(3,18) = 32.663, p < .001; F2(3,123) = 22.448, p < .001]. To explore the source of this interaction, we carried out individual analyses on each condition separately. We first analyzed the semantic control condition. We expected these pairs to prime since many experiments have shown that semantically related words prime robustly (Moss et al., 1995). As predicted, subjects’ RTs were faster in the related than the unrelated condition. This difference of 99 msec was highly significant [F1(1,6) = 131.958, p < .001; F2(1,22) = 47.95, p < .001]. There was also significant positive priming for the regulars [F1(1,6) = 54.103, p < .001; F2(1,40) = 64.063, p < .001] and irregulars [F1(1,6) = 96.72, p < .001; F2(1,40) = 127.756, p < .001]. There was no significant positive priming for the phonologically reTyler et al.

87

Table 3. Experiment 1: Reaction Times (msec) for the Mature Controls and Nonfluent Patients Controls

Patients

Condition

Test

Control

Difference

Test

Control

Difference

Regular past (‘‘baked/bake’’)

730

844

114*

1014

1044

30

Irregular past (‘‘bent/bend’’)

715

867

152*

936

1097

161*

Phonological (‘‘gravy/grave’’)

823

799

24

1070

1098

28

Semantic (‘‘cherry/grape’’)

685

784

99*

828

981

153*

*Indicates significant priming ( p < .05) for that condition.

lated pairs; in the items analysis the difference between test and control was not significant [F2(1,21) = 2.085, p = .163] and in the subjects analysis there was a significant inhibitory effect [F1(1,6) = 7.014, p < .05]. This pattern held for each of the control subjects. We then analyzed the two past tense inflected conditions, with prime type (test/control) as the repeated measures factor and regular/irregular inflection as a between-group factor. There was significantly more priming for the irregulars compared to the regulars [F 1 (1,6) = 201.71 p < .001; F 2 (1,80) = 3.921, p = .051].3 Patients The data from the five nonfluent patients were analyzed as a group. We prepared each subject’s data for analysis by first replacing outliers by the cutoff value (mean ± 2 SDs), and excluding from the analysis any item in which there was no lexical decision RT in both the test and control conditions. The mean RTs in each condition are given in Table 4. The data from each patient were combined, and group ANOVAs were carried out. Lexical decision errors on real words ranged from 0% to 4% and from 5% to 18% on nonwords.

The patients produced very different priming patterns for the regulars and irregulars. The irregulars primed strongly [F1(1,4) = 8.197, p = .04; F2(1,40) = 43.681, p < .001] whereas the regulars did not produce significant priming [F1(1,4) = 5.734, p = .075; F2(1,40) = 2.043, p = .161]. This was reflected in a significant priming by regularity interaction in the items analysis [F2(1,80) = 16.343, p < .001] although this was only marginal in the subjects analysis [F1(1,4) = 5.75, p = .075]. However, each subject’s data when analyzed individually showed this same pattern, demonstrating the consistency of the effect (see Table 4). This cannot be interpreted as a phonological priming effect, with pairs of words that are more phonetically similar producing more priming, since the prime– target regular pairs share more phonological similarity than the irregulars and yet they prime less. Moreover, the phonological control condition did not produce priming (F1, F2 < 1), even though the prime–target pairs were specifically selected to maximize their phonological overlap. Turning to the semantically related prime–target pairs, these produced robust priming [F 1 (1,4) = 8.313, p < .05; F 2 (1,22) = 54.836, p < .001] confirming the semantic priming effects that we have observed in other studies with these patients. These effects are numerically and

Table 4. Experiment 1: Reaction Times (msec) for the Individual Patients Test/Control/ Difference Condition

DE

JG

CH

MR

AS

Regular past (‘‘baked/bake’’)

994/1013/19

859/877/18

1237/1247/9

981/1023/42

1017/1064/46

Irregular past (‘‘bent/bend’’)

925/1303/377*

811/873/62*

1057/1195/138*

897/1042/144*

967/1058/91*

Phonological (‘‘gravy/grave’’)

1259/1128/ 131

890/915/25

1231/1338/107

914/921/7

914/921/7

Semantic (‘‘cherry/grape’’)

721/872/151*

663/733/70*

972/1316/343*

850/904/54*

850/904/54*

Each data set was analyzed individually using ANOVA. *Indicates significant priming ( p < .05) for that condition.

