Johanne Paradis' Keynote Article can be read as a concise critical review of the research that focuses on the sometimes strained relationship between bilingualism and specific language impairment (SLI). In my comments I will add some thoughts based on our own research on the learning of Dutch as a second language (L2) by children with SLI.Footnote 1
Our research focuses on successive bilingualism, and our group of bilingual subjects consists of children who have Turkish as their first language (L1) and Dutch as their L2. Turkish is a minority language in the full sense of Paradis' description: although the number of speakers of Turkish as an L1 is large in The Netherlands, there is no teaching of Turkish in primary schools. This situation could be said to lead to “subtractive bilingualism,” and the possibility of a “cumulative effect” of bilingualism and SLI that Paradis discusses is well worth considering for this particular group of subjects. In our study we were not able to include monolingual children with SLI whose native language is Turkish; the issue was therefore addressed with respect to the children's output in Dutch.
LINGUISTIC MODELS
Before discussing the nature of our results, I initially consider the feasibility of the linguistic model that Paradis highlights. Paradis' prime model for a theory that predicts similarity between monolingual and bilingual children with SLI is Rice and Wexler's extended optional infinitive (EOI) theory.
Whether this theory is fully applicable to Dutch is debatable (see the discussion with respect to monolingual SLI in de Jong, Reference De Jong, Levy and Schaeffer2003). To some extent the EOI account is indeed “a consequence of superficial properties of English” (Paradis, Reference Paradis2010 [this issue]). For instance, unlike in English, the verb stem has been proposed as an alternative optional default in Dutch child language (Paradis & Crago, Reference Paradis and Crago2001). The stem, however, is typically in second position, the position for finite verb forms in main clauses. Therefore, it is considered finite. In English, there are no distinct forms for stem and infinitive. In Dutch, the infinitive has an inflectional marker (-en). Thus, we restrict ourselves to the “true” infinitival form and discuss the presence of an EOI stage in Dutch. Infinitives are indeed found widely in early acquisition of Dutch and persist slightly longer in children with SLI (Wexler, Schaeffer, & Bol, Reference Wexler, Schaeffer and Bol2004). Empirically, however, the prediction that “this stage persists until (children with SLI are) at least nine years of age, and never fully resolves” is not supported by group findings on Dutch, although a small minority of individuals with SLI shows persistent use of root infinitives at the age of 8 or 9 years (de Jong, Reference De Jong1999; Orgassa, Reference Orgassa2009). Consequently, the ramifications for patterns of growth (maturation) in Dutch-speaking children with SLI are not clear.
However, one element of the “maturational” theory does ring true for SLI in Dutch, and that is the basic normality of patterns found in child groups. Unlike the now obsolete theory proposed by Gopnik, in which symptoms arose from “absence of grammar,” the EOI account and its affiliates stress underspecification in a nondeviant grammar. In Dutch, this is highly visible in the robust correlation between verb position and verb form, a connection that is not abandoned by language-impaired children either, whether monolingual or bilingual. For bilingual children with SLI, who received Dutch input later than monolinguals and share the intake problem with monolingual children with SLI, this finding was replicated (Orgassa, Reference Orgassa2009). Given this basic normality, the question remains which “language acquisition mechanisms are faulty” and what is different in (bilingual) SLI. After all, Paradis states that a “complete overlap in developmental patterns with finiteness between a TD population . . . and an impaired population would not be felicitous.” We must be able to diagnose children with SLI. The key to this difference seems to be quantitative. What we find is that the error patterns are similar but the error frequency is different. In the Dutch study (in which the ages ranged from 6 to 8 years), we found an SLI effect for verb agreement inflection, but no L2 effect. Apparently, verb inflection is an index of SLI, also in learners of Dutch as an L2. In that sense, there is no “complete overlap.”
A CUMULATIVE EFFECT?
Of the two theoretical positions (which are loosely called representational and processing based) that Paradis compares, one (the maturational view) strikes a positive note about the effect of successive bilingualism, in that the input in the L2 becomes available “when [the children] are neurocognitively more mature, and already have another developing language,” leading to more “creativity in their learning patterns.” In contrast, under the processing view, bilingualism at first glance spells only trouble, because the processing resources in the bilingual must be shared by two languages and the resources in language-impaired children are more limited. The latter position thus predicts a “double effect” of bilingualism and SLI.
Paradis' present way of establishing cumulative effect improves on the original proposal. Paradis (Reference Paradis2007) understood the cumulative effect as a difference between monolingual and bilingual SLI. For successive bilingualism, as she argues now, this is not a valid comparison. After all, by definition, effects of language impairment and bilingualism are both to be expected in a group of bilingual children with SLI. The question is whether one effect exacerbates the other. For that to be the case, the joint effect must be disproportional.
