The study group was formed by 15 bilingual Catalan-Spanish children with SLI (ten males), with a mean age of 8.0 years (SD=0.3) at the time of assessment (Table 1 for demographic data). These children were paired with a control group of 15 age-matched typically developing bilingual Catalan-Spanish children (nine males), with a mean age of 8.2 years (SD=0.2). There were neither gender nor age differences between groups (Table 1).

All children were involved in a longitudinal project conducted in the Balearic Islands, Spain, and fulfilled the established criteria for diagnosing SLI (Leonard, 1998). Schools sent us a brief profile of their pupils at third year of kindergarten (five years of age) showing language problems and without having a history of cognitive, auditory, social or neurological damage. For the initial sample, we also considered other exclusion criteria, such as only presenting with articulatory problems and/or being newly arrived from a non-Catalan speaking community, in order to avoid the sample to include sequential second language learners. Therefore, we pre-selected those bilingual children who presented with only language problems and no other potential associated difficulties.

Then, we assessed their general language profiles (Form: phonology and morphosyntax; Content: lexicon, and Use: pragmatics) using the PLON-R: Prueba del Lenguaje Oral de Navarra Revisada ‘PLON-R: Navarra Oral Language Test Revised’ (Aguinaga, Armentia, Fraile, Olangua, & Uriz, 2004). The linguistic profile according to PLON-R was form and content (phonological-syntactic and lexical-syntactic) in children with SLI, but not pragmatic (see Table 1). The equivalent departing pragmatic capacities between both groups might help to better evaluate their potential underlying difficulties in performing the experimental task, given that there was not an apparent selective deficit in pragmatics among children with SLI. Participants who had moved to another country or did not fulfil standard criteria for SLI were not considered for the follow-up of in longitudinal study.

Non-verbal Intelligence Quotient (IQ) was measured by means of the Wechsler Preschool and Primary Scale of Intelligence (Wechsler & de la Cruz, 2001). Neurological, social and emotional data were obtained from speech therapists at school. We also requested their records related to audition to the Balearic Ministry of Health. This institution conducts an Otoacoustic Emissions (OAE) analysis to all children in Majorca and an audiometric test to those children that fail the OAE at the age of six. Thus, the study group was comprised of children born in Majorca who showed language difficulties, as measured by PLON-R, with an IQ higher than 85, and no evidence of auditory, social, emotional or neurological problems. Children in the SLI group obtained significantly lower scores than controls in terms of the total score in the PLON-R and did not differ with children in the control group in terms of non-verbal IQ (Table 1).

The control group was comprised by children without language problems going to the same classrooms and having the same age and dominant language as children with SLI. Thus, each participant with SLI was initially paired with a control child in terms of schooling, age and dominant language. Language, intelligence, auditory, social, emotional, and neurological variables were also recorded for the control group. Their language of instruction at school was Catalan (see language of participants and language assessment section, for a more detailed explanation). All participants were followed over six years until they reached the age of 12 years, but the present study only includes dependent measures between ages of seven and eight. The referential communication task was administered at age of eight. One year earlier, at the age of 7 years, the TSA: el desarrollo morfosintáctico del niño ‘Child Morphosyntactic Development’ (Aguado, 1989) was administered in order to reconfirm the language difficulties and to obtain a morphosyntactic score. Children in the SLI group obtained significantly lower scores than controls in terms of the total score in the TSA (Table 1).

Participants in the present study were all simultaneous proficient Spanish-Catalan bilinguals (Table 1). With regard to the family language, seven children in the control group mainly spoke Catalan at home and eight spoke Spanish. In the group of children with SLI, four children mainly used Catalan at home, and 11 used Spanish. There were no differences in terms of language use at home (Table 1).

With respect to the language of assessment during the referential communication task, we allowed children to choose the language of response, and the interviewer adapted his/her language according to the children's preferences. In so doing, seven children from the control group answered in Catalan, six answered in Spanish and two responded in both languages. In the SLI group, two children answered in Catalan, 11 in Spanish and two did so in both languages. There were no differences in terms of language choice for the referential communication task (Table 1).

Finally, the TSA (Aguado, 1989) was administered in Catalan (Table 1). As stated above, morphosyntactic structures of Catalan and Spanish are very similar, and some children answered by using both languages. Code mixing is frequent in bilingual communities and thus, we considered correct answers provided in Catalan and also provided in Spanish when the morphosyntactic structure used was appropriate. Mixing languages while preserving the correct syntactic structure was also considered correct.

The referential communication task used was based on the system created by Krauss and Glucksberg (1969) were two subjects sit at a table facing each other separated by an opaque screen, which prevents visual contact between them. In this case, the child is the speaker and the adult has two roles: the role of listener and the role of regulating the communicative behaviour of the child, re-establishing communicative interaction when necessary. The child was presented with a picture (297cm × 490cm) of a room with different furniture, as well as ten objects, such as an airplane, a car and a little book, among others, which were located in different places around the room (spatial description task, see Appendix A). Four of the objects in the picture were unique and non-ambiguous, and six were ambiguous (i.e. the same objects, yet in different colours or sizes, such as a big book or a little book). The adult had the same picture display, but the objects were not placed on the corresponding site. The task of the child (the speaker) was to describe the arrangement of the objects as placed in the picture and the adult (the listener) had to place each object in his/her picture, attempting to reconstruct their arrangement based on the speaker's instructions. The aim of the task was to produce two identical pictures at the end. For this task, the following instructions were given:

“Now I have given you a sheet with many objects in the room. You must explain to me exactly where those things are. I have a picture of the same room, yet with no objects, and I will have to paste them in. When you are ready, we can start.”