88

Journal of Cognitive Neuroscience

Volume 14, Number 1

Combined Analyses To compare directly the patterns of priming for controls and patients, we carried out further ANOVAs with subject group (controls vs. patients) as a variable. For the regular past tense, there was a significant priming by subject group interaction [F2(1,80) = 10.547, p < .001] reflecting the fact that only the control subjects show priming for the regulars. In contrast, the irregular past tense primed for both subject groups (F < 1 for the interaction). This difference in priming for the two groups as a function of regularity produced a significant three-way interaction between priming, regularity, and subject group [F2(1,160) = 5.942, p < .01]. None of the other contrasts showed significant differences between patients and controls. Experiment 2: Past Tense Elicitation Task The four HSE patients with semantic impairments could not perform the on-line priming study because of difficulties in making on-line lexical decisions, so we developed a past tense elicitation task that they could perform without difficulty. We chose sets of regular and irregular verbs matched on frequency. A two-sentence context was constructed for each verb, where the second sentence was incomplete and required a continuation using a past tense verb. For example: (a) ‘‘My nose sometimes ‘bleeds.’ Last night, it....’’; (b) ‘‘Our children like to ‘climb’ trees. Last week, they....’’ Healthy subjects all continue sentence (a) with ‘‘bled,’’ and (b) with ‘‘climbed.’’ The control subjects made virtually no errors. The HSE patients ranged in accuracy from 83–73% correct overall. They all made more errors on the irregulars than on the regulars [regulars = 92%; irregulars = 65% correct; t(6) = 3.591, p = .011]. The individual patient scores are given in Figure 4. The patients were also tested on the version of the elicitation test described in Ullman et al. (1997) in which subjects are also presented with sentences for completion. In addition to the regular and irregular past tense conditions, there is also a set of nonwords, constructed to be similar phonologically to real regular words. When tested on these materials, the HSE patients produced 80% of the

100 Percent Correct

statistically very similar to the results for the irregular past tense pairs. This group pattern was shown by each of the patients when their data was analyzed individually. The results from this study for DE and JG replicate their performance on a similar but shorter version of this task reported earlier (Marslen-Wilson & Tyler, 1997, 1998). In addition, the present results show that the pattern of priming effects generalizes from these patients to other nonfluent aphasics.

80 Regulars

60

Irregulars

40 20 0 RL

JH

RC

JBR

Figure 4. Past tense elicitation task: Percent correct for the regular and irregulars for each of the HSE patients.

regulars correctly but only 60% of the irregulars and 48% of the nonwords. Thus, in both elicitation tasks the HSE patients were much more accurate at producing the appropriate regular form of the past tense. This pattern is consistent with that of the semantic dementia patient, ES, whose data we reported in our original paper, and who showed significant priming only for the irregular past tense. It is also consistent with the results reported by Patterson, Lambon-Ralph, Hodges, and McClelland (2001), using the elicitation task on a group of semantic dementia patients.

GENERAL DISCUSSION The studies reported here show a distinct pattern of performance on tests that probe processing of the past tense morphology. The nonfluent aphasic patients do not show significant priming for the regular past tense although the irregular past tense primed significantly. Importantly, priming for the irregulars was accompanied by significant priming for the semantically related words, suggesting that these priming patterns are related and subserved by a common component of the language system whose function has not been impaired. In contrast, HSE patients with a severe semantic impairment showed the opposite pattern in an elicitation task: They were more accurate at producing regular past tense forms compared to irregular forms. These behavioral differences, in which poor performance on the irregulars and on semantic tasks go hand-inhand, whereas impaired performance on the regulars is accompanied by intact semantics, are clearly associated with different neuropathologies. The nonfluent patients all suffered a left MCA infarct. This typically causes medial and lateral frontal damage, involving Broca’s area and underlying structures, including the basal ganglia. An MCA infarct can also damage cortical tissue in the superior and middle temporal cortex but rarely affects the inferior temporal lobe. Four of the patients, DE, JG, CH, and MR all had extensive damage to the left frontal cortex, which compromised most, if not all, of Broca’s area. For AS the damage primarily involved the left basal Tyler et al.

