Reconsidering Early Bilingualism. A Corpus-Based Study of Polish Migrant Children in the United Kingdom.

Premise The present study investigates the language of Polish-English bilingual children who were raised in the United Kingdom and their Polish monolingual counterparts. An attempt is made to model the lexico-grammatical knowledge of both groups using corpus-based grammatical inference in order to establish the source of the impediment of the minority language of the bilingual group. The methodology of corpus linguistics is applied in order to inspect whether it is bilingualism itself that contributes to the transitional lacks in grammatical accuracy (Gathercole, 2007), or rather whether this is an effect of language type (i.e. a typological effect), or whether it is a cumulative effect of both bilingualism and language type. Methodology The approach adopted herein is theory-grounded, meaning that rather than inspecting idiolectal features, like most longitudinal studies of child language have done in the past (cf. Pearson, 2007; 2008), the present study presupposes that a child-type langue (in the de Saussurian sense), or competence (in the Chomskyan sense), exists and can be contrasted with an adult-type langue (or competence). Child output was elicited and analyzed using the narrative analysis approach (Gagarina et al, 2012), meaning that each child participated in an experiment session in which he or she told one short picture story and then retold a story delivered by the experimenter. Participants and Data Samples Language samples were collected and transcribed from three groups of children mean age=65.92 months (SD=13.50). The first group (N=40) consisted of bilingual children of Polish migrant parents raised in the United Kingdom. The second group (N=40) consisted of monolingual children who were born and raised in Poland. The bilingual groups participated in two experiment sessions, meaning that three sets of data were produced and analyzed, bilinguals speaking English (N=40), bilinguals speaking Polish (N=40), and monolinguals speaking Polish. Procedure Following non-verbal intelligence testing, spoken narrative output was elicited from the children using an adaptation of the Multilingual Assessment Instrument for Narratives (Gagarina, 2012), specifically its Polish version (Kiebzak-Mandera et al., 2012). The data (120 transcripts, 40 per group) were then annotated with a highly-lexicalized, constraint-based formalism, and, along with all relevant meta-linguistic information, stored as Extensible Mark-Up Language Files, forming a corpus of child narratives. The linguistic content of the transcripts was segmented into communication units, or T-Units (predicates and their arguments and modifiers), based on the syntactically motivated and consistent criteria proposed by Loban (1976) and expanded by Hughes et al. (1998), Miller (2006), and Strong (1998). Analysis A variety of statistical analyses was performed on the corpus data during the course of the present study. The focal points were lexis, syntactic complexity, utterance make-up, errors of various types, and instances of cross-linguistic influence. All of the measures also contributed to a number of derived statistics. The analysis proper began with sets of two-factor analyses of variance (ANOVA) with replication, all followed up by a post-hoc significance test. In addition to the ANOVAs and their follow-ups, a series of linear regression analyses was performed to inspect the correlations of significant variables. Results Significant differences were found between groups. The general tendency among all results presented in this chapter supports the cumulative effect hypothesis, meaning that the bilingual modality and language type both affect production accuracy in bilingual children. Across the board, monolinguals outperform bilinguals regardless of language spoken in areas such as lexical density, syntactic complexity, and overall grammatical accuracy. However, each time Polish is involved, accuracy is lowered even further. (i.e. bilingual Polish being is most disadvantaged configuration). The effect of socio-economic status measured by maternal education has shown to be a powerful predictor of both accuracy and interference (transfer). No correlations were found between the frequency of cross-linguistic influence units and error count either in the telling or the retelling mode. Due to statistically insignificant differences in accuracy between narrative modes, there are grounds to partially question the hypothesis of modelling between narratives (i.e. that the retold narrative is always an improvement upon the told one). Implications The corpus analysis allowed for the inference of two, clear transfer rules, both of which are overgeneralizations. However, since most of the instances of the application of this rule did not produce errors, but was nevertheless marked by gradient well-formedness, a new category of transfer, benign transfer, was proposed, with the intention that it complement the positive-negative transfer distinction. Furthermore, correlations of error count and input suggest that both accuracy and interference is determined by what the child is exposed to in their home environment. For this reason, parents should adopt a remedial strategy when using the minority language at home. Since linguality and language type both contribute to lowering output accuracy, particular remedial strategies should be adopted when Polish is the minority language of a bilingual child.