Abstract
The aim was to evaluate conceptual scoring of lexical organization in bilingual children with language impairment (BLI) and to compare BLI performance with monolingual children with language impairment (MLI). Word associations were assessed in 15 BLI and 9 MLI children. BLI were assessed in Arabic and Swedish, MLI in Swedish only. A number of syntagmatic (semantic link, different word class) and paradigmatic associations (semantic link, same word class) were calculated. Arabic and Swedish scores were compared with a conceptual score (total number of concepts from both languages). For BLI, the paradigmatic conceptual score was significantly higher than single language scores, confirming the distribution of lexical knowledge across languages. The BLI group had significantly higher conceptual paradigmatic scores than the MLI group. Conceptual scoring may reduce the over-identification of language impairment (LI) and underestimation of lexical knowledge in bilingual populations.
There are conflicting views on whether bilingualism negatively affects language development in children with language impairment (LI). The dominant view on typical bilingual language development is that it differs from monolingual development, in that a bilingual person is not an additive combination of two monolinguals (Grosjean, 2008). Bilingual language development has been reported to enhance the child’s cognitive abilities (Adesope, Lavin, Thompson, & Ungerleider, 2010), whereas monolingual children often outperform bilingual peers on vocabulary tests (Bialystok, Luk, Peets, & Yang, 2010). This vocabulary difference is especially noticeable when the bilingual children are assessed in the majority language and are compared with a monolingual sample (Thordardottir, Rothenberg, Rivard, & Naves, 2006). This study aims at evaluating the utility of conceptual scoring and to enhance our understanding of lexical organization in bilingual children with language impairment (BLI). From a long-term perspective, the study may deepen our understanding of the possible interaction between bilingualism and LI within the domain of lexical organization, that is, the interrelationship of words in the mental lexicon.
Lexical Organization
Lexical skills can be categorized into lexical size (e.g., picture identification and picture naming) and lexical organization (word associations and category generation; Meara, 1996). In a discrete word association task, the child is required to name the first word that comes to mind when a stimulus word is presented. The responses are usually categorized as syntagmatic (different word class with a semantic link) or paradigmatic (same word class with a semantic link) associations. In developmental studies, paradigmatic associations are considered more mature and are more frequently found in older children and adults. Paradigmatic associations are interpreted as a result of deeper word knowledge, characterized by knowledge of more semantic features. The ability to associate paradigmatically has been interpreted both within the broader context of general cognitive development and within structural linguistic theories (for an overview, see Nelson, 1977). In a cross-sectional study of word associations in monolingual (Persian and Swedish) and bilingual (Persian-Swedish) children and adults, Namei (2004) found that younger participants produced more phonologically based associations (e.g., onset and rime) than older participants. The finding was interpreted as characteristic of an earlier developmental stage. However, phonologically based associations were present in all groups, indicating that lexical organization is not governed by age alone. Namei (2004) also found that although the bilingual subjects received less exposure to each language, the amount of paradigmatic associations was similar across the bilingual’s languages and between monolingual and bilingual groups. This finding may imply that the ability to associate paradigmatically is driven to a greater extent by general cognitive development than by exposure to a specific language. Indeed, Ordóñez, Carlo, Snow, and McLaughlin (2002) found support for a cross-linguistic relationship regarding paradigmatic organization in a group of Spanish-English bilingual fourth and fifth graders. The participants in the study showed evidence of transfer of paradigmatic knowledge from Spanish to English. A slight bilingual advantage has also been found in a word association study by Sheng, McGregor, and Marian (2006). In a comparison of Mandarin-English bilingual children and English monolingual children, the bilingual group showed enhanced performance in paradigmatic responses, particularly evident for associations of verbs (Sheng et al., 2006). Taken together, these studies show support for lexical organization being governed by general cognitive development and influenced by language-specific input to a lesser extent.
