Deficits for Semantics and the Irregular Past TenseA Causal Relationship

Lorraine K Tyler1 Emmanuel A Stamatakis1 Roy W Jones3Peter Bright1 Kadia Acres1 and William D Marslen-Wilson2

Abstract

amp The regular and irregular past tense has become a focus forrecent debates about the structure of the language processingsystem asking whether language functions are subserved bydifferent neural and functional mechanisms or whether allprocesses can be accommodated within a single unified systemA critical claim of leading single mechanism accounts is that therelationship between an irregular stem and its past tense form isprimarily semantic and not morphological in nature Thispredicts an obligatory relationship between semantic perform-ance and access to the irregular past tense such that a semanticdeficit necessarily leads to impairments on the irregulars We

tested this claim in a series of studies probing the comprehen-sion and production of regular and irregular past tense forms infour semantic dementia patients all of whom had profoundsemantic deficits In two elicitation tasks and one auditorypriming study we found that three out of the four patients didnot have a deficit for the irregular past tense in spite of theirsemantic deficits This argues against the view that therelationship between irregular past tense forms and their stemsis primarily semantic and more generally against the singlesystem claim that morphological structure can be capturedsolely based on phonological and semantic relationships amp

INTRODUCTION

English inflectional morphology has been at the heart ofrecent debates about the processes required to produceand comprehend language These debates concern theissue of whether language functions are subserved bydifferent neural and functional systems or whether allprocesses can be accommodated within a single unifiedmechanism (Marslen-Wilson amp Tyler 1998 Pinker 1991Rumelhart amp McClelland 1986) The contrast betweenthe irregular and regular forms of the English past tensehas featured heavily in this debate because of the strongcontrast these offer between a highly rule-like processmdashthe regular past tense formed by adding the affix -d tothe verb stem (as in jumpjumped sighsighed)mdashand theunpredictable and idiosyncratic irregular forms (as inthinkthought makemade) applying to a small minorityof English verbs The critical empirical question has beenwhether these linguistic contrasts map onto qualitativeunderlying differences in the way the two types of formare represented and processed

Neuropsychological studies of patients who have se-lective deficits for either the regular or irregular pasttense in English have provided an important new inputto this debate (eg Tyler de Mornay Davies et al 2002

Tyler Randall amp Marslen-Wilson 2002 Patterson Lam-bon Ralph Hodges amp McClelland 2001 Marslen-Wilsonamp Tyler 1997 Ullman et al 1997) This dissociation hasshown up in several different experimental contextsincluding both priming and elicitation tasks Overallthese studies show that nonfluent aphasic patientswith damage to the left inferior frontal cortex have prob-lems with regularly inflected past tense forms but haveessentially normal performance on irregularly inflectedforms These patients have difficulty in producing regu-larly inflected words in elicitation tasks (Ullman et al1997) and they do not show the normal pattern ofpriming for the regulars in priming studies although theirregulars prime (Tyler Randall et al 2002) This re-flects a general disruption of lexical access from regularinflected forms which fail to show significant semanticpriming effects for pairs like jumpedleap although theuninflected stems (jumpleap) prime normally (Long-worth Marslen-Wilson amp Tyler 2002)

In contrast to this pattern of effects showing selectivedeficits with the regulars a different group of patientstypically with temporal lobe damage but with sparing ofthe left inferior frontal cortex show the opposite pat-tern with better performance on regulars comparedwith irregulars in elicitation and reading tasks (TylerRandall et al 2002 Patterson et al 2001 Marslen-Wilson amp Tyler 1997 1998 Ullman et al 1997) Thislatter group of patients almost always exhibits an ac-companying semantic deficit and a prime issue for the

1University of Cambridge 2MRC Cognition amp Brain SciencesUnit 3Research Institute for Care of the Elderly

D 2004 Massachusetts Institute of Technology Journal of Cognitive Neuroscience 167 pp 1159ndash1172

research reported here is to disentangle cause and effectin the relationship between problems with the irregularsand problems with semantics

Dual mechanism accounts straightforwardly explainthe past tense dissociation in terms of selective damageto distinct underlying systems with some variation in thetypes of mechanism being proposed Several authors(Pinker amp Ullman 2002 Ullman 2001 Ullman et al1997 Prasada amp Pinker 1993 Pinker 1991) claim that theregulars are processed by means of a specific rule-basedsystem that adds and strips away inflectional affixes fromtheir stems In contrast the irregulars are learned indi-vidually by rote and stored in a separate knowledgestore On this account problems with the regulars stemfrom damage to the productive regular rule-based sys-tem while deficits for the irregulars arise as a conse-quence of damage to the store of words in the mentallexicon which leaves rule-based processes intact

We have proposed a modified version of the dualroute account which places less emphasis on the regu-larityirregularity distinction per se and more emphasison the role of specialized morphophonological parsingprocesses which allow the segmentation and identifica-tion of stems and affixes (Tyler de Mornay Davies et al2002 Tyler Randall et al 2002 Marslen-Wilson amp Tyler1998 2003) These processes which are associated withleft hemisphere inferior frontal cortex are required forthe analysis of regularly inflected forms in English withtheir stem + affix structure but do not apply to Englishirregular past tenses that have no overt morphophono-logical structure These irregular forms can be mappedmore directly from phonology to meaning supported byleft temporal lobe structures (cf Tyler de MornayDavies et al 2002 Tyler Randall et al 2002)

The evidence for neuropsychological dissociationpresents a much stronger challenge to single mechanismtheories requiring them to show how apparent dissoci-ations in the effects of damage to the brain couldnonetheless be explained within a single mechanismframework The most systematic riposte along theselines has come from the model proposed by Joanisseand Seidenberg (1999) where a single set of hiddenunits learns the mappings among speech input speechoutput and semantics for sets of regular and irregularpast tense forms and their stems The critical propertyof the model is the differential reliance of the regularsand irregulars on the contribution of phonology andsemantics The mapping between stems and their regu-lar past tense forms is more dependent on phonologicalsimilarities between a stem and its inflected form (egopenopened) whereas the irregulars are more depen-dent on the semantic relationship between a stem andits past tense form since the phonological relationshipbetween them is less predictable (eg thinkthought)

This leads to the prediction that relatively selectivedeficits for the regulars will be caused by a phonologicalimpairment This has a greater impact on the regulars

because the mapping between a regular stem and itspast tense form is primarily phonological In contrastdeficits for the irregulars should be a by-product ofdamage to the semantic system This has a dispropor-tionate effect on the irregulars because the mappingbetween an irregular stem and its past tense form ismore reliant upon semantic similarity To test thesepredictions Joanisse and Seidenberg (1999) lesionedthe semantic layer of their network and found that itgenerated a greater deficit for the irregulars Lesioningthe phonological output layer in contrast produced agreater deficit for the regulars

The critical empirical claim of this type of model lateradopted in modified form by McClelland and Patterson(2002) and Patterson et al (2001) is the causal link itspecifies between phonological factors and regular mor-phology on the one hand and semantic factors andirregular morphology on the other In recent researchinto the first of these claims (Tyler Randall et al 2002)we showed that this link did not hold for phonology Inthe research reported here we examine the predictionthat deficits with the irregulars are caused by difficultieswith semantics

Semantics and the English Irregular Past Tense

There is no doubt that deficits in performance on se-mantic tasks following damage to the brain are closelyassociated with difficulties in either producing or com-prehending English irregular past tense forms In anearly report of a past-tense dissociation (Marslen-Wilsonamp Tyler 1997) we described a stroke patient (T S) withbilateral temporal damage who failed to show primingfor either irregularstem pairs (gavegive) or for seman-tically related pairs (swangoose) although regular pasttense priming (jumpedjump) was intact In a secondreport (Marslen-Wilson amp Tyler 1998) the semanticdementia patient E S with a severe semantic deficitand bilateral temporal lobe damage showed exactly thesame pattern also in a priming task

Subsequent reports primarily using elicitation taskshave shown relatively selective irregular past tensedeficits for larger groups of patients Patterson et al(2001) report greater problems in producing irregularsthan regulars for a group of semantic dementia patientswith some evidence that the extent of the problemincreases with the degree of semantic deficit Tyler deMornay Davies et al (2002) report a similar result for agroup of patients with bilateral temporal lobe damageand semantic deficits following herpes simplex enceph-alitis Finally while this was not directly tested it is likelythat similar patterns can be found among the groups ofpatients reported by Ullman et al (1997)

This consistent pattern of co-occurrence of semanticdeficits and problems with the irregulars naturally raisesthe question of the basis of this regularity Potentialanswers to this are highly dependent on theory In the

1160 Journal of Cognitive Neuroscience Volume 16 Number 7

context of the currently predominant single mechanismconnectionist model (McClelland amp Patterson 2002Joanisse amp Seidenberg 1999) this linkage is necessaryand causal The relationship between irregular pasttense forms and their stems as indicated above isviewed as being primarily a semantic relationship whilethe relationship between a regular form and its stem isnot We specifically tested this possibility in earlierresearch with healthy subjects and found little evidenceto support it In two different types of experimentcomparing priming between irregular and stem pairs(gavegive) regular and stem pairs (calledcall) andsemantically related pairs (celloviolin) we found thatregular and irregular pairs patterned together and werequite distinct from the semantic pairs

A first study Marslen-Wilson and Tyler (1998) usedthe delayed auditory priming task to separate out theeffects of morphology and semantics This is a taskwhere semantic priming drops away sharply over timebut morphological priming does not Healthy subjectswere presented with single words for lexical decisionwith an average of 12 items intervening between eachprime word and its target Over these long lags regu-larly inflected past tense forms should prime because oftheir morphological relationship but purely semantical-ly related pairs should not If irregularly inflected pairsshare a semantic rather than morphological relationshipthen they should also not prime In fact the resultsshowed equally strong priming for both regular andirregular inflected past tense forms at long lags but nosemantic priming We argue that this is because bothtypes of past tense share a morphological relationshipwith their stems with priming being based on thereactivation of the same shared stem morpheme

This analysis was supported by the results of a secondstudy looking at the electrophysiological correlates ofpriming in the normal brain (Marslen-Wilson et al2000) Event-related scalp potentials were measuredwhile subjects were performing a cross-modal primingtask with auditory primes and visual targets comparingregulars irregulars and semantically related pairs Allthree conditions elicited the N400 effect that is typicallyobserved in priming tasks However the regulars andirregulars but not the semantically related pairs alsoexhibited a significant left anterior negativity This is atype of activation that is usually associated with linguisticprocesses As in the behavioral study the pattern ofresponses elicited by the irregulars is much more similarto the pattern elicited by the regulars than it is to thesemantically related pairs and has the characteristics of alinguistic relationship rather than a predominantly se-mantic relationship

If as these results indicate the relationship betweenan irregular form and its stem in the undamaged systemis not primarily semantic in nature then we have tolook for an alternative account both for deficits with theirregulars and for the correlation of these deficits with

semantic impairments Where the irregulars are con-cerned and given the assumption that the primaryaccess process involves mapping onto stored lexicalrepresentations it is uncontroversial to assume thatdeficits here will result from damage to the systemsmediating these access processes (with possible furtherrefinements depending on whether this is access forproduction from a semantic or conceptual startingpoint or access for comprehension from a phonologicalor orthographic starting point)

Where the correlation with semantic impairmentsare concerned there are two possibilities One is thatdisruption of lexical access can itself produce problemsin semantic tasks (and vice versa where lexical access isitself driven by semantics as in elicitation tasks) Thesecond is that damage to areas involved in semanticperformance especially those in the temporal lobesand associated structures may be likely to co-occur oroverlap with the kinds of damage that disrupt lexicalaccess per se in its various forms

What all of these possibilities have in common is theimplication that semantic deficits and problems with theirregulars (and more generally problems in lexical ac-cess) can occur independently There is already evi-dence to support the first half of this predictionmdashnamely that problems in accessing the irregulars canoccur in patients who do not have a semantic deficitMiozzo (2003) has recently reported an anomic patient(A W) who has no semantic deficit as measured on avariety of different tasks and yet has a marked impair-ment for irregularly inflected past tense forms in anelicitation task

The research we report here is targeted at the otherhalf of the predictionmdashthat patients can have a se-mantic deficit but without this impairing their perfor-mance with irregulars To address this question wetested four patients who had been diagnosed as havingsemantic dementia based upon the results of behavioralstudies neuropathology and clinical assessment On avariety of behavioral measures these patients all hadsevere semantic deficits We tested the patients on theirability to comprehend and produce the regular andirregular past tense using elicitation and priming tasksThe results show as predicted that there is not anobligatory relationship between a semantic deficit andproblems with the irregular past tense

Patients General Description Background Testsand Neuropathology

Semantic dementia is the term which is used to describepatients whose primary deficit is a progressive loss ofsemantic knowledge Other aspects of language (egsyntax and phonology) tend to be relatively preservedas are nonsemantic cognitive abilities (Hodges Patter-son Oxbury amp Funnell 1992 Snowden Goulding ampNeary 1989) This pattern of deficit is usually associated

Tyler et al 1161

with focal degeneration of bilateral anterior temporalcortex (Snowden amp Neary 2002 Galton et al 2001)Patients typically present with word-finding problems inthe presence of fluent speech and preserved memoryfor daily and autobiographical events The four patientsdescribed below were diagnosed based on their neuro-pathology and clinical and neuropsychological assess-ment by experienced clinical neurology groups assuffering from semantic dementia

1 B S a right-handed man was 67 years old whentested He left school at 16 When he first presented hehad been experiencing a gradual decline in semanticmemory for around 4 years He initially complained ofdifficulties in recognizing the faces of acquaintancesbut by the time of the study he also experienced fre-quent word-finding difficulties He was anomic in spon-taneous speech on word fluency tasks and on picturenaming and his poor performance on a range of tasksshowed him to have a severe semantic deficit In con-trast he was well oriented in time and space and hadintact visualndashspatial skills nonverbal reasoning abilitiesand memory for recent events His speech was fluentand syntactically well formed He did not makephonological errors in spontaneous speech or picturenaming He had good verbal short-term memory asmeasured by digit span1

2 E K a right-handed woman was 60 years oldwhen first tested She left school when she was 15When she was first assessed her word-finding abilitieshad been progressively deteriorating over the previous5 years She was living alone and doing occasionalcooking and cleaning jobs at the time of the study Shewas severely anomic in spontaneous speech wordfluency tasks and picture naming and performed poorlyon comprehension tests using words and pictures Incontrast she was well oriented in time and place andshe had normal episodic memory visual spatial proces-sing and nonverbal reasoning Her speech was fluentand syntactically well formed despite her anomia andshe did not make phonological errors in her sponta-neous speech or picture naming Her digit span wasnormal ( Jefferies Patterson Jones Bateman amp LambonRalph submitted)

3 J T a right-handed man was 66 years old attesting He left school at 16 He was running a smallfarming business at the time of the study He had beenexperiencing worsening word-finding difficulties for4 years prior to testing Although he was severely im-paired on a range of tests of semantic memory hisvisualndashspatial skills nonverbal reasoning abilities andmemory for recent events were largely intact He did notmake phonological errors in spontaneous speech orpicture naming and his digit span was normal1

4 K B a right-handed woman was 64 years oldwhen tested in the current series of studies She hasbeen described in detail by Snowden and Neary (2002)

When first seen in the clinic her family had noticed overthe previous 3 years that she was experiencing word-finding difficulties and had problems recognizing peo-ple and objects although her language was fluent andgrammatically correct and her numerical skills appearedto be intact She was severely anomic in picture-namingtasks and showed a severe semantic deficit across avariety of tasks She had good verbal short-term me-mory as measured by digit span tests At the time oftesting she lived alone and was able to carry out thenormal range of daily living tasks with the exception ofcooking

We carried out a voxel-based morphometric analysisto obtain an estimate of the extent of neuronal loss inthe patients T1-weighted MR scans were obtained forboth patients and controls and were processed in SPM99software (Wellcome Institute of Cognitive Neurologywwwfilionuclacuk) All images were spatially normal-ized and comparisons were carried out between eachpatient and a group of eight age-matched controls Theimages were initially skull stripped and coregistered tothe SPM T1 template utilizing the mutual informationregistration algorithm They were then spatially normal-ized to Talairach and Tournouxrsquos atlas with 12 para-meter linear affine transformations and were finallysmoothed with a 12-mm isotropic Gaussian filter Thevoxelwise comparisons between patients and controlswere carried out in the context of the general linearmodel (Friston et al 1995) as implemented in SPM99The results from these comparisons were subjected toconjunction analysis (Price amp Friston 1997) the resultsof which are presented (at p lt 001 uncorrected) inFigure 1 (Stamatakis amp Tyler 2003)

Figure 1 shows the extent of abnormal tissue in boththe left and right temporal cortices which is consistentwith lesions generally reported for semantic dementiapatients (Snowden amp Neary 2002 Galton et al 2001)The damage was more substantial on the left (2478voxels) than on the right (2118 voxels) In both hemi-spheres the damage extends inferiorly from the tempo-ral pole and fusiform gyrus to the superior temporalgyrus In the left hemisphere there was more extensiveventricular enlargement than in the right and in theright hemisphere there was damage to parahippocam-pal regions Figure 1 also includes representative slicesfrom each patient showing that they all have a degreeof bilateral anterior temporal damage

The four patients were tested on a variety of testsdesigned to assess their cognitive and language abil-ities Not all patients could be tested on all the tests Toassess their general cognitive abilities we used RavensMatrices the Mini-Mental State Examination (MMSE)or the Addenbrookersquos Cognitive Examination (ACE)(Mathuranath Nestor Berrios Rakowicz amp Hodges2000) The ACE consists of a number of subtests thatevaluate attention memory verbal fluency language

1162 Journal of Cognitive Neuroscience Volume 16 Number 7

research reported here is to disentangle cause and effectin the relationship between problems with the irregularsand problems with semantics

Dual mechanism accounts straightforwardly explainthe past tense dissociation in terms of selective damageto distinct underlying systems with some variation in thetypes of mechanism being proposed Several authors(Pinker amp Ullman 2002 Ullman 2001 Ullman et al1997 Prasada amp Pinker 1993 Pinker 1991) claim that theregulars are processed by means of a specific rule-basedsystem that adds and strips away inflectional affixes fromtheir stems In contrast the irregulars are learned indi-vidually by rote and stored in a separate knowledgestore On this account problems with the regulars stemfrom damage to the productive regular rule-based sys-tem while deficits for the irregulars arise as a conse-quence of damage to the store of words in the mentallexicon which leaves rule-based processes intact

We have proposed a modified version of the dualroute account which places less emphasis on the regu-larityirregularity distinction per se and more emphasison the role of specialized morphophonological parsingprocesses which allow the segmentation and identifica-tion of stems and affixes (Tyler de Mornay Davies et al2002 Tyler Randall et al 2002 Marslen-Wilson amp Tyler1998 2003) These processes which are associated withleft hemisphere inferior frontal cortex are required forthe analysis of regularly inflected forms in English withtheir stem + affix structure but do not apply to Englishirregular past tenses that have no overt morphophono-logical structure These irregular forms can be mappedmore directly from phonology to meaning supported byleft temporal lobe structures (cf Tyler de MornayDavies et al 2002 Tyler Randall et al 2002)

The evidence for neuropsychological dissociationpresents a much stronger challenge to single mechanismtheories requiring them to show how apparent dissoci-ations in the effects of damage to the brain couldnonetheless be explained within a single mechanismframework The most systematic riposte along theselines has come from the model proposed by Joanisseand Seidenberg (1999) where a single set of hiddenunits learns the mappings among speech input speechoutput and semantics for sets of regular and irregularpast tense forms and their stems The critical propertyof the model is the differential reliance of the regularsand irregulars on the contribution of phonology andsemantics The mapping between stems and their regu-lar past tense forms is more dependent on phonologicalsimilarities between a stem and its inflected form (egopenopened) whereas the irregulars are more depen-dent on the semantic relationship between a stem andits past tense form since the phonological relationshipbetween them is less predictable (eg thinkthought)

This leads to the prediction that relatively selectivedeficits for the regulars will be caused by a phonologicalimpairment This has a greater impact on the regulars

because the mapping between a regular stem and itspast tense form is primarily phonological In contrastdeficits for the irregulars should be a by-product ofdamage to the semantic system This has a dispropor-tionate effect on the irregulars because the mappingbetween an irregular stem and its past tense form ismore reliant upon semantic similarity To test thesepredictions Joanisse and Seidenberg (1999) lesionedthe semantic layer of their network and found that itgenerated a greater deficit for the irregulars Lesioningthe phonological output layer in contrast produced agreater deficit for the regulars

The critical empirical claim of this type of model lateradopted in modified form by McClelland and Patterson(2002) and Patterson et al (2001) is the causal link itspecifies between phonological factors and regular mor-phology on the one hand and semantic factors andirregular morphology on the other In recent researchinto the first of these claims (Tyler Randall et al 2002)we showed that this link did not hold for phonology Inthe research reported here we examine the predictionthat deficits with the irregulars are caused by difficultieswith semantics

Semantics and the English Irregular Past Tense

There is no doubt that deficits in performance on se-mantic tasks following damage to the brain are closelyassociated with difficulties in either producing or com-prehending English irregular past tense forms In anearly report of a past-tense dissociation (Marslen-Wilsonamp Tyler 1997) we described a stroke patient (T S) withbilateral temporal damage who failed to show primingfor either irregularstem pairs (gavegive) or for seman-tically related pairs (swangoose) although regular pasttense priming (jumpedjump) was intact In a secondreport (Marslen-Wilson amp Tyler 1998) the semanticdementia patient E S with a severe semantic deficitand bilateral temporal lobe damage showed exactly thesame pattern also in a priming task

Subsequent reports primarily using elicitation taskshave shown relatively selective irregular past tensedeficits for larger groups of patients Patterson et al(2001) report greater problems in producing irregularsthan regulars for a group of semantic dementia patientswith some evidence that the extent of the problemincreases with the degree of semantic deficit Tyler deMornay Davies et al (2002) report a similar result for agroup of patients with bilateral temporal lobe damageand semantic deficits following herpes simplex enceph-alitis Finally while this was not directly tested it is likelythat similar patterns can be found among the groups ofpatients reported by Ullman et al (1997)

This consistent pattern of co-occurrence of semanticdeficits and problems with the irregulars naturally raisesthe question of the basis of this regularity Potentialanswers to this are highly dependent on theory In the

1160 Journal of Cognitive Neuroscience Volume 16 Number 7

context of the currently predominant single mechanismconnectionist model (McClelland amp Patterson 2002Joanisse amp Seidenberg 1999) this linkage is necessaryand causal The relationship between irregular pasttense forms and their stems as indicated above isviewed as being primarily a semantic relationship whilethe relationship between a regular form and its stem isnot We specifically tested this possibility in earlierresearch with healthy subjects and found little evidenceto support it In two different types of experimentcomparing priming between irregular and stem pairs(gavegive) regular and stem pairs (calledcall) andsemantically related pairs (celloviolin) we found thatregular and irregular pairs patterned together and werequite distinct from the semantic pairs

A first study Marslen-Wilson and Tyler (1998) usedthe delayed auditory priming task to separate out theeffects of morphology and semantics This is a taskwhere semantic priming drops away sharply over timebut morphological priming does not Healthy subjectswere presented with single words for lexical decisionwith an average of 12 items intervening between eachprime word and its target Over these long lags regu-larly inflected past tense forms should prime because oftheir morphological relationship but purely semantical-ly related pairs should not If irregularly inflected pairsshare a semantic rather than morphological relationshipthen they should also not prime In fact the resultsshowed equally strong priming for both regular andirregular inflected past tense forms at long lags but nosemantic priming We argue that this is because bothtypes of past tense share a morphological relationshipwith their stems with priming being based on thereactivation of the same shared stem morpheme

This analysis was supported by the results of a secondstudy looking at the electrophysiological correlates ofpriming in the normal brain (Marslen-Wilson et al2000) Event-related scalp potentials were measuredwhile subjects were performing a cross-modal primingtask with auditory primes and visual targets comparingregulars irregulars and semantically related pairs Allthree conditions elicited the N400 effect that is typicallyobserved in priming tasks However the regulars andirregulars but not the semantically related pairs alsoexhibited a significant left anterior negativity This is atype of activation that is usually associated with linguisticprocesses As in the behavioral study the pattern ofresponses elicited by the irregulars is much more similarto the pattern elicited by the regulars than it is to thesemantically related pairs and has the characteristics of alinguistic relationship rather than a predominantly se-mantic relationship

If as these results indicate the relationship betweenan irregular form and its stem in the undamaged systemis not primarily semantic in nature then we have tolook for an alternative account both for deficits with theirregulars and for the correlation of these deficits with

semantic impairments Where the irregulars are con-cerned and given the assumption that the primaryaccess process involves mapping onto stored lexicalrepresentations it is uncontroversial to assume thatdeficits here will result from damage to the systemsmediating these access processes (with possible furtherrefinements depending on whether this is access forproduction from a semantic or conceptual startingpoint or access for comprehension from a phonologicalor orthographic starting point)

Where the correlation with semantic impairmentsare concerned there are two possibilities One is thatdisruption of lexical access can itself produce problemsin semantic tasks (and vice versa where lexical access isitself driven by semantics as in elicitation tasks) Thesecond is that damage to areas involved in semanticperformance especially those in the temporal lobesand associated structures may be likely to co-occur oroverlap with the kinds of damage that disrupt lexicalaccess per se in its various forms

What all of these possibilities have in common is theimplication that semantic deficits and problems with theirregulars (and more generally problems in lexical ac-cess) can occur independently There is already evi-dence to support the first half of this predictionmdashnamely that problems in accessing the irregulars canoccur in patients who do not have a semantic deficitMiozzo (2003) has recently reported an anomic patient(A W) who has no semantic deficit as measured on avariety of different tasks and yet has a marked impair-ment for irregularly inflected past tense forms in anelicitation task

The research we report here is targeted at the otherhalf of the predictionmdashthat patients can have a se-mantic deficit but without this impairing their perfor-mance with irregulars To address this question wetested four patients who had been diagnosed as havingsemantic dementia based upon the results of behavioralstudies neuropathology and clinical assessment On avariety of behavioral measures these patients all hadsevere semantic deficits We tested the patients on theirability to comprehend and produce the regular andirregular past tense using elicitation and priming tasksThe results show as predicted that there is not anobligatory relationship between a semantic deficit andproblems with the irregular past tense

Patients General Description Background Testsand Neuropathology

Semantic dementia is the term which is used to describepatients whose primary deficit is a progressive loss ofsemantic knowledge Other aspects of language (egsyntax and phonology) tend to be relatively preservedas are nonsemantic cognitive abilities (Hodges Patter-son Oxbury amp Funnell 1992 Snowden Goulding ampNeary 1989) This pattern of deficit is usually associated

Tyler et al 1161

with focal degeneration of bilateral anterior temporalcortex (Snowden amp Neary 2002 Galton et al 2001)Patients typically present with word-finding problems inthe presence of fluent speech and preserved memoryfor daily and autobiographical events The four patientsdescribed below were diagnosed based on their neuro-pathology and clinical and neuropsychological assess-ment by experienced clinical neurology groups assuffering from semantic dementia

1 B S a right-handed man was 67 years old whentested He left school at 16 When he first presented hehad been experiencing a gradual decline in semanticmemory for around 4 years He initially complained ofdifficulties in recognizing the faces of acquaintancesbut by the time of the study he also experienced fre-quent word-finding difficulties He was anomic in spon-taneous speech on word fluency tasks and on picturenaming and his poor performance on a range of tasksshowed him to have a severe semantic deficit In con-trast he was well oriented in time and space and hadintact visualndashspatial skills nonverbal reasoning abilitiesand memory for recent events His speech was fluentand syntactically well formed He did not makephonological errors in spontaneous speech or picturenaming He had good verbal short-term memory asmeasured by digit span1

2 E K a right-handed woman was 60 years oldwhen first tested She left school when she was 15When she was first assessed her word-finding abilitieshad been progressively deteriorating over the previous5 years She was living alone and doing occasionalcooking and cleaning jobs at the time of the study Shewas severely anomic in spontaneous speech wordfluency tasks and picture naming and performed poorlyon comprehension tests using words and pictures Incontrast she was well oriented in time and place andshe had normal episodic memory visual spatial proces-sing and nonverbal reasoning Her speech was fluentand syntactically well formed despite her anomia andshe did not make phonological errors in her sponta-neous speech or picture naming Her digit span wasnormal ( Jefferies Patterson Jones Bateman amp LambonRalph submitted)

3 J T a right-handed man was 66 years old attesting He left school at 16 He was running a smallfarming business at the time of the study He had beenexperiencing worsening word-finding difficulties for4 years prior to testing Although he was severely im-paired on a range of tests of semantic memory hisvisualndashspatial skills nonverbal reasoning abilities andmemory for recent events were largely intact He did notmake phonological errors in spontaneous speech orpicture naming and his digit span was normal1

4 K B a right-handed woman was 64 years oldwhen tested in the current series of studies She hasbeen described in detail by Snowden and Neary (2002)

When first seen in the clinic her family had noticed overthe previous 3 years that she was experiencing word-finding difficulties and had problems recognizing peo-ple and objects although her language was fluent andgrammatically correct and her numerical skills appearedto be intact She was severely anomic in picture-namingtasks and showed a severe semantic deficit across avariety of tasks She had good verbal short-term me-mory as measured by digit span tests At the time oftesting she lived alone and was able to carry out thenormal range of daily living tasks with the exception ofcooking

