درباره پردازش اشکال منظم و نامنظم افعال و اسامی: شواهد از نروسایکلوژی

Following acquired brain damage, a native English speaking patient (AW) encountered problems accessing phonology in speech production, while her ability to access word meaning appeared to be intact. In a series of tasks, AW was presented either with a verb, and was asked to produce its past tense or past participle (walk → “walked”), or with a noun, and was asked to produce its plural (glove → “gloves”). A stark dissociation was found: while AW responded accurately with regular forms of verbs (walked) and nouns (gloves), performance was significantly less accurate with irregular forms (found; children). The appearance of a selective deficit for irregular forms in conditions of impaired lexical access is in line with dual-mechanism accounts, which proposes that irregular forms are specified in the lexicon whereas regular forms are computed via rule-based mechanisms. In contrast, AW's data are problematic for connectionist accounts that do not posit separate mechanisms for processing regular and irregular forms, including the connectionist model recently proposed by Joanisse and Seidenberg (ProceedingsoftheNationalAcademyofSciencesUSA96 (1999) 7592) which successfully simulated a variety of earlier neuropsychological findings. Analyses of AW's responses shed light on further details of the representation and processing of regular and irregular inflected forms.

In many languages, speakers modify the meaning of a word by changing the suffix appended to the end of the word. In English, for example, information about number (plural/singular) is conveyed by the presence/absence of the suffix -s at the end of nouns, and the suffixes -s, -ing, and -ed at the end of verbs express when the action or state described by the verbs takes place. Suffixation is an extremely productive process that speakers extend to recently introduced words like fax (faxes) and e-mail (e-mailed). But there are exceptions. English provides illustrative cases of such exceptions: a few plural nouns are not produced by adding the suffix -s (teeth, women, fish) and a good number of verbs take a past tense form that does not contain the suffix -ed (ran, sat, went). The occurrence of these irregular forms raises interesting questions about the process of word formation: are the mechanisms for word formation the same for regular and irregular words? If the mechanisms are not the same, how do they differ? Not only do the answers to these questions elucidate the organization of word formation processing (morphology) but they also have important implications for our understanding of the structure and functioning of the mental dictionary (the lexicon). For the past 20 years, the debate about regular and irregular form processing has focused on English past tense inflection, which has become a crucial test case for theories of word formation. A view that has found wide support in linguistics and psycholinguistics is the dual-mechanism account, which holds that different mechanisms are at play with regularly inflected verbs (walk–walked, argue–argued) and irregularly inflected verbs (run–ran, sit–sat) ( Bauer, 1983, Marslen-Wilson and Tyler, 1997, Pinker, 1991 and Ullman, 2001). For regular past tenses, rule-based mechanisms add the suffix -ed to the verb stem (walk+ed → walked). 1 Irregular past tenses are not obtained via rule-based mechanisms, but are stored in the lexicon; consequently, to produce these irregular forms speakers have to access the lexicon. Because the process for producing regular and irregular past tenses follows different principles, it is likely that different areas in the brain support the processing of these verbs (on this point see e.g. Marslen-Wilson & Tyler, 1998). A contrasting view holds that regular and irregular past tenses are computed by a single mechanism. This view was implemented in a number of connectionist simulations (e.g. MacWhinney and Leinbach, 1991, Marchman, 1993, Plunkett and Marchman, 1993 and Rumelhart and McClelland, 1986). These simulations operate on the basis of associative mechanisms established through learning which link input nodes specifying the phonology of a verb stem (e.g. walk) to output units specifying the phonology of the verb's past tense (e.g. walked). The theoretical import of these simulations is twofold: they demonstrate that it is not necessary to postulate different mechanisms for regular and irregular past tenses, nor is it necessary to postulate rule-based mechanisms for the formation of regular past tenses. In support of either of these views, researchers have cited various sorts of data: results from reaction time experiments, computer simulations, normal and abnormal language acquisition, and historical linguistics (for a review, see Clahsen, 1999, Marcus, 2000, Pinker, 1999 and Ullman, 2001). Recently, in an attempt to gather data that may resolve the current debate as well as shed light on the brain mechanisms underlying word processing, researchers have turned their attention to brain-damaged patients with acquired language disorders. Ullman et al. (1997) tested patients with language impairments due to different pathologies in a past tense generation task. Patients were presented with a sentence like “Every day I dig a hole. Just like every day, yesterday I ____ a hole” and asked to complete