The debate over the characterization of specific language impairment (SLI) is fundamental to theoretical linguistics and, more broadly, to the whole of cognitive science. It is built directly out of the pervasive question regarding the extent to which language ability is best considered as a domain-specific set of skills or as the outcome of various domain-general processes. Therefore, an examination of this issue in conjunction with bilingual language acquisition, a situation that naturally entangles both linguistic and cognitive systems, is a powerful forum for exploring these basic theoretical questions. Paradis' Keynote Article is a substantial contribution to this enterprise: it provides a thorough review of the literature on bilingualism and SLI, and in so doing, evaluates the evidence in terms of its consistency with the maturational model that follows from the tradition of domain-specific language acquisition and the limited processing capacity (LPC) theory, a more domain-general approach. Her extensive review of the literature shows that both second language (L2) learning and bilingualism produce language proficiency profiles that are not identical to those found in SLI, and therefore support neither approach. In our view, the problem is in the attempt to dichotomize language ability as being controlled by either domain-specific or domain-general factors. A more inclusive approach to language ability, especially regarding bilingualism and L2 learning, would set out different criteria for evaluating language development other than the strictly linguistic features used in Paradis' analyses. Such an analysis may lead to a clearer identification of how these experiences uniquely affect language outcomes.
The dichotomy between domain-specific and domain-general interpretations of SLI deficits is a foundation of Paradis' Keynote, yet such distinctions play a diminishing role in most of cognitive psychology. The focus instead is on identifying which aspects of skill or expertise mature and which aspects are learned. With this perspective, it becomes possible to talk about maturation and environmentally influenced aspects of both domain-general and domain-specific processing. If a theory can accommodate both mechanisms of predisposition and acquired processes, then we can begin the discussion of the relative contributions of each within a model of language processing. Paradis' Keynote Article offers implicit support for the notion that both types of mechanisms contribute to language development yet her discussion implies that only one can be correct; the problem is in understanding the relation between them in an integrated system.
One way of considering the difference between domain-specific and domain-general aspects of performance is in terms of a distinction between two components of processing that correspond roughly to such traditional distinctions as crystallized and fluid intelligence. Craik and Bialystok (Reference Craik and Bialystok2006) described a framework in which processes associated with representation (cf. crystallized intelligence) and control (cf. fluid intelligence) develop and decline across the life span. Performance in all cognitive domains depends on associated levels of each, yet each has a different role in cognitive performance and a different lifespan trajectory of development and decline. Put simply, representations build up through experience in a specific domain (such as language), remain largely (but not wholly) specific to that domain, and are maintained across the life span. Older adults, for example, not only have linguistic knowledge that is as good as that of young people but often have better vocabularies, reflecting their greater experience (Schaie, Reference Schaie1996). In contrast, control is the set of resource-limited processes that are needed to coordinate, plan, monitor, and execute all complex cognitive activities. These control processes, or executive function, are built up with both maturation and experience, applied broadly across a range of problem domains, and decline with healthy cognitive aging (Baltes, Staudinger, & Lindenberger, Reference Baltes, Staudinger and Lindenberger1999; Daniels, Toth, & Jacoby, Reference Daniels, Toth, Jacoby, Bialystok and Craik2006). Thus, older adults are less able than their younger counterparts to multitask, shift attention, plan complex routines, and coordinate plans. The interaction of these systems is essentially what we mean by “expertise”: rich knowledge and skilled execution.