89

ganglia. The basal ganglia project not only to the left frontal motor areas but also to Broca’s area (Preuss, 1995; Hoover & Strick, 1993) so that damage to the basal ganglia can compromise the cognitive functions subserved by Broca’s area. Our data show that damage to these frontal complexes produces impairments in processing the regular past tense morphology. Consistent with this account, we have some recent data collected on members of the KE family who suffer from a well-documented neurodevelopmental disorder (Vargha Khadem et al., 1998) that affects some of the same cortical regions which can suffer damage following an MCA infarct. In the affected members of the KE family, structural MRI and PET neuroimaging reveals both structural and functional abnormalities in the caudate nucleus, as well as in several motor-related areas of the frontal cortex, including Broca’s area and the left ventral premotor areas. This leads to the expectation that they will have problems with the regular past tense morphology, similar to aphasic patients who have sustained damage to comparable regions of the frontal cortex and the basal ganglia. Four of the affected family members were tested on the priming materials described earlier, using the same task and procedures, and showed a priming pattern that was similar to that of the nonfluent patients. That is, they showed no significant priming for the regulars, but significant priming for the irregulars and for the semantically related pairs (Watkins, 1999). It is worth noting that the same subjects showed no deficits on tests of syntactic performance, indicating that the underlying deficit is primarily phonological rather than syntactic in character. The four HSE patients, who all show the reverse pattern of better performance on regular past tense forms, have extensive bilateral temporal damage involving the inferior temporal lobes. These areas have been strongly implicated in semantic processing. Neuroimaging studies typically show activation in the inferior temporal lobe when subjects are engaged in semantic tasks (Vanderburghe, Price, Wise, Josephs, & Frackowiack, 1996) and damage to these areas, which occurs as a result of HSE, Pick’s disease, and AD, can result in profound semantic impairments (Hodges & Patterson, 1997; Gainotti, Silveri, Daniele, & Gustolini, 1995). There was no significant damage to the frontal cortical areas in the HSE patients, as is shown in Figures 1 and 2. These patterns of behavioral data and neuropathology suggest that two separable but interdependent systems underlie the processing of regular and irregular past tense forms. What is the nature and function of these different systems? A significant clue is provided by the association of preserved access to irregular forms with preserved semantics, and the converse pattern when semantic function is impaired. The apparently most straightforward interpretation of this is that the relationship between an irregular past tense and its stem (such 90

Journal of Cognitive Neuroscience

as ‘‘gave–give’’) is an essentially semantic relation, analogous to the relation between two different but semantically related words (such as ‘‘swan–goose’’). This would be sharply distinguished from the linguistic and morphological relationship between a regular inflected form and its stem, as in ‘‘jumped–jump.’’ Here it is typically assumed that ‘‘jumped’’ is not a separate word in the mental lexicon, but rather that just the stem ‘‘jump’’ is represented, with the inflected form generated when required. Evidence from research into the intact system, however, clearly rules out a semantic account. For example, in an ERP study (Marslen-Wilson et al., 2000) we compared activation patterns for semantic, regular past tense, and irregular past tense prime–target pairs in a lexical decision task. Contrary to the semantic hypothesis, activation patterns for the irregular pairs were much more similar to the regular pairs than they were to the semantic pairs. Irregular and regular pairs shared a left anterior negativity, typically associated with linguistic processing, which was not elicited at all by the purely semantically related pairs.4 In a further study, also with normal young adults, using the delayed repetition priming task, we also find the irregular past tense/stem pairs aligning with the regular pairs and not with the semantic pairs (Marslen-Wilson & Tyler, 1998). Delayed repetition priming is a task in which subjects make lexical decisions to spoken words that were presented sequentially at intervals of 2–3 sec. Crucially, several words normally intervene between prime and target, so that any priming effect needs to operate over intervals of 35 sec or more. Previous research has shown that purely semantic priming does not operate over these long delays, whereas morphological priming is relatively unaffected. If the relationship between an irregular past tense and its stem is semantic rather than morphological, then it should not survive in the delayed repetition task; the converse prediction follows if the relationship is morphological. What we find is that priming between semantically related pairs (‘‘cello/violin’’) does indeed drop away over these long intervals, but that priming between irregular past tenses and their stems (‘‘found/ find’’) remains robustly significant, and no different from the effect for regular past tenses and their stems (‘‘talked/talk’’). On the basis of these and other results we would argue that the underlying relationship between irregular inflected forms and their stems is indeed morphological rather than semantic.5 The irregular form, like the regular form, maps onto the same underlying stem morpheme, and it is the repeated activation of this stem morpheme that gives rise to priming (Marslen-Wilson & Tyler, 1998). If a semantic account must be rejected, then on what basis can we explain the patterns of association and dissociation between regular and irregular past tenses and tasks such as semantic priming. We believe that the Volume 14, Number 1