Alternatively, Peña, Bedore, and Zlatic-Giunta (2002) studied the category generation performance in 5- to 8-year-old Spanish-English bilingual children. The authors assessed the ability to generate words to a slot-filler category (associated with function, similar to syntagmatic organization) and taxonomic categories (showing a hierarchical relationship among words, similar to paradigmatic organization). Results showed that the older children showed a preference for taxonomic categories. Moreover, although all children generated an equal number of words in each language, a large proportion of words was language specific and unique to each language (singlets; Peña et al., 2002). This indicates that bilingual children add different words to their different languages and that the words do not necessarily have translation equivalents (doublets). Thus, the study of Peña et al. (2002) stresses also the importance of specific linguistic input as one of the main driving forces in the development of lexical organization skills. To summarize, earlier research has found support for both linguistic and cognitive effects on the lexical organization of bilingual populations.
Distributed Characteristics and Conceptual Scoring
In clinical bilingual language assessment, differences in language exposure must be taken into account. Differing levels of exposure, and innumerable possible language combinations, pose problems to the calculation of reference values in language assessments and to the separation of language impairment from typical language development (Peña, Bedore, & Fiestas, 2013). The development and psychometric evaluation of a new language measure is a time-consuming process, and clinicians with bilingual children on their everyday caseloads are in urgent need of relevant assessment tools. Instead of awaiting bilingual norms, alternative scoring systems, such as conceptual scoring, can be used to better capture the language proficiency of bilingual children (Bedore, Peña, Garcia, & Cortez, 2005).
With exposure to more than one language, the vocabulary of a bilingual child will be distributed across the child’s languages. These “distributed characteristics” of vocabulary knowledge (Oller & Eilers, 2002) are illustrated by the asymmetric vocabularies common especially among young bilingual learners. Whereas children may know words belonging to their home and family life in their first language, they may lack words for school subjects, which are instead known in their second language. When assessed in only one of the languages, these distributed characteristics can result in an underestimation of lexical ability and an over-identification of lexical problems. One way of addressing this issue is to consider the child’s conceptual vocabulary score instead of the single language score. Conceptual scoring takes into account the total number of concepts known by the child, regardless of the language, and awards points to concepts rather than to words. As an example, a child who knows the word “run” in Swedish and Arabic gets credit for one concept only although two different word forms have been produced. Similarly, a child who knows “dog” in Swedish and “frog” in Arabic gets credit for two concepts, although each word form has only been acquired in one language. Conceptual scoring has been used in several studies of vocabulary size in comparisons of monolingual and bilingual children (Bedore et al., 2005; Core, Hoff, Rumiche, & Senor, 2013; Junker & Stockman, 2002; Patterson, 1998; Pearson, Fernandez, & Oller, 1993; Restrepo, Morgan, & Thompson, 2013; Thordardottir et al., 2006). Conceptual scoring has also been used within the context of lexical organization. Sheng, Peña, Bedore, and Fiestas (2012) examined repeated word associations in Spanish-English 7- to 10-year-old bilingual children with and without LI, calculating both single language and conceptual scores. Conceptual scoring yielded higher scores than either language scored separately. Furthermore, conceptual scoring most markedly separated the LI and typically developing groups, indicating the usefulness of conceptual scoring in clinical decision making (Sheng et al., 2012).