We carried out a voxel-based morphometric analysisto obtain an estimate of the extent of neuronal loss inthe patients T1-weighted MR scans were obtained forboth patients and controls and were processed in SPM99software (Wellcome Institute of Cognitive Neurologywwwfilionuclacuk) All images were spatially normal-ized and comparisons were carried out between eachpatient and a group of eight age-matched controls Theimages were initially skull stripped and coregistered tothe SPM T1 template utilizing the mutual informationregistration algorithm They were then spatially normal-ized to Talairach and Tournouxrsquos atlas with 12 para-meter linear affine transformations and were finallysmoothed with a 12-mm isotropic Gaussian filter Thevoxelwise comparisons between patients and controlswere carried out in the context of the general linearmodel (Friston et al 1995) as implemented in SPM99The results from these comparisons were subjected toconjunction analysis (Price amp Friston 1997) the resultsof which are presented (at p lt 001 uncorrected) inFigure 1 (Stamatakis amp Tyler 2003)

Figure 1 shows the extent of abnormal tissue in boththe left and right temporal cortices which is consistentwith lesions generally reported for semantic dementiapatients (Snowden amp Neary 2002 Galton et al 2001)The damage was more substantial on the left (2478voxels) than on the right (2118 voxels) In both hemi-spheres the damage extends inferiorly from the tempo-ral pole and fusiform gyrus to the superior temporalgyrus In the left hemisphere there was more extensiveventricular enlargement than in the right and in theright hemisphere there was damage to parahippocam-pal regions Figure 1 also includes representative slicesfrom each patient showing that they all have a degreeof bilateral anterior temporal damage

The four patients were tested on a variety of testsdesigned to assess their cognitive and language abil-ities Not all patients could be tested on all the tests Toassess their general cognitive abilities we used RavensMatrices the Mini-Mental State Examination (MMSE)or the Addenbrookersquos Cognitive Examination (ACE)(Mathuranath Nestor Berrios Rakowicz amp Hodges2000) The ACE consists of a number of subtests thatevaluate attention memory verbal fluency language

1162 Journal of Cognitive Neuroscience Volume 16 Number 7

context of the currently predominant single mechanismconnectionist model (McClelland amp Patterson 2002Joanisse amp Seidenberg 1999) this linkage is necessaryand causal The relationship between irregular pasttense forms and their stems as indicated above isviewed as being primarily a semantic relationship whilethe relationship between a regular form and its stem isnot We specifically tested this possibility in earlierresearch with healthy subjects and found little evidenceto support it In two different types of experimentcomparing priming between irregular and stem pairs(gavegive) regular and stem pairs (calledcall) andsemantically related pairs (celloviolin) we found thatregular and irregular pairs patterned together and werequite distinct from the semantic pairs

A first study Marslen-Wilson and Tyler (1998) usedthe delayed auditory priming task to separate out theeffects of morphology and semantics This is a taskwhere semantic priming drops away sharply over timebut morphological priming does not Healthy subjectswere presented with single words for lexical decisionwith an average of 12 items intervening between eachprime word and its target Over these long lags regu-larly inflected past tense forms should prime because oftheir morphological relationship but purely semantical-ly related pairs should not If irregularly inflected pairsshare a semantic rather than morphological relationshipthen they should also not prime In fact the resultsshowed equally strong priming for both regular andirregular inflected past tense forms at long lags but nosemantic priming We argue that this is because bothtypes of past tense share a morphological relationshipwith their stems with priming being based on thereactivation of the same shared stem morpheme

This analysis was supported by the results of a secondstudy looking at the electrophysiological correlates ofpriming in the normal brain (Marslen-Wilson et al2000) Event-related scalp potentials were measuredwhile subjects were performing a cross-modal primingtask with auditory primes and visual targets comparingregulars irregulars and semantically related pairs Allthree conditions elicited the N400 effect that is typicallyobserved in priming tasks However the regulars andirregulars but not the semantically related pairs alsoexhibited a significant left anterior negativity This is atype of activation that is usually associated with linguisticprocesses As in the behavioral study the pattern ofresponses elicited by the irregulars is much more similarto the pattern elicited by the regulars than it is to thesemantically related pairs and has the characteristics of alinguistic relationship rather than a predominantly se-mantic relationship

If as these results indicate the relationship betweenan irregular form and its stem in the undamaged systemis not primarily semantic in nature then we have tolook for an alternative account both for deficits with theirregulars and for the correlation of these deficits with

semantic impairments Where the irregulars are con-cerned and given the assumption that the primaryaccess process involves mapping onto stored lexicalrepresentations it is uncontroversial to assume thatdeficits here will result from damage to the systemsmediating these access processes (with possible furtherrefinements depending on whether this is access forproduction from a semantic or conceptual startingpoint or access for comprehension from a phonologicalor orthographic starting point)

Where the correlation with semantic impairmentsare concerned there are two possibilities One is thatdisruption of lexical access can itself produce problemsin semantic tasks (and vice versa where lexical access isitself driven by semantics as in elicitation tasks) Thesecond is that damage to areas involved in semanticperformance especially those in the temporal lobesand associated structures may be likely to co-occur oroverlap with the kinds of damage that disrupt lexicalaccess per se in its various forms

What all of these possibilities have in common is theimplication that semantic deficits and problems with theirregulars (and more generally problems in lexical ac-cess) can occur independently There is already evi-dence to support the first half of this predictionmdashnamely that problems in accessing the irregulars canoccur in patients who do not have a semantic deficitMiozzo (2003) has recently reported an anomic patient(A W) who has no semantic deficit as measured on avariety of different tasks and yet has a marked impair-ment for irregularly inflected past tense forms in anelicitation task

The research we report here is targeted at the otherhalf of the predictionmdashthat patients can have a se-mantic deficit but without this impairing their perfor-mance with irregulars To address this question wetested four patients who had been diagnosed as havingsemantic dementia based upon the results of behavioralstudies neuropathology and clinical assessment On avariety of behavioral measures these patients all hadsevere semantic deficits We tested the patients on theirability to comprehend and produce the regular andirregular past tense using elicitation and priming tasksThe results show as predicted that there is not anobligatory relationship between a semantic deficit andproblems with the irregular past tense

Patients General Description Background Testsand Neuropathology

Semantic dementia is the term which is used to describepatients whose primary deficit is a progressive loss ofsemantic knowledge Other aspects of language (egsyntax and phonology) tend to be relatively preservedas are nonsemantic cognitive abilities (Hodges Patter-son Oxbury amp Funnell 1992 Snowden Goulding ampNeary 1989) This pattern of deficit is usually associated

Tyler et al 1161

with focal degeneration of bilateral anterior temporalcortex (Snowden amp Neary 2002 Galton et al 2001)Patients typically present with word-finding problems inthe presence of fluent speech and preserved memoryfor daily and autobiographical events The four patientsdescribed below were diagnosed based on their neuro-pathology and clinical and neuropsychological assess-ment by experienced clinical neurology groups assuffering from semantic dementia

1 B S a right-handed man was 67 years old whentested He left school at 16 When he first presented hehad been experiencing a gradual decline in semanticmemory for around 4 years He initially complained ofdifficulties in recognizing the faces of acquaintancesbut by the time of the study he also experienced fre-quent word-finding difficulties He was anomic in spon-taneous speech on word fluency tasks and on picturenaming and his poor performance on a range of tasksshowed him to have a severe semantic deficit In con-trast he was well oriented in time and space and hadintact visualndashspatial skills nonverbal reasoning abilitiesand memory for recent events His speech was fluentand syntactically well formed He did not makephonological errors in spontaneous speech or picturenaming He had good verbal short-term memory asmeasured by digit span1

2 E K a right-handed woman was 60 years oldwhen first tested She left school when she was 15When she was first assessed her word-finding abilitieshad been progressively deteriorating over the previous5 years She was living alone and doing occasionalcooking and cleaning jobs at the time of the study Shewas severely anomic in spontaneous speech wordfluency tasks and picture naming and performed poorlyon comprehension tests using words and pictures Incontrast she was well oriented in time and place andshe had normal episodic memory visual spatial proces-sing and nonverbal reasoning Her speech was fluentand syntactically well formed despite her anomia andshe did not make phonological errors in her sponta-neous speech or picture naming Her digit span wasnormal ( Jefferies Patterson Jones Bateman amp LambonRalph submitted)

3 J T a right-handed man was 66 years old attesting He left school at 16 He was running a smallfarming business at the time of the study He had beenexperiencing worsening word-finding difficulties for4 years prior to testing Although he was severely im-paired on a range of tests of semantic memory hisvisualndashspatial skills nonverbal reasoning abilities andmemory for recent events were largely intact He did notmake phonological errors in spontaneous speech orpicture naming and his digit span was normal1

4 K B a right-handed woman was 64 years oldwhen tested in the current series of studies She hasbeen described in detail by Snowden and Neary (2002)

When first seen in the clinic her family had noticed overthe previous 3 years that she was experiencing word-finding difficulties and had problems recognizing peo-ple and objects although her language was fluent andgrammatically correct and her numerical skills appearedto be intact She was severely anomic in picture-namingtasks and showed a severe semantic deficit across avariety of tasks She had good verbal short-term me-mory as measured by digit span tests At the time oftesting she lived alone and was able to carry out thenormal range of daily living tasks with the exception ofcooking

We carried out a voxel-based morphometric analysisto obtain an estimate of the extent of neuronal loss inthe patients T1-weighted MR scans were obtained forboth patients and controls and were processed in SPM99software (Wellcome Institute of Cognitive Neurologywwwfilionuclacuk) All images were spatially normal-ized and comparisons were carried out between eachpatient and a group of eight age-matched controls Theimages were initially skull stripped and coregistered tothe SPM T1 template utilizing the mutual informationregistration algorithm They were then spatially normal-ized to Talairach and Tournouxrsquos atlas with 12 para-meter linear affine transformations and were finallysmoothed with a 12-mm isotropic Gaussian filter Thevoxelwise comparisons between patients and controlswere carried out in the context of the general linearmodel (Friston et al 1995) as implemented in SPM99The results from these comparisons were subjected toconjunction analysis (Price amp Friston 1997) the resultsof which are presented (at p lt 001 uncorrected) inFigure 1 (Stamatakis amp Tyler 2003)

Figure 1 shows the extent of abnormal tissue in boththe left and right temporal cortices which is consistentwith lesions generally reported for semantic dementiapatients (Snowden amp Neary 2002 Galton et al 2001)The damage was more substantial on the left (2478voxels) than on the right (2118 voxels) In both hemi-spheres the damage extends inferiorly from the tempo-ral pole and fusiform gyrus to the superior temporalgyrus In the left hemisphere there was more extensiveventricular enlargement than in the right and in theright hemisphere there was damage to parahippocam-pal regions Figure 1 also includes representative slicesfrom each patient showing that they all have a degreeof bilateral anterior temporal damage

The four patients were tested on a variety of testsdesigned to assess their cognitive and language abil-ities Not all patients could be tested on all the tests Toassess their general cognitive abilities we used RavensMatrices the Mini-Mental State Examination (MMSE)or the Addenbrookersquos Cognitive Examination (ACE)(Mathuranath Nestor Berrios Rakowicz amp Hodges2000) The ACE consists of a number of subtests thatevaluate attention memory verbal fluency language

1162 Journal of Cognitive Neuroscience Volume 16 Number 7

with focal degeneration of bilateral anterior temporalcortex (Snowden amp Neary 2002 Galton et al 2001)Patients typically present with word-finding problems inthe presence of fluent speech and preserved memoryfor daily and autobiographical events The four patientsdescribed below were diagnosed based on their neuro-pathology and clinical and neuropsychological assess-ment by experienced clinical neurology groups assuffering from semantic dementia

1 B S a right-handed man was 67 years old whentested He left school at 16 When he first presented hehad been experiencing a gradual decline in semanticmemory for around 4 years He initially complained ofdifficulties in recognizing the faces of acquaintancesbut by the time of the study he also experienced fre-quent word-finding difficulties He was anomic in spon-taneous speech on word fluency tasks and on picturenaming and his poor performance on a range of tasksshowed him to have a severe semantic deficit In con-trast he was well oriented in time and space and hadintact visualndashspatial skills nonverbal reasoning abilitiesand memory for recent events His speech was fluentand syntactically well formed He did not makephonological errors in spontaneous speech or picturenaming He had good verbal short-term memory asmeasured by digit span1

2 E K a right-handed woman was 60 years oldwhen first tested She left school when she was 15When she was first assessed her word-finding abilitieshad been progressively deteriorating over the previous5 years She was living alone and doing occasionalcooking and cleaning jobs at the time of the study Shewas severely anomic in spontaneous speech wordfluency tasks and picture naming and performed poorlyon comprehension tests using words and pictures Incontrast she was well oriented in time and place andshe had normal episodic memory visual spatial proces-sing and nonverbal reasoning Her speech was fluentand syntactically well formed despite her anomia andshe did not make phonological errors in her sponta-neous speech or picture naming Her digit span wasnormal ( Jefferies Patterson Jones Bateman amp LambonRalph submitted)

3 J T a right-handed man was 66 years old attesting He left school at 16 He was running a smallfarming business at the time of the study He had beenexperiencing worsening word-finding difficulties for4 years prior to testing Although he was severely im-paired on a range of tests of semantic memory hisvisualndashspatial skills nonverbal reasoning abilities andmemory for recent events were largely intact He did notmake phonological errors in spontaneous speech orpicture naming and his digit span was normal1

4 K B a right-handed woman was 64 years oldwhen tested in the current series of studies She hasbeen described in detail by Snowden and Neary (2002)

When first seen in the clinic her family had noticed overthe previous 3 years that she was experiencing word-finding difficulties and had problems recognizing peo-ple and objects although her language was fluent andgrammatically correct and her numerical skills appearedto be intact She was severely anomic in picture-namingtasks and showed a severe semantic deficit across avariety of tasks She had good verbal short-term me-mory as measured by digit span tests At the time oftesting she lived alone and was able to carry out thenormal range of daily living tasks with the exception ofcooking

We carried out a voxel-based morphometric analysisto obtain an estimate of the extent of neuronal loss inthe patients T1-weighted MR scans were obtained forboth patients and controls and were processed in SPM99software (Wellcome Institute of Cognitive Neurologywwwfilionuclacuk) All images were spatially normal-ized and comparisons were carried out between eachpatient and a group of eight age-matched controls Theimages were initially skull stripped and coregistered tothe SPM T1 template utilizing the mutual informationregistration algorithm They were then spatially normal-ized to Talairach and Tournouxrsquos atlas with 12 para-meter linear affine transformations and were finallysmoothed with a 12-mm isotropic Gaussian filter Thevoxelwise comparisons between patients and controlswere carried out in the context of the general linearmodel (Friston et al 1995) as implemented in SPM99The results from these comparisons were subjected toconjunction analysis (Price amp Friston 1997) the resultsof which are presented (at p lt 001 uncorrected) inFigure 1 (Stamatakis amp Tyler 2003)

Figure 1 shows the extent of abnormal tissue in boththe left and right temporal cortices which is consistentwith lesions generally reported for semantic dementiapatients (Snowden amp Neary 2002 Galton et al 2001)The damage was more substantial on the left (2478voxels) than on the right (2118 voxels) In both hemi-spheres the damage extends inferiorly from the tempo-ral pole and fusiform gyrus to the superior temporalgyrus In the left hemisphere there was more extensiveventricular enlargement than in the right and in theright hemisphere there was damage to parahippocam-pal regions Figure 1 also includes representative slicesfrom each patient showing that they all have a degreeof bilateral anterior temporal damage

The four patients were tested on a variety of testsdesigned to assess their cognitive and language abil-ities Not all patients could be tested on all the tests Toassess their general cognitive abilities we used RavensMatrices the Mini-Mental State Examination (MMSE)or the Addenbrookersquos Cognitive Examination (ACE)(Mathuranath Nestor Berrios Rakowicz amp Hodges2000) The ACE consists of a number of subtests thatevaluate attention memory verbal fluency language

1162 Journal of Cognitive Neuroscience Volume 16 Number 7

and visuospatial skills It is designed to provide an es-timate of the degree to which a patient is cognitively im-paired to be able to differentiate between patients withfronto-temporal dementia (AD these patients have lan-guage deficits in the presence of generally preserved cog-nitive abilities) and patients suffering from Alzheimerrsquosdementia (who have more general cognitive deficits) Ascore of less than 22 indicates fronto-temporal dementiaand a score of greater than 32 indicates AD

We also tested the patients on a variety of languagetests examining their phonology syntax semantics andnaming (see Table 1) The patients were clearly severelyimpaired on picture naming (Test 1) and they showed apattern of surface dyslexia (Test 10) consistent with pre-vious descriptions of semantic dementia patients (Snow-den amp Neary 2002 Galton et al 2001) To test thepatientsrsquo phonological processing we used a variety ofnormed tests from the Psycholinguistic Assessment forLanguage Perception of Aphasics (PALPA) (Kay Lesseramp Coltheart 1992) As the scores on Table 1 show thepatients were relatively unimpaired on all the phonolog-ical tests (Tests 9ndash12) They also showed minimal im-

pairment on a sentencendashpicture-matching task designedto probe syntactic processing (Test 8) In this test thesubjects hear a spoken sentence that they are asked tocompare against a set of three black-and-white line draw-ings The sentences are all semantically lsquolsquoreversiblersquorsquo inthat each actor can potentially function as the agent ofthe action (eg lsquolsquothe woman visits the doctorrsquorsquo) and themain verb in each sentence had a preference to take anoun phrase argument as determined by pretests Halfthe sentences are simple active constructions and halfuse a passive construction One picture correctly de-picted the activity described by the sentence one ofthe foils consisted of a lexical distracter whereas thesecond foil consisted of a reverse role distracter wherethe agent of the action becomes the recipient of theaction (eg in which the doctor visits the woman) Thepatients made only one or two errors on this test andwere within the range of normal controls

In contrast they were clearly impaired on tests oftheir semantic abilities In a wordndashpicture-matching testin which they had to match a word to a picture of anobject from the same semantic category the patients

Figure 1 (A) Result of a conjunction analysis showing regions of significant difference between patients and controls The regions are

superimposed on a normal T1 image displayed in neurological convention (R is Right) and the values on the coronal slices represent Talairach y

coordinates (B) Regions of significant difference between individual patients and controls Sagittal slices are shown at Talairach x = 32 (L) andx = 32 (R) for each patient

Tyler et al 1163

were impaired although control subjects are typically atceiling (Test 2) Similarly they were poor on a propertyverification study on which they had to make a verifica-tion judgment to statements about common objects(eg lsquolsquodoes a cat have whiskersrsquorsquo Test 7) The state-ments were of two types they probed general proper-ties of objects (eg lsquolsquodoes a camel have legsrsquorsquo) anddistinctive properties (eg lsquolsquodoes a camel have ahumprsquorsquo) The patients showed preserved knowledgeof the shared properties of objects but a significantimpairment of distinctive properties a pattern that wehave observed in other patients with semantic deficits(Moss amp Tyler 2000) The patients were also tested onboth versions of the Pyramids and Palm Trees test one

involving written words and the other involving pictures(Howard amp Patterson 1992) They were all significantlyimpaired compared with healthy controls

Finally we tested the patients on three tests of regularand irregular past tense processing Two of these testsrequired the patients to produce regularly and irregu-larly inflected past tense forms when they occurred inspoken sentences while the third experiment testedtheir ability to comprehend the spoken forms of regu-lar and irregular past tense forms and their stems Thislatter test required the patients to make speeded lexicaldecision responses to pairs of spoken words and onlythree of the patients could reliably perform the task Wealso report longitudinal data from one of the patients

Table 1 Neuropsychological and Language Test Scores

Control Norms B S E K (Time 1) E K (Time 2) J T K B

General Cognitive Tests

1 ACE 065 119 183 177 NT

2 Ravens Matrices (max 36) 32 30 26 36 15

3 MMSE (max 30) 27 27 NT 23 11

4 Digit span forward (max 14) 6 6 6 8 9

Language Tests

1 Naming (Bunn Tyler amp Moss 1998)a( correct) 95 38 33 21 11 17

2 Within-category wordndashpicture match ( correct) 100 85 93 88 77 66

3 Category fluencyb na 7 7 4 1 10

4 Letter fluencyc 385 14 7 8 5 8

5 Pyramids and Palm Trees (words max 52)( correct)

gt90 67 69 67 60 60

6 Pyramids and Palm Trees (pictures max 52)( correct)

gt90 63 67 58 67 67

7 Property verification test (Moss amp Tyler 2000)( correct)

Shared 97 94 92 83 70

Distinctive 96 73 67 61 59

8 Syntax spoken sentencendashpicture match ( correct) 99 97 94 91 91 94

9 PALPA word rhyme judgmentspoken version( correct)

na 98 95 88 92 92

10 PALPA word rhyme judgmentswritten version na 72 62 55 59

11 PALPA phonological segmentation of initial sounds( correct)

97 100 100 100 85

12 PALPA phonological segmentation of final sounds( correct)

97 93 90 93 97

na = not applicableaA subset of eight categories four living and four nonliving matched on relevant variables (see Bunn et al 1998) Total = 70 picturesbScore averaged over four categories (animals vehicles tools and fruitsvegetables)cScore averaged over three letters (F A and S)

1164 Journal of Cognitive Neuroscience Volume 16 Number 7

(E K) whose performance we were able to monitorover the course of 14 months on all three tests

RESULTS

Past Tense Elicitation Tasks

The control subjects made virtually no errors on theelicitation task developed at the Centre for Speech andLanguage (CSL) (Table 2) Three out of four patientsalso showed no significant differences in their ability toproduce either regular or irregular past tense formsB S and K B correctly inflected 95 of the regularsand 100 of the irregulars Although E K correctlyinflected slightly more of the regulars (86) than theirregulars (82) when first tested in January 2002 thisdifference was not significant (x2 = 001 p gt 05)When we tested her a year later there was no differ-ence at all between the irregulars and regulars withE K correctly inflecting 86 of each Only J T showeda significant advantage for the regulars correctly in-flecting 100 of them compared with only 55 of theirregulars (x2 = 1048 p lt 05) Given that most of thepatients made so few errors on inflecting the irregularsit was not possible to carry out an analysis of the errortypes Of the 16 errors on the irregulars 14 of theseinvolved pure regularization (eg creep creeped)and 2 involved partial regularization (eg deal

dealded) Both single- and dual-mechanism approacheswould predict this outcome

In further analyses we compared high- and low-frequency words in the two conditions to test a keyclaim of the single mechanism account that patientswith semantic deficits will have particular difficulty withthe low-frequency irregulars (Patterson et al 2001) Thisis claimed to arise because low-frequency irregularsdepend more upon semantics to achieve the mappingbetween present and past tense than high-frequencyirregulars For the frequency analysis we divided eachset of words into two bandsmdashlow and high frequencies(see Table 3) As Table 2 shows the patients were notdifferentially impaired on the low- relative to high-fre-quency irregulars Both K B and B S only made oneerror in the entire test and this was on a high-frequencyregular word In January 2002 E K did show a frequen-cy effect having more difficulty correctly inflecting low-compared with high-frequency words However this didnot interact with regularity she correctly inflected 73of the low-frequency regulars and 67 of the low-frequency irregulars This difference was not significant(x2 = 018 p gt 05) A year later she no longer showeda frequency effect J T was the only patient who clearlyhad more difficulty correctly inflecting low-frequencyirregular words Sixty-eight percent of the high-frequen-cy irregulars were correctly inflected but only 32 of thelow-frequency irregulars

The pattern of results on the CSL elicitation test wasbroadly consistent with those for the Ullman elicitationtask (see Table 2) When first tested E K correctlyinflected 90 of the irregular past tense forms 75 ofthe regulars and 70 of the nonwords The differencebetween the regulars and irregulars was not significant(x2 = 06926 p gt 05) When tested a year later thepattern was slightly but not significantly different Shecorrectly inflected 90 of the regulars 80 of theirregulars and 100 of the nonwords but none of thesedifferences were significant (all x2 gt 005) Similarly B Sshowed no significant difference in accuracy betweenthe regulars and irregulars correctly inflecting 100 ofthe regulars 95 of the irregulars and 100 of thenonwords In contrast J T was more accurate on theregulars (95) and the nonwords (100) than the ir-regulars (75) The difference between the regulars andirregulars although substantial did not reach signifi-cance (x2 = 314 p gt 05)

Intramodal Priming Task

Controls

Twelve healthy controls were tested in this study six oneach version Analyses of their reaction times (RTs)showed a significant main effect of priming F1(110) =835 p lt 001 F2(1124) = 1281 p lt 001 and a maineffect of condition F1(330) = 396 plt 001 F2(3124) =825 p lt 001 due to differences in overall RTs in the

Table 2 Past Tense Elicitation ( Correct)

B S E K(Time 1)

E K(Time 2)

J T K B

CSL Past Tense Elicitation Task

Regulars 95 86 86 100 95

Irregulars 100 82 86 55 100

CSL Past Tense Elicitation Task

Regulars

HF (n = 11) 93 100 91 100 93

LF (n = 11) 100 73 82 100 100

Irregulars

HF (n = 13) 100 85 85 68 100

LF (n = 9) 100 67 89 32 100

Ullman Past Tense Elicitation Task

Regulars 100 75 90 95 nt

Irregulars 95 90 80 75

Nonwords 100 70 100 100

CSL = Centre for Speech Language

Tyler et al 1165

four conditions with RTs in the semantic conditionbeing fastest (944 msec) and RTs in the phonologicalcondition being slowest (1091 msec) More importantlythere was a significant Priming Condition interactionF1(330) = 130 p lt 001 F2(3124) = 224 p lt 001 Toprobe the source of this interaction we analyzed eachcondition separately This analysis showed a significantpriming effect of 101 msec for the regulars F1(110) =384 p lt 001 F2(140) = 400 p lt 001 171 msec forthe irregulars F1(110) = 664 p lt 001 F2(140) =1173 p lt 001 and 109 msec for the semanticallyrelated pairs F1(110) = 138 p lt 01 F2(122) =377 p lt 001 and no priming for the phonologicallyrelated pairs (47 msec) F1(110) = 229 p = 162F2(122) = 258 p = 123 Moreover the controlsshowed significantly more priming for the irregularscompared with the semantic condition F2(162) =5596 p = 021 The mean RTs and standard errors aredisplayed in Figure 2 This pattern of results for thehealthy controls is identical to that produced by adifferent group of controls in a similar earlier experi-ments (Tyler et al 2002)

Patients

Wewere only able to test E K B S and J T in this studyas K B was not available for testing All three patientsproduced mean RTs that were within the normal range(E K at first testing = 1006 msec E K at second testing= 1041 msec B S = 1152 msec J T = 1099 msec)

(Figure 3) We carried out an ANOVA on the combineddata from B S J T and E K and found a main effect ofpriming F2(1123) = 9397 p lt 001 due to faster RTsin the related (mean RT = 1023 msec) compared withthe unrelated condition (mean RT = 1194 msec) and asignificant Priming Condition interaction F2(3123) =6423 p = 001 due to faster RTs in the semantic andirregular conditions (1063 and 1052 msec respectively)compared with the regular (1135 msec) and the phono-logical condition (1185msec) This was very similar to thepattern shown by the control subjects

We then analyzed each condition separately andfound that the priming effect of 283 msec for theirregulars was significant F2(140) = 84176 p lt 001as was the 104-msec priming effect for the semanticcondition F2(122) = 12673 p = 002 The regularsalso primed significantly (mean priming = 185 msec)F2(140) = 20027 p lt 001 whereas the phonologicallyrelated items did not (mean priming = 115 msec)F2(122) = 3120 p = 092 In the direct comparisonbetween the semantic and irregular conditions thepatients showed a significant interaction F2(162) =155 p lt 001 because just like the normal subjectsthey showed more priming for the irregulars Howeverthe patients showed a greater difference in the amountof priming for the irregulars compared with the seman-tic condition (179 msec as opposed to 62 msec forthe controls) In an analysis comparing the primingeffects for patients and controls in the semantic andirregular conditions we found a significant interaction

Table 3 Properties of Stimuli for CSL Past Tense Elicitation Task

Median WordForm Frequency

Median LemmaFrequency

MedianImageability

Low-frequency regulars (n = 11) eg blessblessed 10 28 493

Low-frequency irregulars (n = 9) eg bleedbled 10 20 486

High-frequency regulars (n = 11) eg shareshared 64 179 424

High-frequency irregulars (n = 13) eg growgrew 75 183 435

Figure 2 Auditoryndash

auditory priming study

Mean RTs and standard

errors for 12 controlvolunteers

1166 Journal of Cognitive Neuroscience Volume 16 Number 7

F(164) = 464 p = 035 This interaction is shown inFigure 4

The group pattern was essentially repeated wheneach patientrsquos data were analyzed separately with onequalificationmdashthe semantic priming effect did not reachsignificance for any of the patients In the first testingperiod E K showed a main effect of priming F(1110) =6574 p lt 001 and a significant Priming Conditioninteraction F(3110) = 408 p = 009 Although theirregulars primed robustly (333 msec) F(137) = 4460p lt 001 the priming effect of 100 msec for the seman-tically related pairs was not significant F(154) = 249p = 120 A year later when tested a second time sheshowed a significant overall priming effect F(1115) =9193 p lt 001 and a significant Priming Conditioninteraction F(3115) = 625 p= 001 The irregulars still

primed strongly (346 msec) F(138) = 8229 p lt 001but the semantically related pairs did not prime signifi-cantly (104msec) F(121) = 234 p= 141 Similarly B Sshowed a main effect of priming F(1115) = 56042 p lt001 and a Priming Condition interaction F(3115) =2885 p = 039 The irregulars primed strongly(243 msec) F(140) = 38941 p lt 001 in the absenceof significant semantic priming (105 msec) F(121) =2717 p= 12 J T also showed a main effect of primingF(197) = 7138 p lt 01 and a Priming Conditioninteraction F(197) = 482 p lt 01 The irregularsproduced robust priming (177 msec) F(136) = 1135p lt 01 while the semantic conditions produced amarginally significant priming effect (80 msec) F 117) =442 p = 051