Language ability fits well into this paradigm. Language proficiency requires both rich knowledge of an abstract system (representation) and skilled procedures (control) to produce complex utterances, coordinating linguistic, cognitive, and social parameters. Deficits in either system lead to atypical speech production. L2 learners generally have poor representations of the target language because it is incompletely learned, whereas bilinguals may have poorer vocabulary (Oller, Pearson, & Cobo-Lewis, Reference Oller, Pearson and Cobo-Lewis2007) but enhanced control processes (Bialystok, Reference Bialystok2001). Thus, in typical populations we would expect marked speech patterns from L2 learners but no visible deficit from bilinguals (except for detailed vocabulary knowledge) and possibly greater facility by bilinguals on verbal tasks that specifically require high levels of control. But there is little consensus on the identification of the central deficit in SLI. According to the maturational view, the deficit is representational in that the usual acquired knowledge of linguistic structures is absent; according to the LPC view, the deficit is in control in that the processes responsible for coordinating this complex performance are impaired. Differences in speech between children with SLI and both L2 learners and bilinguals suggest that both types of deficits are involved, undermining the “specific” aspect of SLI. Moreover, the divergence between SLI speech and each of the comparison groups is different, largely because the groups deviate on a different underlying component of language ability.
The analysis of language performance in SLI children undertaken to evaluate the nature of the underlying deficit typically considers one aspect at the expense of the other. For example, Paradis points out that Leonard and colleagues (Reference Leonard, Weismer, Miller, Francis, Tomblin and Kail2007) found that verbal working memory was a strong predictor of language performance among 14-year-olds with language impairment. In this study, five measures of verbal working memory and one measure of nonverbal working memory were used. Insofar as any working memory task uses verbal stimuli to be manipulated, that task has a verbal confound. Children who have difficulty with verbal processing would predictably do poorly on such tasks. Therefore, the task confounds assessment of representation and control, so their influence could not be examined independently. Moreover, through the use of principal components analysis, verbal and nonverbal working memory measures were conflated into a single category, precluding the discovery of an independent contribution of nonverbal processing in the model. For this reason, language-specific deficits overwhelm the possibility of identifying domain-general deficits in control.
Other studies have also investigated verbal working memory (e.g., Marton & Schwartz, Reference Marton and Schwartz2003; Montgomery & Windsor, Reference Montgomery and Windsor2007), but fewer have investigated nonverbal working memory (e.g., Alloway & Archibald, Reference Alloway and Archibald2008; Bavin, Wilson, Maruff, & Sleeman, Reference Bavin, Wilson, Maruff and Sleeman2005; Hoffman & Gillam, Reference Hoffman and Gillam2004; Marton, Reference Marton2008) among children with SLI. The results of these studies are inconclusive, ranging from no deficits (Alloway & Archibald, Reference Alloway and Archibald2008) to deficits (Hoffman & Gillam, Reference Hoffman and Gillam2004; Marton, Reference Marton2008) to deficits on some, but not other, nonverbal tasks (Bavin et al. Reference Bavin, Wilson, Maruff and Sleeman2005). Therefore, it is not yet possible to conclude what the domain-general deficits (if they exist) might be for SLI, and if they do exist, what their impact on language might be. What is lacking, therefore, is a study that examines SLI from an integrated approach that assumes domain-general (control) and domain-specific (representation) processing as separate contributors. Establishing the contributions of nonlinguistic processing is a fundamental first step in assessing the domain-general contributions to SLI, and would form the basis for evaluating the contribution of bilingualism to SLI. Assessments of finite-verb morphology deficits in SLI establish the role of impairment in the representational base for language (cf. comparison with L2 learners) but provide no information on the possibility that domain-general control processes are also affected (cf. comparison with bilinguals). Without independent evidence for this component, assessments of the relation between SLI and bilingualism are inconclusive.
Just as comparisons of specific linguistic features, such as finite-verb morphology, allow an evaluation of the development of the specific representational base in L2 learners and children with SLI, a comparison of executive control processes, notably nonverbal working memory, would motivate an evaluation of the development of control in bilingual children and children with SLI. Although there are preliminary data showing a cognitive advantage in working memory conferred by bilingualism, the size of this effect is still unknown (Feng, Bialystok, & Diamond, Reference Feng, Bialystok and Diamond2007). Considering executive control more broadly, however, there is good evidence for bilingual advantages (for a review, see Bialystok, Reference Bialystok, Kroll and de Groot2005), which may indeed extend to bilingual children with SLI. Two predictions follow regarding the possible effects of general cognition on the language abilities of children with SLI. First, the effect could be additive, and where monolingual children with SLI show a pattern of morphosyntactic deficit, bilingual children with SLI show this deficit to a lesser degree. Second, if the effect is compensatory, it could be that bilingual children with SLI show no additional disadvantage compared to monolingual children with SLI. The compensation model suggests that although there is added stress on the representational system for a child with SLI because of two language systems and lexicons, the control advantages compensate for some of these disadvantages.
