2 Y. Faroqi-Shah / Brain Research Bulletin 74 (2007) 1–13 4.3. Can dissociations be manipulated by experimental variables? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 5. Conclusions and future research considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 Conﬂict of interest. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111. Introduction in the mental lexicon and retrieved from declarative memory, and linked to the left temporo-parietal cortex (Fig. 1a). Func- Few questions about the nature and neural organization tional neuroimaging studies have tested these neuroanatomicalof the human language faculty have generated the kind of predictions using a variety of techniques, such as positronintense debate that word inﬂections have in the past two emission tomography, functional magnetic resonance imag-decades. The controversy over regular and irregular word inﬂec- ing (FMRI), and magnetoencephalography. Some studies havetions, often called the “English past tense debate” , has found supportive evidence for the declarative/procedural modelspanned cognitive neuroscience, linguistics, cognitive psychol- with greater left frontal activation for regulars and greater leftogy, neuropsychology, language development, and artiﬁcial temporal activity for irregulars [17,30,44,58]. However, otherintelligence alike. Many languages have two types of inﬂected neuroimaging studies have found the reverse pattern , orwords (although this dichotomy may be simplistic ): regu- no remarkable differences in the neural processing of reg-lar inﬂections, such as past tense -ed in English (walk-walked, ulars and irregulars [49,54]. Hence, neuroimaging evidenceslip-slipped) and the participle -t in German (tanzen-getanzt for the declarative/procedural model has been inconclusive[dance] ) and irregular inﬂections with non-default changes, .such as sing-sang, sleep-slept in English, or the participle -nin German (trinken-getrunken [drink]). Differing opinions pre-vail regarding the nature of linguistic and cognitive differencesbetween regular and irregular words. Some believe that compre-hending and producing regular inﬂections involves afﬁxation(walk + ed → walked), while irregular inﬂections are retrievedfrom memory (dual mechanism account) [12,41,47,59]. Othersbelieve that all inﬂections are retrieved from memory (singleroute, connectionistic accounts) [10,31,48], or are afﬁxed (dis-tributed morphology) [26,54], or that there is a competitionbetween both memory and afﬁxation routes . Linguistic and neurocognitive distinctions between regularsand irregulars have assumed enormous importance among lan-guage researchers because these succinctly embody a broaderquestion about human language: whether we have two distinctlanguage modules, a mental grammar and a mental lexicon. Themental grammar is a system of combinatorial rules that can beproductively applied to generate sentences as well as complexwords. Regular inﬂections are considered to be archetypes ofthese combinatorial rules. The mental lexicon is essentially alisting of all known words, including irregular inﬂections .Recently, the question is not so much whether the two wordsdiffer in psycholinguistic aspects, but whether these differenceshave neuroanatomical validity.1.1. Neuroanatomical predictions The Declarative/Procedural hypothesis ﬁrst made distinctpredictions about the neural correlates of regular and irreg-ular words . Based on the assumption that grammaticalcombinatorial rules operate for the afﬁxation of regulars, the Fig. 1. Predictions of the various neuroanatomical models: (a) declarative-left frontal cortex, particularly Broca’s area, was implicated in procedural model ; (b) phonology-semantics model ; (c) two-stage verbthe processing and production of regular words. Basal ganglia generation model , not shown are: left insula and hippocampus, also impli-circuits were also identiﬁed in this fronto-striatal procedural cated for regular verbs and right DLPFC, also implicated for irregular verbs. Seenetwork. In contrast, irregular verbs were assumed to be housed text (Section 1.1.) for details.
Y. Faroqi-Shah / Brain Research Bulletin 74 (2007) 1–13 3 A more recent neuroanatomical hypothesis was proposed lars [16,17,28,30,49,54,56]. In the next section, another sourceon the basis of auditory input processing deﬁcits of regulars of evidence for neural differences between regulars and irregu-and irregulars [56,57] (hereafter called the semantics-phonology lars is evaluated.model). Examples of input processing tasks include decidingif auditorily presented stimuli are real words (pless, vaim) 1.2. Neuropsychological dissociationsand same-different judgments with word pairs (sprayed-spray,trade-tray, taught-teach). The dorsal–ventral speech process- A widely cited substantiation for the neuroanatomical pre-ing framework  was used to propose that the presence of dictions of the Declarative/Procedural hypothesis comes fromregular inﬂectional afﬁxes engages the dorsal (fronto-temporal) single and double dissociations in brain damaged populationsnetwork to greater extent than irregular inﬂections . This [41,59]. Although a variety of conditions that predominantlyfronto-temporal network, which includes bilateral superior tem- affect left frontal–striatal structures [Broca’s aphasia, Parkin-poral gyri, left inferior frontal gyrus (LIFG), and left anterior son’s disease, Huntington’s disease] have been compared withcingulate, is implicated for “morphophonological parsing pro- conditions that predominantly affect posterior language areascesses that segment morphologically complex spoken forms (anomic aphasia, Wernicke’s aphasia, Alzheimer’s dementia)into stems and afﬁxes” [56, p. 8380] (sprayed → spray + ed) [58,29,60], Broca’s aphasia has received particular attention(Fig. 1b). Explicit statements about the neuroanatomical cor- as the poster-child in the debate between regular and irregularrelates of irregulars are not made in earlier versions [56,57], inﬂections [8,21,41,43]. Broca’s aphasia typically results fromalthough a recent version  allocates left posterior superior focal brain damage to left hemisphere frontal regions  and istemporal gyrus and middle temporal gyrus for semantic access characterized by reduced speech rate, inaccurate grammar, andof irregular verbs and monomorphemic words. In other words, incorrect usage of verb inﬂections . Using data from sin-this model claims that phonological demands are greater for gle case studies, it was originally proposed that Broca’s aphasicregulars while semantic access is more crucial for processing individuals, owing to frontal lobe damage, show dissociationirregular verbs and monomorphemic words. An FMRI study between regulars and irregulars, with worse performance onusing same-different word pair judgments by the same authors regulars [41,59]. However, an accumulating body of evidencerevealed that regulars and irregulars activated the same left supe- suggests that the relationship between regular verb impairmentrior temporal–frontal network and only varied in the extent of and Broca’s aphasia is not straightforward because some studiesactivation (greater activity for regulars) . have found the reverse pattern of worse performance on irreg- Another recent data-driven proposal is based on FMRI acti- ulars compared to regulars [8,6,45,29,43], while others havevation patterns for covert generation of Spanish regular and found no difference between regulars and irregulars [8,15], or airregular verbs  (referred to as the two-stage verb genera- variety of patterns .tion model hereafter). Both regular and irregular verbs activated Task differences, language differences, and variations in thethe left opercular LIFG, regular verbs showed increased acti- number of patients tested are only a few factors that havevation of left anterior superior temporal gyrus-insula and obscured the relation between regular verb production andhippocampus, while irregular verbs showed increased activa- Broca’s aphasia. For example, it can be questioned if disso-tion bilaterally in the dorsolateral–prefrontal cortex (DLPFC). ciation patterns from cognitively diverse tasks, such as singleThe authors proposed a two-stage model for the generation of word reading, lexical priming, and sentence completion, canverb inﬂections, the ﬁrst stage is language and verb type spe- be pooled together. Is it reasonable to assume that the putativeciﬁc, and recruits the DLPFC for lexical access of irregulars, neural differences between regulars and irregulars will surpassand grammatical–phonological network of left anterior superior neuroanatomical differences from inherently dissimilar tasks,temporal gyrus, insula and hippocampus for regular verbs. The such as reading and sentence completion? Further, unlike irreg-second step is common