Inflectional morphology in a family with inherited specific language impairment

The production of regular and irregular past tense forms was investigated among the members of an English-speaking family with a hereditary disorder of language. Unlike the control subjects, the family members affected by the disorder failed to generate overregularizations (e.g., digged) or novel regular forms (plammed, crived), whereas they did produce novel irregularizations (crive– crove). They showed word frequency effects for regular past tense forms (looked) and had trouble producing regulars and irregulars (looked, dug). This pattern cannot be easily explained by deficits of articulation or of perceptual processing, by previous simulations of impairments to a singlemechanism system, or by the extended optional infinitive hypothesis. We argue that the pattern is consistent with a three-level explanation. First, we posit a grammatical deficit of rules or morphological paradigms. This may be caused by a dysfunction of a frontal/basal-ganglia “procedural memory” system previously implicated in the implicit learning and use of motor and cognitive skills. Second, in contexts requiring inflection in the normal adult grammar, the affected subjects appear to retrieve word forms as a function of their accessibility and conceptual appropriateness (“conceptual selection”). Their acquisition and use of these word forms may rely on a “declarative memory” system previously implicated in the explicit learning and use of facts and events. Third, a compensatory strategy may be at work. Some family members may have explicitly learned a strategy of adding suffix-like endings to forms retrieved by conceptual selection. The morphological errors of young normal children appear to be similar to those of the affected family members, who may have been left stranded with conceptual selection by a specific developmental arrest. The same underlying deficit may also explain the impaired subjects’ difficulties with derivational morphology. Specific language impairment (SLI) may be defined as a developmental disorder of language that cannot be explained by a hearing loss, a general cognitive impairment, an emotional disorder, or environmental deprivation (Bishop, 1992). Although SLI is a heterogeneous disorder (Leonard, 1998), there have  1999 Cambridge University Press 0142-7164/99 $9.50 Applied Psycholinguistics 20:1 52 Ullman & Gopnik: Inflectional morphology in SLI been reports of subgroups of people with SLI whose language impairments are relatively homogeneous (e.g., Adams & Bishop, 1989). Of particular interest are reports of SLI subgroups whose language impairments may be hereditary and, it has been argued, specific to grammar (Clahsen, 1989; Gopnik, 1990b; Gopnik & Crago, 1991; van der Lely, 1996b). The investigation of such SLI subgroups may elucidate four important questions about the psychological, neural, and developmental underpinnings of language. (1) Are the mental lexicon, in which words are stored, and the mental grammar, which specifies how words combine into larger words, phrases, and sentences, subserved by distinct mechanisms (Pinker, 1991) or a common mechanism (Elman, Bates, Johnson, Karmiloff-Smith, Parisi, & Plunkett, 1996)? (2) If these two language capacities are subserved by distinct mechanisms, is the grammar-computing mechanism dedicated to grammar or some aspect of grammar or does it underlie nonlanguage functions as well (see Ullman, Corkin, Coppola, Hickok, Growdon, Koroshetz, & Pinker 1997)? (3) Can the neural systems subserving the lexicon and the grammar be localized and, if so, to where? (4) If the language of grammatically impaired people with SLI resembles that of young normal children, can investigations of SLI language elucidate the structure of child language and the process of normal language acquisition? In addition, the study of SLI subgroups’ language impairments may lead to a better understanding of the nature of SLI itself, perhaps leading to diagnostic and therapeutic advances for the condition. In this article, we address these four questions by investigating the production of English past tense and other inflectional forms in an SLI subgroup of native English-speaking people whose hereditary language impairment, it has been hypothesized, may be specific to grammar (Gopnik, 1990a, 1990b, 1994d; Gopnik & Crago, 1991). We focus on inflectional morphology for three reasons. First, inflectional morphological impairments have been shown to occur in several SLI subgroups of native English speakers (e.g., Leonard, Bortolini, Caselli, McGregor, & Sabbadini, 1992; Rice, Wexler, & Cleave, 1995; van der Lely & Ullman, 1996), including the subgroup discussed in this article. It has also been reported in SLI native speakers of German (Clahsen, 1989), Italian (Leonard, Bortolini et al., 1992), Hebrew (Dromi, Leonard, & Shtieman, 1993), Japanese (Fukuda & Fukuda, 1994; Fukuda & Gopnik, 1994), Greek (Dalalakis, 1994), and Inuktitut (Crago & Allen, 1994). Thus, our findings might potentially be generalized to many subgroups of SLI across a number of languages. Second, inflectional morphology has been extensively studied from psycholinguistic (Kim, Pinker, Prince, & Prasada, 1991; Marcus, Brinkmann, Clahsen, Wiese, & Pinker, 1995; Prasada & Pinker, 1993; Ullman, 1993, 1999), developmental (Marcus et al., 1992), neurological (Ullman, in press; Ullman, Corkin et al., 1997; Ullman, Izvorski, Love, Yee, Swinney, & Hickok, in press), neuroimaging (Jaeger et al., 1996; Ullman, Bergida, & O’Craven, 1997), and electrophysiological (Newman, Neville, & Ullman, 1998; Penke et al., 1997; Weyerts, Penke, Dohrn, Clahsen, & Münte, 1996) perspectives (see Pinker, 1991). This extensive and interdisciplinary research on a relatively simple language system provides a foundation on which to build our understanding of the linguistic impairment of SLI. Applied Psycholinguistics 20:1 53 Ullman & Gopnik: Inflectional morphology in SLI Third, and most important, inflectional morphological transformations such as English past tense are particularly well-suited for testing whether the mental lexicon and the mental grammar may be subserved by distinct mechanisms (Pinker, 1994) or by a single mechanism (Elman et al., 1996). Regularly inflected past tense forms (e.g., look–looked, play–played) are structured according to a set of rules (append an -ed suffix to the stem), whereas irregularly inflected forms do not all follow the same stem-past pattern (e.g., sing–sang, fling–flung, bring–brought). According to a dual mechanism view, irregular past tense forms are retrieved from an associative memory, whereas regular past tense forms are computed in real time by a distinct rule-processing system (Pinker, 1991, 1994). When an irregular form is not successfully retreived, the rule-processing system may take over, resulting in “overregularization” errors (flinged, digged). In contrast, according to a single mechanism view, regular as well as irregular forms are learned in, and computed over, an associative memory. On this perspective, there is no distinct system for rule processing; rather, rules are nothing but descriptions of the regularities in the language (Elman et al., 1996; Plunkett & Marchman, 1993; Rumelhart & McClelland, 1986). Dual mechanisms versus a single mechanism: Previous evidence Several lines of evidence have been presented as support for the dual mechanism theory. First, psycholinguistic studies of adults have suggested that irregular past tense forms are retrieved from associative memory, whereas regular past tense forms are computed in real time by a rule-processing system. If irregular forms are retrieved from associative memory, then those that are encountered frequently or that share memory traces with many other similar-sounding neighboring forms (e.g., sing–sang, ring–rang, spring–sprang) should have stronger memory traces and therefore should be better remembered than those of lower frequency or with few neighbors. Prasada, Pinker, and Snyder (1990) reported that subjects took significantly more time to produce low-frequency than highfrequency past tense forms for irregular verbs, holding stem frequency constant. In contrast, time to production was not significantly longer for low-frequency than high-frequency regular past tense forms, holding stem frequency constant. Ullman (1993, 1999) found that acceptability ratings of irregular past tense forms (blew) correlated with their word frequencies and with a measure of neighborhood size, which was based on the type and token frequencies of similar-sounding verbs (grew, threw). This pattern held even when a measure of stem access was held constant. In contrast, acceptability ratings of regulars (walked) correlated neither with their word frequencies nor with the neighborhood size measure (stalked, balked), holding a measure of access constant. Although single system models may attribute the lack of word frequency effects to the high type frequency of similar-sounding regulars, which may overwhelm the memory traces of individual words (Daugherty & Seidenberg, 1992; Seidenberg & Daugherty, 1992), it is not clear whether single system models could explain the contrasting neighborhood effects (Ullman, 1999). These results are thus taken to suggest that irregulars are retrieved from an associative memory storing distributed representations of their phonological forms, whereas regulars (looked) are computed in real time by a distinct system. Applied Psycholinguistics 20:1 54 Ullman & Gopnik: Inflectional morphology in SLI Second, developmental studies have revealed similar contrasts in children. Two groups of children (approximate mean ages of 7 and 8) showed word frequency effects in the elicited production of irregular but not regular past tense forms (van der Lely & Ullman, submitted). Overregularization rates (blowed) in the spontaneous speech of younger children correlated negatively with past tense frequency (blew) and with the number of similar-sounding irregular verbs (threw, grew) but did not correlate with the number of similar-sounding regular verbs (flowed, rowed) (Marcus et al., 1992). The production of regular and irregular past tense forms has also been examined in people with Williams syndrome, a hereditary developmental disorder associated with severe mental retardation. People with this disorder may have spared syntactic abilities but abnormal lexical retrieval (Bellugi, Wang, & Jernigan, 1994). Young adults with Williams syndrome were shown to be more impaired at producing irregular than regular past tense (blew vs. walked) and plural (mice vs. rats) forms. The majority of their errors on irregulars were overregularizations (blowed, mouses) (Bromberg et al., 1994). These results dissociate irregular and regular inflected forms, and they link irregulars to lexical memory and regulars to syntactic abilities. Third, neurological studies of adults with acquired brain damage have revealed double dissociations