Language Impairment in Bilingual Populations
Language impairment may affect the language form, content, and use of a child with typical nonlinguistic abilities, neurological development, and hearing (Bishop, 2006). A child is considered bilingual when receiving and using two or more languages (Grosjean, 2008). A BLI must therefore show deficits in both languages (Salameh, 2003, Thordardottir, 2015). A large number of studies on BLI have focused on grammar (Paradis, 2010; Verhoeven, Steenge, & van Balkom, 2011), but recently studies of lexical abilities in BLI children have been published (Ebert, Pham, & Kohnert, 2014; Sheng, Bedore, Peña, & Taliancich-Klinger, 2013; Sheng et al., 2012). In the studies by Sheng et al. (2012) and Sheng et al. (2013), BLI children showed less developed lexical organization skills in both their languages compared with bilingual children with typical development. They also displayed poor semantic convergence in comparison with bilingual peers with typical development, that is, the BLI children produced fewer associations that were frequent in the normative sample. As an example, one of the most common responses to “dinner” in the normative sample was “breakfast,” whereas the BLI seldom gave this response. Instead, they usually used more general terms such as “food.” A recent longitudinal study of lexical organization in Arabic-Swedish children with LI (Holmström, Salameh, Nettelbladt, & Dahlgren Sandberg, 2015) provides similar results. The BLI children showed no increase in paradigmatic associations, which is in line with the findings of Sheng et al. (2012) and Sheng et al. (2013). For the Arabic-Swedish children with typical development, who served as a comparison group, the amount of paradigmatic associations increased in both Arabic and Swedish children. Greater differences between the groups were seen in Swedish, possibly due to greater exposure to the majority language and to Swedish being used in school (Holmström et al., 2015). Lexical organization may thus be more affected in BLI compared with bilingual peers with typical language development, irrespective of the language combination.
No significant differences have been found between MLI and BLI regarding language performance. Paradis, Crago, Genesee, and Rice (2003) compared the performance of French and English monolingual children with language impairment (MLI) with that of French-English BLI on grammatical competence. The authors found no differences between the groups, indicating that bilingualism did not aggravate the LI. Similarly, Korkman et al. (2012) compared MLI and BLI on a range of language measures. They found no bilingual disadvantage for Swedish-Finnish BLI, although the results were mixed for different language domains, with vocabulary being more negatively affected by bilingualism. Despite these findings, there is a widespread belief among parents and health care staff that children will have greater difficulties learning two languages than one (Nayeb, Wallby, Westerlund, Salameh, & Sarkadi, 2014), especially in combination with language impairment.
Arabic and Swedish
Swedish, with about nine million speakers (Statistics Sweden, 2013), belongs to the east Scandinavian branch of the Germanic languages (Campbell, 1998). The syntax is similar to those of English and German, yet presents a distinctive feature in its verb second word order. Swedish uses nine vowels, with both long and short forms, and 18 consonants. Compounding enables long words with a variety of syllable structures. Arabic, which belongs to the Semitic branch of the Semito-Hamitic family, consists of different dialects which can differ greatly in speech rate, stress, and realization of words (Versteegh, 2001). Words are constructed by combining a single root (usually of three consonants) with three vowels (long and short forms, altogether six). In contrast to Swedish where spoken and written Swedish are uniform, Arabic spoken and written forms represent distinct varieties (Campbell, 1998). In Sweden today, more than 20% of the school-aged population speak a language other than Swedish at home. Due to current migratory patterns, Arabic is the fastest growing minority in Sweden (Statistics Sweden, 2013).
Aims of the Study
The aim of this study was to evaluate the utility of conceptual scoring to capture the distributed characteristics of lexical organization in BLI children. A further aim was to explore if the BLI group scored similarly to MLI when using conceptual scoring.
In line with previous research and theoretical assumptions of lexical organization being based on the general cognitive development, but also influenced by linguistic input, we hypothesized that the Arabic and Swedish scores would be similar within the bilingual child. Furthermore, the specific linguistic input would lead to different items rendering syntagmatic and paradigmatic associations, differently distributed in Arabic and Swedish. Assuming that vocabulary knowledge is distributed across both languages, the conceptual score was expected to be higher than either single language score. For the between-group comparisons, similar performance between MLI and BLI was expected, especially when using the conceptual score.