The failure of semantic priming effects to reachsignificance for the patients individually as well as thestatistically larger difference between semantic and ir-regular priming for the patients compared with thecontrols indicates a significant impairment in semanticpriming for this population In contrast their primingfor the irregulars seemed to be robustly unimpaired andsignificant for each patient individually

DISCUSSION

The aim of this research was to determine whetherpatients with semantic deficits always have associatedproblems in processing irregularly inflected forms aspredicted by current single mechanism accounts ofneuropsychological double dissociations between En-glish regular and irregular past tenses To address thisissue we tested semantic dementia patients on tasksthat probe the production and comprehension of regu-lar and irregularly inflected past tense forms We testedthe patientsrsquo abilities to produce correctly inflected pasttense forms from their present tense variants in twoelicitation experiments In both studies three of the

Figure 3 Auditoryndashauditory

priming study Mean RTs and

standard errors for the patients

(group data for J T E K andB S)

Figure 4 Auditoryndashauditory priming study Priming effects across

groups for the irregular and semantic conditions

Tyler et al 1167

patients were able to correctly inflect both the regularand irregular past tense Only one patient J T hadsignificantly more difficulty in inflecting the irregularscompared with the regulars This set of results suggeststhat semantic deficits do not necessarily go togetherwith difficulties in producing irregularly inflected pasttense forms

A possible argument against this interpretation of datais that the three patients who did not have difficultiesproducing irregularly inflected words may have mildersemantic deficits than the single patient J T who didhave problems with the irregulars However the datafrom the semantic tests do not suggest that J T wasmore severely semantically impaired than the otherthree patients On the Pyramids and Palm Trees testusing either written words or pictures the patients allscored poorly making between 31 and 42 errorsJ Trsquos error rate was 33 Similarly on a wordndashpicturematching task his error rate (23) was within the rangeof the errors produced by the other patients (7ndash34)

Turning to an assessment of the patientsrsquo ability tocomprehend past tense forms we found that the irreg-ulars primed robustly whether we analyzed the patientsas a group or individually suggesting that their ability toaccess the irregulars from a phonological input was notimpaired even for the patient ( J T) who had difficultiesaccessing irregulars in a production task Indeed theoverall group pattern of priming results looks remark-ably normal Like the controls the patient groupshowed priming for both types of past tense and forsemantically related words while phonologically relatedwords did not prime significantly Moreover like thecontrols the patients showed significantly greater prim-ing for the irregulars compared with the regulars

We have previously interpreted priming between averb form such as bring and its irregularly inflected pasttense form (ie brought) as evidence that the relation-ship between them is morphological rather than seman-tic (Marslen-Wilson amp Tyler 1998 2003) Both forms ofthe verb map onto the same underlying stem mor-pheme just as for the regulars and it is the repeatedactivation of this morpheme that gives rise to primingeffects (both at short and long priming delays) On thisaccount the basis for priming between irregularly in-flected words is different from the processes that un-derpin priming between two semantically related butmorphologically unrelated words (eg coatdress) Inthe latter case semantically related words prime be-cause of semantic overlap between two separate lexicalor morphemic representations This predicts that se-mantic deficits and deficits for the irregular past tensewill not necessarily pattern togethermdashwhich is thepattern of results that we obtain here for the semanticdementia patients

To account for neuropsychological dissociations inwhich patients are disproportionately impaired on theregular past tense we have focused on differences in

phonological processes involved in parsing an inflectedform into its stem and affix We assume that regularlyinflected past tense forms are generated as combina-tions of stems and inflectional affixes (open + ed) andrecognized by a process of phonological parsing wherethe speech input is segmented into a stem and its affixProblems with the regulars arise when this phonologi-cal parsing mechanism is impaired usually followingdamage to the left inferior frontal gyrus (LIFG) andassociated regions (Tyler de Mornay Davies et al2002 Tyler Randall et al 2002) Irregularly inflectedforms are stored just like monomorphemic words suchas table or grass as full-form representations with nointernal phonological structure as combinations ofstems and affixes These words do not need to bephonologically parsed into stems and affixes as part ofthe on-line access process These differences in require-ments for phonological parsing accompanied by dam-age to neural systems implicated in these types ofanalysis gives rise to selective deficits for the regularpast tense

Turning to deficits for the irregulars we assume (asnoted above) that an irregular past tense form has astored lexical form representation linking to the sameunderlying morpheme as other forms of the same wordThis predicts that deficits for the irregulars can arise ifbrain damage either affects stored lexical representa-tions directly or disrupts the mechanisms involved inthe process of mapping from phonology to semantics(Marslen-Wilson amp Tyler 1998) A version of this pro-posal is put forward by Miozzo (2003) in the context ofthe anomic patient A W who had problems with theirregulars in the presence of intact semanticsmdashtheopposite pattern to the one reported here for threeout of our four semantically impaired patients Miozzosuggests that this patient typical of anomic patients ingeneral has impairments in lexical processing asevidenced by her difficulty in retrieving a wordrsquos pho-nology On the hypothesis that irregular past tenseforms must be accessed as whole forms so that alterna-tive access routes involving decompositional mecha-nisms cannot be invoked problems in accessing storedlexical representations should differentially disadvantagethe irregulars compared with the regulars

There are some hints in the priming data that J Tthe one patient who showed difficulty in producingirregularly inflected words may also have phonologicalproblems He made many lexical decision errors (31errors) on the items in the phonological condition andmore than in any other condition (11 in the regular5 in the irregular 12 in the semantic conditions)The other patients did not show this pattern B Smade 8 4 0 and 2 errors in the phonologicalregular irregular and semantic conditions respective-ly E K at Time 1 made 8 8 3 and 2 errorson the phonological regular irregular and semanticconditions respectively and 4 6 1 and 2

1168 Journal of Cognitive Neuroscience Volume 16 Number 7

respectively errors in those conditions at the secondtime of testing Similarly J T had a high error rate(30) on the phonologically related foils in the exper-iment such as gossipgoss These data however areonly suggestive Further research is needed to deter-mine the nature of this potential phonological deficitand whether there is a more general relationshipbetween phonological processing deficits and prob-lems in accessing stored lexical forms

One issue that is raised by these results is why thepatients as a group show any semantic priming at allwhen they each have pronounced semantic deficitsThis is not the first time that semantic dementiapatients have been observed to show semantic primingwhile their performance on standard off-line testsshows them to be profoundly impaired We have pre-viously reported semantic priming in three semanticdementia patients (A M Tyler amp Moss 1998 F MTyler Moss Patterson amp Hodges 1997 P P MossTyler Hodges amp Patterson 1995) Given that thesepatients had been unambiguously diagnosed as having asemantic deficit we had not initially predicted that theywould show priming for semantically related words Thefact that priming can still be obtained albeit at asignificantly reduced level suggests that it is not anall-or-none phenomenon Significant priming effectsmay not require the activation of all of the informationassociated with a concept All that may be required isthat the semantic relationship between the two relatedwords (eg onioncarrot) should be stronger to somecriterial degree than between the two unrelated words(eg deskcarrot) This would allow semantic primingeffects even in a damaged system so long as somesemantic information is preserved especially if it isthe shared properties of concepts that are preservedsince it is these that provide the basis for semanticoverlap between related concepts

It has been independently demonstrated that theshared properties of concepts are indeed preservedfollowing brain damage These properties are morehighly correlated with each other and correlated prop-erties are relatively preserved following damage (Moss ampTyler 2000 Tyler et al 2000 Tyler amp Moss 2001 DevlinGonnerman Andersen amp Seidenberg 1998 McRae deSa amp Seidenberg 1997) Moreover studies with patientswho have semantic deficits have shown that the sharedproperties of concepts tend to be preserved while thedistinctive properties tend to be impaired (Tyler amp Moss2001 Moss Tyler Durrant-Peatfield amp Bunn 1998) Thepatients in the current study show the same pattern Ina verification task they were impaired at verifying thedistinctive properties of concepts while relatively pre-served on their shared properties (Table 1) If as arguedabove it is the coactivation of these shared propertiesthat primarily drives priming in the intact system thenthis would explain the preservation of some degree ofsemantic priming in the semantic dementia population

In summary E K and B S did not show any differ-ential impairment for the irregulars compared with theregulars on either elicitation or priming tasks and K Bdid not show a disproportionate deficit for the irregularscompared with the regulars in the elicitation task OnlyJ T showed an association of semantic deficit and prob-lems with the irregulars Moreover when the patientsdid make errors on the irregulars they did not exhibitthe pattern of effects predicted by the single mechanismaccount (Patterson et al 2001) with a higher proportionof errors on low-frequency items These results showthat there is no necessary association between a seman-tic deficit and problems with the irregular past tenseAlthough the two often do co-occur as we and othershave previously documented they can evidently alsooccur independently of each other This dissociationindicates that a separate set of functional capacities isimplicated here supporting the access of stored lexicalrepresentations An important issue for future researchis to identify the neural infrastructure for these accessprocesses so that we can better determine the condi-tions under which semantic deficits will co-occur withdeficits for the irregulars

The present results together with the results fornormal subjects addressed earlier (Marslen-Wilson ampTyler 1998) argue against the claim that the relation-ship between irregular past tense forms and their stemsis primarily a semantic rather than a morphologicalrelationship This in turn undermines the evidencefor the type of single system account of lexical repre-sentation in which morphological structure can becaptured solely on the basis of phonological and seman-tic relationships (Patterson et al 2001 Joanisse amp Seid-enberg 1999)

METHODS

Subjects

In addition to the four semantic dementia patients agroup of seven control subjects (aged between 56 and69 years) were tested on the elicitation task and a groupof 12 subjects (aged 60ndash73 years) were tested on thepriming study

Past Tense Production Elicitation Studies

We tested the patients on two elicitation experimentsdesigned to probe their ability to generate past tenseforms from their stems In the first elicitation testdescribed in Ullman et al (1997) subjects were pre-sented with sentences containing the present tenseforms of regular and irregular words together with aset of nonwords for past-tense generation We used thistest to directly compare our patients with others re-ported in the literature who have deficits in generatingpast tense forms

Tyler et al 1169

For the second elicitation task we selected a set of22 regular and 22 irregular verbs matched as closely aspossible on form and lemma frequency (Baayen Piepen-brock amp Gulikers 1995) and imageability (Coltheart1981) The median word form frequency was 35 for theregulars and 36 for the irregulars The median lemmafrequency was 92 for the regulars and 110 for theirregulars The median imageability was 485 for theregulars and 470 for the irregulars A two-sentencecontext was created for each verb where the secondsentence was incomplete and required a verb inflectedfor the past tense For example for the sentence lsquolsquoMynose sometimes bleeds Last night it rsquorsquo the correctresponse would be bled The test stimuli were precededby five practice trials The sentences were read aloud tothe subjects who were encouraged to produce a pasttense form appropriate for the sentence

Past Tense Comprehension Auditory Priming Study

To test patientsrsquo abilities to comprehend spoken pasttense forms we designed an intramodal auditoryndashauditory priming study contrasting the regulars andirregulars We chose to use a priming study to avoidsome of the pitfalls associated with testing patients ontasks that make heavy metalinguistic demands In thisexperiment target words which were verb stems (suchas pull sing) were preceded by a regularly or irregu-larly inflected past tense prime (pulled sung) or by anunrelated control word In addition to the two mainexperimental conditions we included two other con-ditions The first was a semantic priming condition inwhich the prime-target pairs were semantically but notphonologically or morphologically related (pighorse)We included this to determine whether the patientswould show evidence of a semantic deficit in the sametype of on-line task as used to assess their performanceon the past tense forms The fourth condition con-sisted of a phonological priming condition in whichthe prime-target pairs were phonologically but notsemantically or morphologically related (paintpain)which acted as a control for the effects of phonologicalsimilarity Since the inflected-stem pairs share a greatdeal of phonological overlap we need to be ableto separate out phonological from morphological prim-ing effects

Materials

We selected 42 prime-target pairs in each of the pasttense conditions (regularirregular) and 24 pairs for the2 control conditions (semanticphonological) In boththe regular and irregular past tense conditions theprimes were all inflected past tense forms (eg fixedglanced stuck brought) and the targets consisted oftheir stems all of which were one or two syllables ( fixglance stick bring) Items in the phonological control

condition consisted of morphologically simple monosyl-labic pairs (eg lamplamb plumpplum) that werenot semantically related The proportion of consonant-vowel-consonant (CVC) and consonant-vowel (CV) over-lap between prime and target was matched to theregular past tense pairs In all of the conditions weavoided homophones The two past tense sets and thephonological set were matched as closely as possible fortest prime control prime and target frequency andnumber of syllables (see Table 4) The fourth conditionconsisted of pairs of semantically related words thatwere either phonologically or morphologically related(eg shorecoast alleylane) For this condition weselected pairs of semantically related words that werehighly semantically related (but not associatively) and ofrelatively high frequency This was to ensure significantsemantic priming effects in control subjects to provide abasis for evaluating semantic priming in the patients Asemantic relatedness pretest was carried out in which15 subjects were asked to rate pairs of words accordingto their degree of semantic relatedness on a scale of 1ndash9where 1 was highly unrelated and 9 was highly relatedThe mean semantic relatedness rating for the prime-target pairs was 7 (SD = 102)

Each prime word was matched to a control word infrequency and number of syllables (see Table 4) Inaddition to the 132 test pairs we included 80 real wordfiller pairs making a total of 212 real word pairs Thefillers were an equal mixture of unrelated simpleprimespast tense targets simple primes(-s) inflectedtargetsx (-s) inflected primessimple targets simpleprimessimple targets and unambiguous noun pairsWe also included 148 legal wordnonword pairs whichwere a mixture of phonologically unrelated prime-targetpairs phonologically related pairs with targets derivedfrom regularly inflected past or present tense formsphonologically related pairs with targets made fromirregularly inflected forms and pairs with irregular verbsas primes

Design and Procedure

Test and filler items were pseudorandomly dispersedthroughout the test list We constructed two lists ofstimuli so that the target word only occurred once per

Table 4 Intramodal Priming Study Median Frequencies ofStimuli

Test Prime Target Control Prime

Regulars 55 6 4

Irregulars 6 7 5

Phonological 14 18 13

Semantic 16 10 15

1170 Journal of Cognitive Neuroscience Volume 16 Number 7

version All of the stimuli were recorded by a femalenative speaker of British English digitized at 22 kHz andmarkers placed at the acoustic onset of each targetword The markers triggered a timing device that mea-sured the subjectrsquos reaction time to the target wordPrimes and targets were recorded onto a DAT tape forpresentation to subjects Subjects heard primendashtargetpairs and were asked to make a lexical decision to thetarget by pressing a response key

Acknowledgments

We thank Matt Lambon-Ralph Beth Jeffries and KaralynPatterson for referring patients E K B S and J T and formaking many test scores available to us and to Julie Snowdenfor allowing us to work with K B We thank Lara Brent and BilliRandall for their help with the analyses and the patients fortheir cooperation This work was supported by an MRC (UK)program grant to L K T

Reprint requests should be sent to Lorraine K Tyler Depart-ment of Experimental Psychology University of CambridgeCambridge CB3 ODX UK or via e-mail lktylercslpsycholcamacuk

Note

1 These data were kindly provided by Elizabeth Jeffries

REFERENCES

Baayen R H Piepenbrock R amp Gulikers L (1995) TheCELEX lexical database Linguistic Data ConsortiumPhiladelphia University of Pennsylvania

Coltheart M (1981) The MRC psycholinguistic databaseQuarterly Journal of Experimental Psychology HumanExperimental Psychology 33A 497ndash505

Bunn E M Tyler L K amp Moss H E (1998) Naming picturesThe role of familiarity and property type reexaminedNeuropsychology 12 367ndash379

Devlin J T Gonnerman L M Andersen E S amp SeidenbergM S (1998) Category specific semantic deficits in focaland widespread brain damage A computational accountJournal of Cognitive Neuroscience 10 77ndash94

Friston K J Holmes A P Poline J-B Grasby B J WilliamsC R Frackowiak R S J amp Turner R (1995) Analysis offMRI time-series revisited Neuroscience 2 45ndash53

Galton C J Patterson K Graham K Lambon-Ralph M AWilliams G Antoun N Sahakian B J amp Hodges J R(2001) Differing patterns of temporal atrophy inAlzheimerrsquos disease and semantic dementia Neurology 57216ndash225

Hodges J R Patterson K Oxbury S amp Funnell E (1992)Semantic dementia Progressive fluent aphasia withtemporal lobe atrophy Brain 115 1783ndash1806

Howard D amp Patterson K (1992) The pyramids and palmtrees test Bury St Edmunds Thames Valley Test

Jefferies E Patterson K Jones R W Bateman D amp LambonRalph M A (submitted) A category-specific advantage fornumbers in verbal short-term memory Evidence fromsemantic dementia

Joanisse M F amp Seidenberg M S (1999) Impairments inverb morphology after brain injury A connectionist model

Proceedings of the National Academy of Sciences USA 967592ndash7597

Kay J Lesser R amp Coltheart M (1992) Psycholinguisticassessments of language processing in aphasia (PALPA)London Erlbaum

Longworth C E Marslen-Wilson W D amp Tyler L K(2002) Non-fluent aphasics show an abnormal time courseof semantic activation from the regular past tense[abstract] Journal of Cognitive Neuroscience 14Supplement 55

Marslen-Wilson W D Csibra G Ford M Hatzakis HGaskell G amp Johnson M H (2000) Associations anddissociations in the processing of regular and irregularverbs Electrophysiological evidence Journal of CognitiveNeuroscience 12 Supplement 55E

Marslen-Wilson W D amp Tyler L K (1997) Dissociating typesof mental computation Nature 387 592ndash594

Marslen-Wilson W D amp Tyler L K (1998) Rulesrepresentations and the English past tense Trends inCognitive Science 2 428ndash435

Marslen-Wilson W D amp Tyler L K (2003) Capturingunderlying differentiation in the human language systemTrends in Cognitive Sciences 7 62ndash63

Mathuranath P S Nestor P J Berrios G E Rakowicz Wamp Hodges J (2000) A brief cognitive test battery todifferentiate Alzheimerrsquos disease and frontotemporaldementia Neurology 55 1613ndash1620

McClelland J L amp Patterson K (2002) Rules or connectionsin past-tense inflections What does the evidence rule outTrends in Cognitive Sciences 6 465ndash472

McRae K de Sa V amp Seidenberg M S (1997) On thenature and scope of featural representations of wordmeaning Journal of Experimental Psychology General126 99ndash130

Miozzo M (2003) On the processing of regular and irregularforms of verbs and nouns Evidence from neuropsychologyCognition 87 101ndash127

Moss H E amp Tyler L K (2000) A progressivecategory-specific semantic deficit for non-living thingsNeuropsychologia 38 60ndash82

Moss H E Tyler L K Durrant-Peatfield M amp Bunn E(1998) Two eyes of a see-through Impaired and intactsemantic knowledge in a case of selective deficit for livingthings Neurocase 4 291ndash310

Moss H E Tyler L K Hodges J R amp Patterson K (1995)Exploring the loss of semantic memory in semanticdementia Evidence from a primed monitoring studyNeuropsychology 91 16ndash26

Patterson K Lambon Ralph M A Hodges J Ramp McClelland J L (2001) Deficits in irregular past-tenseverb morphology associated with degraded semanticknowledge Neuropsychologia 39 709ndash724

Pinker S (1991) Rules of language Science 253 530ndash533Pinker S amp Ullman M T (2002) The past and future of thepast tense Trends in Cognitive Sciences 6 456ndash463

Prasada S amp Pinker S (1993) Generalizations of regular andirregular morphological patterns Language and CognitiveProcesses 8 1ndash56

Price C J amp Friston K J (1997) Cognitive conjunction A newapproach to brain activation experiments Neuroimage 5261ndash270

Rumelhart D E amp McClelland J L (1986) On learningthe past tense of English verbs In J L McClelland amp D ERumelhart (Eds) Parallel distributed processing(pp 217ndash270) Cambridge MIT Press

Snowden J S Goulding P J amp Neary D (1989) Semanticdementia A form of circumscribed cerebral atrophyBehavioral Neurology 2 167ndash182

Tyler et al 1171

were impaired although control subjects are typically atceiling (Test 2) Similarly they were poor on a propertyverification study on which they had to make a verifica-tion judgment to statements about common objects(eg lsquolsquodoes a cat have whiskersrsquorsquo Test 7) The state-ments were of two types they probed general proper-ties of objects (eg lsquolsquodoes a camel have legsrsquorsquo) anddistinctive properties (eg lsquolsquodoes a camel have ahumprsquorsquo) The patients showed preserved knowledgeof the shared properties of objects but a significantimpairment of distinctive properties a pattern that wehave observed in other patients with semantic deficits(Moss amp Tyler 2000) The patients were also tested onboth versions of the Pyramids and Palm Trees test one

involving written words and the other involving pictures(Howard amp Patterson 1992) They were all significantlyimpaired compared with healthy controls

Finally we tested the patients on three tests of regularand irregular past tense processing Two of these testsrequired the patients to produce regularly and irregu-larly inflected past tense forms when they occurred inspoken sentences while the third experiment testedtheir ability to comprehend the spoken forms of regu-lar and irregular past tense forms and their stems Thislatter test required the patients to make speeded lexicaldecision responses to pairs of spoken words and onlythree of the patients could reliably perform the task Wealso report longitudinal data from one of the patients

Table 1 Neuropsychological and Language Test Scores

Control Norms B S E K (Time 1) E K (Time 2) J T K B

General Cognitive Tests

1 ACE 065 119 183 177 NT

2 Ravens Matrices (max 36) 32 30 26 36 15

3 MMSE (max 30) 27 27 NT 23 11

4 Digit span forward (max 14) 6 6 6 8 9

Language Tests

1 Naming (Bunn Tyler amp Moss 1998)a( correct) 95 38 33 21 11 17

2 Within-category wordndashpicture match ( correct) 100 85 93 88 77 66

3 Category fluencyb na 7 7 4 1 10

4 Letter fluencyc 385 14 7 8 5 8

5 Pyramids and Palm Trees (words max 52)( correct)

gt90 67 69 67 60 60

6 Pyramids and Palm Trees (pictures max 52)( correct)

gt90 63 67 58 67 67

7 Property verification test (Moss amp Tyler 2000)( correct)

Shared 97 94 92 83 70

Distinctive 96 73 67 61 59

8 Syntax spoken sentencendashpicture match ( correct) 99 97 94 91 91 94

9 PALPA word rhyme judgmentspoken version( correct)

na 98 95 88 92 92

10 PALPA word rhyme judgmentswritten version na 72 62 55 59

11 PALPA phonological segmentation of initial sounds( correct)

97 100 100 100 85

12 PALPA phonological segmentation of final sounds( correct)

97 93 90 93 97

na = not applicableaA subset of eight categories four living and four nonliving matched on relevant variables (see Bunn et al 1998) Total = 70 picturesbScore averaged over four categories (animals vehicles tools and fruitsvegetables)cScore averaged over three letters (F A and S)

1164 Journal of Cognitive Neuroscience Volume 16 Number 7

(E K) whose performance we were able to monitorover the course of 14 months on all three tests

RESULTS

Past Tense Elicitation Tasks

The control subjects made virtually no errors on theelicitation task developed at the Centre for Speech andLanguage (CSL) (Table 2) Three out of four patientsalso showed no significant differences in their ability toproduce either regular or irregular past tense formsB S and K B correctly inflected 95 of the regularsand 100 of the irregulars Although E K correctlyinflected slightly more of the regulars (86) than theirregulars (82) when first tested in January 2002 thisdifference was not significant (x2 = 001 p gt 05)When we tested her a year later there was no differ-ence at all between the irregulars and regulars withE K correctly inflecting 86 of each Only J T showeda significant advantage for the regulars correctly in-flecting 100 of them compared with only 55 of theirregulars (x2 = 1048 p lt 05) Given that most of thepatients made so few errors on inflecting the irregularsit was not possible to carry out an analysis of the errortypes Of the 16 errors on the irregulars 14 of theseinvolved pure regularization (eg creep creeped)and 2 involved partial regularization (eg deal

dealded) Both single- and dual-mechanism approacheswould predict this outcome

In further analyses we compared high- and low-frequency words in the two conditions to test a keyclaim of the single mechanism account that patientswith semantic deficits will have particular difficulty withthe low-frequency irregulars (Patterson et al 2001) Thisis claimed to arise because low-frequency irregularsdepend more upon semantics to achieve the mappingbetween present and past tense than high-frequencyirregulars For the frequency analysis we divided eachset of words into two bandsmdashlow and high frequencies(see Table 3) As Table 2 shows the patients were notdifferentially impaired on the low- relative to high-fre-quency irregulars Both K B and B S only made oneerror in the entire test and this was on a high-frequencyregular word In January 2002 E K did show a frequen-cy effect having more difficulty correctly inflecting low-compared with high-frequency words However this didnot interact with regularity she correctly inflected 73of the low-frequency regulars and 67 of the low-frequency irregulars This difference was not significant(x2 = 018 p gt 05) A year later she no longer showeda frequency effect J T was the only patient who clearlyhad more difficulty correctly inflecting low-frequencyirregular words Sixty-eight percent of the high-frequen-cy irregulars were correctly inflected but only 32 of thelow-frequency irregulars

The pattern of results on the CSL elicitation test wasbroadly consistent with those for the Ullman elicitationtask (see Table 2) When first tested E K correctlyinflected 90 of the irregular past tense forms 75 ofthe regulars and 70 of the nonwords The differencebetween the regulars and irregulars was not significant(x2 = 06926 p gt 05) When tested a year later thepattern was slightly but not significantly different Shecorrectly inflected 90 of the regulars 80 of theirregulars and 100 of the nonwords but none of thesedifferences were significant (all x2 gt 005) Similarly B Sshowed no significant difference in accuracy betweenthe regulars and irregulars correctly inflecting 100 ofthe regulars 95 of the irregulars and 100 of thenonwords In contrast J T was more accurate on theregulars (95) and the nonwords (100) than the ir-regulars (75) The difference between the regulars andirregulars although substantial did not reach signifi-cance (x2 = 314 p gt 05)

Intramodal Priming Task

Controls

Twelve healthy controls were tested in this study six oneach version Analyses of their reaction times (RTs)showed a significant main effect of priming F1(110) =835 p lt 001 F2(1124) = 1281 p lt 001 and a maineffect of condition F1(330) = 396 plt 001 F2(3124) =825 p lt 001 due to differences in overall RTs in the

Table 2 Past Tense Elicitation ( Correct)

B S E K(Time 1)

E K(Time 2)

J T K B

CSL Past Tense Elicitation Task

Regulars 95 86 86 100 95

Irregulars 100 82 86 55 100

CSL Past Tense Elicitation Task

Regulars

HF (n = 11) 93 100 91 100 93

LF (n = 11) 100 73 82 100 100

Irregulars

HF (n = 13) 100 85 85 68 100

LF (n = 9) 100 67 89 32 100

Ullman Past Tense Elicitation Task

Regulars 100 75 90 95 nt

Irregulars 95 90 80 75

Nonwords 100 70 100 100

CSL = Centre for Speech Language

Tyler et al 1165

four conditions with RTs in the semantic conditionbeing fastest (944 msec) and RTs in the phonologicalcondition being slowest (1091 msec) More importantlythere was a significant Priming Condition interactionF1(330) = 130 p lt 001 F2(3124) = 224 p lt 001 Toprobe the source of this interaction we analyzed eachcondition separately This analysis showed a significantpriming effect of 101 msec for the regulars F1(110) =384 p lt 001 F2(140) = 400 p lt 001 171 msec forthe irregulars F1(110) = 664 p lt 001 F2(140) =1173 p lt 001 and 109 msec for the semanticallyrelated pairs F1(110) = 138 p lt 01 F2(122) =377 p lt 001 and no priming for the phonologicallyrelated pairs (47 msec) F1(110) = 229 p = 162F2(122) = 258 p = 123 Moreover the controlsshowed significantly more priming for the irregularscompared with the semantic condition F2(162) =5596 p = 021 The mean RTs and standard errors aredisplayed in Figure 2 This pattern of results for thehealthy controls is identical to that produced by adifferent group of controls in a similar earlier experi-ments (Tyler et al 2002)

Patients

Wewere only able to test E K B S and J T in this studyas K B was not available for testing All three patientsproduced mean RTs that were within the normal range(E K at first testing = 1006 msec E K at second testing= 1041 msec B S = 1152 msec J T = 1099 msec)

(Figure 3) We carried out an ANOVA on the combineddata from B S J T and E K and found a main effect ofpriming F2(1123) = 9397 p lt 001 due to faster RTsin the related (mean RT = 1023 msec) compared withthe unrelated condition (mean RT = 1194 msec) and asignificant Priming Condition interaction F2(3123) =6423 p = 001 due to faster RTs in the semantic andirregular conditions (1063 and 1052 msec respectively)compared with the regular (1135 msec) and the phono-logical condition (1185msec) This was very similar to thepattern shown by the control subjects

We then analyzed each condition separately andfound that the priming effect of 283 msec for theirregulars was significant F2(140) = 84176 p lt 001as was the 104-msec priming effect for the semanticcondition F2(122) = 12673 p = 002 The regularsalso primed significantly (mean priming = 185 msec)F2(140) = 20027 p lt 001 whereas the phonologicallyrelated items did not (mean priming = 115 msec)F2(122) = 3120 p = 092 In the direct comparisonbetween the semantic and irregular conditions thepatients showed a significant interaction F2(162) =155 p lt 001 because just like the normal subjectsthey showed more priming for the irregulars Howeverthe patients showed a greater difference in the amountof priming for the irregulars compared with the seman-tic condition (179 msec as opposed to 62 msec forthe controls) In an analysis comparing the primingeffects for patients and controls in the semantic andirregular conditions we found a significant interaction

