Introduction For the past couple of decades, the study of verb has become a proving ground for two competing accounts of verb representation. Differences between these two theories are revealed in the manner in which they explain the processing of regular versus irregular verbs. In English, there are two broad classes of verbs distinguished by their inflectional morphology. Regular verbs, constituting the majority of verbs, form their past tense by adding the regular affixes /d/, /ed/, or /t/, depending on the stem. For example, “fix / fixed “, “rain / rained “, and “help / helped .” Therefore, a regular past tense verb is composed of its stem and the affix, such as “open-” and “-ed” in “opened.” Such verbs are phonologically predictable since they all form their past tense in a standard way.
On the other hand, the past tense of irregular verbs is inconsistent, and cannot be predicted by the stem. For example “hit / hit “, “run / ran “, “drive / drove .” These irregulars only number approximately 160 but are among the most commonly encountered. Whereas the formation of regular past tense verbs seems to be governed by rules, it is not the case for irregular verbs. We will review the current dual-mechanism account of past tense inflection, followed by recent neuropsychological evidence, as well as some implications for the connectionist model. The rule-based account The dual-route, or rule-based approach, postulates two separate mechanisms for the past tense inflection of regular and irregular verbs (Marslen-Wilson & Tyler, 1998; Tyler, Randall, & Marslen-Wilson, 2002; Tyler et al.
, 2002). Starting in the 1960’s, research on language development established the early premises for a computational rule-based account the past tense. Studies of language acquisition in children show the past tense as learned in three stages (Be rko, 1958; Ervin & Miller, 1963; Kuczaj, 1977). In Stage 1, it appears that the past tense forms of verbs are learned as items separate from their present tense forms. In essence, the child memorizes a list of distinct items. Stage 2 is characterized by implicit knowledge of a rule for the past tense.
Evidence comes from the observation that children over generalize the rule that applies to regulars, to irregular verbs as well, adding the affix /ed/ to irregular verbs, such as “goed” and “b ringed.” It is in Stage 3 that the child correctly uses regular and irregular past tense. Emphasis has been put on Stage 2, where overgeneralization of the rule to irregular verbs is taken as evidence for a cognitive process in verb representation. The argument is that the misapplication of the rule on irregulars could not have been learned though imitation, since the child is not exposed to such errors; therefore what was learned must have been the rule. Tyler, Randall, and Marslen-Wilson (2002) propose that regulars are processed by phonological parsing, involving the identification of the verb’s stem and affix. This mechanism is distinct from the one that processes irregulars, which are mono morphemic. According to the dual-mechanism hypothesis, the past tense of irregulars are represented individually, whereas regulars are stored as the stem, and the past tense of regulars are accessed by adding the affix to the stem.
Neuropsychological evidence Studies with patients have shown that deficits in the processing of regular and irregular verbs can be dissociated (Tyler et al. , 2002). It is observed that patients with damage to the left inferior frontal cortex have trouble with regulars, and that those with damage to the left inferior temporal cortex have problems with irregulars. As a result, it is inferred that the processing of regulars involves the inferior frontal cortex, especially around Broca’s area, and that irregulars are processed via the inferior temporal cortex. This double dissociation is taken as prima facie evidence for the dual-mechanism hypothesis, and supports the claim that both mechanisms have separate underlying neural pathways. Research into the dissociation of deficits in regular and irregular past tense is derived from data on patients varying in their localization of brain damage.
Damage to Broca’s area and surrounding area, as a consequence of stroke or Parkinson’s disease, result in aphasia and deficits with regulars. Patients with damage to the inferior temporal cortex, as in semantic dementia, Herpes Simplex Encephalitis (HSE), or Alzheimer’s disease, often have difficulty with irregulars. Analyses of past tense deficits have largely used priming as a method of evaluating performance with regulars and irregulars (Marslen-Wilson & Tyler, 1998; Patterson, Lamb on Ralph, Hodges, & McClelland, 2001; Tyler et al. , 2002; Tyler, Randall, & Marslen-Wilson, 2002).
A key study by Tyler et al. (2002) demonstrated dissociations in past tense processing using non-fluent aphasic and semantically impaired HSE patients. Aphasics were probed for regular and irregular deficits by priming in a lexical decision task; for HSE patients, a primed elicitation task was used. The results showed that the aphasic patients displayed significant priming on irregulars and semantically related words, but not on regulars. On the other hand, HSE patients demonstrated significantly better performance on regulars than irregulars.
