3.1 Typical sequential bilingual learners

Children with LLI are defined in reference to peers with typical language development. To make a valid comparison, these peers should have comparable social and educational experiences. For young sequential bilingual learners, then, this reference group is composed of children who have consistent and robust experiences with a single L1 from birth and with an L2 beginning at some point during childhood. In this review, the L2 learners of interest are mainly children from immigrant or indigenous families who learn a home language that is different from the majority community language(s). This type of L2 learner is well represented in almost every country; examples include Spanish or Vietnamese L1 learners in the US, native Punjabi or Cantonese speakers in the UK or Canada, Turkish or Arabic L1 learners in Germany or Sweden and children in Mexico who speak Nahuatl or Mayan as their home language. For many sequential bilingual children, the L1 is considered a minority language either because there are fewer opportunities to use this language or because there is less social prestige associated with L1 proficiency. The linguistic characteristics of this typically developing population provide the reference for bilingual learners with LLI.

By definition, early sequential bilinguals are initially more skilled in their home language. The L1 is not fully mastered when the L2 is introduced and, over time, shifts may occur from relative strength or dominance in the L1 to the L2. Sociolinguistic factors such as the language of education play key roles in this dominance shift (Wright, Taylor & Macarthur Reference Wright, Taylor and Macarthur2000). For many minority L1 speakers, L2 learning begins in early childhood and accelerates with immersion in majority language educational programs, becoming the more proficient language by late childhood or adolescence (e.g., Jia, Aaronson & Wu Reference Jia, Aaronson and Wu2002; Verhoeven Reference Verhoeven2007; Sheng, Lu & Kan Reference Sheng, Lu and Kan2011; Pham & Kohnert Reference Pham and Kohnert2014). For example, Pham & Kohnert (Reference Pham and Kohnert2014) longitudinally tracked growth in lexical access and vocabulary skills among 33 Vietnamese L1 learners in the US. Children began the study, on average, at age 7 years and were tracked annually for four years. At initial testing, performance was better in Vietnamese (L1) across tasks. Both L1 and L2 (English) skills grew across the four-year period, but gains in the L2 significantly outpaced those in the L1; for example, expressive vocabulary scores grew at a rate of 11% per year in English compared to 3% per year in Vietnamese. Ultimately, this growth pattern resulted in a gradual shift to English (L2) dominance by study completion.

Even a brief discussion of the dominance shift must acknowledge potential differences in the timing and extent of the shift across the various subcomponents and skill sets that compose language. For both monolingual and bilingual children, language acquisition is a prolonged process; different language skills or subsystems may experience rapid growth or refinement during particular developmental periods, or in association with certain experiences (Chondrogianni & Marinis Reference Chondrogianni and Marinis2011). Studies investigating both languages of sequential bilingual children find that at different points during the dynamic process of language acquisition, learners may perform some tasks better in the L1 and other tasks better in the L2 (e.g., Ordóñez et al. Reference Ordóñez, Carlo, Snow and McLaughlin2002; Peña, Bedore & Zlatic-Giunta Reference Peña, Bedore and Zlatic-Giunta2002; Kan & Kohnert Reference Kan and Kohnert2008). Therefore, even when there is an overall shift from relative strength in the L1 to the L2, the timing may not be uniform across language tasks and subsystems (Kohnert & Bates Reference Kohnert and Bates2002; Pham & Kohnert Reference Pham and Kohnert2014).

Shifting patterns of absolute and relative differences in L1-L2 performance across tasks and over time are consistent with the complementarity principle of bilingualism (Grosjean Reference Grosjean2010). In its essence, the complementarity principle asserts that proficiency follows use; the languages in bilingual or multilingual speakers may be developed and used for different, albeit complementary, purposes. Therefore, differences in relative areas of functional strength within a bilingual, between the two languages, are to be expected. This point is exemplified by a large cross-sectional study of 3- to 10-year-old children in Canada: Bialystok and colleagues (Bialystok et al. Reference Bialystok, Luk, Peets and Yang2010) found that at all age levels, monolingual groups scored significantly better than their sequential bilingual age peers on an English vocabulary comprehension test. The effect was robust and large (Cohen's d = 0.83 for the difference in scores between the two groups). Researchers were quick to point out, however, that the bilingual group scores were within the normal range of variation. The monolingual-bilingual vocabulary difference was largely confined to words that reflect home life; school vocabulary for children in the two groups was more comparable, underscoring the relevance of the complementarity principle.

An additional relevant feature of developing sequential bilingual learners is the presence and nature of within-child, cross-language associations. Hypothetically, two languages within a sequential bilingual learner may coexist without interaction or be interdependent, with the presence of one language affecting acquisition or use of the other. Mere coexistence should yield no statistical associations between performance on select tasks in the L1 and the L2. In contrast, interdependence should produce negative or positive correspondences between the L1 and the L2 (e.g., De Bot, Lowie & Verspoor Reference De Bot, Lowie and Verspoor2007).

Despite the relative nascence of this investigative line, several studies document significant cross-language associations for sequential bilingual learners at various ages and developmental stages, learning different pairs of languages. For example, Verhoeven (Reference Verhoeven2007) investigated a number of language skills (including receptive and expressive vocabulary, sentence repetition, story comprehension, and grammatical comprehension) in a group of 75 kindergartners learning Turkish as the L1 and Dutch as the L2 in the Netherlands. A factor analysis of the children's performance in both the L1 and the L2 showed a one-factor solution, meaning that for these sequential bilingual learners, there was strong interdependence between their two languages. In addition, both languages also contributed cross-linguistically to the development of early reading skills. In younger children learning a very different pair of languages (Hmong as the L1 and English as the L2), Kan & Kohnert (Reference Kan and Kohnert2008) found a positive association between lexical comprehension (or picture identification) in the L1 and lexical production (or picture naming) in the L2. However, for these same participants the ability to quickly learn a novel sequence of English phonemes to name an unfamiliar object (fast mapping) was negatively correlated with Hmong vocabulary knowledge, suggesting some type of temporary cross-linguistic interference (Kan & Kohnert Reference Kan and Kohnert2008; Reference Kan and Kohnert2012). When retelling stories in each of their languages, preschool-age participants who used more different words in Hmong (L1) also used a more diverse vocabulary in English (L2) (Kohnert, Kan & Conboy Reference Kohnert, Kan and Conboy2010). Finally, a number of studies demonstrate better spoken L1 skills are positively associated with early L2 reading skills for Spanish-English sequential bilinguals in the US (e.g., López & Greenfield Reference López and Greenfield2004; Miller et al. Reference Miller, Heilmann, Nockerts, Iglesias, Fabiano and Francis2006).

In contrast, a recent review and meta-analysis of the relationship between the L1 and the L2 vocabulary scores in school-aged Spanish-English bilingual children (Branum-Martin et al. Reference Branum-Martin, Mehta, Francis, Foorman, Cirino and Miller2009) identified positive, neutral, and negative relationships across various published studies. Meta-analytic results showed correlations near zero between the two languages for individual students, at least when vocabulary is measured using standardized tests (Branum-Martin et al. Reference Branum-Martin, Mehta, Francis, Foorman, Cirino and Miller2009). Across studies, it is clear that the L1 and the L2 are interdependent, but that the strength and direction of L1-L2 relations varies. A number of factors may play into this variability, such as the age of the children studied, the sociolinguistic context of the study, the relative linguistic proficiency of participants, and perhaps the cognitive abilities of the participants (Kan & Kohnert Reference Kan and Kohnert2012; Kohnert Reference Kohnert2013). Sorting out the factors underlying cross-linguistic relationships is a key area of research need, not only to understand typical bilingual development but also to understand how best to support the language growth of bilingual children with LLI.

Thus far, we have considered linguistic characteristics of typically developing sequential bilingual children. The relevant characteristics of these bilingual learners extend beyond strictly linguistic skills, however. Children who are proficiently bilingual have been shown to outperform monolingual peers on both metalinguistic tasks and specific types of cognitive tasks (see Adesope et al. Reference Adesope, Lavin, Thompson and Ungerleider2010; Bialystok, Craik & Luk Reference Bialystok, Craik and Luk2012 for reviews). Metalinguistically, bilingual children show enhanced skills in areas such as phonemic awareness and grammaticality judgment. Cognitively, bilingual children outperform monolinguals on a variety of nonlinguistic tasks, particularly in the area of attentional control (Adesope et al. Reference Adesope, Lavin, Thompson and Ungerleider2010). These advantages have been interpreted as evidence of enhanced cognitive control, resulting from the need to manage two separate language systems (Bialystok et al. Reference Bialystok, Craik and Luk2012). Though most studies have included bilingual children already highly proficient in both languages, at least one recent study showed that attentional control advantages may begin to emerge with as little as one year of systematic L2 exposure (Poarch & van Hell Reference Poarch and van Hell2012). Thus, evidence for cognitive and metalinguistic advantages in sequential bilingual children is mounting, and this evidence may have important implications for bilingual children with LLI.

The purpose of our review of typical sequential bilingual learners was to establish developmental patterns with key implications for bilingual peers with LLI. Several such patterns emerge. Children who speak minority L1s frequently shift to L2 dominance during childhood, with the exact timing varying substantially both across and within children (i.e., different language skills may grow at different rates and show different patterns of dominance). Bilinguals cannot be characterized as the sum of two monolinguals, for multiple reasons; they demonstrate unique profiles of relative cross-linguistic areas of strength, frequently driven by patterns of use, and the presence of cross-linguistic associations is a unique feature of bilinguals. Finally, the presence and interdependence of two languages within one developing child has clear cognitive consequences. Alongside these unique features of young bilingual learners, investigators have documented some performance similarities between monolingual children with LLI and typically developing sequential bilingual children. We turn next to the literature comparing these groups.