Method
Participants
Twenty-four children, ranging in age between 5;11 and 8;2, participated in the study, all with a clinical diagnosis of LI given by a certified speech-language pathologist prior to inclusion in the study. All children in this study were considered to have persistent language impairments, with receptive and expressive deficits compared with the scores obtained by bilingual typically developing children (Holmström et al., 2015). Fifteen children were bilingual Arabic-Swedish children, and nine were monolingual Swedish children. The children had attended Swedish-speaking preschools from the mean age of 1 year 6 months for the bilingual and 1 year 3 months for the monolingual children (see Table 1). As for bilingual children, both languages must be assessed for a LI diagnosis. The children were drawn from a larger group of children (n = 111) participating in a study of lexical development in monolingual and bilingual (Arabic-Swedish) children with both children with LI and children with typical development. They were assessed with a battery of cognitive and linguistic tasks. All children in the study passed a 20 dB pure-tone hearing screening at .5, 1, 2, and 4 kHz prior to testing, and had no known neurological or neuropsychiatric disorders. The procedure of recruitment and testing was performed in accordance with the approval given from the Regional Ethics Review Board for Southern Sweden (Approval No. 2010/717).
Participant Characteristics.
Note. BLI = bilingual children with language impairment; MLI = monolingual children with language impairment; F = female participants; M = male participants; CCC = Children’s Communication Checklist; ITPA = Illinois Test of Psycholinguistics Abilities.
Age equivalents in months. bScores below 30 for the subscales Speech and Grammar and below 140 for Pragmatics indicate language difficulties. cThe score is the sum of the responses of all four categories. Scores below 42 (age 5;11) and 53 (age 8;6) indicate language difficulties.
All parents were given a 15-item questionnaire, online or over the telephone. The questions concerned parental education, children’s early language development, and preschool start and number of years in preschool. This questionnaire has been used in clinical practice in speech-language clinics in Sweden for a longtime. Parental education was scored from 0 to 3 (0 = no formal education, 1 = compulsory schooling, 2 = upper secondary schooling, 3 = university). The mother’s and father’s scores were averaged to obtain a joint parental educational level. We also asked parents to indicate at what age the children were enrolled in preschool and whether the children used any Swedish prior to preschool start. According to information from teacher interviews, all children attended Swedish-speaking schools and after-school youth programs for approximately 45 hr per week. The BLI children were exposed to Swedish in school, except for a weekly hour of Arabic teaching. No bilingual participant was reported to use exclusively Swedish or Arabic. Parents of the bilingual children were also given seven additional questions concerning Arabic and Swedish use and exposure at home as an indication of bilingual proficiency. Data from the parental questionnaires and information from teachers were used to calculate two composite scores, one for Arabic exposure and one for Arabic use. The parents answered questions regarding their children’s language use and exposure when interacting with parents, siblings, other family members, and friends. The information obtained yielded 14 variables (8 for language exposure and 6 for language use) and was summed into percentage exposure to and use of Arabic. The bilingual children were exposed to Arabic with a mean of 48% (SD = 7) and used Arabic 39% (SD = 14) of the time. See Table 1 for participants’ characteristics.
Material
Raven’s colored progressive matrices (Raven, Court, & Raven, 1986) were administered to confirm that all children had cognitive abilities within the normal range. To confirm the LI diagnosis, pretesting of grammar and vocabulary was administered. For the bilingual children, tests were administered in both Arabic and Swedish. The Children’s Communication Checklist (CCC; Nettelbladt, Radeborg, & Sahlén, 2003) was given to the children’s teachers to get information about overall language abilities. The Grammatical assessment of Swedish as a Second language (GrUS; Salameh, 2015) and expressive grammar in Arabic (Salameh, 2003) were given to assess productive grammar, defined as number of correct grammatical structures. The grammatical tests are both based on the processability theory, where an emergence criterion is applied (Pienemann & Håkansson, 1999). Systematic grammatical contrasts are awarded, and deviations from the norm in terms of suffixes are disregarded. Processability theory has been used with many different languages (e.g., Swedish, German, Arabic, and Japanese). Furthermore, receptive vocabulary was assessed with a task consisting of 40 words (16 nouns, 12 verbs, 12 adjectives/adverbs; Olsson et al., 2012). All test items are commonly found in children’s books and were represented by black and white line-drawings. The participants were required to choose one of three pictures corresponding to the word they heard. The test was developed in Swedish and translated into Arabic by an Arabic-speaking speech-pathology student and retranslated into Swedish by a PhD student in Arabic (fluent in both Swedish and Arabic). According to the judgment of the two translators, the target items were culturally appropriate for Arabic-Swedish bilingual children growing up in Sweden and had similar difficulty levels in both languages. The test has also been used in other projects with Swedish-Arabic children in the same age range (Olsson et al., 2012). Overall agreement between the translation and retranslation was 97%. Examples of words are furniture, climb, and throw. Finally a verbal fluency test, a subtest from the Swedish version of Illinois Test of Psycholinguistics Abilities (ITPA; Kirk, McCarthy, & Kirk, 2000), was given. The experimenter pronounced a category word, and the child was asked to generate words belonging to this category during 1 min. Four categories were given: any word, outdoors, food, and body parts. For detailed information about participant characteristics and test results, see Table 1.