most plausible account is in terms of the distinct phonological properties of regular and irregular past tenses in English (Marslen-Wilson & Tyler, 1998). As noted above, we assume that regular inflectional forms are not stored as full forms, but are generated as required, as combinations of stems and inflectional affixes (as in ‘‘talk’’ + ‘‘ed’’). The recognition of these forms requires a complementary process of phonological parsing, where the spoken input is disassembled into stem and affix. In contrast, for monomorphemic forms, like ‘‘house’’ or ‘‘lamp,’’ both production and recognition involve the access of a stored full form, where no process of phonological parsing is required in order to access this representation. The reason this distinction is relevant here is that irregular inflected forms are also stored full-form representations, with no internal phonological structure as combinations of stems and affixes. Access to these representations does not require them to be phonologically parsed.6 This points to an alternative explanation of why irregular primes pattern with semantic primes for the patient populations, and why these can remain intact when regular priming is disrupted. One of the typical features of our nonfluent patients is that they have problems with processes of phonological assembly and disassembly, especially when words combine with syntactic elements, such as tense markers and plurals, to form inflectionally complex words. This would lead to on-line impairments in the processing of regularly inflected words, making these less effective as primes in the auditory–auditory priming tasks. But for inputs that do not require this kind of phonological disassembly, including both monomorphemic words and irregular past tenses, there would be no on-line impairment, so that normal priming effects could emerge, whether based on semantic overlap or on repeated access to the same morpheme. The converse pattern, where regular inflectional morphology remains intact while semantic processes and access to irregular past tense forms is disrupted, can be explained in terms of intact systems supporting phonological parsing and morphosyntactic analysis, providing the listener with a different route for accessing lexical representations. These systems appear to be linked to regions of the posterior and inferior frontal lobe, including Broca’s area and the basal ganglia. When these regions are damaged, as is the case for the nonfluent patients tested here, PD patients (Ullman et al., 1997) and the KE family, the processing of regular past tense forms is disrupted. More generally, it may be possible to link these distinctions in underlying neural and functional architecture to the emerging view that a classical ‘‘dorsal’’/ ‘‘ventral’’ analysis can be applied to the human language system (e.g., Hickok & Poeppel, 2000). The ventral system, on this account, involves temporal lobe structures that mediate access (both phono-

logical and orthographic) to stored lexical representations. The dorsal pathway links to auditory– motor systems in the left inferior parietal and frontal areas, important for the analysis and production of complex phonological sequences. The language-specific properties of the English past tense would therefore map differentially onto these two systems, with irregular forms linking into ventral systems optimized for access to stored whole forms, while regular forms require the involvement of intact frontal systems supporting processes of phonological assembly and disassembly. Thus, even if it remains unclear whether this corresponds to differences in the types of mental computation involved, there can be little doubt that the combined behavioral and neuropathological data do implicate a fundamental duality in the processing of language and speech.

METHODS Experiment 1: Past Tense Priming Experiment Subjects We tested seven mature subjects aged between 56 and 69 years of age as control subjects, and the five nonfluent patients described above. Both the controls and the patients were tested on both versions of the materials, with an interval of a month between each testing session.

Materials and Task We selected 42 prime–target pairs in each of the past tense conditions (regular/irregular) and 24 pairs for the two control conditions (semantic/phonological). In Condition 1 the primes were all regularly inflected part tense forms, whereas they were irregular in Condition 2. The targets in all conditions were the uninflected stem forms, of one or two syllables. Condition 3 consisted of pairs of words that were phonologically but not semantically or morphologically related, and in Condition 4 the pairs were semantically but neither phonologically nor morphologically related. For the semantic condition, we selected pairs of semantically related words that were highly related and of relatively high frequency. Apart from its theoretical importance, this condition also served as a paradigm check, since we had observed significant priming effects in these patients in previous studies. To ensure that the semantically related pairs were highly related we carried out a semantic relatedness pretest in which 15 subjects were asked to rate pairs of words according to their degree of semantic relatedness on a scale of 1–9, where 1 was ‘‘highly unrelated’’ and 9 was ‘‘highly related.’’ The mean semantic relatedness rating for the prime–target pairs was 6. The two past tense sets and the phonological set were matched as closely as possible for test prime, control Tyler et al.