to all verb types, involves retrieval of ulars in English (swam, kept, etc.), irregulars in languages, suchgrammatical information, such as tense and recruits the LIFG as Spanish and German contain non-default (albeit quasiregu-(Fig. 1c). lar) afﬁxes, and it is unclear if the brain makes the same binary To summarize, the declarative/procedural model proposes distinction between regulars and irregulars in such languagesleft fronto-striatal involvement for all regulars, and left temporo- .parietal involvement for all irregulars, irrespective of task In addition to the extensive methodological diversity acrossdemands . The semantics-phonology model claims greater studies, there are conceptual and statistical concerns thatfronto-temporal (dorsal) activity for regulars compared to cast doubt on the robustness of regular–irregular dissocia-irregulars, but does not make any speciﬁc statements about tions in aphasic individuals [7,13].1 Neuroanatomical models ofneuroanatomical correlates for irregulars, or for production regular–irregular verb processing imply normal versus impairedtasks [56,57]. Finally, the two-stage verb generation model performance on irregular and regular verbs, respectively, in indi- suggests that LIFG activity is common to both verb viduals with frontal lobe damage [41,59]. However, operationaltypes, DLPFC is increasingly activated for irregulars and a deﬁnitions of normal and impaired performance or what con-temporal-insular-hippocampal network is used for regulars.These predictions are illustrated in Fig. 1. Neuroimaging studieshave produced inconsistent results and are hence inconclusive 1 These conceptual and statistical limitations apply to other areas of neuropsy-regarding the neuroanatomical correlates of regulars and irregu- chological research as well .
4 Y. Faroqi-Shah / Brain Research Bulletin 74 (2007) 1–13stitutes a dissociation are not provided in most studies, a point individual subject data were reported; (3) regular and irregular verb produc-that has been made about most research on neuropsychologi- tion was tested and the scores of regular and irregular verbs were separatelycal dissociations . It has also been emphasized elsewhere reported2 ; (4) either sentence completion, and/or single word repetition tasks were used. Fifteen studies met these criteria. Of these, three studies reportedthat reliable inferences about normal neurocognitive architec- both sentence production and word repetition data, 10 studies reported only sen-ture can be made from neuropsychological dissociations only tence production data and two had only single word repetition. Multiple dataupon statistical comparisons with a normal control group . sets from individual patients were included in the meta-analysis only if eachMost studies of regular–irregular dissociations in aphasia have data set was original to the study.used intra-individual comparisons, without reference to a con-trol group. These so called trend dissociations are considered to 2.1. Coding and data analysesbe weakest form of neuropsychological evidence [52,38]. Fur- All studies were coded for a variety of variables including, language ofther, it has been demonstrated that tasks that are likely to produce testing, the number of stimuli used, raw scores, proportion accuracy, descriptionceiling effects in normal controls (100% accuracy) signiﬁcantly of aphasia proﬁle, description of lesion information, phonological and frequencyinﬂate the occurrence of dissociations (high false positives-Type matching of stimuli, and the overall conclusions of the authors. For sentenceI error] in brain damaged individuals . Finally, conclu- production studies, three task variations have been used, these were separately coded as A, B, and C. Task A referred to a ﬁll in the blank task where the presentsions about neuroanatomical localization of function are limited tense verb was provided in a preceding context (Everyday I rob a bank. Yesterdaywhen patients with similar lesions exhibit reverse dissociation I ), Task B corresponded to ﬁll in the blank that required selectionpatterns. In other words, neuropsychological dissociations are from multiple options (Yesterday I a bank: rob, robbing, robbed),informative only if we can clearly demonstrate that the dissoci- and Task C elicited the sentence in response to a question (The man is stealingation exists. the silverware. What did he do yesterday?). For studies where raw data were not reported, these were computed from the relevant ﬁgures or percentage scores. The authors of one study provided raw1.3. Purpose of this meta-analysis scores via personal communication . Raw scores were statistically compared using the Fisher-exact test since this test considers the number of stimuli in In the presence of mixed evidence and methodological con- calculation of p-values [40,51]. The one-tailed p-values thus derived were used to code if performance of regulars was worse than irregulars since this was thefounds, it is unclear if a single dissociation pattern between predicted direction of dissociation for Broca’s aphasia [41,59].regular and irregular verbs exists in Broca’s aphasia. This ques-tion is important because different neural regions are identiﬁed 3. Resultsfor regular–irregular processing by the declarative/procedural,two-stage verb generation, and auditory input models (Fig. 1), A total of 110 different datasets were obtained across all tasksand it is unclear if existing data favor any one of these models. from 75 patients. Nineteen individuals contributed more thanThis scenario warrants a systematic synthesis of research ﬁnd- one dataset and 16 patients contributed to both sentence pro-ings to ascertain the weight of neuropsychological evidence. The duction and word repetition data. A complete list of includedquestion being evaluated in this meta-analysis is not whether studies and individual subject data are provided in Table 1 forregulars and irregulars can dissociate, rather if they consistently sentence completion and Table 2 for word repetition. The ﬁnd-dissociate in a single direction (regulars worse than irregulars) ings of the two tasks are described in separate sections below.in patients with Broca’s aphasia, and whether this dissociation This is followed by the results of the lesion analysis.is consistently associated with left frontal lobe lesions. A sec-ondary purpose was to explore variables that are likely to result 3.1. Sentence productionin dissociations. Such a synthesis is likely to yield fruitful resultsif patient performance is compared across similar tasks. Since a A total of 78 separate patient data sets, representing 66 dif-majority of studies have used production tasks to compare reg- ferent aphasic patients were obtained for sentence productionular and irregular verbs in Broca’s aphasia, this meta-analysis (Table 1). The number of data sets in which regular verbs werefocuses on two production tasks: sentence production and word signiﬁcantly lower in accuracy compared to irregular verbs (one-repetition. tailed p < 0.05) is 10 (9 unique patients). In 26 data sets (23 unique patients), regular verbs were signiﬁcantly better than2. Materials and methods irregulars in accuracy. In the remaining 42 data sets, there were no signiﬁcant differences between regular and irregular verbs. Published articles or abstracts that compared the production of regular andirregular verbs in English peer-reviewed journals were identiﬁed by searches These proportions are summarized in Fig. 2. Overall, there wasthrough electronic databases. The electronic databases included Science Cita- a high positive correlation between the accuracies of regularstion Index, Medline (Pubmed), PsycInfo, and CINAHL. Combinations of the and irregulars, when computed across all 78 datasets (rs = 0.64,following keywords were used to conduct the search: aphasia, verbs, language p < 000).production, morphology, dissociation, regular, and irregular. The search was Other variables. Forty-nine (out of total 78) data sets camerestricted to research studies published between 1980 and October 2006. Thesearch resulted in a total of 56 reports. The search was further narrowed down from English speakers (33 unique patients). Interestingly, 9 outto 25 articles after reading the abstracts for relevance. Reports that duplicated of the 10 patients who had poorer accuracy for regulars weredata, such as conference presentations and full articles, were included only once.The entire text of these ﬁnal 25 articles was read and the following inclusionarycriteria were used: (1) the study reported original data obtained from patient(s) 2 Authors deﬁnitions of and decisions about what constitutes “regularity” weredescribed as non-ﬂuent, agrammatic or Broca’s aphasic by the author(s); (2) used for the purpose of this analysis.