3.2 Comparing L2 learners to monolingual children with LLI

One prominent research question within the past two decades has been how typically developing sequential bilingual children compare to monolingual peers with and without LLI. That is, a number of studies have conducted three-way group comparisons with typical sequential bilingual children, typical monolingual children, and monolingual children with LLI. This line of work tests the tenets of basic language acquisition theories by comparing the influence of decreased language experience (i.e., sequential bilingual children performing in the L2) to that of impaired language learning (i.e., children with LLI). From a practical standpoint, these three-way comparisons provide an opportunity to identify possible fault lines that could separate typical sequential bilingual learners from age peers with LLI.

Within this line of work, the comparisons have spanned a broad range of linguistic and nonlinguistic skills, including vocabulary, grammar, narrative skills, new word learning, recall of phoneme sequences, and the processing of visual and auditory nonlinguistic information (e.g., Paradis Reference Paradis2005; Gutiérrez-Clellen, Simon-Cereijido & Wagner Reference Gutiérrez-Clellen, Simon-Cereijido and Wagner2008; Windsor et al. Reference Windsor, Kohnert, Loxtercamp and Kan2008; Kohnert, Windsor & Ebert Reference Kohnert, Windsor and Ebert2009; Marinis & Saddy Reference Marinis and Saddy2013). The language tasks are generally administered in the only language of the monolingual comparison groups, which is the L2 of the sequential bilingual participants.

In many of these comparisons, typical sequential bilingual learners perform indistinguishably from their monolingual age peers with LLI on a wide range of language tasks, even when the typical bilingual group has many years of L2 experience (e.g., Kohnert et al. Reference Kohnert, Windsor and Ebert2009). Comparable performance between typically developing Spanish-English sequential bilinguals and English-speaking monolingual children with LLI has been documented on a number of lexical-semantic measures, including standardized tests of receptive vocabulary, speeded picture naming tasks, and measures of verbal recall (Umbel et al. Reference Umbel, Pearson, Fernández and Oller1992; Kohnert et al. Reference Kohnert, Windsor and Ebert2009). Verhoeven et al. (Reference Verhoeven, Steenge and van Balkom2011) reported nearly identical group means on a word definition task in Dutch for monolingual Dutch-speaking children with LLI (n = 137) and for typically developing L2 Dutch learners (whose first languages included Turkish, Moroccan, and Surinamese; n = 321). In contrast to the lack of differentiation between typical bilingual learners and monolinguals with LLI, significant differences between typical bilingual and typical monolingual groups are a frequent finding: that is, when considering only the L2, school-aged typically developing bilinguals know fewer words, name pictures more slowly, and create poorer word definitions (e.g., Kohnert et al. Reference Kohnert, Windsor and Ebert2009; Verhoeven et al. Reference Verhoeven, Steenge and van Balkom2011).

Similar patterns are found outside the realm of lexical-semantic tasks. There has been particular interest in comparisons of grammatical patterns between typical L2 learners and monolingual LLI, due to the prominence of grammar in theories of LLI and in clinical practice. A frequent finding from this literature is that typical L2 learners may make grammatical errors that mirror those made by monolingual children with LLI. For example, patterns of Dutch grammatical errors – including omissions of definite determiners and incorrect adjective-noun agreement – were found to be the same for six- to eight-year-old children with LLI and same-aged peers learning Dutch as the L2 (Orgassa & Weerman Reference Orgassa and Weerman2008). Moreover, similar patterns were found in younger (4- and 5-year-old) typically developing monolingual Dutch-speaking peers, suggesting the error patterns show a typical but delayed sequence of development. In French, both typically developing L2 speakers and monolingual children with LLI tend to omit past and future tense verb markings and copula verbs (Paradis & Crago Reference Paradis and Crago2000). Similar findings exist in Spanish (Gutiérrez-Clellen, Simon-Cereijido & Wagner Reference Gutiérrez-Clellen, Simon-Cereijido and Wagner2008) and Swedish (Håkansson Reference Håkansson2001).

A potentially important factor in comparisons between typical sequential bilinguals and monolinguals with and without LLI is socioeconomic status. Sequential bilingual children frequently live in first- or second-generation immigrant families; in many countries, such families experience fewer employment opportunities and higher rates of poverty (Kerr & Kerr Reference Kerr and Kerr2011). Because poverty is associated with less frequent and less rich language exposure for children (Hart & Risley Reference Hart and Risley1995), socioeconomic status may have a confounding influence on comparisons between monolinguals and sequential bilingual language development if not carefully controlled.

Recently, some researchers have taken a closer look at specific error types to better compare bilingual and LLI development (Chondrogianni & Marinis Reference Chondrogianni and Marinis2012; Marinis & Saddy Reference Marinis and Saddy2013; Armon-Lotem Reference Armon-Lotem2014). For example, Armon-Lotem (Reference Armon-Lotem2014) used sentence completion and sentence repetition tasks to investigate prepositions and verb inflectional morphology in Hebrew. The 5- to 7-year-old participants were typical Hebrew L2 learners, with either English or Russian as their L1, and monolingual Hebrew speaking children with LLI. Results showed that prepositions and verb inflectional morphology were loci of difficulty for both the typical bilingual participants and the Hebrew-only participants with LLI. However, a closer analysis of errors revealed that the number and nature of errors differed for the groups. The typical L2 learners tended to produce errors of commission, characterized largely by substitutions based on L1 features. In contrast, the monolingual LLI errors were characterized as omission of the whole morpheme or feature reduction. Differences in error patterns may surface on grammatical comprehension tasks as well. A detailed analysis of the comprehension of passives by typically developing Turkish-English bilinguals and English-only speakers with LLI revealed group differences in errors and in the time needed to process and interpret sentences (Marinis & Saddy Reference Marinis and Saddy2013). As this body of literature grows, it may be possible to utilize careful analysis of grammatical errors in production, or perhaps error patterns in comprehension, to rule out LLI in L2 learners.

Three-way comparisons have also extended to nonlinguistic processing skills. Given the well-documented subtle processing inefficiencies in monolingual children with LLI (in areas such as processing speed and working memory), it is possible that performance on nonlinguistic cognitive tasks could separate monolingual children with LLI from typically developing monolingual and bilingual peers. Indeed, Kohnert et al. (Reference Kohnert, Windsor and Ebert2009) found that typically developing Spanish-English 8- to 11-year-old bilinguals and monolinguals performed comparably, and better than English-only children with LLI, on a host of experimental nonlinguistic processing tasks. For example, tasks measuring response speed for pushing one of two different buttons in response to a visual cue, or requiring participants to determine if sequences of presented tones were the same or different, separated children with LLI from typical bilingual and monolingual peers. Accuracy in a mental rotation task produced a similar result. Of note, group differences did not reach significance on other tasks (such as pushing a single button in response to an auditory or visual stimulus, or searching for a target digit in an array). There may be a minimal level of task complexity needed to tease out the relatively subtle processing difficulties of children with LLI; further studies in this area should extend and clarify these results.

Finally, it is important to note that although identifying a task that reveals significant group differences in performance between typical L2 learners and monolingual children with LLI advances basic understanding of language, the practical applications are limited. Group-level differences must be verified with investigations of diagnostic accuracy, beginning with sensitivity and specificity analyses at different cutoff points and progressing to the development of positive and negative likelihood ratios (e.g. Dollaghan & Horner Reference Dollaghan and Horner2011). The calculation of positive and negative likelihood ratios advances consideration of diagnostic tasks to the individual level, which is ultimately critical for practical clinical applications.

In addition, standards of comparisons for bilingual children must go beyond monolingual peers. As discussed previously, the bilingual is more than and different than the sum of his or her component languages, and bilingual children may often experience different socioeconomic circumstances than monolingual children. Thus, although comparisons to monolinguals on select tasks may be useful to rule out LLI in some typically developing sequential bilinguals, they cannot be used to identify LLI. Rather, diagnostic tasks must be investigated in samples that also include bilingual children with LLI. Though studies of this population were historically rare, in recent years the literature characterizing sequential bilingual children with LLI as compared to their typically developing bilingual peers has rapidly grown. We review the trends in this emerging literature next, before returning to the questions of accurate identification and assessment of bilingual children with LLI.

4.1 Dual language profiles in bilingual LLI

At the sublexical level, the repetition of nonsense words (known as nonword repetition or NWR) has been investigated in several studies of bilingual children with LLI. Results of these works robustly indicate that bilingual children with LLI repeat nonsense words less accurately than typically developing bilingual peers in both the L1 and the L2, a result found for English-French bilinguals in Canada (Elin Thordardottir & Brandeker Reference Thordardottir and Brandeker2013), Turkish-Dutch bilinguals in the Netherlands (Verhoeven et al. Reference Verhoeven, Steenge and van Balkom2012); and Spanish-English bilinguals in the United States (Gutiérrez-Clellen & Simon-Cereijido Reference Gutiérrez‐Clellen and Simon‐Cereijido2010; Windsor et al. Reference Windsor, Kohnert, Lobitz and Pham2010). NWR is thought to measure phonological memory, though it may tap into multiple linguistic abilities (see Ebert et al. Reference Ebert, Pham and Kohnert2014 for discussion). Thus, NWR appears to be an area of weakness for bilingual children with LLI, as it is for their monolingual peers.