Table 3 Properties of Stimuli for CSL Past Tense Elicitation Task

Median WordForm Frequency

Median LemmaFrequency

MedianImageability

Low-frequency regulars (n = 11) eg blessblessed 10 28 493

Low-frequency irregulars (n = 9) eg bleedbled 10 20 486

High-frequency regulars (n = 11) eg shareshared 64 179 424

High-frequency irregulars (n = 13) eg growgrew 75 183 435

Figure 2 Auditoryndash

auditory priming study

Mean RTs and standard

errors for 12 controlvolunteers

1166 Journal of Cognitive Neuroscience Volume 16 Number 7

F(164) = 464 p = 035 This interaction is shown inFigure 4

The group pattern was essentially repeated wheneach patientrsquos data were analyzed separately with onequalificationmdashthe semantic priming effect did not reachsignificance for any of the patients In the first testingperiod E K showed a main effect of priming F(1110) =6574 p lt 001 and a significant Priming Conditioninteraction F(3110) = 408 p = 009 Although theirregulars primed robustly (333 msec) F(137) = 4460p lt 001 the priming effect of 100 msec for the seman-tically related pairs was not significant F(154) = 249p = 120 A year later when tested a second time sheshowed a significant overall priming effect F(1115) =9193 p lt 001 and a significant Priming Conditioninteraction F(3115) = 625 p= 001 The irregulars still

primed strongly (346 msec) F(138) = 8229 p lt 001but the semantically related pairs did not prime signifi-cantly (104msec) F(121) = 234 p= 141 Similarly B Sshowed a main effect of priming F(1115) = 56042 p lt001 and a Priming Condition interaction F(3115) =2885 p = 039 The irregulars primed strongly(243 msec) F(140) = 38941 p lt 001 in the absenceof significant semantic priming (105 msec) F(121) =2717 p= 12 J T also showed a main effect of primingF(197) = 7138 p lt 01 and a Priming Conditioninteraction F(197) = 482 p lt 01 The irregularsproduced robust priming (177 msec) F(136) = 1135p lt 01 while the semantic conditions produced amarginally significant priming effect (80 msec) F 117) =442 p = 051

The failure of semantic priming effects to reachsignificance for the patients individually as well as thestatistically larger difference between semantic and ir-regular priming for the patients compared with thecontrols indicates a significant impairment in semanticpriming for this population In contrast their primingfor the irregulars seemed to be robustly unimpaired andsignificant for each patient individually

DISCUSSION

The aim of this research was to determine whetherpatients with semantic deficits always have associatedproblems in processing irregularly inflected forms aspredicted by current single mechanism accounts ofneuropsychological double dissociations between En-glish regular and irregular past tenses To address thisissue we tested semantic dementia patients on tasksthat probe the production and comprehension of regu-lar and irregularly inflected past tense forms We testedthe patientsrsquo abilities to produce correctly inflected pasttense forms from their present tense variants in twoelicitation experiments In both studies three of the

Figure 3 Auditoryndashauditory

priming study Mean RTs and

standard errors for the patients

(group data for J T E K andB S)

Figure 4 Auditoryndashauditory priming study Priming effects across

groups for the irregular and semantic conditions

Tyler et al 1167

patients were able to correctly inflect both the regularand irregular past tense Only one patient J T hadsignificantly more difficulty in inflecting the irregularscompared with the regulars This set of results suggeststhat semantic deficits do not necessarily go togetherwith difficulties in producing irregularly inflected pasttense forms

A possible argument against this interpretation of datais that the three patients who did not have difficultiesproducing irregularly inflected words may have mildersemantic deficits than the single patient J T who didhave problems with the irregulars However the datafrom the semantic tests do not suggest that J T wasmore severely semantically impaired than the otherthree patients On the Pyramids and Palm Trees testusing either written words or pictures the patients allscored poorly making between 31 and 42 errorsJ Trsquos error rate was 33 Similarly on a wordndashpicturematching task his error rate (23) was within the rangeof the errors produced by the other patients (7ndash34)

Turning to an assessment of the patientsrsquo ability tocomprehend past tense forms we found that the irreg-ulars primed robustly whether we analyzed the patientsas a group or individually suggesting that their ability toaccess the irregulars from a phonological input was notimpaired even for the patient ( J T) who had difficultiesaccessing irregulars in a production task Indeed theoverall group pattern of priming results looks remark-ably normal Like the controls the patient groupshowed priming for both types of past tense and forsemantically related words while phonologically relatedwords did not prime significantly Moreover like thecontrols the patients showed significantly greater prim-ing for the irregulars compared with the regulars

We have previously interpreted priming between averb form such as bring and its irregularly inflected pasttense form (ie brought) as evidence that the relation-ship between them is morphological rather than seman-tic (Marslen-Wilson amp Tyler 1998 2003) Both forms ofthe verb map onto the same underlying stem mor-pheme just as for the regulars and it is the repeatedactivation of this morpheme that gives rise to primingeffects (both at short and long priming delays) On thisaccount the basis for priming between irregularly in-flected words is different from the processes that un-derpin priming between two semantically related butmorphologically unrelated words (eg coatdress) Inthe latter case semantically related words prime be-cause of semantic overlap between two separate lexicalor morphemic representations This predicts that se-mantic deficits and deficits for the irregular past tensewill not necessarily pattern togethermdashwhich is thepattern of results that we obtain here for the semanticdementia patients

To account for neuropsychological dissociations inwhich patients are disproportionately impaired on theregular past tense we have focused on differences in

phonological processes involved in parsing an inflectedform into its stem and affix We assume that regularlyinflected past tense forms are generated as combina-tions of stems and inflectional affixes (open + ed) andrecognized by a process of phonological parsing wherethe speech input is segmented into a stem and its affixProblems with the regulars arise when this phonologi-cal parsing mechanism is impaired usually followingdamage to the left inferior frontal gyrus (LIFG) andassociated regions (Tyler de Mornay Davies et al2002 Tyler Randall et al 2002) Irregularly inflectedforms are stored just like monomorphemic words suchas table or grass as full-form representations with nointernal phonological structure as combinations ofstems and affixes These words do not need to bephonologically parsed into stems and affixes as part ofthe on-line access process These differences in require-ments for phonological parsing accompanied by dam-age to neural systems implicated in these types ofanalysis gives rise to selective deficits for the regularpast tense

Turning to deficits for the irregulars we assume (asnoted above) that an irregular past tense form has astored lexical form representation linking to the sameunderlying morpheme as other forms of the same wordThis predicts that deficits for the irregulars can arise ifbrain damage either affects stored lexical representa-tions directly or disrupts the mechanisms involved inthe process of mapping from phonology to semantics(Marslen-Wilson amp Tyler 1998) A version of this pro-posal is put forward by Miozzo (2003) in the context ofthe anomic patient A W who had problems with theirregulars in the presence of intact semanticsmdashtheopposite pattern to the one reported here for threeout of our four semantically impaired patients Miozzosuggests that this patient typical of anomic patients ingeneral has impairments in lexical processing asevidenced by her difficulty in retrieving a wordrsquos pho-nology On the hypothesis that irregular past tenseforms must be accessed as whole forms so that alterna-tive access routes involving decompositional mecha-nisms cannot be invoked problems in accessing storedlexical representations should differentially disadvantagethe irregulars compared with the regulars

There are some hints in the priming data that J Tthe one patient who showed difficulty in producingirregularly inflected words may also have phonologicalproblems He made many lexical decision errors (31errors) on the items in the phonological condition andmore than in any other condition (11 in the regular5 in the irregular 12 in the semantic conditions)The other patients did not show this pattern B Smade 8 4 0 and 2 errors in the phonologicalregular irregular and semantic conditions respective-ly E K at Time 1 made 8 8 3 and 2 errorson the phonological regular irregular and semanticconditions respectively and 4 6 1 and 2

1168 Journal of Cognitive Neuroscience Volume 16 Number 7

respectively errors in those conditions at the secondtime of testing Similarly J T had a high error rate(30) on the phonologically related foils in the exper-iment such as gossipgoss These data however areonly suggestive Further research is needed to deter-mine the nature of this potential phonological deficitand whether there is a more general relationshipbetween phonological processing deficits and prob-lems in accessing stored lexical forms

One issue that is raised by these results is why thepatients as a group show any semantic priming at allwhen they each have pronounced semantic deficitsThis is not the first time that semantic dementiapatients have been observed to show semantic primingwhile their performance on standard off-line testsshows them to be profoundly impaired We have pre-viously reported semantic priming in three semanticdementia patients (A M Tyler amp Moss 1998 F MTyler Moss Patterson amp Hodges 1997 P P MossTyler Hodges amp Patterson 1995) Given that thesepatients had been unambiguously diagnosed as having asemantic deficit we had not initially predicted that theywould show priming for semantically related words Thefact that priming can still be obtained albeit at asignificantly reduced level suggests that it is not anall-or-none phenomenon Significant priming effectsmay not require the activation of all of the informationassociated with a concept All that may be required isthat the semantic relationship between the two relatedwords (eg onioncarrot) should be stronger to somecriterial degree than between the two unrelated words(eg deskcarrot) This would allow semantic primingeffects even in a damaged system so long as somesemantic information is preserved especially if it isthe shared properties of concepts that are preservedsince it is these that provide the basis for semanticoverlap between related concepts

It has been independently demonstrated that theshared properties of concepts are indeed preservedfollowing brain damage These properties are morehighly correlated with each other and correlated prop-erties are relatively preserved following damage (Moss ampTyler 2000 Tyler et al 2000 Tyler amp Moss 2001 DevlinGonnerman Andersen amp Seidenberg 1998 McRae deSa amp Seidenberg 1997) Moreover studies with patientswho have semantic deficits have shown that the sharedproperties of concepts tend to be preserved while thedistinctive properties tend to be impaired (Tyler amp Moss2001 Moss Tyler Durrant-Peatfield amp Bunn 1998) Thepatients in the current study show the same pattern Ina verification task they were impaired at verifying thedistinctive properties of concepts while relatively pre-served on their shared properties (Table 1) If as arguedabove it is the coactivation of these shared propertiesthat primarily drives priming in the intact system thenthis would explain the preservation of some degree ofsemantic priming in the semantic dementia population

In summary E K and B S did not show any differ-ential impairment for the irregulars compared with theregulars on either elicitation or priming tasks and K Bdid not show a disproportionate deficit for the irregularscompared with the regulars in the elicitation task OnlyJ T showed an association of semantic deficit and prob-lems with the irregulars Moreover when the patientsdid make errors on the irregulars they did not exhibitthe pattern of effects predicted by the single mechanismaccount (Patterson et al 2001) with a higher proportionof errors on low-frequency items These results showthat there is no necessary association between a seman-tic deficit and problems with the irregular past tenseAlthough the two often do co-occur as we and othershave previously documented they can evidently alsooccur independently of each other This dissociationindicates that a separate set of functional capacities isimplicated here supporting the access of stored lexicalrepresentations An important issue for future researchis to identify the neural infrastructure for these accessprocesses so that we can better determine the condi-tions under which semantic deficits will co-occur withdeficits for the irregulars

The present results together with the results fornormal subjects addressed earlier (Marslen-Wilson ampTyler 1998) argue against the claim that the relation-ship between irregular past tense forms and their stemsis primarily a semantic rather than a morphologicalrelationship This in turn undermines the evidencefor the type of single system account of lexical repre-sentation in which morphological structure can becaptured solely on the basis of phonological and seman-tic relationships (Patterson et al 2001 Joanisse amp Seid-enberg 1999)

METHODS

Subjects

In addition to the four semantic dementia patients agroup of seven control subjects (aged between 56 and69 years) were tested on the elicitation task and a groupof 12 subjects (aged 60ndash73 years) were tested on thepriming study

Past Tense Production Elicitation Studies

We tested the patients on two elicitation experimentsdesigned to probe their ability to generate past tenseforms from their stems In the first elicitation testdescribed in Ullman et al (1997) subjects were pre-sented with sentences containing the present tenseforms of regular and irregular words together with aset of nonwords for past-tense generation We used thistest to directly compare our patients with others re-ported in the literature who have deficits in generatingpast tense forms

Tyler et al 1169

For the second elicitation task we selected a set of22 regular and 22 irregular verbs matched as closely aspossible on form and lemma frequency (Baayen Piepen-brock amp Gulikers 1995) and imageability (Coltheart1981) The median word form frequency was 35 for theregulars and 36 for the irregulars The median lemmafrequency was 92 for the regulars and 110 for theirregulars The median imageability was 485 for theregulars and 470 for the irregulars A two-sentencecontext was created for each verb where the secondsentence was incomplete and required a verb inflectedfor the past tense For example for the sentence lsquolsquoMynose sometimes bleeds Last night it rsquorsquo the correctresponse would be bled The test stimuli were precededby five practice trials The sentences were read aloud tothe subjects who were encouraged to produce a pasttense form appropriate for the sentence

Past Tense Comprehension Auditory Priming Study

To test patientsrsquo abilities to comprehend spoken pasttense forms we designed an intramodal auditoryndashauditory priming study contrasting the regulars andirregulars We chose to use a priming study to avoidsome of the pitfalls associated with testing patients ontasks that make heavy metalinguistic demands In thisexperiment target words which were verb stems (suchas pull sing) were preceded by a regularly or irregu-larly inflected past tense prime (pulled sung) or by anunrelated control word In addition to the two mainexperimental conditions we included two other con-ditions The first was a semantic priming condition inwhich the prime-target pairs were semantically but notphonologically or morphologically related (pighorse)We included this to determine whether the patientswould show evidence of a semantic deficit in the sametype of on-line task as used to assess their performanceon the past tense forms The fourth condition con-sisted of a phonological priming condition in whichthe prime-target pairs were phonologically but notsemantically or morphologically related (paintpain)which acted as a control for the effects of phonologicalsimilarity Since the inflected-stem pairs share a greatdeal of phonological overlap we need to be ableto separate out phonological from morphological prim-ing effects

Materials

We selected 42 prime-target pairs in each of the pasttense conditions (regularirregular) and 24 pairs for the2 control conditions (semanticphonological) In boththe regular and irregular past tense conditions theprimes were all inflected past tense forms (eg fixedglanced stuck brought) and the targets consisted oftheir stems all of which were one or two syllables ( fixglance stick bring) Items in the phonological control

condition consisted of morphologically simple monosyl-labic pairs (eg lamplamb plumpplum) that werenot semantically related The proportion of consonant-vowel-consonant (CVC) and consonant-vowel (CV) over-lap between prime and target was matched to theregular past tense pairs In all of the conditions weavoided homophones The two past tense sets and thephonological set were matched as closely as possible fortest prime control prime and target frequency andnumber of syllables (see Table 4) The fourth conditionconsisted of pairs of semantically related words thatwere either phonologically or morphologically related(eg shorecoast alleylane) For this condition weselected pairs of semantically related words that werehighly semantically related (but not associatively) and ofrelatively high frequency This was to ensure significantsemantic priming effects in control subjects to provide abasis for evaluating semantic priming in the patients Asemantic relatedness pretest was carried out in which15 subjects were asked to rate pairs of words accordingto their degree of semantic relatedness on a scale of 1ndash9where 1 was highly unrelated and 9 was highly relatedThe mean semantic relatedness rating for the prime-target pairs was 7 (SD = 102)

Each prime word was matched to a control word infrequency and number of syllables (see Table 4) Inaddition to the 132 test pairs we included 80 real wordfiller pairs making a total of 212 real word pairs Thefillers were an equal mixture of unrelated simpleprimespast tense targets simple primes(-s) inflectedtargetsx (-s) inflected primessimple targets simpleprimessimple targets and unambiguous noun pairsWe also included 148 legal wordnonword pairs whichwere a mixture of phonologically unrelated prime-targetpairs phonologically related pairs with targets derivedfrom regularly inflected past or present tense formsphonologically related pairs with targets made fromirregularly inflected forms and pairs with irregular verbsas primes

Design and Procedure

Test and filler items were pseudorandomly dispersedthroughout the test list We constructed two lists ofstimuli so that the target word only occurred once per

Table 4 Intramodal Priming Study Median Frequencies ofStimuli

Test Prime Target Control Prime

Regulars 55 6 4

Irregulars 6 7 5

Phonological 14 18 13

Semantic 16 10 15

1170 Journal of Cognitive Neuroscience Volume 16 Number 7

version All of the stimuli were recorded by a femalenative speaker of British English digitized at 22 kHz andmarkers placed at the acoustic onset of each targetword The markers triggered a timing device that mea-sured the subjectrsquos reaction time to the target wordPrimes and targets were recorded onto a DAT tape forpresentation to subjects Subjects heard primendashtargetpairs and were asked to make a lexical decision to thetarget by pressing a response key

Acknowledgments

We thank Matt Lambon-Ralph Beth Jeffries and KaralynPatterson for referring patients E K B S and J T and formaking many test scores available to us and to Julie Snowdenfor allowing us to work with K B We thank Lara Brent and BilliRandall for their help with the analyses and the patients fortheir cooperation This work was supported by an MRC (UK)program grant to L K T

Reprint requests should be sent to Lorraine K Tyler Depart-ment of Experimental Psychology University of CambridgeCambridge CB3 ODX UK or via e-mail lktylercslpsycholcamacuk

Note

1 These data were kindly provided by Elizabeth Jeffries

REFERENCES

Baayen R H Piepenbrock R amp Gulikers L (1995) TheCELEX lexical database Linguistic Data ConsortiumPhiladelphia University of Pennsylvania

Coltheart M (1981) The MRC psycholinguistic databaseQuarterly Journal of Experimental Psychology HumanExperimental Psychology 33A 497ndash505

Bunn E M Tyler L K amp Moss H E (1998) Naming picturesThe role of familiarity and property type reexaminedNeuropsychology 12 367ndash379

Devlin J T Gonnerman L M Andersen E S amp SeidenbergM S (1998) Category specific semantic deficits in focaland widespread brain damage A computational accountJournal of Cognitive Neuroscience 10 77ndash94

Friston K J Holmes A P Poline J-B Grasby B J WilliamsC R Frackowiak R S J amp Turner R (1995) Analysis offMRI time-series revisited Neuroscience 2 45ndash53

Galton C J Patterson K Graham K Lambon-Ralph M AWilliams G Antoun N Sahakian B J amp Hodges J R(2001) Differing patterns of temporal atrophy inAlzheimerrsquos disease and semantic dementia Neurology 57216ndash225

Hodges J R Patterson K Oxbury S amp Funnell E (1992)Semantic dementia Progressive fluent aphasia withtemporal lobe atrophy Brain 115 1783ndash1806

Howard D amp Patterson K (1992) The pyramids and palmtrees test Bury St Edmunds Thames Valley Test

Jefferies E Patterson K Jones R W Bateman D amp LambonRalph M A (submitted) A category-specific advantage fornumbers in verbal short-term memory Evidence fromsemantic dementia

Joanisse M F amp Seidenberg M S (1999) Impairments inverb morphology after brain injury A connectionist model

Proceedings of the National Academy of Sciences USA 967592ndash7597

Kay J Lesser R amp Coltheart M (1992) Psycholinguisticassessments of language processing in aphasia (PALPA)London Erlbaum

Longworth C E Marslen-Wilson W D amp Tyler L K(2002) Non-fluent aphasics show an abnormal time courseof semantic activation from the regular past tense[abstract] Journal of Cognitive Neuroscience 14Supplement 55

Marslen-Wilson W D Csibra G Ford M Hatzakis HGaskell G amp Johnson M H (2000) Associations anddissociations in the processing of regular and irregularverbs Electrophysiological evidence Journal of CognitiveNeuroscience 12 Supplement 55E

Marslen-Wilson W D amp Tyler L K (1997) Dissociating typesof mental computation Nature 387 592ndash594

Marslen-Wilson W D amp Tyler L K (1998) Rulesrepresentations and the English past tense Trends inCognitive Science 2 428ndash435

Marslen-Wilson W D amp Tyler L K (2003) Capturingunderlying differentiation in the human language systemTrends in Cognitive Sciences 7 62ndash63

Mathuranath P S Nestor P J Berrios G E Rakowicz Wamp Hodges J (2000) A brief cognitive test battery todifferentiate Alzheimerrsquos disease and frontotemporaldementia Neurology 55 1613ndash1620

McClelland J L amp Patterson K (2002) Rules or connectionsin past-tense inflections What does the evidence rule outTrends in Cognitive Sciences 6 465ndash472

McRae K de Sa V amp Seidenberg M S (1997) On thenature and scope of featural representations of wordmeaning Journal of Experimental Psychology General126 99ndash130

Miozzo M (2003) On the processing of regular and irregularforms of verbs and nouns Evidence from neuropsychologyCognition 87 101ndash127

Moss H E amp Tyler L K (2000) A progressivecategory-specific semantic deficit for non-living thingsNeuropsychologia 38 60ndash82

Moss H E Tyler L K Durrant-Peatfield M amp Bunn E(1998) Two eyes of a see-through Impaired and intactsemantic knowledge in a case of selective deficit for livingthings Neurocase 4 291ndash310

Moss H E Tyler L K Hodges J R amp Patterson K (1995)Exploring the loss of semantic memory in semanticdementia Evidence from a primed monitoring studyNeuropsychology 91 16ndash26

Patterson K Lambon Ralph M A Hodges J Ramp McClelland J L (2001) Deficits in irregular past-tenseverb morphology associated with degraded semanticknowledge Neuropsychologia 39 709ndash724

Pinker S (1991) Rules of language Science 253 530ndash533Pinker S amp Ullman M T (2002) The past and future of thepast tense Trends in Cognitive Sciences 6 456ndash463

Prasada S amp Pinker S (1993) Generalizations of regular andirregular morphological patterns Language and CognitiveProcesses 8 1ndash56

Price C J amp Friston K J (1997) Cognitive conjunction A newapproach to brain activation experiments Neuroimage 5261ndash270

Rumelhart D E amp McClelland J L (1986) On learningthe past tense of English verbs In J L McClelland amp D ERumelhart (Eds) Parallel distributed processing(pp 217ndash270) Cambridge MIT Press

Snowden J S Goulding P J amp Neary D (1989) Semanticdementia A form of circumscribed cerebral atrophyBehavioral Neurology 2 167ndash182

Tyler et al 1171

(E K) whose performance we were able to monitorover the course of 14 months on all three tests

RESULTS

Past Tense Elicitation Tasks

The control subjects made virtually no errors on theelicitation task developed at the Centre for Speech andLanguage (CSL) (Table 2) Three out of four patientsalso showed no significant differences in their ability toproduce either regular or irregular past tense formsB S and K B correctly inflected 95 of the regularsand 100 of the irregulars Although E K correctlyinflected slightly more of the regulars (86) than theirregulars (82) when first tested in January 2002 thisdifference was not significant (x2 = 001 p gt 05)When we tested her a year later there was no differ-ence at all between the irregulars and regulars withE K correctly inflecting 86 of each Only J T showeda significant advantage for the regulars correctly in-flecting 100 of them compared with only 55 of theirregulars (x2 = 1048 p lt 05) Given that most of thepatients made so few errors on inflecting the irregularsit was not possible to carry out an analysis of the errortypes Of the 16 errors on the irregulars 14 of theseinvolved pure regularization (eg creep creeped)and 2 involved partial regularization (eg deal

dealded) Both single- and dual-mechanism approacheswould predict this outcome

In further analyses we compared high- and low-frequency words in the two conditions to test a keyclaim of the single mechanism account that patientswith semantic deficits will have particular difficulty withthe low-frequency irregulars (Patterson et al 2001) Thisis claimed to arise because low-frequency irregularsdepend more upon semantics to achieve the mappingbetween present and past tense than high-frequencyirregulars For the frequency analysis we divided eachset of words into two bandsmdashlow and high frequencies(see Table 3) As Table 2 shows the patients were notdifferentially impaired on the low- relative to high-fre-quency irregulars Both K B and B S only made oneerror in the entire test and this was on a high-frequencyregular word In January 2002 E K did show a frequen-cy effect having more difficulty correctly inflecting low-compared with high-frequency words However this didnot interact with regularity she correctly inflected 73of the low-frequency regulars and 67 of the low-frequency irregulars This difference was not significant(x2 = 018 p gt 05) A year later she no longer showeda frequency effect J T was the only patient who clearlyhad more difficulty correctly inflecting low-frequencyirregular words Sixty-eight percent of the high-frequen-cy irregulars were correctly inflected but only 32 of thelow-frequency irregulars

The pattern of results on the CSL elicitation test wasbroadly consistent with those for the Ullman elicitationtask (see Table 2) When first tested E K correctlyinflected 90 of the irregular past tense forms 75 ofthe regulars and 70 of the nonwords The differencebetween the regulars and irregulars was not significant(x2 = 06926 p gt 05) When tested a year later thepattern was slightly but not significantly different Shecorrectly inflected 90 of the regulars 80 of theirregulars and 100 of the nonwords but none of thesedifferences were significant (all x2 gt 005) Similarly B Sshowed no significant difference in accuracy betweenthe regulars and irregulars correctly inflecting 100 ofthe regulars 95 of the irregulars and 100 of thenonwords In contrast J T was more accurate on theregulars (95) and the nonwords (100) than the ir-regulars (75) The difference between the regulars andirregulars although substantial did not reach signifi-cance (x2 = 314 p gt 05)

Intramodal Priming Task

Controls

Twelve healthy controls were tested in this study six oneach version Analyses of their reaction times (RTs)showed a significant main effect of priming F1(110) =835 p lt 001 F2(1124) = 1281 p lt 001 and a maineffect of condition F1(330) = 396 plt 001 F2(3124) =825 p lt 001 due to differences in overall RTs in the

Table 2 Past Tense Elicitation ( Correct)

B S E K(Time 1)

E K(Time 2)

J T K B

CSL Past Tense Elicitation Task

Regulars 95 86 86 100 95

Irregulars 100 82 86 55 100

CSL Past Tense Elicitation Task

Regulars

HF (n = 11) 93 100 91 100 93

LF (n = 11) 100 73 82 100 100

Irregulars

HF (n = 13) 100 85 85 68 100

LF (n = 9) 100 67 89 32 100

Ullman Past Tense Elicitation Task

Regulars 100 75 90 95 nt

Irregulars 95 90 80 75

Nonwords 100 70 100 100

CSL = Centre for Speech Language

Tyler et al 1165

four conditions with RTs in the semantic conditionbeing fastest (944 msec) and RTs in the phonologicalcondition being slowest (1091 msec) More importantlythere was a significant Priming Condition interactionF1(330) = 130 p lt 001 F2(3124) = 224 p lt 001 Toprobe the source of this interaction we analyzed eachcondition separately This analysis showed a significantpriming effect of 101 msec for the regulars F1(110) =384 p lt 001 F2(140) = 400 p lt 001 171 msec forthe irregulars F1(110) = 664 p lt 001 F2(140) =1173 p lt 001 and 109 msec for the semanticallyrelated pairs F1(110) = 138 p lt 01 F2(122) =377 p lt 001 and no priming for the phonologicallyrelated pairs (47 msec) F1(110) = 229 p = 162F2(122) = 258 p = 123 Moreover the controlsshowed significantly more priming for the irregularscompared with the semantic condition F2(162) =5596 p = 021 The mean RTs and standard errors aredisplayed in Figure 2 This pattern of results for thehealthy controls is identical to that produced by adifferent group of controls in a similar earlier experi-ments (Tyler et al 2002)

Patients

Wewere only able to test E K B S and J T in this studyas K B was not available for testing All three patientsproduced mean RTs that were within the normal range(E K at first testing = 1006 msec E K at second testing= 1041 msec B S = 1152 msec J T = 1099 msec)

(Figure 3) We carried out an ANOVA on the combineddata from B S J T and E K and found a main effect ofpriming F2(1123) = 9397 p lt 001 due to faster RTsin the related (mean RT = 1023 msec) compared withthe unrelated condition (mean RT = 1194 msec) and asignificant Priming Condition interaction F2(3123) =6423 p = 001 due to faster RTs in the semantic andirregular conditions (1063 and 1052 msec respectively)compared with the regular (1135 msec) and the phono-logical condition (1185msec) This was very similar to thepattern shown by the control subjects

We then analyzed each condition separately andfound that the priming effect of 283 msec for theirregulars was significant F2(140) = 84176 p lt 001as was the 104-msec priming effect for the semanticcondition F2(122) = 12673 p = 002 The regularsalso primed significantly (mean priming = 185 msec)F2(140) = 20027 p lt 001 whereas the phonologicallyrelated items did not (mean priming = 115 msec)F2(122) = 3120 p = 092 In the direct comparisonbetween the semantic and irregular conditions thepatients showed a significant interaction F2(162) =155 p lt 001 because just like the normal subjectsthey showed more priming for the irregulars Howeverthe patients showed a greater difference in the amountof priming for the irregulars compared with the seman-tic condition (179 msec as opposed to 62 msec forthe controls) In an analysis comparing the primingeffects for patients and controls in the semantic andirregular conditions we found a significant interaction