between the use of regular and irregular forms. Patients who have temporal lobe lesions and relatively spared frontal/basal-ganglia structures and who demonstrate impairments remembering words or facts (those with posterior aphasia or Alzheimer’s disease) have more trouble producing irregular than regular English past tense forms (Ullman, in press; Ullman et al., 1993, 1994, in press; Ullman, Corkin et al., 1997) and Italian present tense and past participle forms (Cappa & Ullman, 1998). In contrast, patients who have frontal/basal-ganglia damage and relatively spared temporal lobe structures and who may have impairments of syntactic processing and/or learning and use of motor and cognitive skills (those with anterior aphasia or Parkinson’s disease) show the opposite pattern (Ullman, in press; Ullman et al., 1993, 1994, in press; Ullman, Corkin et al., 1997). These double dissociations have been taken to suggest that lexical memory is part of a “declarative memory” system rooted in the temporal lobes and previously implicated in the memory for facts and events, whereas grammatical rules are processed at least in part by a “procedural memory” system rooted in frontal/basal-ganglia structures and previously implicated in the implicit learning and use of motor and cognitive skills (Ullman, in press; Ullman, Corkin et al., 1997). The existence of distinct neural underpinnings for regular and irregular past tense processing is further strengthened by double dissociations in priming. Marslen-Wilson and Tyler (1997) reported that one aphasic subject showed priming between past tense and stem forms for regulars (e.g., jumped primed jump) but not irregulars (e.g., gave did not prime give), whereas two aphasic subjects showed the opposite pattern. Ullman and his colleagues found that basal ganglia degeneration leading to the suppression of movement (hypokinesia in patients with Parkinson’s or Huntington’s disease) also leads to rule suppression (resulting in the omission of the past tense -ed suffix), whereas basal ganglia degeneration leading to excess Applied Psycholinguistics 20:1 55 Ullman & Gopnik: Inflectional morphology in SLI movement (hyperkinesia in patients with Huntington’s disease) also leads to excess rule use (the production of forms like dugged and walkeded) (Ullman, in press; Ullman, Corkin et al., 1997). This contrast suggests that basal ganglia circuitry contributes to grammatical rule processing, and that the well-studied basal ganglia circuits underlying motor programming may play a comparable role in rule programming. Fourth, neuroimaging investigations have revealed intriguing dissociations between the production of regular and irregular past tense forms. In a positron emission tomography (PET) study, Jaeger et al. (1996) asked healthy men to read out loud lists of irregular, regular, and novel verb stems and to produce their past tense forms. When past tense production was compared to verb stem reading, left temporal and temporo-parietal regions yielded greater statistical significance for irregular than regular or novel verbs, whereas a left prefrontal region was associated with greater statistical significance for regular and novel verbs. Unfortunately, this contrast is problematic for several respects. First, it was not found when past tense production conditions were compared to a rest condition. Second, activation differences found from a comparison of two conditions could be caused by an increase in one condition or a decrease in the other; in the absence of an examination of activation decreases, these cannot be distinguished. Third, the blocking of large numbers of items, a design required by the PET technology, might allow subjects to use a strategy to produce the regulars, all of which undergo -ed suffixation, but not the irregulars, each of which requires a particular stem-past transformation. In a functional magnetic resonance imaging (fMRI) study, five healthy righthanded men were asked to produce the past tense forms of regular and irregular verbs (Ullman, Bergida et al., 1997). The subjects showed similar patterns of activation. The production of irregulars but not regulars yielded a substantial activation decrease, compared to fixation, in temporal/temporo-parietal regions. In prefrontal cortex, regulars but not irregulars showed an activation decrease, compared to fixation. Although the specific causes of these activation decreases remain to be investigated, the double dissociations suggest that irregulars and regulars have distinct neural underpinnings linked to temporal and frontal regions. Like the PET study, in this fMRI study, regular and irregular stimuli were grouped separately, although in blocks of only 10 verbs. Thus, the findings must be treated with caution. Fifth, in two electrophysiological studies of healthy German subjects (Penke et al., 1997; Weyerts et al., 1996), distinct patterns of event-related potentials (ERPs) were found for regular and irregular inflection of verbs and nouns. An ERP study of healthy adults found that violations of regular but not irregular English past tense forms yielded a left frontal negativity, whereas violations of irregulars but not regulars elicited a more posterior negativity (Newman, Neville, & Ullman, 1998). Although these lines of evidence have been taken as strong support for a dual system view, the controversy between the two theories has continued. In particular, single system proponents have argued that many findings taken to support the dual