At the lexical level, LLI is reflected in reduced performance on a number of linguistic measures. First, bilingual children with LLI are likely to show reduced vocabulary in both languages relative to their typically developing peers, as measured in picture identification and naming assessments (e.g., Gibson, Peña & Bedore Reference Gibson, Peña and Bedore2014). Reduced vocabulary size among bilingual children with LLI is also reflected in smaller lexical diversity in connected discourse. For example, when 5- to 7-year-old English-Hebrew bilingual children living in Israel narrated familiar stories to picture stimuli, those with LLI produced significantly fewer different words, and had smaller type-token ratios (or the ratio of unique words to total words produced) than children without LLI (Iluz-Cohen & Walters Reference Iluz-Cohen and Walters2012). Group membership accounted for 22.4% and 36.7% of the variance in these lexical measures. Similar results were found for Spanish-English bilinguals in the US (Simon-Cereijido & Gutiérrez-Clellen Reference Simon-Cereijido and Gutiérrez-Clellen2009); children with LLI produced fewer different words in both L1 and L2 narratives than unaffected peers. We do note that in a few studies, measures of lexical diversity have not shown significant between-group differences for LLI versus typically developing groups (Ooi & Wong Reference Ooi and Wong2012; Jacobsen & Walden Reference Jacobson and Walden2013), though the lack of significance may relate to sample size, methods of language sampling, or other differences among the various participant groups.

In addition, there is evidence that bilingual children with LLI show reduced vocabulary depth, or overall semantic knowledge, in comparison to typically developing bilingual peers. Sheng and colleagues (Sheng et al. Reference Sheng, Peña, Bedore and Fiestas2012; Reference Sheng, Bedore, Penã and Taliancich-Klinger2013) have shown that school-aged Spanish-English bilingual children with LLI demonstrate relatively poor semantic depth in comparison to their typically developing peers. Specific deficits include weaker semantic networks, as measured by children's ability to generate words associated with a target lexical item, and poor semantic convergence, as measured by the ability to generate expected associations for the lexical target (rather than unusual responses). Such differences were evident in each language, as well as when languages were combined using conceptual scoring.

At the sentence level, dual-language weaknesses for bilingual children with LLI are expected to manifest in increased grammatical errors, shorter utterances, and reduced sentence comprehension. Grammatical deficits have been a traditional focus of the LLI literature for bilinguals, as they are for monolinguals, though a substantial percentage of studies have examined these deficits in only one language (e.g., Gutiérrez-Clellen et al. Reference Gutiérrez-Clellen, Simon-Cereijido and Wagner2008; Verhoeven et al. Reference Verhoeven, Steenge and van Balkom2011; Blom et al. Reference Blom, De Jong, Orgassa, Baker and Weerman2013). Again, the focus here is on those that have considered both languages. For example, Swedish investigators considered grammatical development in Arabic-Swedish bilingual children aged 4 to 7 years (Håkansson, Salameh & Nettelbladt Reference Håkansson, Salameh and Nettelbladt2003; Salameh, Håkansson & Nettelbladt Reference Salameh, Håkansson and Nettelbladt2004). A series of tasks in Arabic and in Swedish, designed to elicit specific grammatical structures, were administered to ten children with LLI and ten typically developing bilingual peers matched for age and parental education (Håkansson et al. Reference Håkansson, Salameh and Nettelbladt2003). Results showed that the children with LLI, like their typically developing peers, acquired grammatical structures in a predictable order in both languages. Unlike their typical peers, however, the children with LLI were at earlier stages of grammatical development in both Swedish and Arabic. The same children were then followed for a year to examine rates of grammatical development (Salameh et al. Reference Salameh, Håkansson and Nettelbladt2004); as a group, the children with LLI continued to progress slowly through the expected sequence of grammatical development in both languages, with individual instances of regression in one language or the other.

Further evidence of sentence-level linguistic deficits in bilingual children with LLI can be found in analyses of utterance length and complexity. Works that have compared mean length of utterance (MLU) in bilingual children with LLI to their typically developing bilingual peers have consistently found significantly shorter MLU for children with LLI (Sanz-Torrent et al. Reference Sanz-Torrent, Serrat, Andreu and Serra2008; Ooi & Wong Reference Ooi and Wong2012; Jacobsen & Walden Reference Jacobson and Walden2013). For example, Ooi & Wong (Reference Ooi and Wong2012) found that the presence of LLI accounted for 32% of the variance in English MLU for a group of 18 Cantonese-English bilingual preschoolers. Although it is perhaps unsurprising that this classic indicator of language development would reflect LLI in bilingual children, additional work in this area may reveal more nuanced patterns of deficit. The degree to which bilingual children with LLI show shorter spontaneous MLUs is likely to depend on specific language pairs, patterns of input, and age. For example, there may be a greater difference in English MLU between children with and without LLI than in Spanish MLU, due to language-specific trajectories of MLU growth (Bedore et al. Reference Bedore, Peña, Gillam and Ho2010).

Finally, a limited number of studies have considered the overall quality of discourse produced by bilingual children with LLI. In a recent longitudinal study, Squires et al. (Reference Squires, Lugo‐Neris, Peña, Bedore, Bohman and Gillam2014) asked a group of 21 Spanish-English bilingual children with LLI and a set of 21 matched peers without LLI to retell stories in English and in Spanish to wordless picture books at two time points: kindergarten and first grade. The stories were scored for both microstructural (e.g., mental and linguistic verbs, subordinating conjunctions, and elaborated noun phrases) and macrostructural (e.g., character, setting, and initiating events) elements. The typically developing children received higher macrostructural scores at both time points and in both languages. In addition, the scores of the typically developing children grew at a faster rate than those of the children with LLI.

In contrast, Iluz-Cohen & Walters (Reference Iluz-Cohen and Walters2012) found no differences in story macrostructure between English-Hebrew bilinguals with and without LLI. Children in this study were of comparable age (5 and 6 years old) to those studied by Squires et al. (Reference Squires, Lugo‐Neris, Peña, Bedore, Bohman and Gillam2014); however, they did not receive a model of the story before telling it, perhaps contributing to the conflicting cross-study results. The small number of participants (nine children with LLI and eight without) studied by Iluz-Cohen & Walters (Reference Iluz-Cohen and Walters2012) may also have contributed to the insignificant differences in story macrostructure, though significant between-group differences existed on other measures in this study. It is also possible that the severity of LLI within the impaired group differed across studies; LLI severity has not yet been considered systematically for bilingual children, and will be an important point for future research. There is some evidence of reduced story comprehension in both languages for bilingual children with LLI (Verhoeven et al. Reference Verhoeven, Steenge and van Balkom2012), suggesting that discourse-level deficits are found receptively as well as expressively. Additional investigations to clarify differing results in expressive story macrostructure, to confirm story comprehension deficits, and to consider other forms of discourse (such as expository tasks for older children) will help to complete the linguistic profile of bilingual children with LLI.

An additional aspect of the linguistic profile is language growth over time. For typical minority sequential bilingual learners, differential rates of L1 and L2 growth are expected, often resulting in greater proficiency in the L2 when that is the majority community and educational language. At the same time, the L1, or home language, is maintained or continues to develop, albeit at a somewhat slower pace. This may not be the case for minority L1 learners with LLI. Some reports suggest that, in the absence of robust and systematic support, there is a greater risk of L1 loss or decline over time for minority bilingual children with LLI (Restrepo & Kruth Reference Restrepo and Kruth2000; Salameh et al. Reference Salameh, Håkansson and Nettelbladt2004). Ebert et al. (Reference Ebert, Pham and Kohnert2014) considered changes in lexical processing skills associated with age in a cross-sectional sample of Spanish-English bilingual children with LLI, aged 6 to 11 years. Like their typically developing bilingual age peers, this group of children showed stronger performance in the L2 than in the L1 on tasks such as receptive vocabulary testing, expressive vocabulary testing, and rapid automatic naming. Correlations between age and task performance showed significantly better L2 skills in older children, but minimal association between age and L1 skill. That is, lexical skills in L1 appeared to plateau for many participants. Sheng et al. (Reference Sheng, Bedore, Penã and Taliancich-Klinger2013) also suggested a possible plateau in L1 skills for school-aged bilingual children with LLI, again for Spanish-English bilingual children in the US.

Developmental stage may be an important facet of these results. Longitudinal studies of Spanish-English preschoolers with LLI have shown growth in L1 and the L2 over time (Gutiérrez-Clellen, Simon-Cereijido & Sweet Reference Gutiérrez-Clellen, Simon-Cereijido and Sweet2012; Simon-Cereijido, Gutiérrez-Clellen & Sweet Reference Simon-Cereijido, Gutiérrez-Clellen and Sweet2013) rather than plateaus or loss. Other possible factors in language growth for children with LLI include sociolinguistic context, typological similarity between languages, and the presence of intervention to support language growth (Gutiérrez-Clellen et al. Reference Gutiérrez-Clellen, Simon-Cereijido and Sweet2012; Simon-Cereijido et al. Reference Simon-Cereijido, Gutiérrez-Clellen and Sweet2013). Additional studies that vary these parameters (e.g., studies of different language pairs, in different countries, and with varying levels of support for each language) will help to establish the dual-language growth patterns of bilingual children with LLI. Ultimately, treatment studies may be most valuable in establishing language growth over time for bilingual learners with LLI, as all of these parameters can be varied within a longitudinal design.