Table 3 Properties of Stimuli for CSL Past Tense Elicitation Task

Median WordForm Frequency

Median LemmaFrequency

MedianImageability

Low-frequency regulars (n = 11) eg blessblessed 10 28 493

Low-frequency irregulars (n = 9) eg bleedbled 10 20 486

High-frequency regulars (n = 11) eg shareshared 64 179 424

High-frequency irregulars (n = 13) eg growgrew 75 183 435

Figure 2 Auditoryndash

auditory priming study

Mean RTs and standard

errors for 12 controlvolunteers

1166 Journal of Cognitive Neuroscience Volume 16 Number 7

F(164) = 464 p = 035 This interaction is shown inFigure 4

The group pattern was essentially repeated wheneach patientrsquos data were analyzed separately with onequalificationmdashthe semantic priming effect did not reachsignificance for any of the patients In the first testingperiod E K showed a main effect of priming F(1110) =6574 p lt 001 and a significant Priming Conditioninteraction F(3110) = 408 p = 009 Although theirregulars primed robustly (333 msec) F(137) = 4460p lt 001 the priming effect of 100 msec for the seman-tically related pairs was not significant F(154) = 249p = 120 A year later when tested a second time sheshowed a significant overall priming effect F(1115) =9193 p lt 001 and a significant Priming Conditioninteraction F(3115) = 625 p= 001 The irregulars still

primed strongly (346 msec) F(138) = 8229 p lt 001but the semantically related pairs did not prime signifi-cantly (104msec) F(121) = 234 p= 141 Similarly B Sshowed a main effect of priming F(1115) = 56042 p lt001 and a Priming Condition interaction F(3115) =2885 p = 039 The irregulars primed strongly(243 msec) F(140) = 38941 p lt 001 in the absenceof significant semantic priming (105 msec) F(121) =2717 p= 12 J T also showed a main effect of primingF(197) = 7138 p lt 01 and a Priming Conditioninteraction F(197) = 482 p lt 01 The irregularsproduced robust priming (177 msec) F(136) = 1135p lt 01 while the semantic conditions produced amarginally significant priming effect (80 msec) F 117) =442 p = 051

The failure of semantic priming effects to reachsignificance for the patients individually as well as thestatistically larger difference between semantic and ir-regular priming for the patients compared with thecontrols indicates a significant impairment in semanticpriming for this population In contrast their primingfor the irregulars seemed to be robustly unimpaired andsignificant for each patient individually

DISCUSSION

The aim of this research was to determine whetherpatients with semantic deficits always have associatedproblems in processing irregularly inflected forms aspredicted by current single mechanism accounts ofneuropsychological double dissociations between En-glish regular and irregular past tenses To address thisissue we tested semantic dementia patients on tasksthat probe the production and comprehension of regu-lar and irregularly inflected past tense forms We testedthe patientsrsquo abilities to produce correctly inflected pasttense forms from their present tense variants in twoelicitation experiments In both studies three of the

Figure 3 Auditoryndashauditory

priming study Mean RTs and

standard errors for the patients

(group data for J T E K andB S)

Figure 4 Auditoryndashauditory priming study Priming effects across

groups for the irregular and semantic conditions

Tyler et al 1167

patients were able to correctly inflect both the regularand irregular past tense Only one patient J T hadsignificantly more difficulty in inflecting the irregularscompared with the regulars This set of results suggeststhat semantic deficits do not necessarily go togetherwith difficulties in producing irregularly inflected pasttense forms

A possible argument against this interpretation of datais that the three patients who did not have difficultiesproducing irregularly inflected words may have mildersemantic deficits than the single patient J T who didhave problems with the irregulars However the datafrom the semantic tests do not suggest that J T wasmore severely semantically impaired than the otherthree patients On the Pyramids and Palm Trees testusing either written words or pictures the patients allscored poorly making between 31 and 42 errorsJ Trsquos error rate was 33 Similarly on a wordndashpicturematching task his error rate (23) was within the rangeof the errors produced by the other patients (7ndash34)

Turning to an assessment of the patientsrsquo ability tocomprehend past tense forms we found that the irreg-ulars primed robustly whether we analyzed the patientsas a group or individually suggesting that their ability toaccess the irregulars from a phonological input was notimpaired even for the patient ( J T) who had difficultiesaccessing irregulars in a production task Indeed theoverall group pattern of priming results looks remark-ably normal Like the controls the patient groupshowed priming for both types of past tense and forsemantically related words while phonologically relatedwords did not prime significantly Moreover like thecontrols the patients showed significantly greater prim-ing for the irregulars compared with the regulars

We have previously interpreted priming between averb form such as bring and its irregularly inflected pasttense form (ie brought) as evidence that the relation-ship between them is morphological rather than seman-tic (Marslen-Wilson amp Tyler 1998 2003) Both forms ofthe verb map onto the same underlying stem mor-pheme just as for the regulars and it is the repeatedactivation of this morpheme that gives rise to primingeffects (both at short and long priming delays) On thisaccount the basis for priming between irregularly in-flected words is different from the processes that un-derpin priming between two semantically related butmorphologically unrelated words (eg coatdress) Inthe latter case semantically related words prime be-cause of semantic overlap between two separate lexicalor morphemic representations This predicts that se-mantic deficits and deficits for the irregular past tensewill not necessarily pattern togethermdashwhich is thepattern of results that we obtain here for the semanticdementia patients

To account for neuropsychological dissociations inwhich patients are disproportionately impaired on theregular past tense we have focused on differences in

phonological processes involved in parsing an inflectedform into its stem and affix We assume that regularlyinflected past tense forms are generated as combina-tions of stems and inflectional affixes (open + ed) andrecognized by a process of phonological parsing wherethe speech input is segmented into a stem and its affixProblems with the regulars arise when this phonologi-cal parsing mechanism is impaired usually followingdamage to the left inferior frontal gyrus (LIFG) andassociated regions (Tyler de Mornay Davies et al2002 Tyler Randall et al 2002) Irregularly inflectedforms are stored just like monomorphemic words suchas table or grass as full-form representations with nointernal phonological structure as combinations ofstems and affixes These words do not need to bephonologically parsed into stems and affixes as part ofthe on-line access process These differences in require-ments for phonological parsing accompanied by dam-age to neural systems implicated in these types ofanalysis gives rise to selective deficits for the regularpast tense

Turning to deficits for the irregulars we assume (asnoted above) that an irregular past tense form has astored lexical form representation linking to the sameunderlying morpheme as other forms of the same wordThis predicts that deficits for the irregulars can arise ifbrain damage either affects stored lexical representa-tions directly or disrupts the mechanisms involved inthe process of mapping from phonology to semantics(Marslen-Wilson amp Tyler 1998) A version of this pro-posal is put forward by Miozzo (2003) in the context ofthe anomic patient A W who had problems with theirregulars in the presence of intact semanticsmdashtheopposite pattern to the one reported here for threeout of our four semantically impaired patients Miozzosuggests that this patient typical of anomic patients ingeneral has impairments in lexical processing asevidenced by her difficulty in retrieving a wordrsquos pho-nology On the hypothesis that irregular past tenseforms must be accessed as whole forms so that alterna-tive access routes involving decompositional mecha-nisms cannot be invoked problems in accessing storedlexical representations should differentially disadvantagethe irregulars compared with the regulars

There are some hints in the priming data that J Tthe one patient who showed difficulty in producingirregularly inflected words may also have phonologicalproblems He made many lexical decision errors (31errors) on the items in the phonological condition andmore than in any other condition (11 in the regular5 in the irregular 12 in the semantic conditions)The other patients did not show this pattern B Smade 8 4 0 and 2 errors in the phonologicalregular irregular and semantic conditions respective-ly E K at Time 1 made 8 8 3 and 2 errorson the phonological regular irregular and semanticconditions respectively and 4 6 1 and 2

1168 Journal of Cognitive Neuroscience Volume 16 Number 7

respectively errors in those conditions at the secondtime of testing Similarly J T had a high error rate(30) on the phonologically related foils in the exper-iment such as gossipgoss These data however areonly suggestive Further research is needed to deter-mine the nature of this potential phonological deficitand whether there is a more general relationshipbetween phonological processing deficits and prob-lems in accessing stored lexical forms

One issue that is raised by these results is why thepatients as a group show any semantic priming at allwhen they each have pronounced semantic deficitsThis is not the first time that semantic dementiapatients have been observed to show semantic primingwhile their performance on standard off-line testsshows them to be profoundly impaired We have pre-viously reported semantic priming in three semanticdementia patients (A M Tyler amp Moss 1998 F MTyler Moss Patterson amp Hodges 1997 P P MossTyler Hodges amp Patterson 1995) Given that thesepatients had been unambiguously diagnosed as having asemantic deficit we had not initially predicted that theywould show priming for semantically related words Thefact that priming can still be obtained albeit at asignificantly reduced level suggests that it is not anall-or-none phenomenon Significant priming effectsmay not require the activation of all of the informationassociated with a concept All that may be required isthat the semantic relationship between the two relatedwords (eg onioncarrot) should be stronger to somecriterial degree than between the two unrelated words(eg deskcarrot) This would allow semantic primingeffects even in a damaged system so long as somesemantic information is preserved especially if it isthe shared properties of concepts that are preservedsince it is these that provide the basis for semanticoverlap between related concepts

It has been independently demonstrated that theshared properties of concepts are indeed preservedfollowing brain damage These properties are morehighly correlated with each other and correlated prop-erties are relatively preserved following damage (Moss ampTyler 2000 Tyler et al 2000 Tyler amp Moss 2001 DevlinGonnerman Andersen amp Seidenberg 1998 McRae deSa amp Seidenberg 1997) Moreover studies with patientswho have semantic deficits have shown that the sharedproperties of concepts tend to be preserved while thedistinctive properties tend to be impaired (Tyler amp Moss2001 Moss Tyler Durrant-Peatfield amp Bunn 1998) Thepatients in the current study show the same pattern Ina verification task they were impaired at verifying thedistinctive properties of concepts while relatively pre-served on their shared properties (Table 1) If as arguedabove it is the coactivation of these shared propertiesthat primarily drives priming in the intact system thenthis would explain the preservation of some degree ofsemantic priming in the semantic dementia population

In summary E K and B S did not show any differ-ential impairment for the irregulars compared with theregulars on either elicitation or priming tasks and K Bdid not show a disproportionate deficit for the irregularscompared with the regulars in the elicitation task OnlyJ T showed an association of semantic deficit and prob-lems with the irregulars Moreover when the patientsdid make errors on the irregulars they did not exhibitthe pattern of effects predicted by the single mechanismaccount (Patterson et al 2001) with a higher proportionof errors on low-frequency items These results showthat there is no necessary association between a seman-tic deficit and problems with the irregular past tenseAlthough the two often do co-occur as we and othershave previously documented they can evidently alsooccur independently of each other This dissociationindicates that a separate set of functional capacities isimplicated here supporting the access of stored lexicalrepresentations An important issue for future researchis to identify the neural infrastructure for these accessprocesses so that we can better determine the condi-tions under which semantic deficits will co-occur withdeficits for the irregulars

The present results together with the results fornormal subjects addressed earlier (Marslen-Wilson ampTyler 1998) argue against the claim that the relation-ship between irregular past tense forms and their stemsis primarily a semantic rather than a morphologicalrelationship This in turn undermines the evidencefor the type of single system account of lexical repre-sentation in which morphological structure can becaptured solely on the basis of phonological and seman-tic relationships (Patterson et al 2001 Joanisse amp Seid-enberg 1999)

METHODS

Subjects

In addition to the four semantic dementia patients agroup of seven control subjects (aged between 56 and69 years) were tested on the elicitation task and a groupof 12 subjects (aged 60ndash73 years) were tested on thepriming study

Past Tense Production Elicitation Studies

We tested the patients on two elicitation experimentsdesigned to probe their ability to generate past tenseforms from their stems In the first elicitation testdescribed in Ullman et al (1997) subjects were pre-sented with sentences containing the present tenseforms of regular and irregular words together with aset of nonwords for past-tense generation We used thistest to directly compare our patients with others re-ported in the literature who have deficits in generatingpast tense forms

Tyler et al 1169

For the second elicitation task we selected a set of22 regular and 22 irregular verbs matched as closely aspossible on form and lemma frequency (Baayen Piepen-brock amp Gulikers 1995) and imageability (Coltheart1981) The median word form frequency was 35 for theregulars and 36 for the irregulars The median lemmafrequency was 92 for the regulars and 110 for theirregulars The median imageability was 485 for theregulars and 470 for the irregulars A two-sentencecontext was created for each verb where the secondsentence was incomplete and required a verb inflectedfor the past tense For example for the sentence lsquolsquoMynose sometimes bleeds Last night it rsquorsquo the correctresponse would be bled The test stimuli were precededby five practice trials The sentences were read aloud tothe subjects who were encouraged to produce a pasttense form appropriate for the sentence

Past Tense Comprehension Auditory Priming Study

To test patientsrsquo abilities to comprehend spoken pasttense forms we designed an intramodal auditoryndashauditory priming study contrasting the regulars andirregulars We chose to use a priming study to avoidsome of the pitfalls associated with testing patients ontasks that make heavy metalinguistic demands In thisexperiment target words which were verb stems (suchas pull sing) were preceded by a regularly or irregu-larly inflected past tense prime (pulled sung) or by anunrelated control word In addition to the two mainexperimental conditions we included two other con-ditions The first was a semantic priming condition inwhich the prime-target pairs were semantically but notphonologically or morphologically related (pighorse)We included this to determine whether the patientswould show evidence of a semantic deficit in the sametype of on-line task as used to assess their performanceon the past tense forms The fourth condition con-sisted of a phonological priming condition in whichthe prime-target pairs were phonologically but notsemantically or morphologically related (paintpain)which acted as a control for the effects of phonologicalsimilarity Since the inflected-stem pairs share a greatdeal of phonological overlap we need to be ableto separate out phonological from morphological prim-ing effects

Materials

We selected 42 prime-target pairs in each of the pasttense conditions (regularirregular) and 24 pairs for the2 control conditions (semanticphonological) In boththe regular and irregular past tense conditions theprimes were all inflected past tense forms (eg fixedglanced stuck brought) and the targets consisted oftheir stems all of which were one or two syllables ( fixglance stick bring) Items in the phonological control

condition consisted of morphologically simple monosyl-labic pairs (eg lamplamb plumpplum) that werenot semantically related The proportion of consonant-vowel-consonant (CVC) and consonant-vowel (CV) over-lap between prime and target was matched to theregular past tense pairs In all of the conditions weavoided homophones The two past tense sets and thephonological set were matched as closely as possible fortest prime control prime and target frequency andnumber of syllables (see Table 4) The fourth conditionconsisted of pairs of semantically related words thatwere either phonologically or morphologically related(eg shorecoast alleylane) For this condition weselected pairs of semantically related words that werehighly semantically related (but not associatively) and ofrelatively high frequency This was to ensure significantsemantic priming effects in control subjects to provide abasis for evaluating semantic priming in the patients Asemantic relatedness pretest was carried out in which15 subjects were asked to rate pairs of words accordingto their degree of semantic relatedness on a scale of 1ndash9where 1 was highly unrelated and 9 was highly relatedThe mean semantic relatedness rating for the prime-target pairs was 7 (SD = 102)

Each prime word was matched to a control word infrequency and number of syllables (see Table 4) Inaddition to the 132 test pairs we included 80 real wordfiller pairs making a total of 212 real word pairs Thefillers were an equal mixture of unrelated simpleprimespast tense targets simple primes(-s) inflectedtargetsx (-s) inflected primessimple targets simpleprimessimple targets and unambiguous noun pairsWe also included 148 legal wordnonword pairs whichwere a mixture of phonologically unrelated prime-targetpairs phonologically related pairs with targets derivedfrom regularly inflected past or present tense formsphonologically related pairs with targets made fromirregularly inflected forms and pairs with irregular verbsas primes

Design and Procedure

Test and filler items were pseudorandomly dispersedthroughout the test list We constructed two lists ofstimuli so that the target word only occurred once per

Table 4 Intramodal Priming Study Median Frequencies ofStimuli

Test Prime Target Control Prime

Regulars 55 6 4

Irregulars 6 7 5

Phonological 14 18 13

Semantic 16 10 15

1170 Journal of Cognitive Neuroscience Volume 16 Number 7

version All of the stimuli were recorded by a femalenative speaker of British English digitized at 22 kHz andmarkers placed at the acoustic onset of each targetword The markers triggered a timing device that mea-sured the subjectrsquos reaction time to the target wordPrimes and targets were recorded onto a DAT tape forpresentation to subjects Subjects heard primendashtargetpairs and were asked to make a lexical decision to thetarget by pressing a response key

Acknowledgments

We thank Matt Lambon-Ralph Beth Jeffries and KaralynPatterson for referring patients E K B S and J T and formaking many test scores available to us and to Julie Snowdenfor allowing us to work with K B We thank Lara Brent and BilliRandall for their help with the analyses and the patients fortheir cooperation This work was supported by an MRC (UK)program grant to L K T

Reprint requests should be sent to Lorraine K Tyler Depart-ment of Experimental Psychology University of CambridgeCambridge CB3 ODX UK or via e-mail lktylercslpsycholcamacuk

Note

1 These data were kindly provided by Elizabeth Jeffries

REFERENCES

Baayen R H Piepenbrock R amp Gulikers L (1995) TheCELEX lexical database Linguistic Data ConsortiumPhiladelphia University of Pennsylvania

Coltheart M (1981) The MRC psycholinguistic databaseQuarterly Journal of Experimental Psychology HumanExperimental Psychology 33A 497ndash505

Bunn E M Tyler L K amp Moss H E (1998) Naming picturesThe role of familiarity and property type reexaminedNeuropsychology 12 367ndash379

Devlin J T Gonnerman L M Andersen E S amp SeidenbergM S (1998) Category specific semantic deficits in focaland widespread brain damage A computational accountJournal of Cognitive Neuroscience 10 77ndash94

Friston K J Holmes A P Poline J-B Grasby B J WilliamsC R Frackowiak R S J amp Turner R (1995) Analysis offMRI time-series revisited Neuroscience 2 45ndash53

Galton C J Patterson K Graham K Lambon-Ralph M AWilliams G Antoun N Sahakian B J amp Hodges J R(2001) Differing patterns of temporal atrophy inAlzheimerrsquos disease and semantic dementia Neurology 57216ndash225

Hodges J R Patterson K Oxbury S amp Funnell E (1992)Semantic dementia Progressive fluent aphasia withtemporal lobe atrophy Brain 115 1783ndash1806

Howard D amp Patterson K (1992) The pyramids and palmtrees test Bury St Edmunds Thames Valley Test

Jefferies E Patterson K Jones R W Bateman D amp LambonRalph M A (submitted) A category-specific advantage fornumbers in verbal short-term memory Evidence fromsemantic dementia

Joanisse M F amp Seidenberg M S (1999) Impairments inverb morphology after brain injury A connectionist model

Proceedings of the National Academy of Sciences USA 967592ndash7597

Kay J Lesser R amp Coltheart M (1992) Psycholinguisticassessments of language processing in aphasia (PALPA)London Erlbaum

Longworth C E Marslen-Wilson W D amp Tyler L K(2002) Non-fluent aphasics show an abnormal time courseof semantic activation from the regular past tense[abstract] Journal of Cognitive Neuroscience 14Supplement 55

Marslen-Wilson W D Csibra G Ford M Hatzakis HGaskell G amp Johnson M H (2000) Associations anddissociations in the processing of regular and irregularverbs Electrophysiological evidence Journal of CognitiveNeuroscience 12 Supplement 55E

Marslen-Wilson W D amp Tyler L K (1997) Dissociating typesof mental computation Nature 387 592ndash594

Marslen-Wilson W D amp Tyler L K (1998) Rulesrepresentations and the English past tense Trends inCognitive Science 2 428ndash435

Marslen-Wilson W D amp Tyler L K (2003) Capturingunderlying differentiation in the human language systemTrends in Cognitive Sciences 7 62ndash63

Mathuranath P S Nestor P J Berrios G E Rakowicz Wamp Hodges J (2000) A brief cognitive test battery todifferentiate Alzheimerrsquos disease and frontotemporaldementia Neurology 55 1613ndash1620

McClelland J L amp Patterson K (2002) Rules or connectionsin past-tense inflections What does the evidence rule outTrends in Cognitive Sciences 6 465ndash472

McRae K de Sa V amp Seidenberg M S (1997) On thenature and scope of featural representations of wordmeaning Journal of Experimental Psychology General126 99ndash130

Miozzo M (2003) On the processing of regular and irregularforms of verbs and nouns Evidence from neuropsychologyCognition 87 101ndash127

Moss H E amp Tyler L K (2000) A progressivecategory-specific semantic deficit for non-living thingsNeuropsychologia 38 60ndash82

Moss H E Tyler L K Durrant-Peatfield M amp Bunn E(1998) Two eyes of a see-through Impaired and intactsemantic knowledge in a case of selective deficit for livingthings Neurocase 4 291ndash310

Moss H E Tyler L K Hodges J R amp Patterson K (1995)Exploring the loss of semantic memory in semanticdementia Evidence from a primed monitoring studyNeuropsychology 91 16ndash26

Patterson K Lambon Ralph M A Hodges J Ramp McClelland J L (2001) Deficits in irregular past-tenseverb morphology associated with degraded semanticknowledge Neuropsychologia 39 709ndash724

Pinker S (1991) Rules of language Science 253 530ndash533Pinker S amp Ullman M T (2002) The past and future of thepast tense Trends in Cognitive Sciences 6 456ndash463

Prasada S amp Pinker S (1993) Generalizations of regular andirregular morphological patterns Language and CognitiveProcesses 8 1ndash56

Price C J amp Friston K J (1997) Cognitive conjunction A newapproach to brain activation experiments Neuroimage 5261ndash270

Rumelhart D E amp McClelland J L (1986) On learningthe past tense of English verbs In J L McClelland amp D ERumelhart (Eds) Parallel distributed processing(pp 217ndash270) Cambridge MIT Press

Snowden J S Goulding P J amp Neary D (1989) Semanticdementia A form of circumscribed cerebral atrophyBehavioral Neurology 2 167ndash182

Tyler et al 1171

four conditions with RTs in the semantic conditionbeing fastest (944 msec) and RTs in the phonologicalcondition being slowest (1091 msec) More importantlythere was a significant Priming Condition interactionF1(330) = 130 p lt 001 F2(3124) = 224 p lt 001 Toprobe the source of this interaction we analyzed eachcondition separately This analysis showed a significantpriming effect of 101 msec for the regulars F1(110) =384 p lt 001 F2(140) = 400 p lt 001 171 msec forthe irregulars F1(110) = 664 p lt 001 F2(140) =1173 p lt 001 and 109 msec for the semanticallyrelated pairs F1(110) = 138 p lt 01 F2(122) =377 p lt 001 and no priming for the phonologicallyrelated pairs (47 msec) F1(110) = 229 p = 162F2(122) = 258 p = 123 Moreover the controlsshowed significantly more priming for the irregularscompared with the semantic condition F2(162) =5596 p = 021 The mean RTs and standard errors aredisplayed in Figure 2 This pattern of results for thehealthy controls is identical to that produced by adifferent group of controls in a similar earlier experi-ments (Tyler et al 2002)

Patients

Wewere only able to test E K B S and J T in this studyas K B was not available for testing All three patientsproduced mean RTs that were within the normal range(E K at first testing = 1006 msec E K at second testing= 1041 msec B S = 1152 msec J T = 1099 msec)

(Figure 3) We carried out an ANOVA on the combineddata from B S J T and E K and found a main effect ofpriming F2(1123) = 9397 p lt 001 due to faster RTsin the related (mean RT = 1023 msec) compared withthe unrelated condition (mean RT = 1194 msec) and asignificant Priming Condition interaction F2(3123) =6423 p = 001 due to faster RTs in the semantic andirregular conditions (1063 and 1052 msec respectively)compared with the regular (1135 msec) and the phono-logical condition (1185msec) This was very similar to thepattern shown by the control subjects

We then analyzed each condition separately andfound that the priming effect of 283 msec for theirregulars was significant F2(140) = 84176 p lt 001as was the 104-msec priming effect for the semanticcondition F2(122) = 12673 p = 002 The regularsalso primed significantly (mean priming = 185 msec)F2(140) = 20027 p lt 001 whereas the phonologicallyrelated items did not (mean priming = 115 msec)F2(122) = 3120 p = 092 In the direct comparisonbetween the semantic and irregular conditions thepatients showed a significant interaction F2(162) =155 p lt 001 because just like the normal subjectsthey showed more priming for the irregulars Howeverthe patients showed a greater difference in the amountof priming for the irregulars compared with the seman-tic condition (179 msec as opposed to 62 msec forthe controls) In an analysis comparing the primingeffects for patients and controls in the semantic andirregular conditions we found a significant interaction

Table 3 Properties of Stimuli for CSL Past Tense Elicitation Task

Median WordForm Frequency

Median LemmaFrequency

MedianImageability

Low-frequency regulars (n = 11) eg blessblessed 10 28 493

Low-frequency irregulars (n = 9) eg bleedbled 10 20 486

High-frequency regulars (n = 11) eg shareshared 64 179 424

High-frequency irregulars (n = 13) eg growgrew 75 183 435

Figure 2 Auditoryndash

auditory priming study

Mean RTs and standard

errors for 12 controlvolunteers

1166 Journal of Cognitive Neuroscience Volume 16 Number 7

F(164) = 464 p = 035 This interaction is shown inFigure 4

The group pattern was essentially repeated wheneach patientrsquos data were analyzed separately with onequalificationmdashthe semantic priming effect did not reachsignificance for any of the patients In the first testingperiod E K showed a main effect of priming F(1110) =6574 p lt 001 and a significant Priming Conditioninteraction F(3110) = 408 p = 009 Although theirregulars primed robustly (333 msec) F(137) = 4460p lt 001 the priming effect of 100 msec for the seman-tically related pairs was not significant F(154) = 249p = 120 A year later when tested a second time sheshowed a significant overall priming effect F(1115) =9193 p lt 001 and a significant Priming Conditioninteraction F(3115) = 625 p= 001 The irregulars still

primed strongly (346 msec) F(138) = 8229 p lt 001but the semantically related pairs did not prime signifi-cantly (104msec) F(121) = 234 p= 141 Similarly B Sshowed a main effect of priming F(1115) = 56042 p lt001 and a Priming Condition interaction F(3115) =2885 p = 039 The irregulars primed strongly(243 msec) F(140) = 38941 p lt 001 in the absenceof significant semantic priming (105 msec) F(121) =2717 p= 12 J T also showed a main effect of primingF(197) = 7138 p lt 01 and a Priming Conditioninteraction F(197) = 482 p lt 01 The irregularsproduced robust priming (177 msec) F(136) = 1135p lt 01 while the semantic conditions produced amarginally significant priming effect (80 msec) F 117) =442 p = 051

The failure of semantic priming effects to reachsignificance for the patients individually as well as thestatistically larger difference between semantic and ir-regular priming for the patients compared with thecontrols indicates a significant impairment in semanticpriming for this population In contrast their primingfor the irregulars seemed to be robustly unimpaired andsignificant for each patient individually

DISCUSSION

The aim of this research was to determine whetherpatients with semantic deficits always have associatedproblems in processing irregularly inflected forms aspredicted by current single mechanism accounts ofneuropsychological double dissociations between En-glish regular and irregular past tenses To address thisissue we tested semantic dementia patients on tasksthat probe the production and comprehension of regu-lar and irregularly inflected past tense forms We testedthe patientsrsquo abilities to produce correctly inflected pasttense forms from their present tense variants in twoelicitation experiments In both studies three of the

Figure 3 Auditoryndashauditory

priming study Mean RTs and

standard errors for the patients

(group data for J T E K andB S)

Figure 4 Auditoryndashauditory priming study Priming effects across

groups for the irregular and semantic conditions

Tyler et al 1167

patients were able to correctly inflect both the regularand irregular past tense Only one patient J T hadsignificantly more difficulty in inflecting the irregularscompared with the regulars This set of results suggeststhat semantic deficits do not necessarily go togetherwith difficulties in producing irregularly inflected pasttense forms

A possible argument against this interpretation of datais that the three patients who did not have difficultiesproducing irregularly inflected words may have mildersemantic deficits than the single patient J T who didhave problems with the irregulars However the datafrom the semantic tests do not suggest that J T wasmore severely semantically impaired than the otherthree patients On the Pyramids and Palm Trees testusing either written words or pictures the patients allscored poorly making between 31 and 42 errorsJ Trsquos error rate was 33 Similarly on a wordndashpicturematching task his error rate (23) was within the rangeof the errors produced by the other patients (7ndash34)

Turning to an assessment of the patientsrsquo ability tocomprehend past tense forms we found that the irreg-ulars primed robustly whether we analyzed the patientsas a group or individually suggesting that their ability toaccess the irregulars from a phonological input was notimpaired even for the patient ( J T) who had difficultiesaccessing irregulars in a production task Indeed theoverall group pattern of priming results looks remark-ably normal Like the controls the patient groupshowed priming for both types of past tense and forsemantically related words while phonologically relatedwords did not prime significantly Moreover like thecontrols the patients showed significantly greater prim-ing for the irregulars compared with the regulars

We have previously interpreted priming between averb form such as bring and its irregularly inflected pasttense form (ie brought) as evidence that the relation-ship between them is morphological rather than seman-tic (Marslen-Wilson amp Tyler 1998 2003) Both forms ofthe verb map onto the same underlying stem mor-pheme just as for the regulars and it is the repeatedactivation of this morpheme that gives rise to primingeffects (both at short and long priming delays) On thisaccount the basis for priming between irregularly in-flected words is different from the processes that un-derpin priming between two semantically related butmorphologically unrelated words (eg coatdress) Inthe latter case semantically related words prime be-cause of semantic overlap between two separate lexicalor morphemic representations This predicts that se-mantic deficits and deficits for the irregular past tensewill not necessarily pattern togethermdashwhich is thepattern of results that we obtain here for the semanticdementia patients

To account for neuropsychological dissociations inwhich patients are disproportionately impaired on theregular past tense we have focused on differences in

phonological processes involved in parsing an inflectedform into its stem and affix We assume that regularlyinflected past tense forms are generated as combina-tions of stems and inflectional affixes (open + ed) andrecognized by a process of phonological parsing wherethe speech input is segmented into a stem and its affixProblems with the regulars arise when this phonologi-cal parsing mechanism is impaired usually followingdamage to the left inferior frontal gyrus (LIFG) andassociated regions (Tyler de Mornay Davies et al2002 Tyler Randall et al 2002) Irregularly inflectedforms are stored just like monomorphemic words suchas table or grass as full-form representations with nointernal phonological structure as combinations ofstems and affixes These words do not need to bephonologically parsed into stems and affixes as part ofthe on-line access process These differences in require-ments for phonological parsing accompanied by dam-age to neural systems implicated in these types ofanalysis gives rise to selective deficits for the regularpast tense