Earlier imaging of both patient groups’ brains indicated inferior frontal damage in aphasics, and inferior temporal damage in HSE patients. Connectionism Taken together, these data make the case for a dissociation of regular and irregular processing, relating different neural damage to distinct deficits, although they do not provide a clear explanation as to which linguistic mechanism is involved in those deficits. Proponents of the dual-route hypothesis argue that the dissociation is evidence that regulars are processed by a rule-based mechanism, and that irregulars are stored in whole form, with their past tense represented separately. However, the dissociation can also be accounted for by a connectionist model of verb representation. In a pioneering study, Rumelhart and McClelland (1986) demonstrated how the over regularization of regular past tense inflection to irregulars, observed in children, could be simulated using a parallel distributed processing (PDP) model. This alternate account of past tense processing postulates a single mechanism underlying both regular and irregular verbs.
More recent elaborations on the PDP model presented by Joanisse and Seidenberg (1999) have shown that the dissociation of regular and irregular verb deficits can be simulated using an artificial neural network. The simulation involved connections between speech input units, speech output units, and semantic units. After the training trials of the network, it was found that lesion ing connections in the semantic layer produced larger deficits in irregulars than regulars. This simulation of semantic impairment is consistent with the observation that patients with deficits in the processing of irregulars also have problems with semantics (Patterson et al.
, 2001). In an attempt to account for the opposite case, where impairment is greater on regulars than on irregulars, severe “lesions” in the phonological (speech) layer were introduced. This had the effect of impairing performance on regulars, irregulars, and non-words, but deficits were greater for regulars and non-words than for irregulars. Therefore, it would appear that the dissociation of deficits on regulars and irregulars can be accounted for by both the dual-route and connectionist model of verb representation. The single-route connectionist model explains problems with regulars as a result of severe phonological impairment, and difficulties with irregulars as a consequence of semantic impairments; all within a single system. Discussion Any progress on the debate between proponents of the dual-route system and connectionist requires further neuropsychological investigations.
Since both accounts of the dissociation between regular and irregular past tense are consistent with current neuro pathological studies, more recent research has focused on specific aspects of each account. More specifically, the connectionist model of Joanisse and Seidenberg (1999) shows that phonological impairments and semantic impairments result in deficits on regulars and irregulars, respectively. A study by Marslen-Wilson and Tyler (1998) comparing semantic priming and morphological priming suggests that the former is not involved in priming regular or irregular past tense verbs. The experiment was based on the duration of different types of priming in normal subjects.
Since morphological priming lasts longer than semantic priming, it would be expected that priming over relatively long periods be morphological. This is what the results revealed. In addition, event-related potentials (ERP) of normal adults during priming tasks show that the patterns of activation for priming with regular and irregular pairs are both more similar to each other compared to priming with semantically related pairs. It appears that it is deficits in morphological processing, and not semantic impairments, that are involved in the lack of priming for irregular verbs. On the other hand, the issue concerning phonological impairments underlying deficits with regulars is more complex.
Joanisse and Seidenberg (1999) suggest that greater deficits on regulars are due to severe phonological impairment. However, it is expected that such phonological impairment in general would result in deficits to most aspects of a patient’s linguistic abilities. ss good point To address this issue, Tyler, Randall, and Marslen-Wilson (2002) analyzed aphasic patients’ phonological impairments and have found that phonological properties of irregulars and regulars, or difficulties in acoustic-phonetic judgement, could not account for the dissociation. Instead, they propose that the inflection of regular past tense involves processes of phonological parsing, and that it is the disruption of those mechanisms that result in regular deficits. Therefore, it seems that the phonological impairment hypothesis of Joanisse and Seidenberg has some merit. However, severe general phonological impairment is too broad an explanation, and it appears that a deficit in phonological parsing is a clearer and more precise alternative explanation of aphasics’ problems with regulars.
Currently, Tyler, Randall, and Marslen-Wilson’s model, which proposes that phonological parsing is involved in the processing of regular past tense inflection, and that irregular past tense verbs are stored separately in whole form, seems to be the most plausible of the two. Conclusion So far, debates between dual-route and connectionist views of past tense inflection stem from the fact that dissociations of regular and irregular deficits can be accounted for by both models. Although current neuropsychological research slightly favours the dual-route hypothesis, there is significant similarities in explanatory potential of both accounts. Perhaps subsequent studies will address certain difficulties with the connectionist model.
It is possible that both models accurately describe verb representation, differing only in their level of analysis. Connectionism could be considered a simplistic low-level modeling of neurons, and the dual-route system may represent a higher-level explanation of linguistic function. Additional research needs to done on specific aspects of both accounts, emphasizing differences between them, or perhaps even attempting to reconcile them.