the sentence by providing the past tense form of the verb, “dug” in this example. Discrepancies were observed in patients' abilities to produce regular vs. irregular past tenses. Aphasic patients with posterior lesions and word-finding problems along with a group of patients diagnosed with Alzheimer's disease performed worse with irregularly inflected verbs. A contrasting pattern was reported for one aphasic patient with an anterior lesion and for a group of patients with Parkinson's disease; they encountered greater problems producing regular as opposed to irregular past tenses. Moreover, these patients performed differently than normal controls when asked to generate the past tense of novel verbs (vask, tunch): while normal controls typically added the suffixed -ed (“vasked”, “tunched”), the patients did so far less frequently. Similar dissociations have now been observed in a variety of tasks. Marslen-Wilson and Tyler, 1997 and Marslen-Wilson and Tyler, 1998 have documented selective deficits in the comprehension of regular or irregular past tenses with the priming paradigm. Four patients with acquired language deficits heard two words (a prime and a target) and decided whether the second word corresponded to a familiar word. Various prime–target pairs were used: morphologically related (jumped–jump), semantically related (swan–goose), and unrelated (locked–jump). For two patients, priming (faster responses than those for unrelated pairs) appeared in the presence of pairs formed by semantically related words and by irregular past tenses (found–find). For two other patients, facilitation only appeared in the presence of regular past tenses (walked–walk). Similar dissociations have been observed in reading ( Ullman et al., 1997 and Ullman et al., in press), and in a judgment task in which patients were asked to rate the ‘goodness’ of correct and incorrect past tense forms (e.g. dug/digged, rob/rob; Ullman et al., in press). In sum, the dissociation documented between regular and irregular past tenses both in speech production and comprehension has been interpreted as supporting the principle claim of the dual-mechanism account: that regular and irregular forms are processed by distinct (and neuroanatomically segregated) mechanisms (see also Bullinaria & Chater, 1995). The neuropsychological data have been considered to be incompatible with the view, endorsed by single-mechanism theories and implemented in connectionist networks, that regularly and irregularly inflected verbs recruit identical processes. Joanisse and Seidenberg (1999) challenged the claim that the neuropsychological evidence is irreconcilable with models that do not incorporate distinct processes for regular and irregular inflections. They implemented a connectionist model that simulated different tasks with past tenses, including the elicitation task devised by Ullman et al. (1997). The model includes units devoted to the encoding of verb meaning2 and verb phonology, respectively (see Fig. 1). The model also incorporates separate phonological units for speech input and speech output. A key feature of the model is that regular and irregular past tenses are processed similarly, in the sense that identical mechanisms are implicated in the processing of both classes of verbs. This feature does not mean that semantic and phonological information are equally critical for the production of regular, irregular, and novel past tense forms. Because a novel verb like wug does not have any meaning, semantic information cannot contribute to the generation of the past tense form of wug, which has to be derived “by analogy” from the phonology of known verbs. In contrast, semantic information is crucial for irregularly inflected verbs; the generation of their past tense form depends on the establishment of a link between their semantic representation and their representations in both input and output phonology. In this way, the past tense of irregularly inflected verbs will not conform to the statistically dominant pattern. To simulate the neuropsychological data, Joanisse and Seidenberg (1999) selectively damaged different components of their connectionist model. Damage to the semantic units impaired the generation of regular, irregular, and novel past tenses, but the largest impairment appeared with irregular forms. Damage to the phonological units also affected the processing of all three types of verbs, 3 but novel verbs were impacted most severely. Joanisse and Seidenberg (1999) were therefore able to reproduce (at least part of) the dissociations reported in neuropsychology with regular, irregular, and novel past tenses. The fact that these dissociations emerged in a connectionist network that does not explicitly incorporate specific mechanisms for regular and irregular inflections calls into question the claim that the neuropsychological data are incompatible with this sort of model. Of course, the ability to replicate the neuropsychological dissociation comes with an added cost: semantics is called upon for the processing of the regular/irregular verb distinction that has traditionally been considered an idiosyncrasy of language with relevance only for language processing. By contrast, the dual-mechanism account supports the opposing view that the regular/irregular verb distinction is confined to the realm of language processing.