Turning to deficits for the irregulars we assume (asnoted above) that an irregular past tense form has astored lexical form representation linking to the sameunderlying morpheme as other forms of the same wordThis predicts that deficits for the irregulars can arise ifbrain damage either affects stored lexical representa-tions directly or disrupts the mechanisms involved inthe process of mapping from phonology to semantics(Marslen-Wilson amp Tyler 1998) A version of this pro-posal is put forward by Miozzo (2003) in the context ofthe anomic patient A W who had problems with theirregulars in the presence of intact semanticsmdashtheopposite pattern to the one reported here for threeout of our four semantically impaired patients Miozzosuggests that this patient typical of anomic patients ingeneral has impairments in lexical processing asevidenced by her difficulty in retrieving a wordrsquos pho-nology On the hypothesis that irregular past tenseforms must be accessed as whole forms so that alterna-tive access routes involving decompositional mecha-nisms cannot be invoked problems in accessing storedlexical representations should differentially disadvantagethe irregulars compared with the regulars

There are some hints in the priming data that J Tthe one patient who showed difficulty in producingirregularly inflected words may also have phonologicalproblems He made many lexical decision errors (31errors) on the items in the phonological condition andmore than in any other condition (11 in the regular5 in the irregular 12 in the semantic conditions)The other patients did not show this pattern B Smade 8 4 0 and 2 errors in the phonologicalregular irregular and semantic conditions respective-ly E K at Time 1 made 8 8 3 and 2 errorson the phonological regular irregular and semanticconditions respectively and 4 6 1 and 2

1168 Journal of Cognitive Neuroscience Volume 16 Number 7

respectively errors in those conditions at the secondtime of testing Similarly J T had a high error rate(30) on the phonologically related foils in the exper-iment such as gossipgoss These data however areonly suggestive Further research is needed to deter-mine the nature of this potential phonological deficitand whether there is a more general relationshipbetween phonological processing deficits and prob-lems in accessing stored lexical forms

One issue that is raised by these results is why thepatients as a group show any semantic priming at allwhen they each have pronounced semantic deficitsThis is not the first time that semantic dementiapatients have been observed to show semantic primingwhile their performance on standard off-line testsshows them to be profoundly impaired We have pre-viously reported semantic priming in three semanticdementia patients (A M Tyler amp Moss 1998 F MTyler Moss Patterson amp Hodges 1997 P P MossTyler Hodges amp Patterson 1995) Given that thesepatients had been unambiguously diagnosed as having asemantic deficit we had not initially predicted that theywould show priming for semantically related words Thefact that priming can still be obtained albeit at asignificantly reduced level suggests that it is not anall-or-none phenomenon Significant priming effectsmay not require the activation of all of the informationassociated with a concept All that may be required isthat the semantic relationship between the two relatedwords (eg onioncarrot) should be stronger to somecriterial degree than between the two unrelated words(eg deskcarrot) This would allow semantic primingeffects even in a damaged system so long as somesemantic information is preserved especially if it isthe shared properties of concepts that are preservedsince it is these that provide the basis for semanticoverlap between related concepts

It has been independently demonstrated that theshared properties of concepts are indeed preservedfollowing brain damage These properties are morehighly correlated with each other and correlated prop-erties are relatively preserved following damage (Moss ampTyler 2000 Tyler et al 2000 Tyler amp Moss 2001 DevlinGonnerman Andersen amp Seidenberg 1998 McRae deSa amp Seidenberg 1997) Moreover studies with patientswho have semantic deficits have shown that the sharedproperties of concepts tend to be preserved while thedistinctive properties tend to be impaired (Tyler amp Moss2001 Moss Tyler Durrant-Peatfield amp Bunn 1998) Thepatients in the current study show the same pattern Ina verification task they were impaired at verifying thedistinctive properties of concepts while relatively pre-served on their shared properties (Table 1) If as arguedabove it is the coactivation of these shared propertiesthat primarily drives priming in the intact system thenthis would explain the preservation of some degree ofsemantic priming in the semantic dementia population

In summary E K and B S did not show any differ-ential impairment for the irregulars compared with theregulars on either elicitation or priming tasks and K Bdid not show a disproportionate deficit for the irregularscompared with the regulars in the elicitation task OnlyJ T showed an association of semantic deficit and prob-lems with the irregulars Moreover when the patientsdid make errors on the irregulars they did not exhibitthe pattern of effects predicted by the single mechanismaccount (Patterson et al 2001) with a higher proportionof errors on low-frequency items These results showthat there is no necessary association between a seman-tic deficit and problems with the irregular past tenseAlthough the two often do co-occur as we and othershave previously documented they can evidently alsooccur independently of each other This dissociationindicates that a separate set of functional capacities isimplicated here supporting the access of stored lexicalrepresentations An important issue for future researchis to identify the neural infrastructure for these accessprocesses so that we can better determine the condi-tions under which semantic deficits will co-occur withdeficits for the irregulars

The present results together with the results fornormal subjects addressed earlier (Marslen-Wilson ampTyler 1998) argue against the claim that the relation-ship between irregular past tense forms and their stemsis primarily a semantic rather than a morphologicalrelationship This in turn undermines the evidencefor the type of single system account of lexical repre-sentation in which morphological structure can becaptured solely on the basis of phonological and seman-tic relationships (Patterson et al 2001 Joanisse amp Seid-enberg 1999)

METHODS

Subjects

In addition to the four semantic dementia patients agroup of seven control subjects (aged between 56 and69 years) were tested on the elicitation task and a groupof 12 subjects (aged 60ndash73 years) were tested on thepriming study

Past Tense Production Elicitation Studies

We tested the patients on two elicitation experimentsdesigned to probe their ability to generate past tenseforms from their stems In the first elicitation testdescribed in Ullman et al (1997) subjects were pre-sented with sentences containing the present tenseforms of regular and irregular words together with aset of nonwords for past-tense generation We used thistest to directly compare our patients with others re-ported in the literature who have deficits in generatingpast tense forms

Tyler et al 1169

For the second elicitation task we selected a set of22 regular and 22 irregular verbs matched as closely aspossible on form and lemma frequency (Baayen Piepen-brock amp Gulikers 1995) and imageability (Coltheart1981) The median word form frequency was 35 for theregulars and 36 for the irregulars The median lemmafrequency was 92 for the regulars and 110 for theirregulars The median imageability was 485 for theregulars and 470 for the irregulars A two-sentencecontext was created for each verb where the secondsentence was incomplete and required a verb inflectedfor the past tense For example for the sentence lsquolsquoMynose sometimes bleeds Last night it rsquorsquo the correctresponse would be bled The test stimuli were precededby five practice trials The sentences were read aloud tothe subjects who were encouraged to produce a pasttense form appropriate for the sentence

Past Tense Comprehension Auditory Priming Study

To test patientsrsquo abilities to comprehend spoken pasttense forms we designed an intramodal auditoryndashauditory priming study contrasting the regulars andirregulars We chose to use a priming study to avoidsome of the pitfalls associated with testing patients ontasks that make heavy metalinguistic demands In thisexperiment target words which were verb stems (suchas pull sing) were preceded by a regularly or irregu-larly inflected past tense prime (pulled sung) or by anunrelated control word In addition to the two mainexperimental conditions we included two other con-ditions The first was a semantic priming condition inwhich the prime-target pairs were semantically but notphonologically or morphologically related (pighorse)We included this to determine whether the patientswould show evidence of a semantic deficit in the sametype of on-line task as used to assess their performanceon the past tense forms The fourth condition con-sisted of a phonological priming condition in whichthe prime-target pairs were phonologically but notsemantically or morphologically related (paintpain)which acted as a control for the effects of phonologicalsimilarity Since the inflected-stem pairs share a greatdeal of phonological overlap we need to be ableto separate out phonological from morphological prim-ing effects

Materials

We selected 42 prime-target pairs in each of the pasttense conditions (regularirregular) and 24 pairs for the2 control conditions (semanticphonological) In boththe regular and irregular past tense conditions theprimes were all inflected past tense forms (eg fixedglanced stuck brought) and the targets consisted oftheir stems all of which were one or two syllables ( fixglance stick bring) Items in the phonological control

condition consisted of morphologically simple monosyl-labic pairs (eg lamplamb plumpplum) that werenot semantically related The proportion of consonant-vowel-consonant (CVC) and consonant-vowel (CV) over-lap between prime and target was matched to theregular past tense pairs In all of the conditions weavoided homophones The two past tense sets and thephonological set were matched as closely as possible fortest prime control prime and target frequency andnumber of syllables (see Table 4) The fourth conditionconsisted of pairs of semantically related words thatwere either phonologically or morphologically related(eg shorecoast alleylane) For this condition weselected pairs of semantically related words that werehighly semantically related (but not associatively) and ofrelatively high frequency This was to ensure significantsemantic priming effects in control subjects to provide abasis for evaluating semantic priming in the patients Asemantic relatedness pretest was carried out in which15 subjects were asked to rate pairs of words accordingto their degree of semantic relatedness on a scale of 1ndash9where 1 was highly unrelated and 9 was highly relatedThe mean semantic relatedness rating for the prime-target pairs was 7 (SD = 102)

Each prime word was matched to a control word infrequency and number of syllables (see Table 4) Inaddition to the 132 test pairs we included 80 real wordfiller pairs making a total of 212 real word pairs Thefillers were an equal mixture of unrelated simpleprimespast tense targets simple primes(-s) inflectedtargetsx (-s) inflected primessimple targets simpleprimessimple targets and unambiguous noun pairsWe also included 148 legal wordnonword pairs whichwere a mixture of phonologically unrelated prime-targetpairs phonologically related pairs with targets derivedfrom regularly inflected past or present tense formsphonologically related pairs with targets made fromirregularly inflected forms and pairs with irregular verbsas primes

Design and Procedure

Test and filler items were pseudorandomly dispersedthroughout the test list We constructed two lists ofstimuli so that the target word only occurred once per

Table 4 Intramodal Priming Study Median Frequencies ofStimuli

Test Prime Target Control Prime

Regulars 55 6 4

Irregulars 6 7 5

Phonological 14 18 13

Semantic 16 10 15

1170 Journal of Cognitive Neuroscience Volume 16 Number 7

version All of the stimuli were recorded by a femalenative speaker of British English digitized at 22 kHz andmarkers placed at the acoustic onset of each targetword The markers triggered a timing device that mea-sured the subjectrsquos reaction time to the target wordPrimes and targets were recorded onto a DAT tape forpresentation to subjects Subjects heard primendashtargetpairs and were asked to make a lexical decision to thetarget by pressing a response key

Acknowledgments

We thank Matt Lambon-Ralph Beth Jeffries and KaralynPatterson for referring patients E K B S and J T and formaking many test scores available to us and to Julie Snowdenfor allowing us to work with K B We thank Lara Brent and BilliRandall for their help with the analyses and the patients fortheir cooperation This work was supported by an MRC (UK)program grant to L K T

Reprint requests should be sent to Lorraine K Tyler Depart-ment of Experimental Psychology University of CambridgeCambridge CB3 ODX UK or via e-mail lktylercslpsycholcamacuk

Note

1 These data were kindly provided by Elizabeth Jeffries

REFERENCES

Baayen R H Piepenbrock R amp Gulikers L (1995) TheCELEX lexical database Linguistic Data ConsortiumPhiladelphia University of Pennsylvania

Coltheart M (1981) The MRC psycholinguistic databaseQuarterly Journal of Experimental Psychology HumanExperimental Psychology 33A 497ndash505

Bunn E M Tyler L K amp Moss H E (1998) Naming picturesThe role of familiarity and property type reexaminedNeuropsychology 12 367ndash379

Devlin J T Gonnerman L M Andersen E S amp SeidenbergM S (1998) Category specific semantic deficits in focaland widespread brain damage A computational accountJournal of Cognitive Neuroscience 10 77ndash94

Friston K J Holmes A P Poline J-B Grasby B J WilliamsC R Frackowiak R S J amp Turner R (1995) Analysis offMRI time-series revisited Neuroscience 2 45ndash53

Galton C J Patterson K Graham K Lambon-Ralph M AWilliams G Antoun N Sahakian B J amp Hodges J R(2001) Differing patterns of temporal atrophy inAlzheimerrsquos disease and semantic dementia Neurology 57216ndash225

Hodges J R Patterson K Oxbury S amp Funnell E (1992)Semantic dementia Progressive fluent aphasia withtemporal lobe atrophy Brain 115 1783ndash1806

Howard D amp Patterson K (1992) The pyramids and palmtrees test Bury St Edmunds Thames Valley Test

Jefferies E Patterson K Jones R W Bateman D amp LambonRalph M A (submitted) A category-specific advantage fornumbers in verbal short-term memory Evidence fromsemantic dementia

Joanisse M F amp Seidenberg M S (1999) Impairments inverb morphology after brain injury A connectionist model

Proceedings of the National Academy of Sciences USA 967592ndash7597

Kay J Lesser R amp Coltheart M (1992) Psycholinguisticassessments of language processing in aphasia (PALPA)London Erlbaum

Longworth C E Marslen-Wilson W D amp Tyler L K(2002) Non-fluent aphasics show an abnormal time courseof semantic activation from the regular past tense[abstract] Journal of Cognitive Neuroscience 14Supplement 55

Marslen-Wilson W D Csibra G Ford M Hatzakis HGaskell G amp Johnson M H (2000) Associations anddissociations in the processing of regular and irregularverbs Electrophysiological evidence Journal of CognitiveNeuroscience 12 Supplement 55E

Marslen-Wilson W D amp Tyler L K (1997) Dissociating typesof mental computation Nature 387 592ndash594

Marslen-Wilson W D amp Tyler L K (1998) Rulesrepresentations and the English past tense Trends inCognitive Science 2 428ndash435

Marslen-Wilson W D amp Tyler L K (2003) Capturingunderlying differentiation in the human language systemTrends in Cognitive Sciences 7 62ndash63

Mathuranath P S Nestor P J Berrios G E Rakowicz Wamp Hodges J (2000) A brief cognitive test battery todifferentiate Alzheimerrsquos disease and frontotemporaldementia Neurology 55 1613ndash1620

McClelland J L amp Patterson K (2002) Rules or connectionsin past-tense inflections What does the evidence rule outTrends in Cognitive Sciences 6 465ndash472

McRae K de Sa V amp Seidenberg M S (1997) On thenature and scope of featural representations of wordmeaning Journal of Experimental Psychology General126 99ndash130

Miozzo M (2003) On the processing of regular and irregularforms of verbs and nouns Evidence from neuropsychologyCognition 87 101ndash127

Moss H E amp Tyler L K (2000) A progressivecategory-specific semantic deficit for non-living thingsNeuropsychologia 38 60ndash82

Moss H E Tyler L K Durrant-Peatfield M amp Bunn E(1998) Two eyes of a see-through Impaired and intactsemantic knowledge in a case of selective deficit for livingthings Neurocase 4 291ndash310

Moss H E Tyler L K Hodges J R amp Patterson K (1995)Exploring the loss of semantic memory in semanticdementia Evidence from a primed monitoring studyNeuropsychology 91 16ndash26

Patterson K Lambon Ralph M A Hodges J Ramp McClelland J L (2001) Deficits in irregular past-tenseverb morphology associated with degraded semanticknowledge Neuropsychologia 39 709ndash724

Pinker S (1991) Rules of language Science 253 530ndash533Pinker S amp Ullman M T (2002) The past and future of thepast tense Trends in Cognitive Sciences 6 456ndash463

Prasada S amp Pinker S (1993) Generalizations of regular andirregular morphological patterns Language and CognitiveProcesses 8 1ndash56

Price C J amp Friston K J (1997) Cognitive conjunction A newapproach to brain activation experiments Neuroimage 5261ndash270

Rumelhart D E amp McClelland J L (1986) On learningthe past tense of English verbs In J L McClelland amp D ERumelhart (Eds) Parallel distributed processing(pp 217ndash270) Cambridge MIT Press

Snowden J S Goulding P J amp Neary D (1989) Semanticdementia A form of circumscribed cerebral atrophyBehavioral Neurology 2 167ndash182

Tyler et al 1171

F(164) = 464 p = 035 This interaction is shown inFigure 4

The group pattern was essentially repeated wheneach patientrsquos data were analyzed separately with onequalificationmdashthe semantic priming effect did not reachsignificance for any of the patients In the first testingperiod E K showed a main effect of priming F(1110) =6574 p lt 001 and a significant Priming Conditioninteraction F(3110) = 408 p = 009 Although theirregulars primed robustly (333 msec) F(137) = 4460p lt 001 the priming effect of 100 msec for the seman-tically related pairs was not significant F(154) = 249p = 120 A year later when tested a second time sheshowed a significant overall priming effect F(1115) =9193 p lt 001 and a significant Priming Conditioninteraction F(3115) = 625 p= 001 The irregulars still

primed strongly (346 msec) F(138) = 8229 p lt 001but the semantically related pairs did not prime signifi-cantly (104msec) F(121) = 234 p= 141 Similarly B Sshowed a main effect of priming F(1115) = 56042 p lt001 and a Priming Condition interaction F(3115) =2885 p = 039 The irregulars primed strongly(243 msec) F(140) = 38941 p lt 001 in the absenceof significant semantic priming (105 msec) F(121) =2717 p= 12 J T also showed a main effect of primingF(197) = 7138 p lt 01 and a Priming Conditioninteraction F(197) = 482 p lt 01 The irregularsproduced robust priming (177 msec) F(136) = 1135p lt 01 while the semantic conditions produced amarginally significant priming effect (80 msec) F 117) =442 p = 051

The failure of semantic priming effects to reachsignificance for the patients individually as well as thestatistically larger difference between semantic and ir-regular priming for the patients compared with thecontrols indicates a significant impairment in semanticpriming for this population In contrast their primingfor the irregulars seemed to be robustly unimpaired andsignificant for each patient individually

DISCUSSION

The aim of this research was to determine whetherpatients with semantic deficits always have associatedproblems in processing irregularly inflected forms aspredicted by current single mechanism accounts ofneuropsychological double dissociations between En-glish regular and irregular past tenses To address thisissue we tested semantic dementia patients on tasksthat probe the production and comprehension of regu-lar and irregularly inflected past tense forms We testedthe patientsrsquo abilities to produce correctly inflected pasttense forms from their present tense variants in twoelicitation experiments In both studies three of the

Figure 3 Auditoryndashauditory

priming study Mean RTs and

standard errors for the patients

(group data for J T E K andB S)

Figure 4 Auditoryndashauditory priming study Priming effects across

groups for the irregular and semantic conditions

Tyler et al 1167

patients were able to correctly inflect both the regularand irregular past tense Only one patient J T hadsignificantly more difficulty in inflecting the irregularscompared with the regulars This set of results suggeststhat semantic deficits do not necessarily go togetherwith difficulties in producing irregularly inflected pasttense forms

A possible argument against this interpretation of datais that the three patients who did not have difficultiesproducing irregularly inflected words may have mildersemantic deficits than the single patient J T who didhave problems with the irregulars However the datafrom the semantic tests do not suggest that J T wasmore severely semantically impaired than the otherthree patients On the Pyramids and Palm Trees testusing either written words or pictures the patients allscored poorly making between 31 and 42 errorsJ Trsquos error rate was 33 Similarly on a wordndashpicturematching task his error rate (23) was within the rangeof the errors produced by the other patients (7ndash34)

Turning to an assessment of the patientsrsquo ability tocomprehend past tense forms we found that the irreg-ulars primed robustly whether we analyzed the patientsas a group or individually suggesting that their ability toaccess the irregulars from a phonological input was notimpaired even for the patient ( J T) who had difficultiesaccessing irregulars in a production task Indeed theoverall group pattern of priming results looks remark-ably normal Like the controls the patient groupshowed priming for both types of past tense and forsemantically related words while phonologically relatedwords did not prime significantly Moreover like thecontrols the patients showed significantly greater prim-ing for the irregulars compared with the regulars

We have previously interpreted priming between averb form such as bring and its irregularly inflected pasttense form (ie brought) as evidence that the relation-ship between them is morphological rather than seman-tic (Marslen-Wilson amp Tyler 1998 2003) Both forms ofthe verb map onto the same underlying stem mor-pheme just as for the regulars and it is the repeatedactivation of this morpheme that gives rise to primingeffects (both at short and long priming delays) On thisaccount the basis for priming between irregularly in-flected words is different from the processes that un-derpin priming between two semantically related butmorphologically unrelated words (eg coatdress) Inthe latter case semantically related words prime be-cause of semantic overlap between two separate lexicalor morphemic representations This predicts that se-mantic deficits and deficits for the irregular past tensewill not necessarily pattern togethermdashwhich is thepattern of results that we obtain here for the semanticdementia patients

To account for neuropsychological dissociations inwhich patients are disproportionately impaired on theregular past tense we have focused on differences in

phonological processes involved in parsing an inflectedform into its stem and affix We assume that regularlyinflected past tense forms are generated as combina-tions of stems and inflectional affixes (open + ed) andrecognized by a process of phonological parsing wherethe speech input is segmented into a stem and its affixProblems with the regulars arise when this phonologi-cal parsing mechanism is impaired usually followingdamage to the left inferior frontal gyrus (LIFG) andassociated regions (Tyler de Mornay Davies et al2002 Tyler Randall et al 2002) Irregularly inflectedforms are stored just like monomorphemic words suchas table or grass as full-form representations with nointernal phonological structure as combinations ofstems and affixes These words do not need to bephonologically parsed into stems and affixes as part ofthe on-line access process These differences in require-ments for phonological parsing accompanied by dam-age to neural systems implicated in these types ofanalysis gives rise to selective deficits for the regularpast tense

Turning to deficits for the irregulars we assume (asnoted above) that an irregular past tense form has astored lexical form representation linking to the sameunderlying morpheme as other forms of the same wordThis predicts that deficits for the irregulars can arise ifbrain damage either affects stored lexical representa-tions directly or disrupts the mechanisms involved inthe process of mapping from phonology to semantics(Marslen-Wilson amp Tyler 1998) A version of this pro-posal is put forward by Miozzo (2003) in the context ofthe anomic patient A W who had problems with theirregulars in the presence of intact semanticsmdashtheopposite pattern to the one reported here for threeout of our four semantically impaired patients Miozzosuggests that this patient typical of anomic patients ingeneral has impairments in lexical processing asevidenced by her difficulty in retrieving a wordrsquos pho-nology On the hypothesis that irregular past tenseforms must be accessed as whole forms so that alterna-tive access routes involving decompositional mecha-nisms cannot be invoked problems in accessing storedlexical representations should differentially disadvantagethe irregulars compared with the regulars

There are some hints in the priming data that J Tthe one patient who showed difficulty in producingirregularly inflected words may also have phonologicalproblems He made many lexical decision errors (31errors) on the items in the phonological condition andmore than in any other condition (11 in the regular5 in the irregular 12 in the semantic conditions)The other patients did not show this pattern B Smade 8 4 0 and 2 errors in the phonologicalregular irregular and semantic conditions respective-ly E K at Time 1 made 8 8 3 and 2 errorson the phonological regular irregular and semanticconditions respectively and 4 6 1 and 2

1168 Journal of Cognitive Neuroscience Volume 16 Number 7

respectively errors in those conditions at the secondtime of testing Similarly J T had a high error rate(30) on the phonologically related foils in the exper-iment such as gossipgoss These data however areonly suggestive Further research is needed to deter-mine the nature of this potential phonological deficitand whether there is a more general relationshipbetween phonological processing deficits and prob-lems in accessing stored lexical forms

One issue that is raised by these results is why thepatients as a group show any semantic priming at allwhen they each have pronounced semantic deficitsThis is not the first time that semantic dementiapatients have been observed to show semantic primingwhile their performance on standard off-line testsshows them to be profoundly impaired We have pre-viously reported semantic priming in three semanticdementia patients (A M Tyler amp Moss 1998 F MTyler Moss Patterson amp Hodges 1997 P P MossTyler Hodges amp Patterson 1995) Given that thesepatients had been unambiguously diagnosed as having asemantic deficit we had not initially predicted that theywould show priming for semantically related words Thefact that priming can still be obtained albeit at asignificantly reduced level suggests that it is not anall-or-none phenomenon Significant priming effectsmay not require the activation of all of the informationassociated with a concept All that may be required isthat the semantic relationship between the two relatedwords (eg onioncarrot) should be stronger to somecriterial degree than between the two unrelated words(eg deskcarrot) This would allow semantic primingeffects even in a damaged system so long as somesemantic information is preserved especially if it isthe shared properties of concepts that are preservedsince it is these that provide the basis for semanticoverlap between related concepts

It has been independently demonstrated that theshared properties of concepts are indeed preservedfollowing brain damage These properties are morehighly correlated with each other and correlated prop-erties are relatively preserved following damage (Moss ampTyler 2000 Tyler et al 2000 Tyler amp Moss 2001 DevlinGonnerman Andersen amp Seidenberg 1998 McRae deSa amp Seidenberg 1997) Moreover studies with patientswho have semantic deficits have shown that the sharedproperties of concepts tend to be preserved while thedistinctive properties tend to be impaired (Tyler amp Moss2001 Moss Tyler Durrant-Peatfield amp Bunn 1998) Thepatients in the current study show the same pattern Ina verification task they were impaired at verifying thedistinctive properties of concepts while relatively pre-served on their shared properties (Table 1) If as arguedabove it is the coactivation of these shared propertiesthat primarily drives priming in the intact system thenthis would explain the preservation of some degree ofsemantic priming in the semantic dementia population

In summary E K and B S did not show any differ-ential impairment for the irregulars compared with theregulars on either elicitation or priming tasks and K Bdid not show a disproportionate deficit for the irregularscompared with the regulars in the elicitation task OnlyJ T showed an association of semantic deficit and prob-lems with the irregulars Moreover when the patientsdid make errors on the irregulars they did not exhibitthe pattern of effects predicted by the single mechanismaccount (Patterson et al 2001) with a higher proportionof errors on low-frequency items These results showthat there is no necessary association between a seman-tic deficit and problems with the irregular past tenseAlthough the two often do co-occur as we and othershave previously documented they can evidently alsooccur independently of each other This dissociationindicates that a separate set of functional capacities isimplicated here supporting the access of stored lexicalrepresentations An important issue for future researchis to identify the neural infrastructure for these accessprocesses so that we can better determine the condi-tions under which semantic deficits will co-occur withdeficits for the irregulars

The present results together with the results fornormal subjects addressed earlier (Marslen-Wilson ampTyler 1998) argue against the claim that the relation-ship between irregular past tense forms and their stemsis primarily a semantic rather than a morphologicalrelationship This in turn undermines the evidencefor the type of single system account of lexical repre-sentation in which morphological structure can becaptured solely on the basis of phonological and seman-tic relationships (Patterson et al 2001 Joanisse amp Seid-enberg 1999)

METHODS

Subjects

In addition to the four semantic dementia patients agroup of seven control subjects (aged between 56 and69 years) were tested on the elicitation task and a groupof 12 subjects (aged 60ndash73 years) were tested on thepriming study

Past Tense Production Elicitation Studies

We tested the patients on two elicitation experimentsdesigned to probe their ability to generate past tenseforms from their stems In the first elicitation testdescribed in Ullman et al (1997) subjects were pre-sented with sentences containing the present tenseforms of regular and irregular words together with aset of nonwords for past-tense generation We used thistest to directly compare our patients with others re-ported in the literature who have deficits in generatingpast tense forms

Tyler et al 1169

For the second elicitation task we selected a set of22 regular and 22 irregular verbs matched as closely aspossible on form and lemma frequency (Baayen Piepen-brock amp Gulikers 1995) and imageability (Coltheart1981) The median word form frequency was 35 for theregulars and 36 for the irregulars The median lemmafrequency was 92 for the regulars and 110 for theirregulars The median imageability was 485 for theregulars and 470 for the irregulars A two-sentencecontext was created for each verb where the secondsentence was incomplete and required a verb inflectedfor the past tense For example for the sentence lsquolsquoMynose sometimes bleeds Last night it rsquorsquo the correctresponse would be bled The test stimuli were precededby five practice trials The sentences were read aloud tothe subjects who were encouraged to produce a pasttense form appropriate for the sentence

Past Tense Comprehension Auditory Priming Study

To test patientsrsquo abilities to comprehend spoken pasttense forms we designed an intramodal auditoryndashauditory priming study contrasting the regulars andirregulars We chose to use a priming study to avoidsome of the pitfalls associated with testing patients ontasks that make heavy metalinguistic demands In thisexperiment target words which were verb stems (suchas pull sing) were preceded by a regularly or irregu-larly inflected past tense prime (pulled sung) or by anunrelated control word In addition to the two mainexperimental conditions we included two other con-ditions The first was a semantic priming condition inwhich the prime-target pairs were semantically but notphonologically or morphologically related (pighorse)We included this to determine whether the patientswould show evidence of a semantic deficit in the sametype of on-line task as used to assess their performanceon the past tense forms The fourth condition con-sisted of a phonological priming condition in whichthe prime-target pairs were phonologically but notsemantically or morphologically related (paintpain)which acted as a control for the effects of phonologicalsimilarity Since the inflected-stem pairs share a greatdeal of phonological overlap we need to be ableto separate out phonological from morphological prim-ing effects