Overall, the literature to date has clearly established the presence of weaknesses in both languages of sequential bilingual children with LLI, presumably due to a more general inefficiency in the intrinsic language learning system. In bilingual LLI, differences in relative L1 and L2 skills within individuals are also expected over time and across language subsystems, as they are with typical bilingual learners. Additional research is needed to expand our understanding of bilingual LLI, considering different language pairs and social contexts. Attention to individual variation, including disorder severity, and within-learner interactions between the L1 and the L2 should prove valuable. For example, investigations of how these factors influence rates of code-switching in bilingual LLI (e.g., Gutiérrez-Clellen, Simon-Cereijido & Leone Reference Gutiérrez-Clellen, Simon-Cereijido and Leone2009; Greene, Peña & Bedore Reference Gutiérrez-Clellen, Simon-Cereijido and Sweet2012) highlight their complex interaction as well as the need for further careful study.

Profiles of bilingual LLI children necessarily encompass both the L1 and the L2, but may also need to extend beyond language. Basic cognitive processing differences have been implicated in LLI as well as in typical bilingualism. As such, we next consider cognition in bilingual LLI.

4.2 Cognition and bilingual LLI

A small but growing literature has begun to consider the cognitive characteristics of bilingual children with LLI (Kohnert, Windsor & Pham Reference Kohnert, Windsor and Ebert2009; Engel de Abreu, Cruz-Santos & Puglisi Reference Engel de Abreu, Cruz-Santos and Puglisi2014). Presumably, the subtle nonlinguistic cognitive weaknesses present in monolingual children with LLI would also be present in affected bilingual children, but until recently few studies had considered this question. Engel de Abreu et al. (Reference Engel de Abreu, Cruz-Santos and Puglisi2014) recently compared the verbal working memory, visuospatial working memory, selective attention, and attentional control skills of 15 Portuguese-Luxembourgish children with LLI to a group of 33 typically developing bilingual peers. The children with LLI fell below their peers on one verbal working memory measure (digit span) and the attentional control task (a flanker task), though not on the visuospatial working memory tasks or the selective attention task (a search task). Due to the relatively small samples, the comparisons which did reach significance showed very large effect sizes (e.g., d = 0.99 for the group difference on the attentional control task). In a larger sample of children (n = 233) within the US, Kohnert, Windsor & Pham (Reference Kohnert, Windsor and Ebert2009) found that Spanish-English bilingual children with LLI scored below typically developing peers on nonlinguistic measures of tonal working memory and visual processing speed. Moreover, these measures helped distinguish between children with and without LLI when both mono- and bilingual children were included in the sample.

As work on cognitive characteristics of bilingual children with LLI moves forward, it will be crucial to consider the intersection of cognitive advantages associated with bilingualism (Bialystok, Craik & Luk Reference Bialystok, Craik and Luk2012) with the subtle areas of weakness present in children with LLI. It is not yet clear if the apparent cognitive advantage of bilingualism could act as a buffer against the cognitive weaknesses associated with LLI. In Engel de Abreu's recent (Reference Engel de Abreu, Cruz-Santos and Puglisi2014) work with Portuguese-Luxembourgish bilinguals, the bilingual children with LLI performed equivalently to a group of typically developing monolingual Portuguese speakers on the attentional control task, though they showed significant deficits in comparison to bilingual peers. In other words, the pattern of results suggested that bilingual experience could mitigate against impairments associated with LLI. However, it is important to note that the study did not contain a group of monolinguals with LLI, who would provide an important benchmark for this hypothesis.

It is also possible that bilingual cognitive advantages are not manifested in children with LLI, either because their development depends on language proficiency or because they overlap with areas of cognitive weakness in children with LLI. A recent result from English-Hebrew bilingual children (Iluz-Cohen & Armon-Lotem Reference Iluz-Cohen and Armon-Lotem2013) may provide preliminary evidence for this. Preschool-aged participants in the study were classified according to their proficiency in both languages, and children classified as having low proficiency in both languages performed below peers with high proficiency in one or both languages on nonlinguistic assessments of cognitive control. It is important to note, however, that the authors used the term ‘low language proficiency’ rather than LLI to describe the participants and that monolingual children were not included in the study. It is clear that more work is needed to establish the cognitive characteristics of bilingual children with LLI.

We have now considered the linguistic profiles of bilingual children affected by LLI as well as questions of language growth, dominance, and cognition. With this enhanced cognitive, linguistic, and developmental profile established, we shift to the clinical implications of this information. As with monolingual children with LLI, for bilingual children with LLI timely identification, comprehensive assessment, and effective intervention are needed to ameliorate negative long-term effects. The key difference, of course, is that monolingual children with LLI need one language to be successful in their different communicative environments and bilingual children with LLI need two.

5.1 Identification: clinical markers

Accurate identification of LLI may be difficult, even for children learning only a single language. For children learning two languages, the difficulty may be greatly magnified (see Armon-Lotem, de Jong & Meir Reference Armon-Lotem, de Jong and Meir2015; Kohnert Reference Kohnert2013 for discussion). One approach to improving diagnostic accuracy is to identify clinical markers, or tasks that are both sensitive and specific to LLI across different developmental stages and languages. For monolingual children with LLI, three major candidates for clinical markers have emerged: poor accuracy on NWR; poor repetition of sentences (called sentence repetition, sentence imitation, or recalling sentences); and omission of tense-marking morphemes (see Archibald & Joanisse Reference Archibald and Joanisse2009). All three of these characteristics have been extended to the bilingual LLI population, albeit in slightly different ways.

The omission of tense-marking morphemes is a characteristic of LLI that is largely restricted to monolingual English-speaking populations. Indeed, as discussed earlier, there is some evidence that typically developing L2 learners of English exhibit the same grammatical characteristics (e.g., Paradis Reference Paradis2005), and that specific grammatical profiles of LLI differ cross-linguistically. However, there have been several recent investigations of the validity of specific grammatical markers for the identification of LLI in L2 learners in both English and a variety of other languages. Based on narrative data in Dutch from 7- to 9-year-old children, Verhoeven et al. (Reference Verhoeven, Steenge and van Balkom2011) argued that omission of an agreement marker in the third-person singular verb form is a possible clinical marker of LLI in both monolingual and bilingual children learning the language (cf., Blom et al. Reference Blom, De Jong, Orgassa, Baker and Weerman2013). The conclusion was based on four-way group comparisons of error rates between monolingual children with and without LLI as well as bilingual children learning L2 Dutch. Both groups with LLI made more of this type of agreement error than their unaffected peers, whereas the same pattern was not evident on other types of errors (such as past tense marking). In a similar vein, Rothweiler, Clahsen and colleagues (Rothweiler, Chilla & Clahsen Reference Rothweiler, Chilla and Clahsen2012; Clahsen et al. Reference Clahsen, Rothweiler, Sterner and Chilla2014) have argued that subject-verb agreement errors constitute a marker of LLI in both monolingual and bilingual (L2) learners of German, based on the longitudinal development of grammar in small groups of German-speaking children with LLI.

Grammatical markers of LLI in L2 learners of English have also been considered, but results have differed according to developmental stage of the participants and the language pairs studied. For preschool children aged 3 to 6 years learning Cantonese and English in Malaysia, MLU and an overall rating of syntactic development (Index of Productive Syntax, or IPSyn) from spontaneous language samples showed promise in discriminating children with and without LLI (Ooi & Wong Reference Ooi and Wong2012). For school-age children (aged 8 to 10) living in New York City, USA, and acquiring Spanish and English, failure to mark past tense verbs characterized a group of children with LLI (compared to younger, typically developing bilingual children; Jacobson & Livert Reference Jacobson and Livert2010). Further work with the same population suggested that word and morpheme omissions in both languages (Jacobson & Walden Reference Jacobson and Walden2013) predicted children's LLI status.

Though the search for grammatical markers of LLI in bilingual children is an active area of research, it is clear that there are a number of challenges to be met. First, the literature to date is largely based on small samples of children, as well as on group-level comparisons; fewer studies have included individual diagnostic accuracy measures (see Ooi & Wong Reference Ooi and Wong2012; Morgan, Restrepo & Auza Reference Morgan, Restrepo and Auza2013 for examples). In some cases, application of diagnostic accuracy measures has led to the opposite conclusion from group-level studies (Blom et al. Reference Blom, De Jong, Orgassa, Baker and Weerman2013). More work is needed to establish diagnostic accuracy, even for markers identified as promising. Moreover, it appears likely that not only will specific markers need to be identified and verified for each language, there may also be unique markers for different ages (cf., Jacobson & Livert Reference Jacobson and Livert2010; Ooi & Wong Reference Ooi and Wong2012), for different language pairs (even given the same L2), and for bilingual versus monolingual children (Morgan et al. Reference Morgan, Restrepo and Auza2013). Thus, while it is important to understand grammatical characteristics of sequential bilingual learners with and without LLI, the pursuit of clinical markers may face fewer challenges in areas less susceptible to variation in experience.