Outcome Measure
Word associations were assessed with a shortened version of the Kent–Rosanoff list (Kent & Rosanoff, 1910), translated into Arabic (Salameh & Nettelbladt, 2013) and Swedish (Namei, 2002), consisting of 50 words (35 nouns and 15 adjectives). The Arabic list was translated back into Swedish by an Arabic language teacher, with a 94% agreement. The words included in this shortened list (Johansson & Wahlstrand, 2010) are all of high frequency and well known to bilingual Arabic-Swedish children according to an earlier study (Mikoczy & Nyman, 2008). The proportion of nouns and adjectives is the same as for the original list. Words that were excluded from the original list were words to which the participants of the Mikoczy and Nyman (2008) study gave very few or no responses (e.g., words such as priest, bible, whisky, and tobacco).
Procedure
The Arabic and Swedish versions of the test battery were given to the bilingual children in two sessions, with language order counterbalanced between sessions, with a mean of 22 days between sessions. The same tests in Swedish were given to the MLI group. A trained graduate student in speech-language pathology, fluent in Arabic and Swedish, performed the Arabic assessment, and the first author and a trained graduate student in speech-language pathology performed the Swedish assessment. Testing took place in a quiet room at the children’s schools. If the child did not understand the test procedures, instructions were repeated once. If children were hesitant to respond, verbal encouragement was given. In the grammatical task, children were instructed to complete sentences with correct grammatical forms. For the assessment of receptive vocabulary, participants were required to choose one of three pictures corresponding to a word they heard. For the word association task, participants were introduced to the task by switching roles with the test administrator, presenting the adult with a verbal stimulus (e.g., dog) and the adult responding with a paradigmatic association (e.g., cat). When the child understood the procedure, the real test started. The test administrator gave the stimuli verbally, one at a time, and asked the child to say the first word that came to mind. If the child hesitated, the child was encouraged to make the response by an “ok, have a try.” The children’s associations were recorded verbatim on a scoring form. Each session lasted approximately 1 hr, and the child could take a break when necessary.
Scoring
The Swedish version of CCC (Nettelbladt, Radeborg, & Sahlén, 2003) consists of 79 items, subdivided into 11 subscales. The subscales Speech, Grammar, and Pragmatics are reported here. For nonrelated words, each trial yielded 2 points if correctly repeated and 1 point if the child needed to hear it one more time. The maximum score was 12. The Swedish version of GrUS has a five-level scale with increasing difficulty and a maximum score of 22. The Arabic version comprises three levels (with higher levels requiring knowledge of written Arabic), with a maximum score of 12. For the receptive vocabulary test (maximum 40), both single language and conceptual scores were calculated. The conceptual score consisted of unique concepts known in either language. For example, if the word envelope is known in Arabic and the word fight is known in Swedish, the child receives 1 point in each single language score but 2 points in the conceptual score. If the word envelope is known in both Arabic and Swedish, the child will still be awarded only 1 point in the conceptual score. For the verbal fluency task, all correctly produced words were added together. Results of all assessment are presented in raw scores.