These predictions depend on the relationship between the domain-specific representations and the domain-general control processes. That is, different aspects of language may or may not be affected by working memory or other types of control processing. Paradis frames her discussion of finite-verb morphology through the domain-specific perspective, but the cognitive explanations from the domain-general perspective are not focused on a specific linguistic feature. For example, Leonard and colleagues (Reference Leonard, Weismer, Miller, Francis, Tomblin and Kail2007) used a composite language score in their model, thereby precluding the possibility of comparable data for finite-verb morphology, or any other specific aspect of language processing. For this perspective to be a valid counterargument to the maturational view, it must have a way of accounting for more specific linguistic features (e.g., finite-verb morphology) through domain-general processing. To truly test the domain-specific model, an integrated approach that is targeted around linguistic forms known to have varying working memory demands would be required. It may be that linguistic categories, such as finite-verb morphology, do not, in fact, have the psychological reality that is assumed by the domain-specific approach. Alternately, it may be that finite-verb morphology is unaffected by domain-general processing, but another aspect of linguistic processing is indeed affected.
What we know from Paradis' Keynote Article is that a linguistic analysis of specific features shows that there are both commonalities and differences among TD children, children with SLI, L2 learners, and bilinguals. What we do not know is how to characterize the similarities and differences in a way that allows us to move beyond the data and explain the fundamental processes of language acquisition and use for all children. To do that, a more inclusive model that considers representation and control elements that reflect knowledge, experience, and cognitive ability in a coherent and integrated framework will be necessary.
You have
Access
The debate over the characterization of specific language impairment (SLI) is fundamental to theoretical linguistics and, more broadly, to the whole of cognitive science. It is built directly out of the pervasive question regarding the extent to which language ability is best considered as a domain-specific set of skills or as the outcome of various domain-general processes. Therefore, an examination of this issue in conjunction with bilingual language acquisition, a situation that naturally entangles both linguistic and cognitive systems, is a powerful forum for exploring these basic theoretical questions. Paradis' Keynote Article is a substantial contribution to this enterprise: it provides a thorough review of the literature on bilingualism and SLI, and in so doing, evaluates the evidence in terms of its consistency with the maturational model that follows from the tradition of domain-specific language acquisition and the limited processing capacity (LPC) theory, a more domain-general approach. Her extensive review of the literature shows that both second language (L2) learning and bilingualism produce language proficiency profiles that are not identical to those found in SLI, and therefore support neither approach. In our view, the problem is in the attempt to dichotomize language ability as being controlled by either domain-specific or domain-general factors. A more inclusive approach to language ability, especially regarding bilingualism and L2 learning, would set out different criteria for evaluating language development other than the strictly linguistic features used in Paradis' analyses. Such an analysis may lead to a clearer identification of how these experiences uniquely affect language outcomes.
The dichotomy between domain-specific and domain-general interpretations of SLI deficits is a foundation of Paradis' Keynote, yet such distinctions play a diminishing role in most of cognitive psychology. The focus instead is on identifying which aspects of skill or expertise mature and which aspects are learned. With this perspective, it becomes possible to talk about maturation and environmentally influenced aspects of both domain-general and domain-specific processing. If a theory can accommodate both mechanisms of predisposition and acquired processes, then we can begin the discussion of the relative contributions of each within a model of language processing. Paradis' Keynote Article offers implicit support for the notion that both types of mechanisms contribute to language development yet her discussion implies that only one can be correct; the problem is in understanding the relation between them in an integrated system.