Table 1Studies of sentence production sorted by language and chronological orderReference Language Task/stimuli Patient Lesion # Stimuli Accuracy p One-tail R < I? p < 0.05 R I R IUllman et al., p. 271 (Table 3)  English B FCL L frontal, insula, white matter 20 20 0.20 0.69 0.00 YesBird et al., p. 507 (Fig. 1), p. 509 BB L 44 44 0.00 0.08 0.06 No (Fig. 2)  IJ L 44 44 0.07 0.23 0.03 Yes IB L frontal temporal parietal 44 44 0.09 0.14 0.37 No DE L mca 44 44 0.20 0.43 0.02 Yes AB L mca 44 44 0.24 0.36 0.18 No English BF MB L frontoparietal, R frontal 44 44 0.18 0.39 0.03 Yes RT L 44 44 0.07 0.34 0.00 Yes VC L frontoparietal 44 44 0.20 0.36 0.08 No GN L mca incl. Broca’s, occipital 44 44 0.28 0.51 0.02 Yes JL L frontal temporal parietal 44 44 0.69 0.84 0.07 No Y. Faroqi-Shah / Brain Research Bulletin 74 (2007) 1–13 English BPF BB 34 34 0.18 0.12 0.85 No IJ 34 34 0.06 0.06 0.69 No IB 34 34 0.26 0.09 0.98 No DE 34 34 0.21 0.26 0.39 No AB 34 34 0.24 0.18 0.82 No MB 34 34 0.24 0.26 0.53 No RT 34 34 0.09 0.24 0.09 No GN 34 34 0.59 0.05 1.00 No JL 34 34 0.79 0.62 0.97 No B RC L frontal incl. Broca’s, insula, 112 154 0.43 0.18 1.00 NoShapiro and Caramazza, p. 1193  English BF RC internal capsule, putamen 26 26 0.46 0.08 1.00 No GD L 44 44 0.67 0.75 0.24 No PG L mca, parietal 44 44 0.54 0.29 1.09 No DC L parietal 44 44 0.83 0.75 0.91 NoLambon Ralph et al., p. 110 (Table 4) JS L 44 44 0.75 0.54 0.99 No English BPF  DM L subarachanoid hemorrhage 44 44 0.96 0.67 1.00 No AB 44 44 0.25 0.33 0.24 No GN 44 44 0.33 0.63 0.01 Yes JL 44 44 0.71 0.83 0.10 NoUllman et al., p. 202 (Table 5)  English B (F) RBA L Broca’s, temporal isthmus 20 16 0.20 0.25 0.51 NoDruks p. 1006  English BF MC L 36 36 0.42 0.36 0.77 No P1/B6 21 39 0.43 0.49 0.44 No P2 21 39 0.71 0.05 1.00 No P3/B8 21 39 0.71 0.31 1.00 NoFix and Thompson, p. 167 (Table 1)* English BF P5 21 39 0.29 0.62 0.02 Yes  P6 21 39 0.60 0.16 1.00 No P7/B1 21 39 0.57 0.49 0.81 No P8 21 39 0.38 0.41 0.52 No B1/P7 L mca 15 15 0.50 0.60 0.43 No B2 L mca 15 15 0.53 0.67 0.36 No B3 L mca 15 15 0.53 0.53 0.64 No B4 L mca 15 15 0.67 0.53 0.87 NoFaroqi-Shah and Thompson (Fig. 2)* English AF 5 
6Table 1 (Continued )Reference Language Task/stimuli Patient Lesion # Stimuli Accuracy p One-tail R < I? p < 0.05 R I R I B5 L mca 15 15 0.40 0.53 0.36 No B6/P1 L mca 15 15 0.33 0.33 0.65 No B7 L mca 15 15 0.47 0.33 0.87 No B8/P3 L mca 15 15 0.47 0.93 0.01 Yes B9 L mca 15 15 0.40 0.53 0.43 No B10 L mca, basal ganglia 15 15 0.87 0.80 0.83 NoPenke and Westerman, p. 568 (Fig. 2) Dutch B MG L 67 51 1.00 0.60 1.00 No  CO L frontal 30 30 1.00 0.72 1.00 No HU L dorsal mca 30 30 1.00 0.82 1.00 No Z 67 51 0.92 0.58 1.00 No KE L internal capsule 30 30 0.97 0.74 1.00 No AR L fronto temporal 30 30 0.93 0.74 0.99 No Y. Faroqi-Shah / Brain Research Bulletin 74 (2007) 1–13 JW L 67 51 0.98 0.82 1.00 No KL L mca 30 30 0.97 0.90 0.94 No NI L mca 30 30 0.97 0.93 0.88 No AD L mca 67 51 0.92 0.91 0.67 No AN L 67 51 0.93 0.96 0.35 No RE 30 30 0.81 0.81 0.63 No MT L frontal area of mca 39 39 1.00 0.35 1.00 No PB L mca 39 39 1.00 0.50 1.00 No MB L frontal area of mca 39 39 1.00 0.65 1.00 No KH L mca 39 39 1.00 0.91 1.00 No WW L mca 39 39 0.99 0.90 1.00 NoPenke et al., p. 228 (Fig. 1)  German B HR L mca 39 39 0.99 0.71 1.00 No JZ L mca 39 39 0.99 0.57 1.00 No AH L frontal area of mca 39 39 0.91 0.57 1.00 No GB L mca incl. Broca’s 39 39 0.85 0.80 0.81 No MJ L 39 39 0.89 0.98 0.18 No FW L 39 39 0.60 0.65 0.41 NoTsapkini et al., p. 276  Greek C SK L temporal, internal capsule 11 8 0.27 0.625 0.14 NoLaiacona and Caramazza, p. 116 Italian B MR Frontoparietal, ant. and sup. temporal 204 156 0.67 0.52 1.00 Yes (Fig. 5), p. 117 (Fig. 6)  JM L fronto temporal subcortical, Rt. 34 30 0.824 0.437 1.00 No Spanish BBalaguer, p. 217–218, p. 220 (Fig. 4) periventricular $ MP L mca 28 25 0.90 0.524 1.00 No JM 34 39 0.912 0.713 1.00 No Catalan B MP 35 18 0.80 0.417 1.00 NoMean 0.59 0.51Standard deviation 0.31 0.26Standard error of mean 0.04 0.03Abbreviations: A, ﬁll blank sentence completion; ant., anterior; B, multiple choice sentence completion; C, response to question; F, frequency matching of regular–irregular; I = irregular; L, left hemisphere; MCA,middle cerebral artery; P, phonological matching of regular–irregulars (syllable structure); post., posterior; R, regular; R < I, accuracy of regulars is less than irregulars; Rt., right; sup., superior; $, past and presentelicited; (* ) three participants overlapped in these two studies and their corresponding codes are listed.Disclaimer: In some cases, p-values may vary marginally from those reported by authors since raw data were computed from ﬁgures.