For the Kent–Rosanoff list, responses were coded into one of three categories of word associations: syntagmatic associations (semantic link but not from the same word class, for example, bread–eat), paradigmatic associations (semantic link and same word class, for example, table–chair), and other associations (phonological associations, repetitions, translations, code-switches, all without semantic link to the stimulus word, for example, fruit–gazelle, and no answer).
Syntagmatic and paradigmatic associations in Arabic and Swedish were individually counted and comprised the single language scores. A conceptual score (for syntagmatic and paradigmatic associations separately) was also calculated. In the BLI group, participants were awarded conceptual points for associations in Swedish, Arabic, or both languages. For example, for the stimulus word table, the association eat in Swedish constitutes a syntagmatic association, whereas the association chair in Arabic constitutes a paradigmatic association. The paradigmatic conceptual score, representing the higher order association from either language, is thus paradigmatic. For each item, responses were compared and analyzed regarding whether a paradigmatic response was given in Swedish (Swedish singlet), Arabic (Arabic singlet) or the same association was given in both languages (paradigmatic response in both Swedish and Arabic, doublet). The same procedure was carried out for syntagmatic associations. For the MLI group, assessed only in Swedish, the conceptual and Swedish scores were the same. For an example of the scoring procedure, see Table 2.
Example of Monolingual and Conceptual Scoring and the Calculation of Singlets and Doublets.
Note. All responses are translated into English. Ar = Arabic; Sw = Swedish; Conc = conceptual; P = paradigmatic; S = syntagmatic; O = other.
Data Analyses
To ensure normal distribution of the variables, Kolmogorov–Smirnov tests were conducted. All dependent variables were normally distributed. Effect sizes (Cohen’s d; Cohen, 1988) were calculated and interpreted accordingly.
Results
Within-Group Comparisons
To assess the variation between performances in Arabic and Swedish and to compare the single language Arabic and Swedish scores with the conceptual, three paired-samples t tests (Arabic-Swedish; Arabic-Conceptual; Swedish-Conceptual) for the BLI group were conducted. After a Bonferroni correction, the significance level was set at .017 (.05/3).
Syntagmatic associations
There was a significant difference between single language Arabic and Swedish syntagmatic scores, t(14) = 2.84, p = .013, d = .73, with fewer syntagmatic associations in Arabic. A significant difference was found between the single language Arabic score and the conceptual score, with more syntagmatic associations in the conceptual score, t(14) = 3.74, p = .002, d = .97. No significant difference was found between the single language Swedish score and conceptual score, t(14) = .42, p = .68, d = .11. Descriptive statistics (means, standard deviations, and range) are shown in Table 3. For syntagmatic associations, Swedish singlets were the most common (M = 9.93, SD = 6.06), followed by Arabic singlets (M = 4.53, SD = 3.04). Doublets (a syntagmatic response to the same item in both Arabic and Swedish) were produced sparingly (M = 3.87, SD = 3.83).
Mean Swedish, Arabic, Conceptual, and Total Scores (and Standard Deviations) on the Kent–Rosanoff List for the Bilingual and Monolingual Children With Language Impairment.
Note. BLI = bilingual children with language impairment; MLI = monolingual children with language impairment.
Paradigmatic associations
There was no significant difference between single language Arabic and Swedish scores, t(14) = .14, p = .89, d = .036. The conceptual score differed significantly from both the single language Arabic and Swedish scores, Arabic, t(14) = 4.77, p < .001, d = 1.23; Swedish, t(14) = 5.18, p < .001, d = 1.34, with higher conceptual than single language scores. The amount of Arabic and Swedish paradigmatic singlets was similar, Arabic (M = 8.40, SD = 6.16); Swedish (M = 8.40, SD = 7.11), whereas paradigmatic doublets were less common (M = 5.20, SD = 4.92).
Between-Group Comparisons
Four independent t tests (conceptual and single language Swedish scores for syntagmatic and paradigmatic associations, respectively) were conducted, and the significance level was set at .0125 (.05/4) to avoid inflated significance values.