Materials

We selected 42 prime-target pairs in each of the pasttense conditions (regularirregular) and 24 pairs for the2 control conditions (semanticphonological) In boththe regular and irregular past tense conditions theprimes were all inflected past tense forms (eg fixedglanced stuck brought) and the targets consisted oftheir stems all of which were one or two syllables ( fixglance stick bring) Items in the phonological control

condition consisted of morphologically simple monosyl-labic pairs (eg lamplamb plumpplum) that werenot semantically related The proportion of consonant-vowel-consonant (CVC) and consonant-vowel (CV) over-lap between prime and target was matched to theregular past tense pairs In all of the conditions weavoided homophones The two past tense sets and thephonological set were matched as closely as possible fortest prime control prime and target frequency andnumber of syllables (see Table 4) The fourth conditionconsisted of pairs of semantically related words thatwere either phonologically or morphologically related(eg shorecoast alleylane) For this condition weselected pairs of semantically related words that werehighly semantically related (but not associatively) and ofrelatively high frequency This was to ensure significantsemantic priming effects in control subjects to provide abasis for evaluating semantic priming in the patients Asemantic relatedness pretest was carried out in which15 subjects were asked to rate pairs of words accordingto their degree of semantic relatedness on a scale of 1ndash9where 1 was highly unrelated and 9 was highly relatedThe mean semantic relatedness rating for the prime-target pairs was 7 (SD = 102)

Each prime word was matched to a control word infrequency and number of syllables (see Table 4) Inaddition to the 132 test pairs we included 80 real wordfiller pairs making a total of 212 real word pairs Thefillers were an equal mixture of unrelated simpleprimespast tense targets simple primes(-s) inflectedtargetsx (-s) inflected primessimple targets simpleprimessimple targets and unambiguous noun pairsWe also included 148 legal wordnonword pairs whichwere a mixture of phonologically unrelated prime-targetpairs phonologically related pairs with targets derivedfrom regularly inflected past or present tense formsphonologically related pairs with targets made fromirregularly inflected forms and pairs with irregular verbsas primes

Design and Procedure

Test and filler items were pseudorandomly dispersedthroughout the test list We constructed two lists ofstimuli so that the target word only occurred once per

Table 4 Intramodal Priming Study Median Frequencies ofStimuli

Test Prime Target Control Prime

Regulars 55 6 4

Irregulars 6 7 5

Phonological 14 18 13

Semantic 16 10 15

1170 Journal of Cognitive Neuroscience Volume 16 Number 7

version All of the stimuli were recorded by a femalenative speaker of British English digitized at 22 kHz andmarkers placed at the acoustic onset of each targetword The markers triggered a timing device that mea-sured the subjectrsquos reaction time to the target wordPrimes and targets were recorded onto a DAT tape forpresentation to subjects Subjects heard primendashtargetpairs and were asked to make a lexical decision to thetarget by pressing a response key

Acknowledgments

We thank Matt Lambon-Ralph Beth Jeffries and KaralynPatterson for referring patients E K B S and J T and formaking many test scores available to us and to Julie Snowdenfor allowing us to work with K B We thank Lara Brent and BilliRandall for their help with the analyses and the patients fortheir cooperation This work was supported by an MRC (UK)program grant to L K T

Reprint requests should be sent to Lorraine K Tyler Depart-ment of Experimental Psychology University of CambridgeCambridge CB3 ODX UK or via e-mail lktylercslpsycholcamacuk

Note

1 These data were kindly provided by Elizabeth Jeffries

REFERENCES

Baayen R H Piepenbrock R amp Gulikers L (1995) TheCELEX lexical database Linguistic Data ConsortiumPhiladelphia University of Pennsylvania

Coltheart M (1981) The MRC psycholinguistic databaseQuarterly Journal of Experimental Psychology HumanExperimental Psychology 33A 497ndash505

Bunn E M Tyler L K amp Moss H E (1998) Naming picturesThe role of familiarity and property type reexaminedNeuropsychology 12 367ndash379

Devlin J T Gonnerman L M Andersen E S amp SeidenbergM S (1998) Category specific semantic deficits in focaland widespread brain damage A computational accountJournal of Cognitive Neuroscience 10 77ndash94

Friston K J Holmes A P Poline J-B Grasby B J WilliamsC R Frackowiak R S J amp Turner R (1995) Analysis offMRI time-series revisited Neuroscience 2 45ndash53

Galton C J Patterson K Graham K Lambon-Ralph M AWilliams G Antoun N Sahakian B J amp Hodges J R(2001) Differing patterns of temporal atrophy inAlzheimerrsquos disease and semantic dementia Neurology 57216ndash225

Hodges J R Patterson K Oxbury S amp Funnell E (1992)Semantic dementia Progressive fluent aphasia withtemporal lobe atrophy Brain 115 1783ndash1806

Howard D amp Patterson K (1992) The pyramids and palmtrees test Bury St Edmunds Thames Valley Test

Jefferies E Patterson K Jones R W Bateman D amp LambonRalph M A (submitted) A category-specific advantage fornumbers in verbal short-term memory Evidence fromsemantic dementia

Joanisse M F amp Seidenberg M S (1999) Impairments inverb morphology after brain injury A connectionist model

Proceedings of the National Academy of Sciences USA 967592ndash7597

Kay J Lesser R amp Coltheart M (1992) Psycholinguisticassessments of language processing in aphasia (PALPA)London Erlbaum

Longworth C E Marslen-Wilson W D amp Tyler L K(2002) Non-fluent aphasics show an abnormal time courseof semantic activation from the regular past tense[abstract] Journal of Cognitive Neuroscience 14Supplement 55

Marslen-Wilson W D Csibra G Ford M Hatzakis HGaskell G amp Johnson M H (2000) Associations anddissociations in the processing of regular and irregularverbs Electrophysiological evidence Journal of CognitiveNeuroscience 12 Supplement 55E

Marslen-Wilson W D amp Tyler L K (1997) Dissociating typesof mental computation Nature 387 592ndash594

Marslen-Wilson W D amp Tyler L K (1998) Rulesrepresentations and the English past tense Trends inCognitive Science 2 428ndash435

Marslen-Wilson W D amp Tyler L K (2003) Capturingunderlying differentiation in the human language systemTrends in Cognitive Sciences 7 62ndash63

Mathuranath P S Nestor P J Berrios G E Rakowicz Wamp Hodges J (2000) A brief cognitive test battery todifferentiate Alzheimerrsquos disease and frontotemporaldementia Neurology 55 1613ndash1620

McClelland J L amp Patterson K (2002) Rules or connectionsin past-tense inflections What does the evidence rule outTrends in Cognitive Sciences 6 465ndash472

McRae K de Sa V amp Seidenberg M S (1997) On thenature and scope of featural representations of wordmeaning Journal of Experimental Psychology General126 99ndash130

Miozzo M (2003) On the processing of regular and irregularforms of verbs and nouns Evidence from neuropsychologyCognition 87 101ndash127

Moss H E amp Tyler L K (2000) A progressivecategory-specific semantic deficit for non-living thingsNeuropsychologia 38 60ndash82

Moss H E Tyler L K Durrant-Peatfield M amp Bunn E(1998) Two eyes of a see-through Impaired and intactsemantic knowledge in a case of selective deficit for livingthings Neurocase 4 291ndash310

Moss H E Tyler L K Hodges J R amp Patterson K (1995)Exploring the loss of semantic memory in semanticdementia Evidence from a primed monitoring studyNeuropsychology 91 16ndash26

Patterson K Lambon Ralph M A Hodges J Ramp McClelland J L (2001) Deficits in irregular past-tenseverb morphology associated with degraded semanticknowledge Neuropsychologia 39 709ndash724

Pinker S (1991) Rules of language Science 253 530ndash533Pinker S amp Ullman M T (2002) The past and future of thepast tense Trends in Cognitive Sciences 6 456ndash463

Prasada S amp Pinker S (1993) Generalizations of regular andirregular morphological patterns Language and CognitiveProcesses 8 1ndash56

Price C J amp Friston K J (1997) Cognitive conjunction A newapproach to brain activation experiments Neuroimage 5261ndash270

Rumelhart D E amp McClelland J L (1986) On learningthe past tense of English verbs In J L McClelland amp D ERumelhart (Eds) Parallel distributed processing(pp 217ndash270) Cambridge MIT Press

Snowden J S Goulding P J amp Neary D (1989) Semanticdementia A form of circumscribed cerebral atrophyBehavioral Neurology 2 167ndash182

Tyler et al 1171

patients were able to correctly inflect both the regularand irregular past tense Only one patient J T hadsignificantly more difficulty in inflecting the irregularscompared with the regulars This set of results suggeststhat semantic deficits do not necessarily go togetherwith difficulties in producing irregularly inflected pasttense forms

A possible argument against this interpretation of datais that the three patients who did not have difficultiesproducing irregularly inflected words may have mildersemantic deficits than the single patient J T who didhave problems with the irregulars However the datafrom the semantic tests do not suggest that J T wasmore severely semantically impaired than the otherthree patients On the Pyramids and Palm Trees testusing either written words or pictures the patients allscored poorly making between 31 and 42 errorsJ Trsquos error rate was 33 Similarly on a wordndashpicturematching task his error rate (23) was within the rangeof the errors produced by the other patients (7ndash34)

Turning to an assessment of the patientsrsquo ability tocomprehend past tense forms we found that the irreg-ulars primed robustly whether we analyzed the patientsas a group or individually suggesting that their ability toaccess the irregulars from a phonological input was notimpaired even for the patient ( J T) who had difficultiesaccessing irregulars in a production task Indeed theoverall group pattern of priming results looks remark-ably normal Like the controls the patient groupshowed priming for both types of past tense and forsemantically related words while phonologically relatedwords did not prime significantly Moreover like thecontrols the patients showed significantly greater prim-ing for the irregulars compared with the regulars

We have previously interpreted priming between averb form such as bring and its irregularly inflected pasttense form (ie brought) as evidence that the relation-ship between them is morphological rather than seman-tic (Marslen-Wilson amp Tyler 1998 2003) Both forms ofthe verb map onto the same underlying stem mor-pheme just as for the regulars and it is the repeatedactivation of this morpheme that gives rise to primingeffects (both at short and long priming delays) On thisaccount the basis for priming between irregularly in-flected words is different from the processes that un-derpin priming between two semantically related butmorphologically unrelated words (eg coatdress) Inthe latter case semantically related words prime be-cause of semantic overlap between two separate lexicalor morphemic representations This predicts that se-mantic deficits and deficits for the irregular past tensewill not necessarily pattern togethermdashwhich is thepattern of results that we obtain here for the semanticdementia patients

To account for neuropsychological dissociations inwhich patients are disproportionately impaired on theregular past tense we have focused on differences in

phonological processes involved in parsing an inflectedform into its stem and affix We assume that regularlyinflected past tense forms are generated as combina-tions of stems and inflectional affixes (open + ed) andrecognized by a process of phonological parsing wherethe speech input is segmented into a stem and its affixProblems with the regulars arise when this phonologi-cal parsing mechanism is impaired usually followingdamage to the left inferior frontal gyrus (LIFG) andassociated regions (Tyler de Mornay Davies et al2002 Tyler Randall et al 2002) Irregularly inflectedforms are stored just like monomorphemic words suchas table or grass as full-form representations with nointernal phonological structure as combinations ofstems and affixes These words do not need to bephonologically parsed into stems and affixes as part ofthe on-line access process These differences in require-ments for phonological parsing accompanied by dam-age to neural systems implicated in these types ofanalysis gives rise to selective deficits for the regularpast tense

Turning to deficits for the irregulars we assume (asnoted above) that an irregular past tense form has astored lexical form representation linking to the sameunderlying morpheme as other forms of the same wordThis predicts that deficits for the irregulars can arise ifbrain damage either affects stored lexical representa-tions directly or disrupts the mechanisms involved inthe process of mapping from phonology to semantics(Marslen-Wilson amp Tyler 1998) A version of this pro-posal is put forward by Miozzo (2003) in the context ofthe anomic patient A W who had problems with theirregulars in the presence of intact semanticsmdashtheopposite pattern to the one reported here for threeout of our four semantically impaired patients Miozzosuggests that this patient typical of anomic patients ingeneral has impairments in lexical processing asevidenced by her difficulty in retrieving a wordrsquos pho-nology On the hypothesis that irregular past tenseforms must be accessed as whole forms so that alterna-tive access routes involving decompositional mecha-nisms cannot be invoked problems in accessing storedlexical representations should differentially disadvantagethe irregulars compared with the regulars

There are some hints in the priming data that J Tthe one patient who showed difficulty in producingirregularly inflected words may also have phonologicalproblems He made many lexical decision errors (31errors) on the items in the phonological condition andmore than in any other condition (11 in the regular5 in the irregular 12 in the semantic conditions)The other patients did not show this pattern B Smade 8 4 0 and 2 errors in the phonologicalregular irregular and semantic conditions respective-ly E K at Time 1 made 8 8 3 and 2 errorson the phonological regular irregular and semanticconditions respectively and 4 6 1 and 2

1168 Journal of Cognitive Neuroscience Volume 16 Number 7

respectively errors in those conditions at the secondtime of testing Similarly J T had a high error rate(30) on the phonologically related foils in the exper-iment such as gossipgoss These data however areonly suggestive Further research is needed to deter-mine the nature of this potential phonological deficitand whether there is a more general relationshipbetween phonological processing deficits and prob-lems in accessing stored lexical forms

One issue that is raised by these results is why thepatients as a group show any semantic priming at allwhen they each have pronounced semantic deficitsThis is not the first time that semantic dementiapatients have been observed to show semantic primingwhile their performance on standard off-line testsshows them to be profoundly impaired We have pre-viously reported semantic priming in three semanticdementia patients (A M Tyler amp Moss 1998 F MTyler Moss Patterson amp Hodges 1997 P P MossTyler Hodges amp Patterson 1995) Given that thesepatients had been unambiguously diagnosed as having asemantic deficit we had not initially predicted that theywould show priming for semantically related words Thefact that priming can still be obtained albeit at asignificantly reduced level suggests that it is not anall-or-none phenomenon Significant priming effectsmay not require the activation of all of the informationassociated with a concept All that may be required isthat the semantic relationship between the two relatedwords (eg onioncarrot) should be stronger to somecriterial degree than between the two unrelated words(eg deskcarrot) This would allow semantic primingeffects even in a damaged system so long as somesemantic information is preserved especially if it isthe shared properties of concepts that are preservedsince it is these that provide the basis for semanticoverlap between related concepts

It has been independently demonstrated that theshared properties of concepts are indeed preservedfollowing brain damage These properties are morehighly correlated with each other and correlated prop-erties are relatively preserved following damage (Moss ampTyler 2000 Tyler et al 2000 Tyler amp Moss 2001 DevlinGonnerman Andersen amp Seidenberg 1998 McRae deSa amp Seidenberg 1997) Moreover studies with patientswho have semantic deficits have shown that the sharedproperties of concepts tend to be preserved while thedistinctive properties tend to be impaired (Tyler amp Moss2001 Moss Tyler Durrant-Peatfield amp Bunn 1998) Thepatients in the current study show the same pattern Ina verification task they were impaired at verifying thedistinctive properties of concepts while relatively pre-served on their shared properties (Table 1) If as arguedabove it is the coactivation of these shared propertiesthat primarily drives priming in the intact system thenthis would explain the preservation of some degree ofsemantic priming in the semantic dementia population

In summary E K and B S did not show any differ-ential impairment for the irregulars compared with theregulars on either elicitation or priming tasks and K Bdid not show a disproportionate deficit for the irregularscompared with the regulars in the elicitation task OnlyJ T showed an association of semantic deficit and prob-lems with the irregulars Moreover when the patientsdid make errors on the irregulars they did not exhibitthe pattern of effects predicted by the single mechanismaccount (Patterson et al 2001) with a higher proportionof errors on low-frequency items These results showthat there is no necessary association between a seman-tic deficit and problems with the irregular past tenseAlthough the two often do co-occur as we and othershave previously documented they can evidently alsooccur independently of each other This dissociationindicates that a separate set of functional capacities isimplicated here supporting the access of stored lexicalrepresentations An important issue for future researchis to identify the neural infrastructure for these accessprocesses so that we can better determine the condi-tions under which semantic deficits will co-occur withdeficits for the irregulars

The present results together with the results fornormal subjects addressed earlier (Marslen-Wilson ampTyler 1998) argue against the claim that the relation-ship between irregular past tense forms and their stemsis primarily a semantic rather than a morphologicalrelationship This in turn undermines the evidencefor the type of single system account of lexical repre-sentation in which morphological structure can becaptured solely on the basis of phonological and seman-tic relationships (Patterson et al 2001 Joanisse amp Seid-enberg 1999)

METHODS

Subjects

In addition to the four semantic dementia patients agroup of seven control subjects (aged between 56 and69 years) were tested on the elicitation task and a groupof 12 subjects (aged 60ndash73 years) were tested on thepriming study

Past Tense Production Elicitation Studies

We tested the patients on two elicitation experimentsdesigned to probe their ability to generate past tenseforms from their stems In the first elicitation testdescribed in Ullman et al (1997) subjects were pre-sented with sentences containing the present tenseforms of regular and irregular words together with aset of nonwords for past-tense generation We used thistest to directly compare our patients with others re-ported in the literature who have deficits in generatingpast tense forms

Tyler et al 1169

For the second elicitation task we selected a set of22 regular and 22 irregular verbs matched as closely aspossible on form and lemma frequency (Baayen Piepen-brock amp Gulikers 1995) and imageability (Coltheart1981) The median word form frequency was 35 for theregulars and 36 for the irregulars The median lemmafrequency was 92 for the regulars and 110 for theirregulars The median imageability was 485 for theregulars and 470 for the irregulars A two-sentencecontext was created for each verb where the secondsentence was incomplete and required a verb inflectedfor the past tense For example for the sentence lsquolsquoMynose sometimes bleeds Last night it rsquorsquo the correctresponse would be bled The test stimuli were precededby five practice trials The sentences were read aloud tothe subjects who were encouraged to produce a pasttense form appropriate for the sentence

Past Tense Comprehension Auditory Priming Study

To test patientsrsquo abilities to comprehend spoken pasttense forms we designed an intramodal auditoryndashauditory priming study contrasting the regulars andirregulars We chose to use a priming study to avoidsome of the pitfalls associated with testing patients ontasks that make heavy metalinguistic demands In thisexperiment target words which were verb stems (suchas pull sing) were preceded by a regularly or irregu-larly inflected past tense prime (pulled sung) or by anunrelated control word In addition to the two mainexperimental conditions we included two other con-ditions The first was a semantic priming condition inwhich the prime-target pairs were semantically but notphonologically or morphologically related (pighorse)We included this to determine whether the patientswould show evidence of a semantic deficit in the sametype of on-line task as used to assess their performanceon the past tense forms The fourth condition con-sisted of a phonological priming condition in whichthe prime-target pairs were phonologically but notsemantically or morphologically related (paintpain)which acted as a control for the effects of phonologicalsimilarity Since the inflected-stem pairs share a greatdeal of phonological overlap we need to be ableto separate out phonological from morphological prim-ing effects

Materials

We selected 42 prime-target pairs in each of the pasttense conditions (regularirregular) and 24 pairs for the2 control conditions (semanticphonological) In boththe regular and irregular past tense conditions theprimes were all inflected past tense forms (eg fixedglanced stuck brought) and the targets consisted oftheir stems all of which were one or two syllables ( fixglance stick bring) Items in the phonological control

condition consisted of morphologically simple monosyl-labic pairs (eg lamplamb plumpplum) that werenot semantically related The proportion of consonant-vowel-consonant (CVC) and consonant-vowel (CV) over-lap between prime and target was matched to theregular past tense pairs In all of the conditions weavoided homophones The two past tense sets and thephonological set were matched as closely as possible fortest prime control prime and target frequency andnumber of syllables (see Table 4) The fourth conditionconsisted of pairs of semantically related words thatwere either phonologically or morphologically related(eg shorecoast alleylane) For this condition weselected pairs of semantically related words that werehighly semantically related (but not associatively) and ofrelatively high frequency This was to ensure significantsemantic priming effects in control subjects to provide abasis for evaluating semantic priming in the patients Asemantic relatedness pretest was carried out in which15 subjects were asked to rate pairs of words accordingto their degree of semantic relatedness on a scale of 1ndash9where 1 was highly unrelated and 9 was highly relatedThe mean semantic relatedness rating for the prime-target pairs was 7 (SD = 102)

Each prime word was matched to a control word infrequency and number of syllables (see Table 4) Inaddition to the 132 test pairs we included 80 real wordfiller pairs making a total of 212 real word pairs Thefillers were an equal mixture of unrelated simpleprimespast tense targets simple primes(-s) inflectedtargetsx (-s) inflected primessimple targets simpleprimessimple targets and unambiguous noun pairsWe also included 148 legal wordnonword pairs whichwere a mixture of phonologically unrelated prime-targetpairs phonologically related pairs with targets derivedfrom regularly inflected past or present tense formsphonologically related pairs with targets made fromirregularly inflected forms and pairs with irregular verbsas primes

Design and Procedure

Test and filler items were pseudorandomly dispersedthroughout the test list We constructed two lists ofstimuli so that the target word only occurred once per

Table 4 Intramodal Priming Study Median Frequencies ofStimuli

Test Prime Target Control Prime

Regulars 55 6 4

Irregulars 6 7 5

Phonological 14 18 13

Semantic 16 10 15

1170 Journal of Cognitive Neuroscience Volume 16 Number 7

version All of the stimuli were recorded by a femalenative speaker of British English digitized at 22 kHz andmarkers placed at the acoustic onset of each targetword The markers triggered a timing device that mea-sured the subjectrsquos reaction time to the target wordPrimes and targets were recorded onto a DAT tape forpresentation to subjects Subjects heard primendashtargetpairs and were asked to make a lexical decision to thetarget by pressing a response key

Acknowledgments

We thank Matt Lambon-Ralph Beth Jeffries and KaralynPatterson for referring patients E K B S and J T and formaking many test scores available to us and to Julie Snowdenfor allowing us to work with K B We thank Lara Brent and BilliRandall for their help with the analyses and the patients fortheir cooperation This work was supported by an MRC (UK)program grant to L K T

Reprint requests should be sent to Lorraine K Tyler Depart-ment of Experimental Psychology University of CambridgeCambridge CB3 ODX UK or via e-mail lktylercslpsycholcamacuk

Note

1 These data were kindly provided by Elizabeth Jeffries

REFERENCES

Baayen R H Piepenbrock R amp Gulikers L (1995) TheCELEX lexical database Linguistic Data ConsortiumPhiladelphia University of Pennsylvania

Coltheart M (1981) The MRC psycholinguistic databaseQuarterly Journal of Experimental Psychology HumanExperimental Psychology 33A 497ndash505

Bunn E M Tyler L K amp Moss H E (1998) Naming picturesThe role of familiarity and property type reexaminedNeuropsychology 12 367ndash379

Devlin J T Gonnerman L M Andersen E S amp SeidenbergM S (1998) Category specific semantic deficits in focaland widespread brain damage A computational accountJournal of Cognitive Neuroscience 10 77ndash94

Friston K J Holmes A P Poline J-B Grasby B J WilliamsC R Frackowiak R S J amp Turner R (1995) Analysis offMRI time-series revisited Neuroscience 2 45ndash53

Galton C J Patterson K Graham K Lambon-Ralph M AWilliams G Antoun N Sahakian B J amp Hodges J R(2001) Differing patterns of temporal atrophy inAlzheimerrsquos disease and semantic dementia Neurology 57216ndash225

Hodges J R Patterson K Oxbury S amp Funnell E (1992)Semantic dementia Progressive fluent aphasia withtemporal lobe atrophy Brain 115 1783ndash1806

Howard D amp Patterson K (1992) The pyramids and palmtrees test Bury St Edmunds Thames Valley Test

Jefferies E Patterson K Jones R W Bateman D amp LambonRalph M A (submitted) A category-specific advantage fornumbers in verbal short-term memory Evidence fromsemantic dementia

Joanisse M F amp Seidenberg M S (1999) Impairments inverb morphology after brain injury A connectionist model

Proceedings of the National Academy of Sciences USA 967592ndash7597

Kay J Lesser R amp Coltheart M (1992) Psycholinguisticassessments of language processing in aphasia (PALPA)London Erlbaum

Longworth C E Marslen-Wilson W D amp Tyler L K(2002) Non-fluent aphasics show an abnormal time courseof semantic activation from the regular past tense[abstract] Journal of Cognitive Neuroscience 14Supplement 55

Marslen-Wilson W D Csibra G Ford M Hatzakis HGaskell G amp Johnson M H (2000) Associations anddissociations in the processing of regular and irregularverbs Electrophysiological evidence Journal of CognitiveNeuroscience 12 Supplement 55E

Marslen-Wilson W D amp Tyler L K (1997) Dissociating typesof mental computation Nature 387 592ndash594

Marslen-Wilson W D amp Tyler L K (1998) Rulesrepresentations and the English past tense Trends inCognitive Science 2 428ndash435

Marslen-Wilson W D amp Tyler L K (2003) Capturingunderlying differentiation in the human language systemTrends in Cognitive Sciences 7 62ndash63

Mathuranath P S Nestor P J Berrios G E Rakowicz Wamp Hodges J (2000) A brief cognitive test battery todifferentiate Alzheimerrsquos disease and frontotemporaldementia Neurology 55 1613ndash1620

McClelland J L amp Patterson K (2002) Rules or connectionsin past-tense inflections What does the evidence rule outTrends in Cognitive Sciences 6 465ndash472

McRae K de Sa V amp Seidenberg M S (1997) On thenature and scope of featural representations of wordmeaning Journal of Experimental Psychology General126 99ndash130

Miozzo M (2003) On the processing of regular and irregularforms of verbs and nouns Evidence from neuropsychologyCognition 87 101ndash127

Moss H E amp Tyler L K (2000) A progressivecategory-specific semantic deficit for non-living thingsNeuropsychologia 38 60ndash82

Moss H E Tyler L K Durrant-Peatfield M amp Bunn E(1998) Two eyes of a see-through Impaired and intactsemantic knowledge in a case of selective deficit for livingthings Neurocase 4 291ndash310

Moss H E Tyler L K Hodges J R amp Patterson K (1995)Exploring the loss of semantic memory in semanticdementia Evidence from a primed monitoring studyNeuropsychology 91 16ndash26

Patterson K Lambon Ralph M A Hodges J Ramp McClelland J L (2001) Deficits in irregular past-tenseverb morphology associated with degraded semanticknowledge Neuropsychologia 39 709ndash724

Pinker S (1991) Rules of language Science 253 530ndash533Pinker S amp Ullman M T (2002) The past and future of thepast tense Trends in Cognitive Sciences 6 456ndash463

Prasada S amp Pinker S (1993) Generalizations of regular andirregular morphological patterns Language and CognitiveProcesses 8 1ndash56

Price C J amp Friston K J (1997) Cognitive conjunction A newapproach to brain activation experiments Neuroimage 5261ndash270

Rumelhart D E amp McClelland J L (1986) On learningthe past tense of English verbs In J L McClelland amp D ERumelhart (Eds) Parallel distributed processing(pp 217ndash270) Cambridge MIT Press

Snowden J S Goulding P J amp Neary D (1989) Semanticdementia A form of circumscribed cerebral atrophyBehavioral Neurology 2 167ndash182

Tyler et al 1171

respectively errors in those conditions at the secondtime of testing Similarly J T had a high error rate(30) on the phonologically related foils in the exper-iment such as gossipgoss These data however areonly suggestive Further research is needed to deter-mine the nature of this potential phonological deficitand whether there is a more general relationshipbetween phonological processing deficits and prob-lems in accessing stored lexical forms

One issue that is raised by these results is why thepatients as a group show any semantic priming at allwhen they each have pronounced semantic deficitsThis is not the first time that semantic dementiapatients have been observed to show semantic primingwhile their performance on standard off-line testsshows them to be profoundly impaired We have pre-viously reported semantic priming in three semanticdementia patients (A M Tyler amp Moss 1998 F MTyler Moss Patterson amp Hodges 1997 P P MossTyler Hodges amp Patterson 1995) Given that thesepatients had been unambiguously diagnosed as having asemantic deficit we had not initially predicted that theywould show priming for semantically related words Thefact that priming can still be obtained albeit at asignificantly reduced level suggests that it is not anall-or-none phenomenon Significant priming effectsmay not require the activation of all of the informationassociated with a concept All that may be required isthat the semantic relationship between the two relatedwords (eg onioncarrot) should be stronger to somecriterial degree than between the two unrelated words(eg deskcarrot) This would allow semantic primingeffects even in a damaged system so long as somesemantic information is preserved especially if it isthe shared properties of concepts that are preservedsince it is these that provide the basis for semanticoverlap between related concepts

It has been independently demonstrated that theshared properties of concepts are indeed preservedfollowing brain damage These properties are morehighly correlated with each other and correlated prop-erties are relatively preserved following damage (Moss ampTyler 2000 Tyler et al 2000 Tyler amp Moss 2001 DevlinGonnerman Andersen amp Seidenberg 1998 McRae deSa amp Seidenberg 1997) Moreover studies with patientswho have semantic deficits have shown that the sharedproperties of concepts tend to be preserved while thedistinctive properties tend to be impaired (Tyler amp Moss2001 Moss Tyler Durrant-Peatfield amp Bunn 1998) Thepatients in the current study show the same pattern Ina verification task they were impaired at verifying thedistinctive properties of concepts while relatively pre-served on their shared properties (Table 1) If as arguedabove it is the coactivation of these shared propertiesthat primarily drives priming in the intact system thenthis would explain the preservation of some degree ofsemantic priming in the semantic dementia population