Table 2Studies of word repetition sorted by language and chronological orderAuthor/data location Language stimuli Patient Lesion # Stimuli Accuracy % p One-tail R<I? p < 0.05 R I R IBadeckar and Caramazza, p. 295 English F FM L post. inf. frontal, inf. 50 50 0.56 0.74 0.00 Yes (Table 9)  parietal, ant. temporal BD L temporo-parietal, R lacunar 10 10 0.60 0.90 0.15 NoKohn and Melvold, p. 337  English JW L frontal–parietal temporal 10 10 0.50 0.90 0.07 NoTsapkini et al., p. 278  Greek SK L temporal, internal capsule 43 43 1.00 0.98 1.00 NoBird et al., p. 507 (Fig. 1), p. 509 English F BB L 44 44 0.13 0.32 0.03 Yes (Fig. 2)  IJ L 44 44 0.15 0.35 0.04 Yes IB L frontal temporal parietal 44 44 0.56 0.82 0.00 Yes DE L mca 44 44 0.15 0.65 0.00 Yes Y. Faroqi-Shah / Brain Research Bulletin 74 (2007) 1–13 AB L mca 44 44 0.58 0.82 0.01 Yes MB L frontoparietal, R frontal 44 44 0.08 0.20 0.00 Yes RT L 44 44 0.75 0.95 0.30 No VC L frontoparietal 44 44 0.74 0.82 0.01 Yes GN L frontal including Broca’s, 44 44 0.76 0.94 0.00 Yes occipital JL L frontal temporal parietal 44 44 0.66 0.96 0.11 No BB 34 34 0.25 0.30 0.47 No IJ 34 34 0.25 0.18 0.87 No IB 34 34 0.90 0.89 0.79 No DE 34 34 0.59 0.50 0.83 No AB 34 34 0.70 0.80 0.29 No English P F MB 34 34 0.20 0.25 0.41 No RT 34 34 0.91 0.91 0.66 No VC 34 34 0.91 0.85 0.87 No GN 34 34 0.81 0.85 0.50 No JL 34 34 0.98 0.92 0.94 NoLambon Ralph et al., p. 112 (Table 6) English F AB L 34 34 0.85 0.97 0.10 No  GN L mca, parietal 34 34 0.82 0.94 0.13 No JL 34 34 0.91 0.91 0.66 No GD L 34 34 0.97 0.97 0.75 No PG L subarachanoid hem. 34 34 0.97 0.94 0.88 No DC 34 34 1.00 0.97 1.00 No JS 34 34 0.94 0.88 0.90 No DM 34 34 1.00 0.91 1.00 NoMean 0.66 0.76Standard deviation 0.30 0.26Standard error of mean 0.05 0.04Abbreviations: ant., anterior; F, frequency matching of regular–irregular; I = irregular; L, left hemisphere; mca, middle cerebral artery; P, phonological matching of regular–irregulars (syllable structure); post.,posterior; R, regular; R < I, accuracy of regulars is less than irregulars; Rt., right; sup., superior.Disclaimer: In some cases, p-values may vary marginally from those reported by authors since raw data were computed from ﬁgures. 7
8 Y. Faroqi-Shah / Brain Research Bulletin 74 (2007) 1–13Fig. 2. Meta-analysis of sentence completion data showing proportion ofpatients with: signiﬁcantly lower accuracy of regulars compared to irregulars Fig. 3. Meta-analysis of repetition data showing proportion of patients with:(R < I), irregulars compared to regulars (R > I), and no signiﬁcant differences signiﬁcantly lower accuracy of regulars compared to irregulars (R < I), irregularsbetween accuracies of regulars and irregulars (R–I ns) (Fisher-exact test, one- compared to regulars (R > I), and no signiﬁcant differences between accuracies oftailed p < 0.05) (N = 78). regulars and irregulars (R–I ns) (Fisher-exact test, one-tailed p < 0.05) (N = 22). to regulars were reported. There was a high positive correla-English speakers. The number of English speakers who showed tion between the accuracies of regulars and irregulars, whenpoorer accuracy for irregulars is also 9 (out of a total of 26 computed across all datasets (rs = 0.8, p < 000).who showed this pattern). Seven (out of a total of 13) reports Other variables. Word repetition data are primarily frommatched regulars and irregulars for frequency of occurrence, English speakers, and only one of the 32 datasets came fromat least for a subset of the stimuli, and only two studies had a non-English speaker . Four out of six reports had somematched phonological aspects of regular and irregular verbs, frequency matching between regulars and irregular verb stim-such as syllable structure. The number of stimuli used ranged uli, while only one study reported matching of phonologicalfrom 8 to 204 [35,55]. To summarize, over half of the sen- variables.tence production datasets are from English speakers. An equalnumber of English speaking patients showed opposite patterns 3.3. Lesion information(regulars < irregulars and irregular < regulars). If all patients whowere worse on regulars were considered, a majority (9/10) were The results of the lesion analysis are reported in Table 3English speakers. Further, only half of the studies matched stim- and Fig. 4. As is evident from Table 3, lesion data from onlyuli for frequency and an even smaller number matched stimuli 31 patients could be entered into the analysis of dissociationsfor phonological complexity. because lesion information was either not reported (N = 9) or described in non-speciﬁc terms, such as left hemisphere or left3.2. Repetition middle cerebral artery (N = 35). Of particular interest to the neu- roanatomical hypotheses being evaluated are patients whose Word repetition included 32 data sets from 26 different apha- lesions were exclusively frontal (or frontoparietal) without anysic patients (see Table 2). A majority of these data sets (N = 28) temporal lobe involvement and vice versa (N = 15 datasets).came from the same group of authors [8,36] whose focus was Table 3 reveals that no speciﬁc dissociation patterns were asso-to demonstrate that variations in stimulus characteristics can ciated with either of these lesion sites.inﬂuence the existence of a dissociation. Nine datasets showedsigniﬁcantly lower accuracy of regular verbs compared to irregu- 3.4. Inter-test consistencylar verbs. The remaining 23 datasets had no signiﬁcant differencebetween regulars and irregulars as shown in Fig. 3. No patients As mentioned earlier, 19 individuals contributed more thanwith signiﬁcantly better performance on irregulars compared one dataset. A comparison of regular–irregular performanceTable 3Regular–irregular verb performance for various lesions # Patients # Datasets R<I R>I R = I, ns PatientsFrontal excl. temporal 9 12 5 6 1 FCL, MB, VC, RC, PG, CO, MT, AH, GNTemporal excl. frontal 2 3 3 SK, BDFrontal + parietal + temporal 5 12 3 9 FM, JW, IB, JL, MRFrontal + temporal 3 3 1 2 RBA, AR, JMOther (subcortical, subarachnoid, etc.) 12Non-speciﬁc description (L mca) 35No lesion information 9 P1-P7, RE, Z