Syntagmatic associations
For the syntagmatic associations, no significant group differences were found, with both groups performing on par, Swedish syntagmatic, t(22) = 1.46, p = .16, d = .61; conceptual syntagmatic, t(22) = 1.70, p = .11, d = .69, yet with moderate effect sizes, with the monolingual group showing a higher number of syntagmatic associations, see Figure 1.
Confidence interval (95%) for syntagmatic and paradigmatic associations for bilingual children with LI and monolingual children with LI.
Paradigmatic associations
There was no significant difference between BLI and MLI children when comparing Swedish paradigmatic associations, t(22) = 2.05, p = .054, d = .79, yet a moderate effect. However, for the conceptual score the BLI group outperformed the MLI group, t(22) = 4.92, p < .001, d = 1.89, with significantly more paradigmatic associations, for group comparisons, see Figure 1.
Discussion
The purpose of this study was twofold: to evaluate the utility of conceptual scoring within the paradigm of lexical organization in a group of bilingual children with LI and to explore whether BLI children scored similarly to monolingual children with LI. The study intends to add to our knowledge of lexical organization being influenced by general cognitive factors. A predominantly cognitive basis for lexical organization would be reflected by similar development of a bilingual person’s both languages as well as similar development for monolingual and bilingual groups. Language-specific factors would, in contrast, result in a difference in performance between the different languages of a bilingual child and also between monolingual and bilingual speakers. The findings should be taken as preliminary evidence in need of further evaluation in larger samples.
Within-Group Comparisons
For syntagmatic associations, a significant difference was found between the BLI children’s single language Arabic and Swedish scores, with a higher number of syntagmatic associations in Swedish. This finding is surprising since syntagmatic associations are viewed as an earlier stage of lexical development, implying that Arabic would be more developed than Swedish in this group of children. However, to fully understand this finding, all associations must be considered. The higher amount of syntagmatic associations (e.g., dark–cave or dark–to sleep) in Swedish may be due to some of the stimulus words in Arabic being less familiar and rendering associations categorized as other (dark–get, dark–trash). The results are in line with those of Ordóñez et al. (2002) who found Spanish-English bilingual children to give more syntagmatic associations in English than in Spanish. Their interpretation was that syntagmatic associations are more influenced by general communication, a skill practiced throughout the day in school. Hearing and using a language may influence the ability to associate syntagmatically. Another possible explanation may pertain to the languages tested. Most word association studies have focused on English. In a recent exception investigating word associations in Greek (Mattheoudakis, 2011), adult speakers tended to give more syntagmatic associations than children, leading to the assumption that different languages influence the organization of the mental lexicon differently. Similar results, with a high amount of syntagmatic associations given by native adult speakers, have been found for Danish (Nissen & Henriksen, 2006), a language closely related typologically to Swedish, also belonging to the Scandinavian branch of Germanic languages.
Comparable amounts of paradigmatic associations were found between the performance in Arabic and Swedish. This is in line with earlier research on bilingual children with typical development. Many scholars (see, for example, Cronin, 2002; Namei, 2002; Ordóñez et al., 2002; Sheng et al., 2006) hypothesize that paradigmatic associations are based on cognitive and metalinguistic knowledge. Therefore, these associations can more easily be transferred between languages than syntagmatic associations, which are based more on specific linguistic input. However, the words which rendered paradigmatic associations were not the same in Arabic and Swedish. As mentioned earlier, other factors, such as frequency of occurrence, age of acquisition, and in which contexts the words are learned, may influence the type of association (Fitzpatrick, Playfoot, Wray, & Wright, 2013). As with all language measures, environmental factors, such as parental education, may also influence the result. The influence of these factors needs to be systematically studied in future research.