One way of considering the difference between domain-specific and domain-general aspects of performance is in terms of a distinction between two components of processing that correspond roughly to such traditional distinctions as crystallized and fluid intelligence. Craik and Bialystok (Reference Craik and Bialystok2006) described a framework in which processes associated with representation (cf. crystallized intelligence) and control (cf. fluid intelligence) develop and decline across the life span. Performance in all cognitive domains depends on associated levels of each, yet each has a different role in cognitive performance and a different lifespan trajectory of development and decline. Put simply, representations build up through experience in a specific domain (such as language), remain largely (but not wholly) specific to that domain, and are maintained across the life span. Older adults, for example, not only have linguistic knowledge that is as good as that of young people but often have better vocabularies, reflecting their greater experience (Schaie, Reference Schaie1996). In contrast, control is the set of resource-limited processes that are needed to coordinate, plan, monitor, and execute all complex cognitive activities. These control processes, or executive function, are built up with both maturation and experience, applied broadly across a range of problem domains, and decline with healthy cognitive aging (Baltes, Staudinger, & Lindenberger, Reference Baltes, Staudinger and Lindenberger1999; Daniels, Toth, & Jacoby, Reference Daniels, Toth, Jacoby, Bialystok and Craik2006). Thus, older adults are less able than their younger counterparts to multitask, shift attention, plan complex routines, and coordinate plans. The interaction of these systems is essentially what we mean by “expertise”: rich knowledge and skilled execution.
Language ability fits well into this paradigm. Language proficiency requires both rich knowledge of an abstract system (representation) and skilled procedures (control) to produce complex utterances, coordinating linguistic, cognitive, and social parameters. Deficits in either system lead to atypical speech production. L2 learners generally have poor representations of the target language because it is incompletely learned, whereas bilinguals may have poorer vocabulary (Oller, Pearson, & Cobo-Lewis, Reference Oller, Pearson and Cobo-Lewis2007) but enhanced control processes (Bialystok, Reference Bialystok2001). Thus, in typical populations we would expect marked speech patterns from L2 learners but no visible deficit from bilinguals (except for detailed vocabulary knowledge) and possibly greater facility by bilinguals on verbal tasks that specifically require high levels of control. But there is little consensus on the identification of the central deficit in SLI. According to the maturational view, the deficit is representational in that the usual acquired knowledge of linguistic structures is absent; according to the LPC view, the deficit is in control in that the processes responsible for coordinating this complex performance are impaired. Differences in speech between children with SLI and both L2 learners and bilinguals suggest that both types of deficits are involved, undermining the “specific” aspect of SLI. Moreover, the divergence between SLI speech and each of the comparison groups is different, largely because the groups deviate on a different underlying component of language ability.
The analysis of language performance in SLI children undertaken to evaluate the nature of the underlying deficit typically considers one aspect at the expense of the other. For example, Paradis points out that Leonard and colleagues (Reference Leonard, Weismer, Miller, Francis, Tomblin and Kail2007) found that verbal working memory was a strong predictor of language performance among 14-year-olds with language impairment. In this study, five measures of verbal working memory and one measure of nonverbal working memory were used. Insofar as any working memory task uses verbal stimuli to be manipulated, that task has a verbal confound. Children who have difficulty with verbal processing would predictably do poorly on such tasks. Therefore, the task confounds assessment of representation and control, so their influence could not be examined independently. Moreover, through the use of principal components analysis, verbal and nonverbal working memory measures were conflated into a single category, precluding the discovery of an independent contribution of nonverbal processing in the model. For this reason, language-specific deficits overwhelm the possibility of identifying domain-general deficits in control.