system perspective are also compatible with single system models (Cottrell & Plunkett, 1991; Daugherty & Seidenberg, 1992; Hare, Elman, & DaughApplied Psycholinguistics 20:1 56 Ullman & Gopnik: Inflectional morphology in SLI erty, 1995; Hoeffner, 1992; Hoeffner & McClelland, 1993; MacWhinney & Leinbach, 1991; Marchman, 1993; Plunkett & Marchman, 1991, 1993; Seidenberg & Hoeffner, in press). It is therefore important to gather additional evidence to help resolve the controversy. SLI and the dual versus single mechanism controversy Under a dual system view, grammar or a component of grammar may be impaired in some SLI subgroups, while lexical memory is left relatively intact. Indeed, a number of different hypotheses have posited an SLI deficit specific to grammar, although these hypotheses differ with respect to the linguistic character of the impairment: Gopnik and colleagues’ feature blindness and feature checking hypotheses (Gopnik, 1990a, 1990b; Gopnik & Crago, 1991); Clahsen’s (1989) missing agreement account; Rice, Wexler, and Cleave’s (1995) extended optional infinitive hypothesis; and van der Lely’s representational deficit for syntactic dependent relationships hypothesis (van der Lely 1996a, 1996b; van der Lely & Stollwerck, 1997). We propose that the learning and use of grammatical rules implicitly learned by normal individuals are dysfunctional in the particular SLI subgroup under study and perhaps in other SLI subgroups as well. Lexical memory, in contrast, is posited to be relatively spared. The subjects affected with the impairment are predicted to have trouble with rule-governed -ed suffixation. They may therefore show a lack of overregularizations (dig–digged) and regularizations of novel verbs (plammed, crived) and may have trouble generating past tense forms of very low-frequency regular verbs. In the absence of intact suffixation rules, they may be forced to memorize regular as well as irregular past tense forms. Unlike normal children (van der Lely & Ullman, submitted) and adults (Prasada et al., 1990; Ullman, 1993, 1999), they would be expected to show frequency effects for regular as well as irregular past tense forms. If regular and irregular forms are stored, similar production rates may be found for the two past tense types. Moreover, if syntactic processing is normally dependent on grammatical rules that are dysfunctional in this population, then the production of all types of past tense forms (irregular as well as regular) for real and novel verbs may be impaired in past tense processing tasks involving syntax. In summary, if grammatical rules are dysfunctional, suffixed forms which are unlikely to have been memorized (novel past tenses and overregularizations) may be difficult to produce; frequency effects should be found for regular and irregular past tense forms; and the production of all past tense forms may be impaired in certain past tense processing tasks because of syntactic deficits. Under a single system view, grammar cannot be selectively impaired in SLI, although different patterns of impairment may be associated with the computation of regular and irregular forms. Marchman (1993) investigated the effects of “lesions” (i.e., the random elimination of hidden units) in a connectionist model of the acquisition and computation of English past tense. Greater damage prior to or during training led to greater impairments in the learning and computation of regular past tense forms, whereas the learning and computation of irregular past tense forms were relatively impervious to damage. Greater damage prior to Applied Psycholinguistics 20:1 57 Ullman & Gopnik: Inflectional morphology in SLI training led to greater deficits in the production of suffixed past tense forms of novel verbs whose stems are dissimilar to the stems of real irregulars (“novel regulars” such as plam–plammed) – although even when the maximum percentage of hidden units (44%) was lesioned, the network correctly suffixed about 40% of these novel verb stems. In contrast, damage did not affect the production of novel “no-change” past tense forms of verbs whose stems are similar to the stems of real no-change irregulars (“novel irregulars” such as scrit–scrit; cf. hit–hit). The result of lesions on novel verbs whose stems resemble irregular vowel-change verbs (e.g., crive–crove; cf. drive–drove) was not discussed, presumably because the intact network produced vowel-change irregularizations (e.g., crove) at a very low rate. Hoeffner and McClelland (1993) attempted to simulate a perceptual processing deficit, which has been hypothesized to underlie SLI and to result in difficulties learning items with low phonological salience, such as the -ed suffix (Leonard, Bortolini et al., 1992; Tallal, Stark, & Mellits, 1985). They presented a three-layer network with “normal” regular and irregular items and with items whose phonological input to the model was weakened. The “impaired” network was only slightly less accurate than the normal network at producing irregular past tense forms, but it was significantly worse at producing regular past tense forms. In summary, if these connectionist simulations are taken as models of SLI (see Marchman & Weismer, 1994), they predict that people with SLI should have more difficulty producing regular than irregular past tense forms. Marchman’s (1993) simulations also predicted impaired performance at the production of novel regularizations (plammed). Neither study made predictions about frequency effects.