The results from our project (cf. Orgassa, Reference Orgassa2009) contradict that. The SLI and L2 effects were measured separately. However, in the comparison between bilingual SLI and monolingual typically developing (TD) groups (two of the groups in our study) the two effects are necessarily conflated. If the effect were cumulative, it should be visible here. However, the difference between these two groups did not exceed the sum of the two effects measured separately (i.e., between groups that only differ on a single factor: bilingualism or impairment status).
Paradis proposes (differential) effect size as a measure that can discriminate between additional and cumulative effects. Paradis tests this rationale by reanalyzing data from Steenge's (Reference Steenge2006) study on bilingual children with SLI in The Netherlands. For the “morphology task,” Steenge found, in 7- and 8-year-olds, L2 and SLI effects, including a difference between bilingual and monolingual SLI (the key comparison in Paradis, Reference Paradis2007; Steenge's conceptualization of “additionally disadvantaged” resembles that of the double effect in Paradis' Keynote. In that sense, Steenge's results are a good candidate for Paradis, Reference Paradis2007). Unfortunately, for other reasons the figures from Steenge (Reference Steenge2006) cannot serve as an adequate test of cumulative effect in bilingual SLI. This is because of the selection of items. The aim of Steenge's morphology task, which Paradis recalculates in figure 4 in her Keynote, was “to assess whether a child could derive plural nouns from single nouns and past participles from single present tenses” (Steenge, Reference Steenge2006, p. 19; for single, read: singular, JdJ). Those morphological processes hardly represent proven clinical markers of SLI (number on nouns and, to a lesser extent, participles are relative strengths in the language profile) and they do not include tense or agreement marking (cf. Oetting & Rice, Reference Oetting and Rice1993). All this does not subtract from Paradis' methodological point that effect size differences constitute a valid measure, and it may yet be the case that a more vulnerable domain yields different results.
CROSS-DOMAIN DIFFERENCES
Is double effect even the key issue? Remember Paradis' suggestion that “in nonadditive contexts, cumulative effects might be longstanding, or even permanent, causing incomplete acquisition of some difficult structures.” I recognize the risk of incomplete acquisition and would argue that this possibility exists even in the absence of a cumulative effect. The Amsterdam study can illustrate that. In the results a sharp difference was found between the two linguistic domains included in the study. In the determiner phrase (DP) domain a clear L2 effect was found. Adjectival agreement and determiner selection are two phenomena that are intrinsically related and proved to be more problematic for bilingual children than for monolinguals. In addition, there was an SLI effect: children with SLI performed poorly on these measures. These effects were additive; one did not exacerbate the other. In contrast, as we saw before, verb agreement inflection is influenced by language impairment only, not by the bilingual status of the learner.
The bilingual children with SLI in our study were long past the age at which they could qualify as late talkers. For verb inflection, they produced a relatively high number of correct forms. In this domain it is possible that “one language bootstraps the learning of a second one” according to Paradis' Keynote. Both the children's languages have verb agreement.
In contrast, the production of correct adjectives and determiners was much lower, involving, for instance, the near exclusive use of one definite determiner, the common, rather than the neuter. Here, because of the absence of the relevant linguistic elements in Turkish, transfer from L1 was not possible. In short, in the DP domain, incomplete acquisition is inevitable. Even if the effect is not cumulative, the SLI and L2 effects in this domain conspire to prevent mastery. Note that for this particular domain full mastery by TD L2 learners is also improbable.
SLI research has devoted a lot of attention to cross-domain differences. Paradis herself draws our attention to the dissociation between linguistic domains or markers as a hallmark of SLI: tense marking versus nontense marking.
In bilingual SLI, another factor enters the picture and with that, a different kind of dissociation. The problems described above in the DP domain are very typical of L2 learners of Dutch. There is a dissociation in L2 learners: IP (inflection) versus DP.
We must find a way of modeling these two dissociations to accommodate for a profile of bilingual SLI. Our challenge is to establish the L2 and SLI ingredients of the language profile in bilingual SLI. It is promising that in verb inflection the SLI effect can be separated from the L2 effect. The DP domain illustrates the opposite situation: the two effects go together. For domains where the two effects are entangled the adoption of bilingual norms may be a way of identifying SLI in the presence of L2 effects, as Paradis suggests. After all, in domains that are vulnerable in L2, children with SLI still lag behind their TD bilingual peers.