Y. Faroqi-Shah / Brain Research Bulletin 74 (2007) 1–13 9 and irregulars (65 datasets), worse performance on irregulars (26 datasets) and worse performance on regulars (9 datasets). The most frequent pattern in both sentence production and word repetition tasks was that of no signiﬁcant difference between regulars and irregulars (Figs. 2 and 3). It is also noteworthy that only ﬁve datasets out of a total of 110 had a difference of greater than 50% between regulars and irregulars (sentences: FCL, GN, MT, PB; repetition: DE). Among these ﬁve datasets, there was no consistent pattern. In other words, classical dissoci- ations, the strongest form of neuropsychological evidence , were hardly found. Moreover, a signiﬁcantly high positive cor- relation was found between regular and irregular verb accuracy. The observed heterogeneity, predominant pattern of no differ- Fig. 4. Proportion of patients with various lesion sites. ence, and high positive correlation, argue against the presence of a clear and consistent pattern of dissociation between regularsacross the different datasets of each of these patients revealed and irregulars in Broca’s aphasia.intra-individual variations in dissociation patterns. In 13 cases, It is evident that the declarative/procedural  and seman-the proﬁle changed from regular < irregular to no signiﬁcant dif- tic/phonology [31,32,42,56,57] models, both of which predictference between regulars and irregulars. A majority of these worse performance on regular compared to irregular verbs inpatients were reported by the same group of authors whose Broca’s aphasia, are not supported. The semantics/phonologyprimary focus was to demonstrate how dissociation patterns [31,32] model, assumes that individuals with Broca’s aphasiaare vulnerable to stimulus manipulations [8,36]. A patient have phonological deﬁcits and hence difﬁculty with regular(B8/P3) who participated in studies by two different authors verb production. Since the presence of phonological deﬁcits[21,22], was signiﬁcantly worse on regulars in one study was not documented by most authors, it may be preliminarywhile being signiﬁcantly worse on irregulars in another study to reject its claims on the basis of lack of evidence. In a study(Table 1). These two studies used different task variations. of two ﬂuent and two non-ﬂuent aphasic individuals with con-The performance of ﬁve other patients remained unchanged ﬁrmed phonological deﬁcits, repetition of regular verbs wasacross multiple datasets. To summarize, these 19 individuals worse than irregulars in all four patients . In another study,who contributed multiple datasets, demonstrate poor inter- controlling for phonological difﬁculty eliminated prior disso-test consistency. This indicates that dissociation patterns are ciations between regulars and irregulars in non-ﬂuent aphasicsusceptible change across testing sessions and experimental patients . Hence, it appears that phonological complexity isvariations. a stimulus artifact that needs to be controlled in future stud- ies of regular–irregular verb processing. Given that a majority4. Discussion of data sets found no regular–irregular differences, the data are best accommodated by viewpoints that assume more similarities This meta-analysis of published research on regular–irregular rather than differences between regular and irregular verb pro-verb production in individuals with non-ﬂuent/Broca’s aphasia duction, such as the two-stage model [5,6,37]. The second stagewas conducted to address three questions. (1) Is there a con- of the two-stage model is assumed to be common to both irreg-sistent pattern of dissociation between regulars and irregulars? ular and regular verbs by involving retrieval of morphosyntactic(2) Are speciﬁc dissociation patterns associated with speciﬁc information, such as tense [5,6]. Further, this second stage islesions? In particular, is the pattern of worse performance on claimed to be processed by the LIFG, a brain area whose lesionsregulars associated with frontal lobe lesions? More speciﬁcally, are commonly associated with Broca’s aphasia.we compared the predictions of the declarative-procedural , The neuroanatomical models that were developed on thesemantic-phonology  and two-stage  models, and (3) is idea that regulars are impaired in Broca’s aphasia have largelyit possible to delineate any variables that inﬂuence dissocia- ignored the fact that some aspects of regular morphology aretion patterns? These questions are addressed in the following relatively spared in Broca’s aphasia [41,57]. For example, pro-sections. gressive forms of verbs (kicking, sleeping), verbs that inﬂect to agree with the preceding noun (The cat sleeps versus The cats4.1. Is there a dissociation between regular and irregular sleep), and plural noun morphology (dogs) are all regular andverbs? are produced with relatively high accuracy [23,24]. This means that models that make a binary distinction between regulars and The most evident aspect of the data presented is the great irregulars while failing to incorporate the whole spectrum ofvariability across patients in the production of regular and irreg- regular morphological operations are inadequate. Any modelular verbs, although all patients were described as individuals of the neurocognitive differences between regular and irregularwith non-ﬂuent/Broca’s aphasia. Combined across the 110 data morphology is incomplete if it does not account for the relativesets of sentence production and word repetition, all three pos- differences among these various regular morphological formssible patterns were observed: no difference between regulars [35,53].