The remaining two proposed markers, NWR and sentence repetition, have the potential to be more universal because a single criterion (i.e., poor repetition accuracy) would be used across languages. Moreover, both NWR and sentence repetition have been classified as language-based processing tasks (e.g., Elin Thordardottir & Brandeker Reference Thordardottir and Brandeker2013). Such tasks are designed to reflect real-time proficiency in the learning, recall, or manipulation of either novel or very familiar language stimuli, and therefore to draw less heavily on accrued linguistic knowledge (Thordardottir & Brandeker Reference Thordardottir and Brandeker2013). In theory, such tasks could level the playing field for bilingual and monolingual learners. In particular, both NWR and sentence repetition may tap aspects of memory related to language learning (Ebert Reference Ebert2014), reducing the emphasis on accrued linguistic knowledge.

As we have discussed, there is indeed a substantial literature demonstrating poor performance on NWR tasks for sequential bilingual children with LLI. Furthermore, a number of studies on NWR have extended beyond group-level comparisons to individual diagnostic accuracy measures. NWR was identified as promising in a recent meta-analysis of the diagnostic accuracy of various language tasks for the identification of LLI in Spanish-English bilingual children in the US (Dollaghan & Horner Reference Dollaghan and Horner2011); positive and likelihood ratios for the task (+ 9.0 and 0.2, respectively) indicate good diagnostic utility for the task. In Canada, Paradis and colleagues (Paradis, Schneider & Sorenson Duncan Reference Paradis, Schneider and Sorenson Duncan2013) found NWR performance in the L2 to discriminate between bilingual children with and without LLI; children in this study spoke a wide variety of L1s (including Assyrian, Farsi, Punjabi, and Somali) bolstering the diagnostic utility of NWR across a diverse set of language learning situations.

There is also an emerging literature on the diagnostic utility of sentence repetition for bilingual children (e.g., Komeili & Marshall Reference Komeili and Marshall2013; Ziethe, Eysholdt & Doellinger Reference Ziethe, Eysholdt and Doellinger2013). In a retrospective analysis of language tasks for mono- and bilingual children in Germany, Ziethe and colleagues (Reference Ziethe, Eysholdt and Doellinger2013) found significant group differences in sentence repetition performance between groups with and without LLI; the authors argue that sentence repetition shows promise as a marker of LLI in bilingual as well as monolingual children, though the study did not provide effect sizes or diagnostic accuracy calculations. Thordardottir & Brandeker (Reference Thordardottir and Brandeker2013) used a French sentence repetition test to identify LLI in 5-year-old children living in Canada. Study groups included monolingual French-speaking children with and without LLI as well as bilingual children (learning French as the L2) with and without LLI. Sentence repetition showed good sensitivity to LLI for the monolingual (93%) and bilingual (92%) children as well as adequate specificity for the monolingual children (86%); however, specificity within the bilingual group was poor at 57%. In other words, this study found that typically developing bilingual children may be falsely identified as having LLI when only sentence repetition is used.

This last finding highlights an important caveat in the literature on the use of NWR and sentence repetition to identify bilingual children with LLI. Both tasks have shown some promise for accurate identification of bilingual children with LLI and, as language-based processing measures, both have the potential to reduce bias in assessment by reducing emphasis on past experience. However, it is important to emphasize that these tasks appear to be less influenced by previous linguistic experience, rather than un-influenced by this experience. In the same study, Thordardottir & Brandeker (Reference Thordardottir and Brandeker2013) found that, while sentence repetition and NWR were less influenced by language-specific experience than vocabulary scores, performance on these tasks was still significantly related to the percentage of time a child spent using the language of the task. Similar findings of lower performance for bilingual children than monolingual peers (with and without LLI) can be found for Finnish-Swedish bilingual kindergartners on sentence repetition (Westman et al. Reference Westman, Korkman, Mickos and Byring2008) and for Spanish-English bilingual children on NWR (Windsor et al. Reference Windsor, Kohnert, Lobitz and Pham2010). At present, strong performance on these tasks may effectively rule out LLI for typical bilingual children (Kohnert et al. Reference Kohnert, Yim, Nett, Kan and Duran2005; Gutiérrez-Clellen & Simon-Cereijido Reference Gutiérrez‐Clellen and Simon‐Cereijido2010; Windsor et al. Reference Windsor, Kohnert, Lobitz and Pham2010) but poor performance may need to be combined with other indicators to provide an unequivocal positive identification of LLI in bilingual children. Even for the most studied group of bilingual children with LLI, Spanish-English speakers in the United States, Dollaghan & Horner's (Reference Dollaghan and Horner2011) meta-analysis found no single measure to be optimal for separating children with LLI from typical peers.

Continued research in the area of clinical makers and the identification of LLI in bilinguals could provide new options. For example, a recent international initiative in Europe aimed to develop a set of LLI identification tasks across a wide array of languages (see Armon-Lotem et al. Reference Armon-Lotem, de Jong and Meir2015 for a summary of this work), with promising candidates identified. However, it must be acknowledged that perhaps no single task will ever serve as an adequate clinical marker in LLI, given the heterogeneity of children with LLI as well as developing bilinguals. Creating composite clinical markers – i.e., groups of tasks, perhaps implemented in both languages, which jointly possess adequate sensitivity and specificity – may be a more valid approach for identifying LLI in bilingual children. It is possible that identification composites could include nonlinguistic tasks as well; such assessments do completely strip away the effects of linguistic experience and have shown promise in improving identification of LLI in diverse groups (Kohnert et al. Reference Kohnert, Windsor and Ebert2009), but much more work is needed to establish these tasks.

5.2 Alternatives to clinical markers in the identification of LLI

It may also be that an alternative to clinical markers is a better approach to the identification of LLI. In monolingual speakers, performance on standardized, norm-referenced assessments has been proposed as the ‘gold standard’ for identification of LLI (Tomblin et al. Reference Tomblin, Records, Buckwalter, Zhang, Smith and O'Brien1997) and is commonly used in clinical scenarios. However, this is not yet the case for sequential bilingual children. There has unquestionably been progress in developing norm-referenced assessments for specific, high-density groups (e.g., in the US, creation of the Bilingual English-Spanish Assessment, Peña et al. Reference Peña, Gillam and Bedore2014). However, the tremendous heterogeneity of sequential bilingual children combined with the resources necessary for adequate standardization of these instruments may render this approach impractical on a universal level. Instruments normed on monolingual populations do have value in ruling out LLI, as bilingual children who are able to score within monolingual norms in one or more languages are not affected by LLI. Use of monolingual instruments to positively identify LLI, or to characterize a bilingual child's linguistic strengths and weaknesses, is a dubious (albeit common) clinical practice (De Lamo White & Jin Reference De Lamo White and Jin2011; Williams & McLeod Reference Williams and McLeod2012).

Instead, at present the ‘gold standard’ for identification of bilingual children with LLI appears to be clinician, teacher, and parent concern with language development (e.g., Peña, Gillam & Bedore Reference Peña, Gillam and Bedore2014). Many of the studies reviewed thus far have used clinician judgment (e.g., Peña et al. Reference Peña, Gillam and Bedore2014), identification through the educational system (e.g. Engel de Abreu et al. Reference Engel de Abreu, Cruz-Santos and Puglisi2014), and/or parent concern with language development (e.g., Gutiérrez-Clellen et al., Reference Gutiérrez-Clellen, Simon-Cereijido and Wagner2008) to identify bilingual children with LLI for their study. In addition, there is literature to support the diagnostic accuracy of parental concern (Restrepo Reference Restrepo1998; Paradis et al. Reference Paradis, Schneider and Sorenson Duncan2013). For example, although Paradis et al. (Reference Paradis, Schneider and Sorenson Duncan2013) found NWR to have good diagnostic accuracy in the diverse group of Canadian bilingual children referenced above, a parental questionnaire regarding language development had better diagnostic accuracy than NWR or any other single language measure. The diagnostic utility of parental concern and clinician judgment should not be discounted.

There is one final, important alternative to clinical markers and norm-referenced assessments in the identification of LLI for sequential bilingual children. Dynamic assessment is a cross-disciplinary evaluation approach based in Vygotsky's learning theory (Lantolf Reference Lantolf2009). It focuses on the process rather than the product of learning, potentially isolating the construct of interest in LLI (i.e., language learning ability) (e.g., Kapantzoglou, Restrepo & Thompson Reference Kapantzoglou, Restrepo and Thompson2012; Martin Reference Martin2012). Though a variety of specific techniques may fall under the umbrella term dynamic assessment, all integrate an aspect of teaching into the assessment. The primary outcome of the assessment may be a comparison of the child's performance before and after the teaching phase (e.g., Peña, Iglesias & Lidz Reference Peña, Iglesias and Lidz2001); the examiner's rating of modifiability, meaning how easily the child responds to the teaching (Kapantzoglou et al. Reference Kapantzoglou, Restrepo and Thompson2012); an assessment of the child's learning strategies or motivation (e.g., Ukrainetz et al. Reference Ukrainetz, Harpell, Walsh and Coyle2000); or a combination of these measures.