You have
Access
Johanne Paradis' Keynote Article can be read as a concise critical review of the research that focuses on the sometimes strained relationship between bilingualism and specific language impairment (SLI). In my comments I will add some thoughts based on our own research on the learning of Dutch as a second language (L2) by children with SLI.Footnote 1
Our research focuses on successive bilingualism, and our group of bilingual subjects consists of children who have Turkish as their first language (L1) and Dutch as their L2. Turkish is a minority language in the full sense of Paradis' description: although the number of speakers of Turkish as an L1 is large in The Netherlands, there is no teaching of Turkish in primary schools. This situation could be said to lead to “subtractive bilingualism,” and the possibility of a “cumulative effect” of bilingualism and SLI that Paradis discusses is well worth considering for this particular group of subjects. In our study we were not able to include monolingual children with SLI whose native language is Turkish; the issue was therefore addressed with respect to the children's output in Dutch.
LINGUISTIC MODELS
Before discussing the nature of our results, I initially consider the feasibility of the linguistic model that Paradis highlights. Paradis' prime model for a theory that predicts similarity between monolingual and bilingual children with SLI is Rice and Wexler's extended optional infinitive (EOI) theory.
Whether this theory is fully applicable to Dutch is debatable (see the discussion with respect to monolingual SLI in de Jong, Reference De Jong, Levy and Schaeffer2003). To some extent the EOI account is indeed “a consequence of superficial properties of English” (Paradis, Reference Paradis2010 [this issue]). For instance, unlike in English, the verb stem has been proposed as an alternative optional default in Dutch child language (Paradis & Crago, Reference Paradis and Crago2001). The stem, however, is typically in second position, the position for finite verb forms in main clauses. Therefore, it is considered finite. In English, there are no distinct forms for stem and infinitive. In Dutch, the infinitive has an inflectional marker (-en). Thus, we restrict ourselves to the “true” infinitival form and discuss the presence of an EOI stage in Dutch. Infinitives are indeed found widely in early acquisition of Dutch and persist slightly longer in children with SLI (Wexler, Schaeffer, & Bol, Reference Wexler, Schaeffer and Bol2004). Empirically, however, the prediction that “this stage persists until (children with SLI are) at least nine years of age, and never fully resolves” is not supported by group findings on Dutch, although a small minority of individuals with SLI shows persistent use of root infinitives at the age of 8 or 9 years (de Jong, Reference De Jong1999; Orgassa, Reference Orgassa2009). Consequently, the ramifications for patterns of growth (maturation) in Dutch-speaking children with SLI are not clear.
However, one element of the “maturational” theory does ring true for SLI in Dutch, and that is the basic normality of patterns found in child groups. Unlike the now obsolete theory proposed by Gopnik, in which symptoms arose from “absence of grammar,” the EOI account and its affiliates stress underspecification in a nondeviant grammar. In Dutch, this is highly visible in the robust correlation between verb position and verb form, a connection that is not abandoned by language-impaired children either, whether monolingual or bilingual. For bilingual children with SLI, who received Dutch input later than monolinguals and share the intake problem with monolingual children with SLI, this finding was replicated (Orgassa, Reference Orgassa2009). Given this basic normality, the question remains which “language acquisition mechanisms are faulty” and what is different in (bilingual) SLI. After all, Paradis states that a “complete overlap in developmental patterns with finiteness between a TD population . . . and an impaired population would not be felicitous.” We must be able to diagnose children with SLI. The key to this difference seems to be quantitative. What we find is that the error patterns are similar but the error frequency is different. In the Dutch study (in which the ages ranged from 6 to 8 years), we found an SLI effect for verb agreement inflection, but no L2 effect. Apparently, verb inflection is an index of SLI, also in learners of Dutch as an L2. In that sense, there is no “complete overlap.”
A CUMULATIVE EFFECT?
Of the two theoretical positions (which are loosely called representational and processing based) that Paradis compares, one (the maturational view) strikes a positive note about the effect of successive bilingualism, in that the input in the L2 becomes available “when [the children] are neurocognitively more mature, and already have another developing language,” leading to more “creativity in their learning patterns.” In contrast, under the processing view, bilingualism at first glance spells only trouble, because the processing resources in the bilingual must be shared by two languages and the resources in language-impaired children are more limited. The latter position thus predicts a “double effect” of bilingualism and SLI.
Paradis' present way of establishing cumulative effect improves on the original proposal. Paradis (Reference Paradis2007) understood the cumulative effect as a difference between monolingual and bilingual SLI. For successive bilingualism, as she argues now, this is not a valid comparison. After all, by definition, effects of language impairment and bilingualism are both to be expected in a group of bilingual children with SLI. The question is whether one effect exacerbates the other. For that to be the case, the joint effect must be disproportional.