10 Y. Faroqi-Shah / Brain Research Bulletin 74 (2007) 1–134.2. Are speciﬁc lesions sites associated with speciﬁc that a majority of patients who contributed more than onedissociation patterns? dataset showed variations in the relative performance of reg- ulars and irregulars across the datasets (Section 3.4). This Not surprisingly, lesions of the left hemisphere were reported intra-individual variability questions the robustness of previ-for all the participants. A variety of lesions, including cor- ously reported “regular–irregular dissociations” and highlightstical and subcortical regions, were reported. Given that the the importance of demonstrating test–retest consistency in neu-LIFG is frequently credited with the mental grammar and ropsychological data, especially when such data are used tohence production of regular verbs [41,59], the most interest- frame theories about the normal neural processing of regularing group constitutes patients with exclusively frontal lobe (or and irregular verbs [56,59].frontal–parietal) lesions. We are making the liberal assump- Secondly, the data suggest that at least some reported disso-tion that the lesion was inclusive of the LIFG in all patients ciations may have been an artifact of stimulus manipulations.with reported frontal lobe lesions. Considerable variability in This point is especially made by a few studies in whichregular–irregular performance is observed across the 17 patients apparent dissociations between regular and irregular verb pro-who had some form of frontal lobe lesion (see Table 3), duction were eliminated when verbs with matched phonologicalwith approximately equal numbers of datasets showing regu- complexity (the number of syllables, the presence of con-lars worse and better than irregulars. Hence, the data do not sonant clusters) were used [8,36]. It was emphasized that,support a special connection between LIFG lesions and a reg- because a majority of regularly inﬂected verbs are phonolog-ular verb deﬁcit. The three patients who had temporal lobe ically complex in English, failure to control for phonologicallesions with the sparing of frontal lobe showed no signiﬁ- complexity disproportionately affects the production of regularcant differences between regulars and irregulars. Recall that the verbs. Phonological factors seem to inﬂuence the occurrencedeclarative/procedural and semantic/phonology models predict of regular–irregular dissociations in another manner. Aphasicpoor performance on irregulars in temporal lobe lesions, while individuals with phonological deﬁcits have a greater difﬁcultythe two-stage model predicts poor regular verb performance for producing regular verb inﬂections in English, even when they dosuch patients. To summarize, there seems to be no evident cor- not have Broca’s aphasia . Hence, the pattern of worse per-relation between any pattern of performance and speciﬁc frontal formance on regulars is not invariably linked to Broca’s aphasia,and temporal lesion sites. but may be inﬂuenced by phonological deﬁcits  and stimulus These complex lesion-deﬁcit ﬁndings are actually con- characteristics [8,36]. There may also be an interaction betweensistent with neuroimaging studies of regular–irregular verbs modality (reading versus repetition) for phonological variables.that have found a wide variety of activation patterns In a previous study, matching phonological complexity in a read-[16,17,28,30,49,54,56]. No two-brain imaging studies agree on ing task (links versus lynx) still showed a worse performance forthe regions activated for regular and irregular verbs, perhaps due the morphologically complex word (links) [3,4]. A variety ofto methodological variations. In other words, neuropsycholog- frequency measures including afﬁx frequency  lexeme fre-ical and neuroimaging data suggest that the neural generation quency , and stem frequency  have also been shownof regulars and irregulars is far more complex than that pre- to inﬂuence accuracy of verb morphology. As Tables 1 and 2dicted by any current neuroanatomical model. One similarity show, a majority of the studies did not control frequency mea-across most brain imaging studies is the reported activation of sures between regular and irregular stimuli, thereby weakeningfrontal regions for both regular and irregular verbs. As for left the ﬁndings. Task variations are another factor that need carefulfrontal lobe involvement, the two-stage model  is the only consideration as is evident with patient B6/P1 [21,22] in Table 1neuroanatomical model that speciﬁcally implicates both regular and also in other studies [8,34].and irregular verb processing for this region. Hence, one may Finally, the reviewed studies had at least two methodolog-conclude (though not without reservations) that the ﬁnding of ical weaknesses that are likely to inﬂate the occurrence ofhigh correlation between regular and irregular verb scores in dissociations: use of experimental tasks that produced ceilingBroca’s aphasia ﬁts best with the two-stage model. performance in unimpaired controls, and the failure to use a A cautionary note about the present lesion analysis is that matched control group [13,14,38,52].less than half of all patients could contribute to the lesion anal-ysis because lesion data in other studies were not reported in 5. Conclusions and future research considerationsterms that were speciﬁc enough for inclusion. It is quite possi-ble that the null result may be because the lesion descriptions are The following statements can be made on the basis of the 110too vague and non-speciﬁc to reveal any correlations. Perhaps datasets reviewed: (1) there was no consistent pattern of disso-this question needs to be addressed using a more ﬁne-grained ciation between regular and irregular verbs in persons describedapproach, such as voxel-based lesion analysis . as non-ﬂuent or Broca’s aphasia, at least for sentence and word repetition tasks reviewed in this meta-analysis3 ; (2) the most fre-4.3. Can dissociations be manipulated by experimental quent pattern was that of no difference in performance; (3) thevariables? At least three distinct points can be made regarding this 3 The semantics/phonology model is primarily based on auditory same differ-question. First, a non-trivial ﬁnding of this meta-analysis was ent discrimination and lexical priming .