91

Table 5. Experiment 1: Median Frequencies of Stimuli for Priming Study Condition

Number

Prime Frequency

Target Frequency

Control Prime Frequency

Regular past e.g., ‘‘pulled/pull’’

42

13

15

12

Irregular past e.g., ‘‘slept/sleep’’

42

13

12

12

Semantic e.g., ‘‘skirt/dress’’

24

43

72

50

Phonological e.g., ‘‘pillow/pill’’

24

15

20

14

Frequencies from CELEX (Baayen, Piepenbroek, & Guilikers, 1995).

prime, and target frequency and number of syllables (see Table 5). Each prime word was matched to a control word in frequency and number of syllables (see Table 3). In addition to the 132 test pairs, we included 80 real word fillers, making a total of 212 real words. The fillers were an equal mixture of simple primes/past tense targets, simple primes/-s inflected targets, -s inflected primes/ simple targets, simple primes/simple targets, and unambiguous noun pairs. We also included 212 legal nonwords, which were a mixture of phonologically unrelated prime–target pairs, phonologically related pairs with targets made from regularly inflected past or present tense forms, phonologically related pairs with targets made from irregularly inflected forms, and pairs with irregular verbs as primes. We constructed two versions of the material so that the target word only occurred once per version. All of the stimuli were recorded by a female native speaker of British English, digitized at 22 kHz and markers placed at the acoustic onset of each target word. The markers served to trigger a timing device that measured the subject’s reaction time to the target word. Primes and targets were recorded onto a digital audiotape for presentation to subjects. Subjects heard prime–target pairs and were asked to make a lexical decision to the target by pressing a response key.

Experiment 2: Past Tense Elicitation Task Subjects We tested eight mature control subjects, ranging in age from 59 to 71 years, and the four HSE patients. Since each of the agrammatics had a severe output problem, they were unable to perform the task satisfactorily. Materials and Task For this experiment, we selected a set of 26 regular and 27 irregular verbs matched on frequency (regular: median lemma frequency = 9, median wordform frequency = 10; frequency irregular: median lemma frequency = 7, median wordform frequency = 8). A two-sentence con92

Journal of Cognitive Neuroscience

text was constructed for each verb where the second sentence was incomplete and required a continuation using a past tense verb. For example: (a) ‘‘My nose sometimes ‘bleeds.’ Last night, it. . ..’’; (b) ‘‘Our children like to ‘‘climb’’ trees. Last week, they. . ..’’ Healthy subjects all continue sentence (a) with ‘‘bled’’ and (b) with ‘‘climbed.’’ Acknowledgments We thank Mike Cooper for helping with the studies reported here and Dr F. Manes for his assistance with the lesion tracings. This research was supported by an MRC programme grant to L.K.T. and W.M.W. Reprint requests should be sent to Lorraine K. Tyler, Centre for Speech and Language, Department of Experimental Psychology, University of Cambridge, Downing Street, Cambridge CB2 3EB, UK, or via e-mail: [email protected].

Notes 1. The claim that irregulars are accessed and stored as whole forms does not exclude the possibility that subregularities among the set of irregulars may be picked up by the language learner, and can affect performance in some tasks (such as elicited responses to nonce primes). 2. In this study, we also reported data from a third nonfluent patient with bilateral damage. This patient, TS, suffered repeated further hemorrhages and could not be tested further. 3. However, it is worth noting that this is not a reliable interaction across all subject groups. When we tested a group of young controls (undergraduates) they showed the same pattern of significant priming for the semantically related pairs and for both sets of inflected pairs but not for the phonological conditions. In addition, they did not show a significant interaction [Type  Priming: F1(1,18) = 1.614, p = .2201; F2(1,83) = 3.213, p = .077]. 4. Consistent with the account developed here, the ERP patterns for the regulars and irregulars showed some further differentiation, with earlier and stronger temporal activation for the irregulars, and more robust left frontal activation for the regulars. 5. We thank an anonymous reviewer for reminding us that there are additional linguistic reasons for concluding that the irregular pairs are not just semantically related. For example, the phenomenon of ‘‘blocking,’’ where each verb only can have one past tense form, does not apply to pairs of semantically related words (thus, ‘‘came’’ blocks ‘‘comed’’, but ‘‘booted’’ does not block ‘‘kicked’’). Volume 14, Number 1