The results from our project (cf. Orgassa, Reference Orgassa2009) contradict that. The SLI and L2 effects were measured separately. However, in the comparison between bilingual SLI and monolingual typically developing (TD) groups (two of the groups in our study) the two effects are necessarily conflated. If the effect were cumulative, it should be visible here. However, the difference between these two groups did not exceed the sum of the two effects measured separately (i.e., between groups that only differ on a single factor: bilingualism or impairment status).
Paradis proposes (differential) effect size as a measure that can discriminate between additional and cumulative effects. Paradis tests this rationale by reanalyzing data from Steenge's (Reference Steenge2006) study on bilingual children with SLI in The Netherlands. For the “morphology task,” Steenge found, in 7- and 8-year-olds, L2 and SLI effects, including a difference between bilingual and monolingual SLI (the key comparison in Paradis, Reference Paradis2007; Steenge's conceptualization of “additionally disadvantaged” resembles that of the double effect in Paradis' Keynote. In that sense, Steenge's results are a good candidate for Paradis, Reference Paradis2007). Unfortunately, for other reasons the figures from Steenge (Reference Steenge2006) cannot serve as an adequate test of cumulative effect in bilingual SLI. This is because of the selection of items. The aim of Steenge's morphology task, which Paradis recalculates in figure 4 in her Keynote, was “to assess whether a child could derive plural nouns from single nouns and past participles from single present tenses” (Steenge, Reference Steenge2006, p. 19; for single, read: singular, JdJ). Those morphological processes hardly represent proven clinical markers of SLI (number on nouns and, to a lesser extent, participles are relative strengths in the language profile) and they do not include tense or agreement marking (cf. Oetting & Rice, Reference Oetting and Rice1993). All this does not subtract from Paradis' methodological point that effect size differences constitute a valid measure, and it may yet be the case that a more vulnerable domain yields different results.
CROSS-DOMAIN DIFFERENCES
Is double effect even the key issue? Remember Paradis' suggestion that “in nonadditive contexts, cumulative effects might be longstanding, or even permanent, causing incomplete acquisition of some difficult structures.” I recognize the risk of incomplete acquisition and would argue that this possibility exists even in the absence of a cumulative effect. The Amsterdam study can illustrate that. In the results a sharp difference was found between the two linguistic domains included in the study. In the determiner phrase (DP) domain a clear L2 effect was found. Adjectival agreement and determiner selection are two phenomena that are intrinsically related and proved to be more problematic for bilingual children than for monolinguals. In addition, there was an SLI effect: children with SLI performed poorly on these measures. These effects were additive; one did not exacerbate the other. In contrast, as we saw before, verb agreement inflection is influenced by language impairment only, not by the bilingual status of the learner.
The bilingual children with SLI in our study were long past the age at which they could qualify as late talkers. For verb inflection, they produced a relatively high number of correct forms. In this domain it is possible that “one language bootstraps the learning of a second one” according to Paradis' Keynote. Both the children's languages have verb agreement.
In contrast, the production of correct adjectives and determiners was much lower, involving, for instance, the near exclusive use of one definite determiner, the common, rather than the neuter. Here, because of the absence of the relevant linguistic elements in Turkish, transfer from L1 was not possible. In short, in the DP domain, incomplete acquisition is inevitable. Even if the effect is not cumulative, the SLI and L2 effects in this domain conspire to prevent mastery. Note that for this particular domain full mastery by TD L2 learners is also improbable.
SLI research has devoted a lot of attention to cross-domain differences. Paradis herself draws our attention to the dissociation between linguistic domains or markers as a hallmark of SLI: tense marking versus nontense marking.
In bilingual SLI, another factor enters the picture and with that, a different kind of dissociation. The problems described above in the DP domain are very typical of L2 learners of Dutch. There is a dissociation in L2 learners: IP (inflection) versus DP.
We must find a way of modeling these two dissociations to accommodate for a profile of bilingual SLI. Our challenge is to establish the L2 and SLI ingredients of the language profile in bilingual SLI. It is promising that in verb inflection the SLI effect can be separated from the L2 effect. The DP domain illustrates the opposite situation: the two effects go together. For domains where the two effects are entangled the adoption of bilingual norms may be a way of identifying SLI in the presence of L2 effects, as Paradis suggests. After all, in domains that are vulnerable in L2, children with SLI still lag behind their TD bilingual peers.