Y. Faroqi-Shah / Brain Research Bulletin 74 (2007) 1–13 11magnitude of difference between regular and irregular produc- number of stimuli, task demands, inter-session consistency, andtion accuracy was minimal, with only ﬁve individuals showing associated deﬁcits when using neuropsychological data to for-a difference of greater than 50%; (4) a wide variety of lesions mulate theories of the neural organization of language.were reported; (5) lesion sites were not associated with speciﬁc Secondly, a frequent logical fallacy when drawing interpre-performance patterns in any consistent manner; and (6) inter-test tations about normal language function from studies of aphasicvariability was found in nearly two-thirds of patients who had individuals is the over-simpliﬁed assumption that all Broca’smultiple datasets. aphasic individuals have a lesion of the Broca’s area, and there- The absence of a consistent dissociation and lesion-deﬁcit fore observed deﬁcits reﬂect functions of a normal Broca’s area.correlation can be interpreted in different ways. First, the results A ﬁne-grained documentation of the extent of lesion is war-may be taken at face value to mean that there is indeed no ranted in every study that makes speciﬁc claims about normaldissociation between regulars and irregulars in Broca’s apha- neural architecture. There are previous instances when a studysia. Regular and irregular verbs may fail to dissociate because of Broca’s aphasic individuals is cited elsewhere as a study ofa complex network of multiple, overlapping brain regions are patients with lesions of LIFG, even though the original authorsinvolved in producing both verb types, and this was demon- did not report LIFG lesions (e.g. [5, p. 883]). In fact, onlystrated by brain imaging research [16,17,28,30,32,49,54,56]. 50–60% of patients with lesions to Broca’s area have persistentAlternatively, these verbs may fail to dissociate because the Broca’s aphasia and about 15% of right-handed Broca’s aphasicpsycholinguistic dichotomy between regulars and irregulars individuals do not have lesions of Broca’s area . Also givenassumed by aphasiologists is overly simplistic [26,54,22,19]. the heterogeneity of the deﬁnition of Broca’s area across studiesArguments for a gradation between regulars and irregulars , neuroanatomical models that are based on patient perfor-include the presence of subregularities among irregulars (sang, mance need to explicitly deﬁne their “Broca’s area” and providerang, drank versus slept, kept, wept) , afﬁx-like patterns in ﬁne-grained lesion information.some irregulars (kept → kep + t) [22,26], presence of afﬁxes Third, future models also need to incorporate a more com-in irregular verbs of other languages [6,43], and “morphologi- plete range of morphological operations, such as noun plurals,cally simple” behavior of high frequency regulars . As some subject verb agreement, and non-ﬁnite verbs. This is war-authors have previously pointed out, regular and irregular past ranted because recent research suggests that the trouble intense verbs have more commonalities (morphosemantic notion Broca’s aphasia may not be morphological regularity per se,of past tense, syntactic structure of the sentence, experimen- but rather an effect of grammatical class [35,53], or mor-tal task demands, etc.) than the one morphological difference phosemantic aspects, such as tense marking [21,23,61]. Finally,[5,6,37]. language speciﬁc effects need to be considered since this meta- As for the three neuroanatomical viewpoints presented analysis revealed a high proportion of English speakers in theearlier, the existing data are incompatible with the declarative- group that performed worse on regular verbs. Currently, onlyprocedural model , but fairly consistent with the two-stage the two-stage model accommodates language speciﬁc patterns.model . The two-stage model assumes some degree of neu- It is noteworthy that recent neuroanatomical hypotheses areroanatomical overlap between regulars and irregulars and hence moving in the right direction and becoming increasingly sophis-the failure of regulars and irregulars to completely dissoci- ticated in their empirically driven descriptions of processesate. Some aspects of the results may be consistent with the and neural regions involved in the generation of regular andsemantic-phonology model, especially if one makes the con- irregular verbs [5,33,57]. To conclude, this meta-analysis doesnection between regulars and phonological skills rather than not claim that regular and irregular verbs cannot dissociatebetween regulars and frontal lobe. Another neuroanatomical in non-ﬂuent/Broca’s aphasia. Rather, the ﬁnding is that non-model that is a viable alternative is the memory, uniﬁcation, ﬂuent/Broca’s aphasia is not associated with a consistent patterncontrol (MUC) framework , which allocates the role of or direction of dissociation.uniﬁcation to the LIFG. Uniﬁcation is deﬁned as combininglinguistic elements into larger units and it is recognized that uni- Conﬂict of interestﬁcation operations occur in parallel in semantic, syntactic, andphonological domains. Although the author does not explicitly None.talk about morphology, it can be seen how both regulars andirregulars involve the same semantic and syntactic uniﬁcation Acknowledgementsoperations, differing only in morphological uniﬁcation. There-fore LIFG lesions would impair both regular and irregular verb The author wishes to thank Rachel Mont, Adrianna Naim and“uniﬁcation” operations.4 Jennifer Maultasch for help with data coding. I also thank two There are several important considerations for future anonymous reviewers for their helpful comments on an earlierresearch. First, the analysis underlines the relevance of metic- version of this article.ulous control and documentation of confounding variables, References 4 In its present form, the MUC framework  is drawn from comprehension  M. Alegre, P. Gordon, Frequency effects and the representational status ofdata. regular inﬂections, J. Mem. Lang. 40 (1999) 41–61.
12 Y. Faroqi-Shah / Brain Research Bulletin 74 (2007) 1–13  H. Baayen, R. Schreuder, War and peace: morphemes and full forms in a  A.L. Inglis, Is ‘time’ up for the irregularity of tense in aphasia, in relation noninteractive activation parallel dual-route model, Brain Lang. 68 (1999) to semantics and the temporal lobe? Brain Lang. 95 (2005) 189. 27–32.  J.J. Jaeger, A.H. Lockwood, D.L. Kemmerer, J. Van Valin, B.W. Mur-  W. Badeckar, A. Caramazza, The analysis of morphological errors in a case phy, H.G. Khalak, A positron emission tomographic study of regular of acquired dyslexia, Brain Lang. 32 (1987) 278–305. and irregular verb morphology in English, Language 72 (1996) 451–  W. Badeckar, A. Caramazza, Morphological composition in the lexical 493. output system, Cogn. Neurosci. 8 (1991) 335–367.  M.F. Joanisse, M.S. Seidenberg, Impairments in verb morphology after  R.D. Balaguer, A. Rodriguez-Fornells, M. Rotte, J. Bahlmann, H.J. Heinze, brain injury: a connectionist model, Proc. Natl. Acad. Sci. U.S.A. 96 (1999) T.F. Munte, Neural circuits subserving the retrieval of stems and grammat- 7592–7597. ical features in regular and irregular verbs, Hum. Brain Map. 27 (2006)  M. Joanisse, M. Seidenberg, Imaging the past: neural activation in frontal 874–888. and temporal regions during regular and irregular past tense processing,  R.D. Balaguer, Regular and irregular morphology and is relationship with Cogn. Affect. Behav. Neurosci. 5 (2005) 282–296. agrammatism: evidence from two Spanish-Catalan bilinguals, Brain Lang.  S. Kohn, J. Melvold, Effects of morphological complexity on phonologi- 91 (2004) 212–222. cal output deﬁcits in ﬂuent and nonﬂuent aphasia, Brain Lang. 73 (2000)  E. Bates, M. Appelbaum, J. Salcedo, A.P. Saygin, L. Pizzamiglio, 323–346. Quantifying dissociations in neuropsychological research, J. Clin. Exp.  P. Kok, A. van Doorn, H. Kolk, Inﬂection and computational load in agram- Neuropsychol. 