6. This seems to apply across the board to the English irregulars. Even though there is a subclass of irregular forms (labeled by linguists as ‘‘semiweak’’) where some form of suffix is still present (as in forms like ‘‘kept’’, ‘‘built,’’ etc.), we have found no evidence in earlier research that this leads to differences in access or representation relative to other types of irregulars (Marslen-Wilson et al., 1995).

REFERENCES Ashburner, J., & Frison, K. (2000). Voxel-based morphometry— The methods. Neuroimage, 11, 805–821. Baayen, R. H., Piepenbroek, R., & Guilikers, L. (1995). The CELEX lexical database [CD-ROM]. Philadelphia: Linguistic Data Consortium, University of Pennsylvania. Black, M., Nickels, L., & Byng, S. (1991). Patterns of sentence processing deficit: Processing simple sentences can be a complex matter. Journal of Neurolinguistics, 6, 79–101. Bunn, E., Tyler, L. K., & Moss, H. E. (1997). Category-specific deficits: The role of familiarity re-examined. Neuropsychology, 12, 367–379. Bybee, J. L., & Modor, C. L. (1983). Morphological classes as natural categories. Language, 59, 251–270. Gainotti, G., Silveri, M. C., Daniele, A., & Gustolini, L. (1995). Neuroanatomical correlates of category-specific impairments: A critical survey. Memory, 3/5, 247–64. Gitelman, D., Ashburner, J., Friston, K., Tyler, L. K., & Price, C. (2001). Voxel-based morphometry of herpes simplex encephalitis. Neuroimage, 13, 623–631. Hickok, G., & Poeppel, D. (2000). Towards a functional neuroanatomy of speech perception. Trends in Cognitive Sciences, 4, 131–138. Hodges, J., & Patterson, K. (1997). Semantic memory disorders. Trends in Cognitive Science, 1, 68–72. Hoover, J. E., & Strick, P. L. (1993). Multiple output channels in the basal ganglia. Science, 259, 819–821. Howard. D., & Patterson, K. (1992). The pyramids and palm trees test. Thomas Valley Test Company. Bury St. Edmunds. Joanisse, M. F., & Seidenberg, M. S. (1999). Impairments in verb morphology after brain injury: A connectionist model. Proceedings of the National Academy of Sciences, U.S.A., 96, 7592–7597. Kay, J., Lesser, R., & Coltheart, M. (1992). Psycholinguistic assessments of language processing in aphasia (PALPA). London: Erlbaum. Kempley, S. T., & Morton, J. (1982). The effects of priming with regularly and irregularly related words in auditory word recognition. British Journal of Psychology, 73, 441–454. Marchman, V. A. (1993). Constraints on plasticity in a connectionist model of the English past tense. Journal of Cognitive Neuroscience, 5, 215–234. Marslen-Wilson, W. D., Csibra, G., Ford, M., Hatzakis, H., Gaskell, G., & Johnson, M. H. (2000). Associations and dissociations in the processing of regular and irregular verbs: ERP evidence. Cognitive Neuroscience Conference, San Francisco, April. Marslen-Wilson, W. D., Hare, M., & Older, L. (1993). Inflectional morphology and phonological regularity in the English mental lexicon. In Proceedings of the 15th Annual Meeting of the Cognitive Science Society (p. 945). New Jersey: Erlbaum. Marslen-Wilson, W. D., Hare, M, & Older, L. (1995). Priming and blocking in English inflectional morphology. Paper