Thus, the child had to tell where to place the objects in the picture, as correctly as possible, in order to put them into the right places. When the child produced an ambiguous or an incorrect message, the adult had to regulate the behaviour of the child so that he/she could improve the informative content of the message as many times as he/she considered necessary in order to reach a correct description. For this purpose, the adult could use strong regulations, to request further information, or weak regulations, in order to spur the child to expand on the information, yet without directly asking the child to specify the object or the location of the referent. This specific task is an adaptation of “organization of a room” by Boada and Forns (2004). A group of trained undergraduate Spanish-Catalan bilingual students administered all tasks at the children's schools. All did not know whether the participants belonged to the study group or to the control group. Every examiner assessed a child with SLI and his/her paired control, in order to minimise differences between examiners.

A photograph displayed the two sheets at the end of each session, to record the exact position of each object. Then, language samples were transcribed and coded in CHAT format (MacWhinney, 2000). Codes classify messages by their type (initial, final or reformulated), by their accuracy (correct, erroneous, ambiguous or omitted) and by the object (referent) specification (attributes) and its location (spatial elements). The coding was checked by two independent reviewers and levels of conformity were established in order to ensure an optimal level of reliability (see Reliability section). The specific codes of the referential communication task, the types of errors and adult regulations can be consulted in Appendix B.

Before an agreement was reached by the two external evaluators, two reviewers assigned codes to eight transcriptions randomly selected, four for each group of children, and used the relay software from CHILDES project (MacWhinney, 2000) to determine the levels of concordance between them. The level of reliability between the two coders was 0.83.

Independent sample comparisons were carried out between the group of children with SLI and the control group using non-parametric tests, because each group included less than 30 participants. Because of the high number of a priori planned comparisons, the Bonferroni correction was applied separately for each analysis, thus allowing us to reduce Type I error. This procedure resulted in a cut-off criterion for significance at .05, .03, .02, .02 and .03 for measures of message quality, type of correct messages, inappropriate elements, type of inappropriate messages, and communicative exchange. Effect sizes were calculated with r and we have interpreted these results using Cohen's (1988) categories: small effect size r=.1, medium r=.3, and large r=.5.

First, we show an independent sample analysis between groups for the total number of messages produced (Table 2) in order to show that total messages were equivalent between both groups and thus comparisons were not biased by different production rates. The messages were classified as initial, final and reformulations. As it can be seen in Table 2, there were no significant differences in the initial number, final number and the number of reformulations between children with SLI and age-matched controls at age of eight.

Regarding results of the message quality in terms of correct messages produced by the child to the listener (see Figure 1), we found significant differences between children with SLI and controls. Specifically, the control group produced significantly more correct messages than children with SLI, with a medium-to-large effect size (U=44.00, p=.01, r=.49). Considering the type of messages produced, this statistically significant difference was true for both initial and final messages, with medium-to-large effect sizes (Table 2).

Figure 1.

Median and standard deviation of message quality (total number of initial and final correct messages) in children with SLI and age-matched controls in the referential communication task.

(0,04MB).

With respect to the type of inappropriate description of the referents (object identification, object specification – attributes- and object location), results showed no significant differences in any of the measured variables (Table 3). Finally, with respect to the type of inappropriate messages (Table 3), none of the measures (ambiguous, omitted or erroneous message) showed differences between both groups.

Our results support hypothesis 1, as we found a lower number of correct messages conveyed by children with SLI during the referential communication task. This might be seen as a communicative ineffectiveness, which can be observed both at the initial message and also at the final message, after the regulations had been provided. Thus, children with SLI could not take advantage (in form of reformulations) from adult exchanges/demands (regulations) in order to produce a correct final message. In addition, any of the inappropriate type of elements in the message– object, object specification or location–, nor any type of inappropriate message–ambiguous, omitted or erroneous message–seemed to characterize the lower number of correct responses provided by children with SLI. Therefore, the lower number of correct messages produced by children with SLI, even with an equivalent baseline number of delivered messages, was explained by the sum of the rates of the different inappropriate elements rather than a disruption in a particular information provided (either regarding the referent or regarding the type of inappropriate message).

Following hypothesis 2, we expected a higher number of adult regulations (weak and strong) addressed to children with SLI, and results (see Table 4) did not show significant differences in any of the variables analysed. Thus, results reject our hypothesis 2.

In order to test hypothesis 3, we conducted three standard regression analyses with morphosyntactic score (TSA) at the age of seven as the predictor variable for the number of correct messages in the referential communication task at the age of eight. The first regression analysis was conducted with the whole sample of participants; the second included only children with SLI, and the third was comprised by children in the control group. Results with all participants showed that the morphosyntactic score at the age of seven was not a significant predictor of the number of correct messages in the referential communication task evaluated one year later (R2 = .13, ΔR2 = .09, F(1,28) = 4.03, B = .24, β = .35, p = .06). The regression model was not statistically significant and explained a low percentage of the variance (9%) observed in the number of correct messages produced by children. The results of the regression analysis including with the SLI group and the control separately group showed even lower values as compared to the first regression model (sli group: R2 = .01, ΔR2 = -.06, F(1,13) = 0.15, B = .09, β =.11, p =.70; control group: R2 = .03, ΔR2 = –.05, F(1,13) = 0.36, B = – .17, β = –.16, p = .56). Following hypothesis 3, morphosyntactic scores did not play a significant role in the referential communication task and thus other abilities, perhaps outside the linguistic domain, might be involved referential communication.