25 (2003) 1128–1153. matic speech, Brain Lang. 102 (2007) 273–283.  H. Bird, M.A. Lambon Ralph, M.S. Seidenberg, J.L. McClelland, K.  M. Laiacona, A. Caramazza, The noun/verb dissociation in language Patterson, Deﬁcits in phonology and past-tense morphology: what’s the production: varieties of causes, Cogn. Neuropsychol. 21 (2004) 103– connection? J. Mem. Lang. 48 (2003) 502–526. 123.  J. Bybee, D.I. Slobin, Rules and schemas in the development and use of  M.A. Lambon Ralph, N. Braber, J.L. McClelland, K. Patterson, What the English past tense, Language 58 (1982) 265–289. underlies the neuropsychological pattern of irregular > regular past-tense J. Bybee, Regular morphology and the lexicon, Lang. Cogn. Processes 10 verb production? Brain Lang. 93 (2005) 106–119. (1995) 425–455.  A. Laudanna, Regular versus irregular inﬂection: a question of levels, J.G. Centeno, L.K. Obler, Agrammatic verb errors in Spanish speakers and Behav. Brain Sci. (1998) 1028–1029. their normal discourse correlates, J. Neuroling. 14 (2001) 349.  K.R. Laws, T.M. Gail, V.M. Leeson, J.R. Crawford, When is category H. Clahsen, M. Almazan, Syntax and morphology in William’s syndrome, speciﬁc in Alzheimer’s disease? Cortex 41 (2005) 452–463. Cognition 68 (1998) 167–198.  R. Lindenberg, H. Fangerau, R.J. Seitz, “Broca’s area” as a collective term? J.R. Crawford, P.H. Garthwaite, Detecting dissociations in single-case stud- Brain Lang. 102 (2007) 22–29. ies: type I errors, statistical power and the classical versus strong distinction,  http://www.matforsk.no/ola/ﬁsher.htm. Neuropsychologia 44 (2006) 2249.  W.D. Marslen-Wilson, L.K. Tyler, Dissociating types of mental computa- J.R. Crawford, P.H. Garthwaite, C. Gray, Wanted: fully operational deﬁni- tion, Nature 387 (1997) 592–594. tions of dissociations in single case studies, Cortex 39 (2003) 357–370.  W.D. Marslen-Wilson, L.K. Tyler, Morphology, language and the brain: A.R. Damasio, Aphasia. New Engl. J. Med. 326 (1992) 531–539. the decompositional substrate for language comprehension, Philos. Trans. M.H. Davis, F. Meunier, W.D. Marslen-Wilson, Neural responses to mor- R. Soc. B 362 (2007) 823–836. phological, syntactic, and semantic properties of single words: an FMRI  M. Penke, G. Westermann, Broca’s area and inﬂectional morphology: study, Brain Lang. 89 (2004) 439. evidence from Broca’s aphasia and computer modeling, Cortex (2006) R.P. Dhond, K. Marinkovic, A.M. Dale, T. Witzel, E. Halgren, Spa- 563–576. tiotemporal maps of past-tense verb inﬂection, Neuroimage 19 (2003) 91–  M. Penke, H. Weyerts, M. Gross, E. Zander, T.F. Munte, H. Clahsen, How 100. the brain processes complex words: an event-related potential study of N.F. Dronkers, J.K. Shapiro, B. Redfern, R.T. Knight, The role of Broca’s German verb inﬂections, Cogn. Brain Res. 6 (1997) 37–52. area in Broca’s aphasia, J. Clin. Exp. Neuropsychol. 14 (1992) 52–53.  M. Penke, U. Janssen, M. Krause, The representation of inﬂectional mor- J. Druks, Morpho-syntactic and morpho-phonological deﬁcits in the pro- phology: evidence from Broca’s aphasia, Brain Lang. 68 (1999) 225–232. duction of regularly and irregularly inﬂected verbs, Aphasiology 20 (2006)  S. Pinker, M.T. Ullman, The past and future of the past tense, Trends Cogn. 993–1017. Sci. 6 (2002) 456–463. Y. Faroqi-Shah, C.K. Thompson, Semantic, lexical, and phonological inﬂu-  S. Pinker, The Language Instinct, William Morrow, New York, 1994. ences on the production of verb inﬂections in agrammatic aphasia, Brain  D. Rumelhart, J. McClelland, On learning the past tenses of English Lang. 89 (2004) 484–498. verbs: implicit rules or parallel distributed processing, in: D. Rumel- Y. Faroqi-Shah, C.K. Thompson, Verb inﬂections in agrammatic aphasia: hart, J. McClelland (Eds.), Parallel Distributed Processing: Explorations encoding of tense features, J. Mem. Lang. 56 (2007) 129–151. in the Microstructure of Cognition, MIT Press, Cambridge, MA, S. Fix, C.K. Thompson, Morphophonological structure and agrammatic 1986. regular/irregular past-tense production, Brain Lang. 99 (2006) 166–167.  M. Sach, R.J. Seitz, P. Indefrey, Uniﬁed inﬂectional processing of regular N.a. Friedmann, Y. Grodzinksy, Tense and agreement in agrammatic pro- and irregular verbs: a PET study, Neuroreport 15 (2004) 533–537. duction: pruning the syntactic tree, Brain Lang. 56 (1997) 397–425.  E.M. Saffran, R.S. Berndt, M.F. Schwartz, The quantative analysis of H. Goodglass, J.A. Christiansen, R. Gallagher, Comparison of morphology agrammatic production: procedure and data, Brain Lang. 37 (1989) and syntax in free narrative and structured tests: ﬂuent versus nonﬂuent 440–479. aphasics, Cortex 29 (1993) 377–407.  S. Seigel, J.N. Castellan, Nonparametric Statistics for the Behavioral Sci- P. Hagoort, On Broca, brain, and binding: a new framework, Trends Cogn. ences, second ed., McGraw Hill, Boston, 1988. Sci. 9 (2005) 416–423.  T. Shallice, From Neuropsychology to Mental Structure, Cambridge Uni- M. Halle, A. Marantz, Distributed morphology and the pieces of inﬂection, versity Press, Cambridge, UK, 1988. in: K. Halle, S.J. Keyser (Eds.), The View from Building 29: Essays in  K. Shapiro, A. Caramazza, Grammatical processing of nouns and verbs in Linguistics in Honor of Sylvian Bromberger, MIT Press, Massachusetts, the left frontal cortex? Neuropsychologia 41 (2003) 1189–1198. 1993, pp. 111–175.  L. Stockall, A. Marantz, A single route, full decomposition model of G. Hickok, D. Poeppel, Towards a functional neuroanatomy of speech morphological complexity: MEG evidence, Mental Lexicon 1 (2006) 85– perception, Trends Cogn. Sci. 4 (2000) 131–138. 123. P. Indefrey, C. Brown, P. Hagoort, H. Herzog, M. Sach, R.J. Seitz, A PET  K. Tsapkini, G. Jarema, E. Kehayia, A morphological processing deﬁcit study of cerebral activation patterns induced by verb inﬂection, NeuroIm- in verbs but not in nouns: a case study in a highly inﬂected language, J. age 5 (1997) S548. Neuroling. 15 (2002) 265–288.
Y. Faroqi-Shah / Brain Research Bulletin 74 (2007) 1–13 13 L.K. Tyler, W.D. Marslen-Wilson, E.A. Stamatakis, Differentiating lexical  M.T. Ullman, S. Corkin, M. Coppola, G. Hickok, A neural dissociation form, meaning, and structure in the neural language system, Proc. Natl. within language: evidence that the mental dictionary is part of declara- Acad. Sci. U.S.A. 102 (2005) 8375–8380. tive memory, and that grammatical rules are processed by the procedural L.K. Tyler, E.A. Stamatakis, B. Post, B. Randall, W. Marslen-Wilson, Tem- system, J. Cogn. Neurosci. 9 (1997) 266–276. poral and frontal systems in speech comprehension: an FMRI study of past  M.T. Ullman, R. Pancheva, T. Love, E. Yee, D. Swinney, G. Hickok, tense processing, Neuropsychologia 43 (2005) 1963–1974. Neural correlates of lexicon and grammar: evidence from the produc- M.T. Ullman, R. Bergida, M. O’ Craven, Distinct fMRI activation pat- tion, reading, and judgment of inﬂection in aphasia, Brain Lang. 93 (2005) terns for regular and irregular past tense, in: Paper Presented at the Third 185. International Conference on Functional Mapping of the Human Brain,  M. Wenzlaff, H. Clahsen, Tense and agreement in German agrammatism, Copenhagen, Denmark, 1997. Brain Lang. 89 (2004) 57–68.