This approach is gaining increasing attention for its ability to accurately identify bilingual children with LLI (e.g., Peña et al. Reference Peña, Iglesias and Lidz2001; Kapantzoglou et al. Reference Kapantzoglou, Restrepo and Thompson2012; Martin Reference Martin2012; Hasson et al. Reference Hasson, Camilleri, Jones, Smith and Dodd2013; Peña, Gillam & Bedore Reference Peña, Gillam and Bedore2014). Within the United States, investigations of dynamic assessment for bilingual children with LLI have focused on word learning and story-telling skills. For example, recent work by Peña and colleagues (Reference Peña, Gutiérrez-Clellen, Iglesias, Goldstein and Bedore2014) shows that, when bilingual children are given models and explicit instruction in how to construct high-quality narratives in the L2, typically developing children respond better to the intervention than those with LLI. Clinician ratings of modifiability robustly differentiated between the two groups. In the UK, a structured dynamic assessment tool for preschoolers (Hasson et al. Reference Hasson, Camilleri, Jones, Smith and Dodd2013) has been developed. The assessment contains teaching phases for expressive vocabulary, phonology and sentence structure, and initial work shows promise in differentiating bilingual children with and without LLI (Hasson et al. Reference Hasson, Camilleri, Jones, Smith and Dodd2013). Dynamic assessment appears to be a promising approach, though additional work is needed to test it for learners of L2s other than English, to validate it in larger participant samples, and to verify its accuracy in more typical clinical conditions (including limited testing time and administration by less experienced clinicians).

5.3 Complete assessment

Thus far we have focused primarily on the question of accurate identification of sequential bilingual children with LLI rather than the necessary components of a complete assessment for this population. Our focus has reflected the state of the literature, which also emphasizes accurate identification. But identification is only the first step in clinical service delivery for LLI. A complete assessment will likely include many of the tasks we have already discussed, including interviews or questionnaires to collect parent and teacher views, processing-based tasks such as NWR, static assessments of language knowledge such as standardized tests, and perhaps dynamic assessment procedures as well. An additional important component of a complete assessment is the collection of language samples, such as narratives. Language samples are an ecologically valid task (Costanza-Smith Reference Costanza-Smith2010) and may provide a supportive, familiar context for eliciting a valid view of a child's language production abilities (Cleave et al. Reference Cleave, Girolametto, Chen and Johnson2010). Language samples have already been highlighted in our previous discussion, as a number of the findings regarding dual-language deficits and potential grammatical markers of LLI have stemmed from analyses of language samples. Here we reiterate their clinical utility as an assessment tool; in a single assessment task, they provide a means of analyzing not only grammatical errors but also overall linguistic productivity, verbal fluency, lexical development, and even code-switching between languages (Gutiérrez-Clellen et al. Reference Gutiérrez-Clellen, Simon-Cereijido and Wagner2008; Bedore et al. Reference Bedore, Peña, Gillam and Ho2010). Some evidence suggests that assessment results may differ between standardized tests and language samples (Cleave et al. Reference Cleave, Girolametto, Chen and Johnson2010; Ebert & Scott Reference Ebert and Scott2014), suggesting that they provide a unique source of information in assessment.

Perhaps most importantly, there is a consensus that a complete assessment must consider both languages of the developing bilingual child (Gutiérrez-Clellen & Simon-Cereijido Reference Gutiérrez‐Clellen and Simon‐Cereijido2010; De Lamo White & Jin Reference De Lamo White and Jin2011; Kohnert Reference Kohnert2013). It is clearly unrealistic to expect that a bilingual clinician, fluent in the individual child's L1 and L2, will be available for each assessment of a bilingual child. Instead, consideration of a minority L1 in assessment could take a variety of forms, including detailed interviews with parents and other family members regarding development and current skills in the home language (Paradis, Emmerzael & Sorenson Duncan Reference Paradis, Emmerzael and Sorenson Duncan2010); collaboration with interpreters or bilingual support personnel to directly assess L1 performance (Guiberson & Atkins Reference Guiberson and Atkins2012); or administration of select measures that may be prerecorded by fluent L1 speakers and interpreted by a monolingual clinician (e.g., a sentence comprehension task with a pointing response; an NWR task).

6.1 Benefits of dual-language support for sequential bilingual children with LLI

For children who have received systematic input in a minority home language and are subsequently identified as having LLI, the question of whether to treat one language or two is practically manifested as whether or not to actively treat the minority L1. That is, because there tend to be few clinicians who can provide direct support in the child's L1 and because educational systems tend to emphasize the L2, the majority L2 receives support and the question becomes whether the minority L1 can also be addressed. Numerous rationales for actively supporting the L1 have now been identified. The first reflects the communicative needs of sequential bilingual children. Systematic exposure to a minority L1 implies that the child is a member of a family and possibly a broader community that uses the L1. If the child shifts towards exclusive use of the L2, communication breakdowns with family and community become likely. Though direct evidence from children with LLI is lacking, the broader literature on sequential bilingual language development supports the conclusion that loss of proficiency in the L1 has a host of negative psychosocial consequences (e.g., Phinney et al. Reference Phinney, Romero, Nava and Huang2001). The need for L1 proficiency, combined with the lack of evidence that L1 growth detracts from L2 proficiency, lays a solid foundation for bilingual rather than monolingual treatment for this population.

In fact, some recent literature has suggested that bilingualism is actually beneficial for children with LLI, both in linguistic and nonlinguistic areas. It is possible that exposure to greater linguistic complexity – achieved through experience with two distinct linguistic systems – could enhance language learning for bilingual children with LLI (Roeper Reference Roeper2012). Armon-Lotem (Reference Armon-Lotem2010) offers a summary of evidence that would support this conclusion; the grammatical errors made by bilingual children with LLI have in some cases suggested greater linguistic knowledge than their monolingual impaired peers, perhaps because the bilingual children have been able to draw on knowledge of their L1. Armon-Lotem (Reference Armon-Lotem2010) concludes that bilingualism may in fact be facilitative for children with LLI. Empirical validation of this interesting perspective is an important area for future investigation.

The possibility of bilingual cognitive advantages must also be considered in a discussion of maintaining bilingualism in children with LLI. As previously discussed, bilingual advantages in attentional control and metalinguistic awareness have largely not yet been explored in children with LLI. Should these advantages extend to children with LLI and thus mitigate the cognitive and metalinguistic weaknesses associated with LLI, there would be yet another clear argument for supporting bilingual development in children with LLI.

6.2 Challenges to dual-language support in LLI

Despite the benefits of dual-language support for children with LLI, it is clear that the recommendation is not always adopted by clinical practitioners. Survey data from several countries confirms that the predominant clinical practice remains treatment in one language only (e.g., Jordaan Reference Jordaan2008; Williams & McLeod Reference Williams and McLeod2012). For example, a recent international survey of speech-language pathologists across 13 countries (Jordaan Reference Jordaan2008) showed that bilingual children were receiving monolingual treatment in 87% of cases. Reasons cited for this were a combination of relatively few bilingual service providers (74% of respondents were monolingual) and perceived pressure to use the language of education and the language of the community in order to succeed. Thus, professional resistance appears to be one challenge in the implementation of dual-language treatment for children with LLI. Continuing to educate clinical practitioners – and to provide them with high-quality evidence of the efficacy of dual-language services – remains a crucial priority for the field.

Additional practical challenges exist in the provision of dual-language services for children with LLI. First, if ‘dual-language services’ is interpreted as the provision of direct treatment services in both the L1 and the L2, there is a gross mismatch between the workforce of clinical practitioners and the client base in most countries. That is, the majority of qualified speech-language professionals are monolingual in many countries; even in locations with greater numbers of bilingual service providers, it remains very difficult to match the diversity of minority languages with qualified providers (see Jordaan Reference Jordaan2008; Pham, Kohnert & Mann Reference Pham, Kohnert and Mann2011 for discussion). In other cases, policies preventing direct treatment in a home language may exist, particularly in school-based settings.

6.3 Empirical studies of treatment for bilingual children with LLI

Thus far, our discussion of treatment for bilingual children with LLI has focused on logical arguments and practical obstacles surrounding the language or languages to be targeted in treatment. There is a small but growing empirical literature that directly examines the results of treatments provided to bilingual children with LLI. In reviewing this work, we classify studies based on the measures of treatment progress they employed. Until very recently, treatment studies with bilingual children with LLI measured progress in terms of learning on the linguistic stimuli that were trained in treatment sessions. Most commonly, participants were taught a set of words, and progress was measured in terms of performance on probes of the targeted words. We discuss conclusions from this foundational work first, before considering studies that have used more generalized measures of progress – such as omnibus measures of language, measures derived from spontaneous language samples, and measures of domains not directly targeted in treatment.

6.3.1 Learning of trained stimuli in the L1 and the L2

Table 1 summarizes six published studies that have examined gains on trained stimuli made by sequential bilingual children with LLI within a language treatment program. The table highlights both diversity in the literature (in terms of the participant ages, the delivery method and intensity of treatment, and the treatment methods) and lack thereof (most studies have included Spanish-English bilinguals; vocabulary has been the most common target; and, until very recently, small-group or single-subject studies were the norm). The earliest studies (Perozzi Reference Perozzi1985; Perozzi & Sanchez Reference Perozzi and Sanchez1992; Elin Thordardottir, Ellis Weismer & Smith Reference Thordardottir, Ellis Weismer and Smith1997) targeted vocabulary and reported learned words within the L2 only. Although the methods of instruction differed notably, the conclusion of all three studies was consistent: L2 vocabulary can be acquired as quickly or even more quickly when the words are taught in both the L1 and the L2 (i.e., bilingually) as compared to being taught in one language only. This finding is notable because it contradicts the assumption that the amount of time spent working in the L2 will directly correlate with learning in the L2. In all three studies, the amount of treatment received was held constant across the bilingual and L2-only treatment programs. Thus, less time was spent working on the L2 in the bilingual treatment conditions. This conclusion appears to apply to preschoolers (Perozzi Reference Perozzi1985; Elin Thordardottir et al. Reference Thordardottir, Ellis Weismer and Smith1997) and to young school-aged children (Perozzi & Sanchez Reference Perozzi and Sanchez1992).