presented to the Experimental Psychology Society, London, January 1995. Marslen-Wilson, W. D., & Tyler, L. K. (1997). Dissociating types of mental computation. Nature, 387, 592–594. Marslen-Wilson, W. D., & Tyler L. K. (1998). Rules, representations, and the English past tense. Trends in Cognitive Science, 2, 428–435. Miceli, G., & Caramazza, A. (1988). Dissociation of inflectional and derivational morphology. Brain and Language, 35, 24–65. Moss, H. E., Ostrin, P. K., Tyler, L. K. & Marslen-Wilson, W. D. (1995). Accessing different types of a lexical semantic information—Evidence from priming. Journal of Experimental Psychology: Learning, Memory and Cognition, 21, 863–883. Moss, H. E., & Tyler, L. K. (2000). A progressive categoryspecific semantic deficit for non-living things. Neuropsychologia, 38, 60–82. Moss, H. E., Tyler, L. K., Durrant-Peatfield, M., & Bunn, E. (1998). ‘‘Two eyes of a see-through’’: Impaired and intact semantic knowledge in a case of selective deficit for living things. NeuroCase, 4, 291–310. Ostrin, R. K., & Tyler, L. K. (1995). Dissociations of lexical function: Semantics, syntax and morphology. Cognitive Neuropsychology, 12, 345–391. Patterson, K., Lambon-Ralph, M. A., & McClelland, J. (2001). Deficits in irregular past tense verb morphology associated with degraded semantic knowledge. Neuropsychologia, 39, 709–724. Patterson, K., & Marcel, A. (1977). Aphasia, dyslexia and the phonological coding of written words. Quarterly Journal of Experimental Psychology, 29, 307–318. Plaut, D. C., & Shallice, T. (1993). Perseverative and semantic influences on visual object naming errors in optic aphasia—A connectionist account. Journal of Cognitive Neuroscience, 5, 89–117. Plunkett, K., & Marchman, V. (1991). U-shaped learning and frequency effects in a multilayered perceptron: Implications for child language acquisition. Cognition, 38, 43–102. Prasada, S., & Pinker, S. (1993). Generalization of regular and irregular morphological patterns. Language and Cognitive Processes, 8, 1–56. Preuss, T. (1995). The argument from animals to humans in cognitive neuroscience. In M. Gazzaniga (Ed.), The cognitive neurosciences. Cambridge: MIT Press. Price, C. J., Howard, D., Patterson, K., Warburton, E., Friston, K., & Frackowiack, R. (1998). A functional neuroimaging description of two deep dyslexic patients. Journal of Cognitive Neuroscience, 10, 303–315. Spreen, O., & Benton, A. L. (1969). Neurosensory centre comprehensive examination for aphasia. Victoria, BC: Neuropsychology Laboratory, University of Victoria. Stanners, R., Neiser, J., Hernon, W., & Hall, R. (1979). Memory representation for morphologically related words. Journal of Verbal Learning and Verbal Behavior, 18, 399–412. Tyler, L. K. (1985). Real-time comprehension processes in agrammatism: A case study. Brain and Language, 26, 259–275. Tyler, L. K. (1992). Spoken language comprehension: An experimental approach to normal and disordered processing. Cambridge: MIT Press. Tyler, L. K., Moss, H., & Jennings, F. (1995). Abstract word deficits in aphasia: Evidence from semantic priming. Neuropsychology, 9, 1–10. Tyler, L. K., & Ostrin, R. K. (1994). The processing of simple and complex words in an agrammatic patient: Evidence from priming. Neuropsychologia, 32, 1001–1013. Tyler et al.

93

Ullman, M. T., Corkin, S., Coppola, M., Hickok, G., Growdon, J. H., Koroshetz, W. J., & Pinker, S. (1997). A neural dissociation within language: Evidence that the mental dictionary is part of declarative memory, and that grammatical rules are processed by the procedural system. Journal of Cognitive Neuroscience, 9, 266–276. Vanderburghe, R., Price, C., Wise, R., Josephs, O., & Frackowiack, R. (1996). Functional anatomy of a common semantic system for words and pictures. Nature, 383, 254–6. Vargha Khadem, F., Watkins, K. E., Price, C. J., Ashburner, J.,

94

Journal of Cognitive Neuroscience

Alcock, K. J., Connelly, A., Frackowiak, R. S. J., Friston, K. J., Pembrey, M. E., Mishkin, M., Gadian, D. G., & Passingham, R. E. (1998). Neural basis of an inherited speech and language disorder. Proceedings of the National Academy of Sciences, U.S.A., 95, 12695–12700. Warrington, E. K., & Shallice, T. (1984). Category specific semantic impairments. Brain, 107, 829–854. Watkins, K. (1999). Neuropsychological and neuroimaging investigations of an inherited disorder of speech and language. PhD thesis, University of London.

Volume 14, Number 1