The results further revealed that the conceptual score for paradigmatic associations was significantly higher than either single language score for the BLI group. This confirms that the vocabulary knowledge of the bilingual participants is distributed across Arabic and Swedish, extending the validity of “distributed characteristics” (Oller, Pearson, & Cobo-Lewis, 2007) from lexical size to lexical organization. The results are strengthened by the fact that the data were gathered from two structurally and typologically distant languages, Swedish, a Germanic language, and Arabic, a Semitic language, and are not a result of cognates (word that shares form and meaning in both languages).
Peña et al. (2002) found that almost 70% of responses in category generation were language specific, a result in line with our finding of more singlet than doublet productions. In Peña et al. (2002), the single language score did not capture the full vocabulary knowledge, making comparisons with monolingual norms inadequate and ill-suited for identifying language impairment. Therefore, the results advocate the use of conceptual scoring instead of single language scoring in the language assessment of bilingual children, as previously suggested (Bedore et al., 2005; Sheng et al., 2012). These results also imply that although the ability to associate paradigmatically may be cognitively driven, the linguistic input may influence which words undergo all stages of lexical organization (Namei, 2002).
Between-Group Comparisons
With regard to our second aim, the BLI and MLI groups performed on par on syntagmatic associations. However, the MLI group showed a stronger preference for syntagmatic associations, which may, as discussed earlier, be due to the specific language tested. If the Swedish language is more prone to be syntagmatically organized, then being a monolingual Swedish speaker may, in itself, lead to an increased use of syntagmatic associations. Cross-linguistic data on comparable words need to be examined to more fully understand these results.
For paradigmatic associations scored in Swedish, no significant group difference was found. For the conceptual score, however, we found a significant difference in performance between the groups with the BLI group outperforming the MLI group. We expected to find convergence between the groups, especially when using conceptual scoring. In many previous studies, where monolingual and bilingual children are compared on single language scores, monolingual children, even with LI, outperform bilinguals on vocabulary-related assessments (Windsor & Kohnert, 2004). We will discuss two possible interpretations of this result.
First, although word association can be viewed as a lexical task, most scholars agree that it is highly influenced by cognitive development (Cronin, 2002; Nelson, 1977). The complex relationship between bilingualism and cognitive development has been extensively studied, and numerous studies have found a bilingual advantage in attentional control, working memory, metalinguistic awareness, and symbolic representation skills (Adesope et al., 2010). Therefore, the heightened metalinguistic awareness in our BLI group may be an effect of a bilingual advantage (Paradis, 2010). In an earlier study of monolingual and bilingual children’s performances on word associations, similar results between languages and groups have been found (Sheng et al., 2006). Sheng et al. (2006) used single language scoring and compared children with typical development and found subtle differences. The bilingual children produced more paradigmatic associations in the first-word association trials, a bilingual advantage was seen for verbs, and a trend toward a bilingual advantage was found when best performance in either language was compared. In our study, using conceptual scoring to take full linguistic competence of the bilingual children into account, the bilingual advantage is significant. A further difference between the study of Sheng et al. (2006) and ours is that the interaction of bilingualism and LI, under study in the present investigation, may affect the results differently than bilingualism alone. However, the sample size of our study is small, and more data, from larger samples, need to be gathered to draw firm conclusions.
Second, although all children in this study had a clinical diagnosis of LI prior to inclusion, and diagnoses were confirmed by initial language assessments establishing deficits in both receptive and expressive domains of both languages, a lack of standardized assessment tools for bilingual populations in Sweden raises concerns for clinical decision making. Due to the lack of standardized assessment tools, there is a risk of over-diagnosing LI in bilingual populations, in the present study resulting in a risk of bilingual children with typical development being included in the BLI group. However, assessment of both languages and the use of conceptual scoring reduce this risk.
A tentative conclusion can be drawn that conceptual scoring may be used as a method to capture the distributed characteristics of lexical knowledge in bilingual populations, and therefore facilitate comparisons between monolingual and bilingual populations. The use of conceptual scoring in clinical decision making may thus reduce the risk of over-identification of LI in bilingual populations.
Footnotes
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This study was supported by Sunnerdahl Foundation for the Handicapped.