Table 1

Intervention studies for sequential bilingual children with LLI examining trained stimuli only

Note. Ind = Individual, min = minute, TX = treatment, BI = bilingual, NK = not known. For all studies, N includes only participants who were sequential bilingual children with LLI. Age is reported as year; month.

^a Studies marked with ^a included additional participants (not listed here) who showed typical language development or who spoke only one language.

^b Divided over 4 sub-studies: 1 = 10 sessions in 2 weeks; 2 = 12 sessions in 1 month; 3 = 12 sessions in 1 month; 4 = 11 sessions in 2.5 weeks.

^c The Gutiérrez-Clellen & Simon-Cereijido studies summarized in Table 2 report on the same treatment program and an overlapping participant sample.

More recent studies of vocabulary treatment for bilingual children with LLI have replicated findings of comparable gains in the L2 between L2-only and bilingual treatments; however, these studies have also extended the measurement of treatment gains to the L1. In a large group study of Spanish-English preschoolers with LLI, Restrepo, Morgan & Thompson (Reference Restrepo, Morgan and Thompson2013) compared four treatment conditions crossing content (vocabulary versus math) with language of delivery (bilingual versus L2-only). The vocabulary treatment was delivered within the classroom by trained teachers, and incorporated dialogic book reading and interactive activities supported by techniques such as rich definitions, scaffolding conversations, expanding, and modeling. This vocabulary instruction program was more effective than the math curriculum in both languages. A comparison of L2 outcomes of the bilingual and L2-only vocabulary programs echoed previous literature: outcomes were not statistically different between the programs despite the difference in time spent working on L2 words. In the L1, however, the group that received bilingual vocabulary instruction learned significantly more Spanish words; for example, on the receptive vocabulary probe, children in the bilingual vocabulary group averaged a gain of about 14 words whereas those in the remaining treatment groups averaged 7–8 word gains (out of a possible 45 words).

Pham et al. (Reference Pham, Kohnert and Mann2011) also measured both L1 and L2 gains in a single-subject- design study investigating receptive vocabulary learning in a Vietnamese-English bilingual preschooler. Treatment in this study was delivered in brief (ten-minute) sessions conducted through a combination of computerized audio recordings and visual stimuli with direct support from the classroom teacher; this procedure allowed the teacher to address an L1 she did not speak. Consistent with other studies reviewed here, treatment of words in L2-only and bilingually resulted in comparable accuracy in the L2 (with the participant requiring about five treatment sessions to master five to eight new words). Gains were also seen in the L1; in fact, there was evidence that some words treated in L2-only were understood in the L1 after treatment. The authors suggest that increased attention to task resulting from the treatment may have contributed to this generalization.

The final study in Table 1 (Motsch & Riehemann Reference Motsch and Riehemann2008) is unlike the others in several respects. Instead of vocabulary, treatment in this study targeted morphosyntax, specifically case-marking in German. Perhaps more importantly, the study was not restricted to bilingual children; only about half of the participants were bilingual, and all assessment and treatment took place in the L2, German. However, the study does contain an important finding for the treatment of bilingual children with LLI: there was no interaction between bilingual status and rate of change in the treatment program. Bilingual children gained about 18.5 percentage points in their accuracy of marking the accusative case and 14.6 percentage points for the dative case, in comparison to 28.0 and 13.2 percentage-point average gains for monolingual children. Thus, these school-aged sequential bilingual children with LLI made statistically equivalent progress in treatment targeting L2 morphosyntax as did their monolingual peers, though it should be noted that they began treatment with lower levels of accuracy in grammatical case-marking (in the L2) than their monolingual peers.

The literature on learning of trained linguistic stimuli by bilingual children with LLI clearly supports the conclusion that bilingual vocabulary instruction is at least as effective as L2-only instruction at teaching words in the L2. A variety of treatment approaches appear to be effective in teaching vocabulary, including drill to associate words with picture stimuli (Perozzi Reference Perozzi1985; Perozzi & Sanchez Reference Perozzi and Sanchez1992; Pham et al. Reference Pham, Kohnert and Mann2011), semi-structured play with focused stimulation (Elin Thordardottir et al. Reference Thordardottir, Ellis Weismer and Smith1997), and classroom-based dialogic book reading and interactive instruction (Restrepo et al. Reference Restrepo, Morgan and Thompson2013). Recent results indicate that bilingual vocabulary treatment is more effective for learning words in the L1 than L2-only treatment (Pham et al. Reference Pham, Kohnert and Mann2011; Restrepo et al. Reference Restrepo, Morgan and Thompson2013). Generalization from the L2 to the L1 may be possible, but evidence to date is limited to picture identification contexts (Pham et al. Reference Pham, Kohnert and Mann2011). Finally, school-aged sequential bilinguals receiving morphosyntactic instruction in the L2 acquire the trained L2 morphosyntactic forms, and do so at the same rate as monolingual peers (Motsch & Riehemann Reference Motsch and Riehemann2008).

6.3.2 Generalized outcomes of language treatment

The acquisition of trained linguistic forms is a crucial foundation for any language treatment program. Given the number of linguistic forms to be learned, however, an optimal language treatment program will have a broader impact on children's language knowledge or even their language learning ability. Thus, treatment typically must generalize beyond the trained stimuli in order to have a significant impact on daily communication. The need for generalization is even greater for sequential bilingual children, who must learn the forms of two language systems to communicate successfully. Within the past few years, a handful of publications have addressed the question of whether sequential bilingual children with LLI can generalize beyond trained stimuli by reporting generalized outcomes of language treatment programs. These studies are summarized in Table 2.

Table 2

Intervention studies examining outcomes outside of trained stimuli

Notes. Table summarizes published intervention studies that measured effects outside of trained stimuli.

Multiple publications from the same set of participants are summarized in a single row (e.g., Ebert, Kohnert, et al. Reference Ebert, Pham and Kohnert2014; Pham et al. Reference Pham, Ebert and Kohnert2015).

BI = bilingual, ctrl = control, LLI = language learning impairment, MLU = mean length of utterance, NDW = number of different words, SLP = speech-language pathologist, std = standard, TD = typically developing, TX = treatment.Age is reported as year; month.

^a Participant sample overlaps with the one reported in Restrepo et al. Reference Restrepo, Morgan and Thompson2013.

^b All children were reported to be 4 years of age; no mean or range given.

^c Data from children with LLI not subjected to hypothesis testing. Reported results stem from effect sizes and qualitative analysis, for limited group (N = 20)

With one exception, the studies in Table 2 report results from two large treatment projects conducted with Spanish-English sequential bilingual children within the US. The first project, the VOLAR program (Gutiérrez-Clellen et al. Reference Gutiérrez-Clellen, Simon-Cereijido and Sweet2012; Simon-Cereijido et al. Reference Simon-Cereijido, Gutiérrez-Clellen and Sweet2013; Simon-Cereijido & Gutiérrez-Clellen Reference Simon-Cereijido and Gutiérrez-Clellen2014), has compared L2-only and bilingual treatment programs using academic enrichment activities to target vocabulary and overall language growth. The treatment program is the same as the vocabulary treatment condition reported in Restrepo et al. (Reference Restrepo, Morgan and Thompson2013) but additional language outcomes are reported in these works. Gutiérrez-Clellen et al. (Reference Gutiérrez-Clellen, Simon-Cereijido and Sweet2012) and Simon-Cereijido et al. (Reference Simon-Cereijido, Gutiérrez-Clellen and Sweet2013) report predictors of overall L1 and L2 growth in 188 four-year-old children with LLI, all of whom began the treatment program with little to no proficiency in the L2 (English). Measures of language growth included production of target words in a picture description task, measures of utterance length (MLU) and lexical diversity in narrative language samples, and a sentence repetition task (reported in the L1 only). Growth in L2 MLU was predicted by the language of intervention; that is, children who received bilingual intervention made greater gains in L2 MLU with a large effect size (d = 1.79 for MLU in words). In addition, children who began with higher levels of L1 proficiency (as measured by L1 MLU) showed more rapid growth in L2 MLU; again, this effect was large (d = 0.89). The interaction between treatment program (bilingual vs. L2-only) and initial L1 MLU approached significance, as the difference in growth rate (faster gains for those with higher initial L1 MLUs) was greater in the L2-only condition. Assuming that L1 MLU is a good marker of overall linguistic development in these L1-dominant preschool children, this result suggests that those with less severe language impairment were better able to take advantage of L2 input. In the L1, however, very limited differences between the bilingual and L2-only intervention programs were identified. Instead, children continued to grow in their L1 skills during the nine weeks of intervention and the five months following intervention, regardless of the program of intervention.