In summary E K and B S did not show any differ-ential impairment for the irregulars compared with theregulars on either elicitation or priming tasks and K Bdid not show a disproportionate deficit for the irregularscompared with the regulars in the elicitation task OnlyJ T showed an association of semantic deficit and prob-lems with the irregulars Moreover when the patientsdid make errors on the irregulars they did not exhibitthe pattern of effects predicted by the single mechanismaccount (Patterson et al 2001) with a higher proportionof errors on low-frequency items These results showthat there is no necessary association between a seman-tic deficit and problems with the irregular past tenseAlthough the two often do co-occur as we and othershave previously documented they can evidently alsooccur independently of each other This dissociationindicates that a separate set of functional capacities isimplicated here supporting the access of stored lexicalrepresentations An important issue for future researchis to identify the neural infrastructure for these accessprocesses so that we can better determine the condi-tions under which semantic deficits will co-occur withdeficits for the irregulars

The present results together with the results fornormal subjects addressed earlier (Marslen-Wilson ampTyler 1998) argue against the claim that the relation-ship between irregular past tense forms and their stemsis primarily a semantic rather than a morphologicalrelationship This in turn undermines the evidencefor the type of single system account of lexical repre-sentation in which morphological structure can becaptured solely on the basis of phonological and seman-tic relationships (Patterson et al 2001 Joanisse amp Seid-enberg 1999)

METHODS

Subjects

In addition to the four semantic dementia patients agroup of seven control subjects (aged between 56 and69 years) were tested on the elicitation task and a groupof 12 subjects (aged 60ndash73 years) were tested on thepriming study

Past Tense Production Elicitation Studies

We tested the patients on two elicitation experimentsdesigned to probe their ability to generate past tenseforms from their stems In the first elicitation testdescribed in Ullman et al (1997) subjects were pre-sented with sentences containing the present tenseforms of regular and irregular words together with aset of nonwords for past-tense generation We used thistest to directly compare our patients with others re-ported in the literature who have deficits in generatingpast tense forms

Tyler et al 1169

For the second elicitation task we selected a set of22 regular and 22 irregular verbs matched as closely aspossible on form and lemma frequency (Baayen Piepen-brock amp Gulikers 1995) and imageability (Coltheart1981) The median word form frequency was 35 for theregulars and 36 for the irregulars The median lemmafrequency was 92 for the regulars and 110 for theirregulars The median imageability was 485 for theregulars and 470 for the irregulars A two-sentencecontext was created for each verb where the secondsentence was incomplete and required a verb inflectedfor the past tense For example for the sentence lsquolsquoMynose sometimes bleeds Last night it rsquorsquo the correctresponse would be bled The test stimuli were precededby five practice trials The sentences were read aloud tothe subjects who were encouraged to produce a pasttense form appropriate for the sentence

Past Tense Comprehension Auditory Priming Study

To test patientsrsquo abilities to comprehend spoken pasttense forms we designed an intramodal auditoryndashauditory priming study contrasting the regulars andirregulars We chose to use a priming study to avoidsome of the pitfalls associated with testing patients ontasks that make heavy metalinguistic demands In thisexperiment target words which were verb stems (suchas pull sing) were preceded by a regularly or irregu-larly inflected past tense prime (pulled sung) or by anunrelated control word In addition to the two mainexperimental conditions we included two other con-ditions The first was a semantic priming condition inwhich the prime-target pairs were semantically but notphonologically or morphologically related (pighorse)We included this to determine whether the patientswould show evidence of a semantic deficit in the sametype of on-line task as used to assess their performanceon the past tense forms The fourth condition con-sisted of a phonological priming condition in whichthe prime-target pairs were phonologically but notsemantically or morphologically related (paintpain)which acted as a control for the effects of phonologicalsimilarity Since the inflected-stem pairs share a greatdeal of phonological overlap we need to be ableto separate out phonological from morphological prim-ing effects

Materials

We selected 42 prime-target pairs in each of the pasttense conditions (regularirregular) and 24 pairs for the2 control conditions (semanticphonological) In boththe regular and irregular past tense conditions theprimes were all inflected past tense forms (eg fixedglanced stuck brought) and the targets consisted oftheir stems all of which were one or two syllables ( fixglance stick bring) Items in the phonological control

condition consisted of morphologically simple monosyl-labic pairs (eg lamplamb plumpplum) that werenot semantically related The proportion of consonant-vowel-consonant (CVC) and consonant-vowel (CV) over-lap between prime and target was matched to theregular past tense pairs In all of the conditions weavoided homophones The two past tense sets and thephonological set were matched as closely as possible fortest prime control prime and target frequency andnumber of syllables (see Table 4) The fourth conditionconsisted of pairs of semantically related words thatwere either phonologically or morphologically related(eg shorecoast alleylane) For this condition weselected pairs of semantically related words that werehighly semantically related (but not associatively) and ofrelatively high frequency This was to ensure significantsemantic priming effects in control subjects to provide abasis for evaluating semantic priming in the patients Asemantic relatedness pretest was carried out in which15 subjects were asked to rate pairs of words accordingto their degree of semantic relatedness on a scale of 1ndash9where 1 was highly unrelated and 9 was highly relatedThe mean semantic relatedness rating for the prime-target pairs was 7 (SD = 102)

Each prime word was matched to a control word infrequency and number of syllables (see Table 4) Inaddition to the 132 test pairs we included 80 real wordfiller pairs making a total of 212 real word pairs Thefillers were an equal mixture of unrelated simpleprimespast tense targets simple primes(-s) inflectedtargetsx (-s) inflected primessimple targets simpleprimessimple targets and unambiguous noun pairsWe also included 148 legal wordnonword pairs whichwere a mixture of phonologically unrelated prime-targetpairs phonologically related pairs with targets derivedfrom regularly inflected past or present tense formsphonologically related pairs with targets made fromirregularly inflected forms and pairs with irregular verbsas primes

Design and Procedure

Test and filler items were pseudorandomly dispersedthroughout the test list We constructed two lists ofstimuli so that the target word only occurred once per

Table 4 Intramodal Priming Study Median Frequencies ofStimuli

Test Prime Target Control Prime

Regulars 55 6 4

Irregulars 6 7 5

Phonological 14 18 13

Semantic 16 10 15

1170 Journal of Cognitive Neuroscience Volume 16 Number 7

version All of the stimuli were recorded by a femalenative speaker of British English digitized at 22 kHz andmarkers placed at the acoustic onset of each targetword The markers triggered a timing device that mea-sured the subjectrsquos reaction time to the target wordPrimes and targets were recorded onto a DAT tape forpresentation to subjects Subjects heard primendashtargetpairs and were asked to make a lexical decision to thetarget by pressing a response key

Acknowledgments

We thank Matt Lambon-Ralph Beth Jeffries and KaralynPatterson for referring patients E K B S and J T and formaking many test scores available to us and to Julie Snowdenfor allowing us to work with K B We thank Lara Brent and BilliRandall for their help with the analyses and the patients fortheir cooperation This work was supported by an MRC (UK)program grant to L K T

Reprint requests should be sent to Lorraine K Tyler Depart-ment of Experimental Psychology University of CambridgeCambridge CB3 ODX UK or via e-mail lktylercslpsycholcamacuk

Note

1 These data were kindly provided by Elizabeth Jeffries

REFERENCES

Baayen R H Piepenbrock R amp Gulikers L (1995) TheCELEX lexical database Linguistic Data ConsortiumPhiladelphia University of Pennsylvania

Coltheart M (1981) The MRC psycholinguistic databaseQuarterly Journal of Experimental Psychology HumanExperimental Psychology 33A 497ndash505

Bunn E M Tyler L K amp Moss H E (1998) Naming picturesThe role of familiarity and property type reexaminedNeuropsychology 12 367ndash379

Devlin J T Gonnerman L M Andersen E S amp SeidenbergM S (1998) Category specific semantic deficits in focaland widespread brain damage A computational accountJournal of Cognitive Neuroscience 10 77ndash94

Friston K J Holmes A P Poline J-B Grasby B J WilliamsC R Frackowiak R S J amp Turner R (1995) Analysis offMRI time-series revisited Neuroscience 2 45ndash53

Galton C J Patterson K Graham K Lambon-Ralph M AWilliams G Antoun N Sahakian B J amp Hodges J R(2001) Differing patterns of temporal atrophy inAlzheimerrsquos disease and semantic dementia Neurology 57216ndash225

Hodges J R Patterson K Oxbury S amp Funnell E (1992)Semantic dementia Progressive fluent aphasia withtemporal lobe atrophy Brain 115 1783ndash1806

Howard D amp Patterson K (1992) The pyramids and palmtrees test Bury St Edmunds Thames Valley Test

Jefferies E Patterson K Jones R W Bateman D amp LambonRalph M A (submitted) A category-specific advantage fornumbers in verbal short-term memory Evidence fromsemantic dementia

Joanisse M F amp Seidenberg M S (1999) Impairments inverb morphology after brain injury A connectionist model

Proceedings of the National Academy of Sciences USA 967592ndash7597

Kay J Lesser R amp Coltheart M (1992) Psycholinguisticassessments of language processing in aphasia (PALPA)London Erlbaum

Longworth C E Marslen-Wilson W D amp Tyler L K(2002) Non-fluent aphasics show an abnormal time courseof semantic activation from the regular past tense[abstract] Journal of Cognitive Neuroscience 14Supplement 55

Marslen-Wilson W D Csibra G Ford M Hatzakis HGaskell G amp Johnson M H (2000) Associations anddissociations in the processing of regular and irregularverbs Electrophysiological evidence Journal of CognitiveNeuroscience 12 Supplement 55E

Marslen-Wilson W D amp Tyler L K (1997) Dissociating typesof mental computation Nature 387 592ndash594

Marslen-Wilson W D amp Tyler L K (1998) Rulesrepresentations and the English past tense Trends inCognitive Science 2 428ndash435

Marslen-Wilson W D amp Tyler L K (2003) Capturingunderlying differentiation in the human language systemTrends in Cognitive Sciences 7 62ndash63

Mathuranath P S Nestor P J Berrios G E Rakowicz Wamp Hodges J (2000) A brief cognitive test battery todifferentiate Alzheimerrsquos disease and frontotemporaldementia Neurology 55 1613ndash1620

McClelland J L amp Patterson K (2002) Rules or connectionsin past-tense inflections What does the evidence rule outTrends in Cognitive Sciences 6 465ndash472

McRae K de Sa V amp Seidenberg M S (1997) On thenature and scope of featural representations of wordmeaning Journal of Experimental Psychology General126 99ndash130

Miozzo M (2003) On the processing of regular and irregularforms of verbs and nouns Evidence from neuropsychologyCognition 87 101ndash127

Moss H E amp Tyler L K (2000) A progressivecategory-specific semantic deficit for non-living thingsNeuropsychologia 38 60ndash82

Moss H E Tyler L K Durrant-Peatfield M amp Bunn E(1998) Two eyes of a see-through Impaired and intactsemantic knowledge in a case of selective deficit for livingthings Neurocase 4 291ndash310

Moss H E Tyler L K Hodges J R amp Patterson K (1995)Exploring the loss of semantic memory in semanticdementia Evidence from a primed monitoring studyNeuropsychology 91 16ndash26

Patterson K Lambon Ralph M A Hodges J Ramp McClelland J L (2001) Deficits in irregular past-tenseverb morphology associated with degraded semanticknowledge Neuropsychologia 39 709ndash724

Pinker S (1991) Rules of language Science 253 530ndash533Pinker S amp Ullman M T (2002) The past and future of thepast tense Trends in Cognitive Sciences 6 456ndash463

Prasada S amp Pinker S (1993) Generalizations of regular andirregular morphological patterns Language and CognitiveProcesses 8 1ndash56

Price C J amp Friston K J (1997) Cognitive conjunction A newapproach to brain activation experiments Neuroimage 5261ndash270

Rumelhart D E amp McClelland J L (1986) On learningthe past tense of English verbs In J L McClelland amp D ERumelhart (Eds) Parallel distributed processing(pp 217ndash270) Cambridge MIT Press

Snowden J S Goulding P J amp Neary D (1989) Semanticdementia A form of circumscribed cerebral atrophyBehavioral Neurology 2 167ndash182

Tyler et al 1171

For the second elicitation task we selected a set of22 regular and 22 irregular verbs matched as closely aspossible on form and lemma frequency (Baayen Piepen-brock amp Gulikers 1995) and imageability (Coltheart1981) The median word form frequency was 35 for theregulars and 36 for the irregulars The median lemmafrequency was 92 for the regulars and 110 for theirregulars The median imageability was 485 for theregulars and 470 for the irregulars A two-sentencecontext was created for each verb where the secondsentence was incomplete and required a verb inflectedfor the past tense For example for the sentence lsquolsquoMynose sometimes bleeds Last night it rsquorsquo the correctresponse would be bled The test stimuli were precededby five practice trials The sentences were read aloud tothe subjects who were encouraged to produce a pasttense form appropriate for the sentence

Past Tense Comprehension Auditory Priming Study

To test patientsrsquo abilities to comprehend spoken pasttense forms we designed an intramodal auditoryndashauditory priming study contrasting the regulars andirregulars We chose to use a priming study to avoidsome of the pitfalls associated with testing patients ontasks that make heavy metalinguistic demands In thisexperiment target words which were verb stems (suchas pull sing) were preceded by a regularly or irregu-larly inflected past tense prime (pulled sung) or by anunrelated control word In addition to the two mainexperimental conditions we included two other con-ditions The first was a semantic priming condition inwhich the prime-target pairs were semantically but notphonologically or morphologically related (pighorse)We included this to determine whether the patientswould show evidence of a semantic deficit in the sametype of on-line task as used to assess their performanceon the past tense forms The fourth condition con-sisted of a phonological priming condition in whichthe prime-target pairs were phonologically but notsemantically or morphologically related (paintpain)which acted as a control for the effects of phonologicalsimilarity Since the inflected-stem pairs share a greatdeal of phonological overlap we need to be ableto separate out phonological from morphological prim-ing effects

Materials

We selected 42 prime-target pairs in each of the pasttense conditions (regularirregular) and 24 pairs for the2 control conditions (semanticphonological) In boththe regular and irregular past tense conditions theprimes were all inflected past tense forms (eg fixedglanced stuck brought) and the targets consisted oftheir stems all of which were one or two syllables ( fixglance stick bring) Items in the phonological control

condition consisted of morphologically simple monosyl-labic pairs (eg lamplamb plumpplum) that werenot semantically related The proportion of consonant-vowel-consonant (CVC) and consonant-vowel (CV) over-lap between prime and target was matched to theregular past tense pairs In all of the conditions weavoided homophones The two past tense sets and thephonological set were matched as closely as possible fortest prime control prime and target frequency andnumber of syllables (see Table 4) The fourth conditionconsisted of pairs of semantically related words thatwere either phonologically or morphologically related(eg shorecoast alleylane) For this condition weselected pairs of semantically related words that werehighly semantically related (but not associatively) and ofrelatively high frequency This was to ensure significantsemantic priming effects in control subjects to provide abasis for evaluating semantic priming in the patients Asemantic relatedness pretest was carried out in which15 subjects were asked to rate pairs of words accordingto their degree of semantic relatedness on a scale of 1ndash9where 1 was highly unrelated and 9 was highly relatedThe mean semantic relatedness rating for the prime-target pairs was 7 (SD = 102)

Each prime word was matched to a control word infrequency and number of syllables (see Table 4) Inaddition to the 132 test pairs we included 80 real wordfiller pairs making a total of 212 real word pairs Thefillers were an equal mixture of unrelated simpleprimespast tense targets simple primes(-s) inflectedtargetsx (-s) inflected primessimple targets simpleprimessimple targets and unambiguous noun pairsWe also included 148 legal wordnonword pairs whichwere a mixture of phonologically unrelated prime-targetpairs phonologically related pairs with targets derivedfrom regularly inflected past or present tense formsphonologically related pairs with targets made fromirregularly inflected forms and pairs with irregular verbsas primes

Design and Procedure

Test and filler items were pseudorandomly dispersedthroughout the test list We constructed two lists ofstimuli so that the target word only occurred once per

Table 4 Intramodal Priming Study Median Frequencies ofStimuli

Test Prime Target Control Prime

Regulars 55 6 4

Irregulars 6 7 5

Phonological 14 18 13

Semantic 16 10 15

1170 Journal of Cognitive Neuroscience Volume 16 Number 7

version All of the stimuli were recorded by a femalenative speaker of British English digitized at 22 kHz andmarkers placed at the acoustic onset of each targetword The markers triggered a timing device that mea-sured the subjectrsquos reaction time to the target wordPrimes and targets were recorded onto a DAT tape forpresentation to subjects Subjects heard primendashtargetpairs and were asked to make a lexical decision to thetarget by pressing a response key

Acknowledgments

We thank Matt Lambon-Ralph Beth Jeffries and KaralynPatterson for referring patients E K B S and J T and formaking many test scores available to us and to Julie Snowdenfor allowing us to work with K B We thank Lara Brent and BilliRandall for their help with the analyses and the patients fortheir cooperation This work was supported by an MRC (UK)program grant to L K T

Reprint requests should be sent to Lorraine K Tyler Depart-ment of Experimental Psychology University of CambridgeCambridge CB3 ODX UK or via e-mail lktylercslpsycholcamacuk

Note

1 These data were kindly provided by Elizabeth Jeffries

REFERENCES

Baayen R H Piepenbrock R amp Gulikers L (1995) TheCELEX lexical database Linguistic Data ConsortiumPhiladelphia University of Pennsylvania

Coltheart M (1981) The MRC psycholinguistic databaseQuarterly Journal of Experimental Psychology HumanExperimental Psychology 33A 497ndash505

Bunn E M Tyler L K amp Moss H E (1998) Naming picturesThe role of familiarity and property type reexaminedNeuropsychology 12 367ndash379

Devlin J T Gonnerman L M Andersen E S amp SeidenbergM S (1998) Category specific semantic deficits in focaland widespread brain damage A computational accountJournal of Cognitive Neuroscience 10 77ndash94

Friston K J Holmes A P Poline J-B Grasby B J WilliamsC R Frackowiak R S J amp Turner R (1995) Analysis offMRI time-series revisited Neuroscience 2 45ndash53

Galton C J Patterson K Graham K Lambon-Ralph M AWilliams G Antoun N Sahakian B J amp Hodges J R(2001) Differing patterns of temporal atrophy inAlzheimerrsquos disease and semantic dementia Neurology 57216ndash225

Hodges J R Patterson K Oxbury S amp Funnell E (1992)Semantic dementia Progressive fluent aphasia withtemporal lobe atrophy Brain 115 1783ndash1806

Howard D amp Patterson K (1992) The pyramids and palmtrees test Bury St Edmunds Thames Valley Test

Jefferies E Patterson K Jones R W Bateman D amp LambonRalph M A (submitted) A category-specific advantage fornumbers in verbal short-term memory Evidence fromsemantic dementia

Joanisse M F amp Seidenberg M S (1999) Impairments inverb morphology after brain injury A connectionist model

Proceedings of the National Academy of Sciences USA 967592ndash7597

Kay J Lesser R amp Coltheart M (1992) Psycholinguisticassessments of language processing in aphasia (PALPA)London Erlbaum

Longworth C E Marslen-Wilson W D amp Tyler L K(2002) Non-fluent aphasics show an abnormal time courseof semantic activation from the regular past tense[abstract] Journal of Cognitive Neuroscience 14Supplement 55

Marslen-Wilson W D Csibra G Ford M Hatzakis HGaskell G amp Johnson M H (2000) Associations anddissociations in the processing of regular and irregularverbs Electrophysiological evidence Journal of CognitiveNeuroscience 12 Supplement 55E

Marslen-Wilson W D amp Tyler L K (1997) Dissociating typesof mental computation Nature 387 592ndash594

Marslen-Wilson W D amp Tyler L K (1998) Rulesrepresentations and the English past tense Trends inCognitive Science 2 428ndash435

Marslen-Wilson W D amp Tyler L K (2003) Capturingunderlying differentiation in the human language systemTrends in Cognitive Sciences 7 62ndash63

Mathuranath P S Nestor P J Berrios G E Rakowicz Wamp Hodges J (2000) A brief cognitive test battery todifferentiate Alzheimerrsquos disease and frontotemporaldementia Neurology 55 1613ndash1620

McClelland J L amp Patterson K (2002) Rules or connectionsin past-tense inflections What does the evidence rule outTrends in Cognitive Sciences 6 465ndash472

McRae K de Sa V amp Seidenberg M S (1997) On thenature and scope of featural representations of wordmeaning Journal of Experimental Psychology General126 99ndash130

Miozzo M (2003) On the processing of regular and irregularforms of verbs and nouns Evidence from neuropsychologyCognition 87 101ndash127

Moss H E amp Tyler L K (2000) A progressivecategory-specific semantic deficit for non-living thingsNeuropsychologia 38 60ndash82

Moss H E Tyler L K Durrant-Peatfield M amp Bunn E(1998) Two eyes of a see-through Impaired and intactsemantic knowledge in a case of selective deficit for livingthings Neurocase 4 291ndash310

Moss H E Tyler L K Hodges J R amp Patterson K (1995)Exploring the loss of semantic memory in semanticdementia Evidence from a primed monitoring studyNeuropsychology 91 16ndash26

Patterson K Lambon Ralph M A Hodges J Ramp McClelland J L (2001) Deficits in irregular past-tenseverb morphology associated with degraded semanticknowledge Neuropsychologia 39 709ndash724

Pinker S (1991) Rules of language Science 253 530ndash533Pinker S amp Ullman M T (2002) The past and future of thepast tense Trends in Cognitive Sciences 6 456ndash463

Prasada S amp Pinker S (1993) Generalizations of regular andirregular morphological patterns Language and CognitiveProcesses 8 1ndash56

Price C J amp Friston K J (1997) Cognitive conjunction A newapproach to brain activation experiments Neuroimage 5261ndash270

Rumelhart D E amp McClelland J L (1986) On learningthe past tense of English verbs In J L McClelland amp D ERumelhart (Eds) Parallel distributed processing(pp 217ndash270) Cambridge MIT Press

Snowden J S Goulding P J amp Neary D (1989) Semanticdementia A form of circumscribed cerebral atrophyBehavioral Neurology 2 167ndash182

Tyler et al 1171

version All of the stimuli were recorded by a femalenative speaker of British English digitized at 22 kHz andmarkers placed at the acoustic onset of each targetword The markers triggered a timing device that mea-sured the subjectrsquos reaction time to the target wordPrimes and targets were recorded onto a DAT tape forpresentation to subjects Subjects heard primendashtargetpairs and were asked to make a lexical decision to thetarget by pressing a response key

Acknowledgments

We thank Matt Lambon-Ralph Beth Jeffries and KaralynPatterson for referring patients E K B S and J T and formaking many test scores available to us and to Julie Snowdenfor allowing us to work with K B We thank Lara Brent and BilliRandall for their help with the analyses and the patients fortheir cooperation This work was supported by an MRC (UK)program grant to L K T

Reprint requests should be sent to Lorraine K Tyler Depart-ment of Experimental Psychology University of CambridgeCambridge CB3 ODX UK or via e-mail lktylercslpsycholcamacuk

Note

1 These data were kindly provided by Elizabeth Jeffries

REFERENCES

Baayen R H Piepenbrock R amp Gulikers L (1995) TheCELEX lexical database Linguistic Data ConsortiumPhiladelphia University of Pennsylvania

Coltheart M (1981) The MRC psycholinguistic databaseQuarterly Journal of Experimental Psychology HumanExperimental Psychology 33A 497ndash505

Bunn E M Tyler L K amp Moss H E (1998) Naming picturesThe role of familiarity and property type reexaminedNeuropsychology 12 367ndash379

Devlin J T Gonnerman L M Andersen E S amp SeidenbergM S (1998) Category specific semantic deficits in focaland widespread brain damage A computational accountJournal of Cognitive Neuroscience 10 77ndash94

Friston K J Holmes A P Poline J-B Grasby B J WilliamsC R Frackowiak R S J amp Turner R (1995) Analysis offMRI time-series revisited Neuroscience 2 45ndash53

Galton C J Patterson K Graham K Lambon-Ralph M AWilliams G Antoun N Sahakian B J amp Hodges J R(2001) Differing patterns of temporal atrophy inAlzheimerrsquos disease and semantic dementia Neurology 57216ndash225

Hodges J R Patterson K Oxbury S amp Funnell E (1992)Semantic dementia Progressive fluent aphasia withtemporal lobe atrophy Brain 115 1783ndash1806

Howard D amp Patterson K (1992) The pyramids and palmtrees test Bury St Edmunds Thames Valley Test

Jefferies E Patterson K Jones R W Bateman D amp LambonRalph M A (submitted) A category-specific advantage fornumbers in verbal short-term memory Evidence fromsemantic dementia

Joanisse M F amp Seidenberg M S (1999) Impairments inverb morphology after brain injury A connectionist model

Proceedings of the National Academy of Sciences USA 967592ndash7597

Kay J Lesser R amp Coltheart M (1992) Psycholinguisticassessments of language processing in aphasia (PALPA)London Erlbaum

Longworth C E Marslen-Wilson W D amp Tyler L K(2002) Non-fluent aphasics show an abnormal time courseof semantic activation from the regular past tense[abstract] Journal of Cognitive Neuroscience 14Supplement 55

Marslen-Wilson W D Csibra G Ford M Hatzakis HGaskell G amp Johnson M H (2000) Associations anddissociations in the processing of regular and irregularverbs Electrophysiological evidence Journal of CognitiveNeuroscience 12 Supplement 55E

Marslen-Wilson W D amp Tyler L K (1997) Dissociating typesof mental computation Nature 387 592ndash594

Marslen-Wilson W D amp Tyler L K (1998) Rulesrepresentations and the English past tense Trends inCognitive Science 2 428ndash435

Marslen-Wilson W D amp Tyler L K (2003) Capturingunderlying differentiation in the human language systemTrends in Cognitive Sciences 7 62ndash63

Mathuranath P S Nestor P J Berrios G E Rakowicz Wamp Hodges J (2000) A brief cognitive test battery todifferentiate Alzheimerrsquos disease and frontotemporaldementia Neurology 55 1613ndash1620

McClelland J L amp Patterson K (2002) Rules or connectionsin past-tense inflections What does the evidence rule outTrends in Cognitive Sciences 6 465ndash472

McRae K de Sa V amp Seidenberg M S (1997) On thenature and scope of featural representations of wordmeaning Journal of Experimental Psychology General126 99ndash130

Miozzo M (2003) On the processing of regular and irregularforms of verbs and nouns Evidence from neuropsychologyCognition 87 101ndash127

Moss H E amp Tyler L K (2000) A progressivecategory-specific semantic deficit for non-living thingsNeuropsychologia 38 60ndash82

Moss H E Tyler L K Durrant-Peatfield M amp Bunn E(1998) Two eyes of a see-through Impaired and intactsemantic knowledge in a case of selective deficit for livingthings Neurocase 4 291ndash310

Moss H E Tyler L K Hodges J R amp Patterson K (1995)Exploring the loss of semantic memory in semanticdementia Evidence from a primed monitoring studyNeuropsychology 91 16ndash26

Patterson K Lambon Ralph M A Hodges J Ramp McClelland J L (2001) Deficits in irregular past-tenseverb morphology associated with degraded semanticknowledge Neuropsychologia 39 709ndash724

Pinker S (1991) Rules of language Science 253 530ndash533Pinker S amp Ullman M T (2002) The past and future of thepast tense Trends in Cognitive Sciences 6 456ndash463

Prasada S amp Pinker S (1993) Generalizations of regular andirregular morphological patterns Language and CognitiveProcesses 8 1ndash56

Price C J amp Friston K J (1997) Cognitive conjunction A newapproach to brain activation experiments Neuroimage 5261ndash270

Rumelhart D E amp McClelland J L (1986) On learningthe past tense of English verbs In J L McClelland amp D ERumelhart (Eds) Parallel distributed processing(pp 217ndash270) Cambridge MIT Press

Snowden J S Goulding P J amp Neary D (1989) Semanticdementia A form of circumscribed cerebral atrophyBehavioral Neurology 2 167ndash182

Tyler et al 1171

Snowden J S amp Neary D (2002) Relearning of verballabels in semantic dementia Neuropsychologia 401715ndash1728

Stamatakis E A amp Tyler L K (2003) Detecting lesions onstructural brain images with voxel based methodologiesNeuroimage 19 S1043

Tyler L K de Mornay Davies P Anokhina R Longworth CRandall B amp Marslen-Wilson W D (2002) Dissociations inprocessing past tense morphology Neuropathology andbehavioural studies Journal of Cognitive Neuroscience 1479ndash94

Tyler L K amp Moss H E (1998) Going going gone Implicit and explicit tests of conceptual knowledge in alongitudinal study of semantic dementia Neuropsychologia36 1313ndash1323

Tyler L K amp Moss H E (2001) Towards a distributedaccount of conceptual knowledge Trends in CognitiveScience 5 244ndash252

Tyler L K Moss H E Durrant-Peatfield M amp Levy J (2000)

Conceptual structure and the structure of categories Adistributed account of category-specific deficits Brain andLanguage 75 195ndash231

Tyler L K Moss H E Patterson K amp Hodges J (1997)The gradual deterioration of syntax and semantics in apatient with progressive aphasia Brain and Language 56426ndash477

Tyler L K Randall B amp Marslen-Wilson W D (2002)Phonology and neuropsychology of the English past tenseNeuropsychologia 40 1154ndash1166

Ullman M T (2001) The declarativeprocedural model oflexicon and grammar Journal of Psycholinguistic Research30 37ndash69

Ullman M T Corkin S Coppola M Hickok G GrowdonJ H Koroshetz W J amp Pinker S (1997) A neuraldissociation within language Evidence that the mentaldictionary is part of declarative memory and thatgrammatical rules are processed by the procedural systemJournal of Cognitive Neuroscience 9 266ndash276

1172 Journal of Cognitive Neuroscience Volume 16 Number 7