Other studies have also investigated verbal working memory (e.g., Marton & Schwartz, Reference Marton and Schwartz2003; Montgomery & Windsor, Reference Montgomery and Windsor2007), but fewer have investigated nonverbal working memory (e.g., Alloway & Archibald, Reference Alloway and Archibald2008; Bavin, Wilson, Maruff, & Sleeman, Reference Bavin, Wilson, Maruff and Sleeman2005; Hoffman & Gillam, Reference Hoffman and Gillam2004; Marton, Reference Marton2008) among children with SLI. The results of these studies are inconclusive, ranging from no deficits (Alloway & Archibald, Reference Alloway and Archibald2008) to deficits (Hoffman & Gillam, Reference Hoffman and Gillam2004; Marton, Reference Marton2008) to deficits on some, but not other, nonverbal tasks (Bavin et al. Reference Bavin, Wilson, Maruff and Sleeman2005). Therefore, it is not yet possible to conclude what the domain-general deficits (if they exist) might be for SLI, and if they do exist, what their impact on language might be. What is lacking, therefore, is a study that examines SLI from an integrated approach that assumes domain-general (control) and domain-specific (representation) processing as separate contributors. Establishing the contributions of nonlinguistic processing is a fundamental first step in assessing the domain-general contributions to SLI, and would form the basis for evaluating the contribution of bilingualism to SLI. Assessments of finite-verb morphology deficits in SLI establish the role of impairment in the representational base for language (cf. comparison with L2 learners) but provide no information on the possibility that domain-general control processes are also affected (cf. comparison with bilinguals). Without independent evidence for this component, assessments of the relation between SLI and bilingualism are inconclusive.
Just as comparisons of specific linguistic features, such as finite-verb morphology, allow an evaluation of the development of the specific representational base in L2 learners and children with SLI, a comparison of executive control processes, notably nonverbal working memory, would motivate an evaluation of the development of control in bilingual children and children with SLI. Although there are preliminary data showing a cognitive advantage in working memory conferred by bilingualism, the size of this effect is still unknown (Feng, Bialystok, & Diamond, Reference Feng, Bialystok and Diamond2007). Considering executive control more broadly, however, there is good evidence for bilingual advantages (for a review, see Bialystok, Reference Bialystok, Kroll and de Groot2005), which may indeed extend to bilingual children with SLI. Two predictions follow regarding the possible effects of general cognition on the language abilities of children with SLI. First, the effect could be additive, and where monolingual children with SLI show a pattern of morphosyntactic deficit, bilingual children with SLI show this deficit to a lesser degree. Second, if the effect is compensatory, it could be that bilingual children with SLI show no additional disadvantage compared to monolingual children with SLI. The compensation model suggests that although there is added stress on the representational system for a child with SLI because of two language systems and lexicons, the control advantages compensate for some of these disadvantages.
These predictions depend on the relationship between the domain-specific representations and the domain-general control processes. That is, different aspects of language may or may not be affected by working memory or other types of control processing. Paradis frames her discussion of finite-verb morphology through the domain-specific perspective, but the cognitive explanations from the domain-general perspective are not focused on a specific linguistic feature. For example, Leonard and colleagues (Reference Leonard, Weismer, Miller, Francis, Tomblin and Kail2007) used a composite language score in their model, thereby precluding the possibility of comparable data for finite-verb morphology, or any other specific aspect of language processing. For this perspective to be a valid counterargument to the maturational view, it must have a way of accounting for more specific linguistic features (e.g., finite-verb morphology) through domain-general processing. To truly test the domain-specific model, an integrated approach that is targeted around linguistic forms known to have varying working memory demands would be required. It may be that linguistic categories, such as finite-verb morphology, do not, in fact, have the psychological reality that is assumed by the domain-specific approach. Alternately, it may be that finite-verb morphology is unaffected by domain-general processing, but another aspect of linguistic processing is indeed affected.
What we know from Paradis' Keynote Article is that a linguistic analysis of specific features shows that there are both commonalities and differences among TD children, children with SLI, L2 learners, and bilinguals. What we do not know is how to characterize the similarities and differences in a way that allows us to move beyond the data and explain the fundamental processes of language acquisition and use for all children. To do that, a more inclusive model that considers representation and control elements that reflect knowledge, experience, and cognitive ability in a coherent and integrated framework will be necessary.
ACKNOWLEDGMENTS
Preparation of this article was supported by Grant R01HD052523 from the US National Institutes of Health (to E.B.).