Finally, the same program has now been implemented in school-readiness classrooms for at risk children, which include both typically developing preschoolers and those with LLI (Simon-Cereijido & Gutiérrez-Clellen Reference Simon-Cereijido and Gutiérrez-Clellen2014). This work has compared the experimental VOLAR treatment program (offering bilingual vocabulary and general language support) to a ‘business as usual’ preschool curriculum that offers minimal language support and, although bilingual to some extent, primarily uses the L1 for non-instructional time. Although results for the group of children with LLI have not yet been subjected to hypothesis testing (Simon-Cereijido & Gutiérrez-Clellen Reference Simon-Cereijido and Gutiérrez-Clellen2014), the results for the combined group (of children with and without LLI) remain supportive of the bilingual intervention program: lexical diversity in both the L1 and the L2 improved more for children who received the experimental treatment program than for those who received the ‘business as usual’ curriculum, with medium effect sizes (d = 0.44 and d = 0.43 for the number of different words in the L1 and the L2, respectively). MLU did not differ between the groups in either language; average MLU for both the experimental and control groups increased over time in both languages.

The second project was conducted with 59 school-aged children living in the Midwestern United States (Ebert et al. Reference Ebert, Pham and Kohnert2014; Pham, Ebert & Kohnert Reference Pham, Ebert and Kohnert2015). As in other studies, L2-only treatment was compared to a bilingual treatment program. However, a third treatment condition within this project targeted cognitive processing skills such as attention and speed of processing in the nonlinguistic domain. This treatment condition explored the hypothesis that subtle impairments in nonlinguistic cognitive processing contribute to language learning deficits in LLI (see Ebert et al. Reference Ebert, Pham and Kohnert2014). The nonlinguistic cognitive processing treatment was piloted in a single case experimental design study (Ebert, Rentmeester-Disher & Kohnert Reference Ebert, Rentmeester-Disher and Kohnert2012) in which it showed effects in both the L1 and the L2.

In the three-treatment group study (Ebert et al. Reference Ebert, Pham and Kohnert2014) analyses immediately following treatment showed that children in all three active treatment groups made significant gains in the L2 and some gains in nonlinguistic cognitive processing skills, whereas children in a delayed treatment control group did not. The gains in the L2 were comparable between the L2-only and bilingual treatment groups, with no statistically significant differences and roughly similar effect sizes (averaging d = 0.43 across the L2 outcome measures). In the L1, however, gains were smaller across groups. Effect sizes for the active treatment groups averaged d = 0.18 across L1 outcome measures, and statistically significant gains were seen only by children who received bilingual treatment. Three months following the completion of treatment (Pham et al. Reference Pham, Ebert and Kohnert2015), children continued to show growth on L2 measures and maintenance of the L1 (new growth in the L1 was limited). Though children in the bilingual treatment condition continued to show comparable L2 growth to those in the L2-only condition, their advantage in L1 growth was no longer significant at this follow up.

The final study in Table 2 summarizes a recent group treatment study involving preschool children (aged 4 to 6) living in Montreal (Elin Thordardottir et al. Reference Thordardottir, Cloutier, Ménard, Pelland-Blais and Rvachew2015). Participants in the study spoke a variety of home languages (including Arabic, Japanese, Punjabi, and Tamil) and were learning French as the L2. A total of 29 participants were divided into three groups: one received treatment delivered by a certified speech therapist in the L2 only; one received treatment delivered in the L2 by a certified speech therapist and in the L1 by a parent, within the same session; and one served as a delayed treatment control. Both active treatment groups received a focused stimulation approach targeting vocabulary and syntactic structures, in which targets are modeled repeatedly within meaningful contexts (see Ellis Weismer & Robertson Reference Ellis Weismer, Robertson, McCauley and Fey2006). Outcome measures in this study spanned each of the categories discussed here (i.e., trained linguistic stimuli and generalized linguistic outcomes). On probes of trained vocabulary items in the L2, both treatment groups made more gains than the delayed treatment control, with group membership explaining 46.3% of variance in receptive vocabulary gains and 53.2% of variance in expressive vocabulary gains. On more generalized measures of L2 growth (including a score for the inclusion of grammatical constituents in a story retell task and a set of French language standardized tests), all children made gains but those in active treatment did not significantly outpace those in the delayed treatment control.

Finally, there were no significant gains in MLU for either language. Thus, the L2-only and bilingual interventions were indistinguishable across all outcomes. Though on the surface this result appears contradictory to the previous findings that L1 gains can be achieved through bilingual treatment (Elin Thordardottir et al. Reference Thordardottir, Ellis Weismer and Smith1997; Simon-Cereijido et al. Reference Simon-Cereijido, Gutiérrez-Clellen and Sweet2013; Ebert et al. Reference Ebert, Pham and Kohnert2014), the treatment programs were not comparable in all respects. The L2 intervention was delivered by a trained professional whereas the L1 was supported by parents alone; moreover, the participation of the parents in therapy sessions was reportedly low. As discussed by study authors, it is unclear if the type of parent-participation therapy implemented was a good cultural match for participating parents. In addition, the only index of L1 gains, MLU, did not prove sensitive to gains in either language.

The projects highlighted in this section have taken on the formidable task of demonstrating generalized gains of language treatment in sequential bilingual children with LLI. To some degree, results across the three major projects support similar conclusions. First, gains in the L2 are at least equivalent (Ebert et al. Reference Ebert, Pham and Kohnert2014; Elin Thordardottir et al. Reference Thordardottir, Cloutier, Ménard, Pelland-Blais and Rvachew2015; Pham et al. Reference Pham, Ebert and Kohnert2015) if not superior (Gutiérrez-Clellen et al. Reference Gutiérrez-Clellen, Simon-Cereijido and Sweet2012) for children who receive bilingual treatment in comparison to those who receive L2-only treatment. This message is highly consistent with the literature on learning of trained stimuli and again poses a very strong challenge to the ‘time on task’ hypothesis. Bilingual children with LLI can clearly generalize treatment gains to novel linguistic stimuli and tasks in the L2, and do so after treatment in L2 only, the L1 and the L2, or even treatment that directly targets cognitive processing skills rather than language. Indeed, the latter result highlights the ability of bilingual children with LLI to generalize across domains, from cognition to language. Although more work is needed to determine the conditions under which cross-linguistic and cross-domain generalization is most likely, its occurrence in treatment studies is most consistent with domain-general theories of LLI, which posit direct links between language and cognition, and with theories of bilingual development that highlight relationships across languages (e.g., De Bot et al. Reference De Bot, Lowie and Verspoor2007).

In the L1, however, the interpretation of findings is more complex. It would be expected that children who receive bilingual treatment programs would make larger L1 gains than those who receive no direct treatment in the L1. This prediction has received some empirical support, although results have not always reached statistical significance or been found on all L1 measures. One possible factor in this result is the social context and status of the L1 within the children's family and community. For the two major projects reviewed here that found evidence of superior L1 gains following bilingual treatment, there were notable differences in overall rates of L1 growth: Simon-Cereijido et al. (Reference Simon-Cereijido, Gutiérrez-Clellen and Sweet2013) found significant growth across time for nearly all L1 skills across all groups of participants, whereas Pham et al. (Reference Pham, Ebert and Kohnert2015) found flat rates of L1 growth across measures and treatment conditions. Sociolinguistic factors and age may play a role in this contrast. The preschoolers in Simon-Cereijido et al. (Reference Simon-Cereijido, Gutiérrez-Clellen and Sweet2013) had little to no L2 proficiency at the start of the treatment program, suggesting that they were living in a predominantly L1-speaking environment. In contrast, the school-aged children in Pham et al. (Reference Pham, Ebert and Kohnert2015) attended L2-only schools and were in the process of transitioning to L2 dominance (Ebert et al. Reference Ebert, Pham and Kohnert2014). These very strong social influences may have interacted with the effects of the treatment programs, perhaps obscuring differential effects on L1 growth. More research – particularly work in different sociolinguistic environments and on different language pairs – is needed to clarify these patterns. The results of the third project (Elin Thordardottir et al. Reference Thordardottir, Cloutier, Ménard, Pelland-Blais and Rvachew2015) make clear that not all support of the L1 is automatically sufficient to produce gains in that language.

Regardless, the literature to date does not justify the conclusion that generalization from the L2 to the L1 is automatic. When given L2-only support, or a minimal level of L1 support (Thordardottir et al. Reference Thordardottir, Cloutier, Ménard, Pelland-Blais and Rvachew2015), L1 gains may not be apparent. And, although bilingual children with LLI may be able to make some language gains via improvements in nonlinguistic cognitive processing skills (Ebert, Kohnert et al. Reference Ebert, Pham and Kohnert2014), this mechanism alone may not provide sufficient support for long-term gains in L1 (Pham et al. Reference Pham, Ebert and Kohnert2015). Future work will need to consider the optimal means of supporting the L1, including perhaps a combination of cognitive processing activities and direct treatment of L1 skills. In particular, investigation of effective means of collaboration with parents to support L1 growth is needed. In monolingual children, parent-delivered treatments have been shown to be effective (Roberts & Kaiser Reference Roberts and Kaiser2011). Preliminary treatment evidence with younger simultaneous bilingual children with low vocabulary also shows that parents can implement dialogic reading treatment to bolster L1 lexical skills (Tsybina & Eriks-Brophy Reference Tsybina and Eriks-Brophy2010). For young developing sequential bilinguals, collaboration with parents remains one of the most feasible, logical and ethical means of supporting the L1 (Kohnert et al. Reference Kohnert, Yim, Nett, Kan and Duran2005) and it is crucial that culturally valid collaborative models be developed and tested for treatment effectiveness. Finally, there is a clear need for additional investigations of treatment in children who speak home languages other than Spanish, and in domains other than vocabulary, as the majority of studies to date have focused on these areas.