Highly proficient bilinguals activate overlapping regions in the brain while processing either one of their languages. In addition, it has been shown that both languages of a bilingual speaker are activated simultaneously. This raises the question of how linguistic information is represented to allow correct language selection and what cognitive processes contribute to language disambiguation. This thesis aims to shed light on these issues with a focus on written word processing.
The fMRI study tested the hypothesis that language-specific lexico-semantic knowledge is encoded in distinct but overlapping neural networks. In addition, it was hypothesised that only multivariate analysis would allow detection of these embedded networks. Twenty Welsh-English bilinguals performed a semantic categorisation task on English and Welsh written words. As expected, the univariate general linear model (GLM) analysis revealed little differences in overall activation between the two languages. However, multi-voxel pattern analysis (MVPA) revealed an extensive set of areas within the lexical processing network that show strong sensitivity to language identity. This is the very first fMRI study to unveil positive evidence for the micro-organisation of language networks in bilinguals and to map the areas of the brain that distinguish between languages. Furthermore, preliminary functional connectivity analyses (PPI) revealed some differences in the connectivity profile of English and Welsh that may be attributed to differences in orthographic transparency.
The thesis then addresses the question of what factors may contribute to the selection of language appropriate lexical representations, focusing on the potential role of language specific sublexical orthographic cues. A series of of single- and mixed-language lexical decision tasks were conducted with a total of 73 fluent Welsh-English bilingual and 55 English monolingual control participants. The stimulus set included language-marked and unmarked words and pseudowords in both languages, and the items were matched on a variety of psycholinguistic measures including within- and cross-language neighbourhood statistics. Both bilinguals and monolinguals showed evidence that sublexical markedness features led to language membership activation and as such, could be used as cues for response selection. However, markedness only affected performance in mixed language contexts. In addition, the direction of the effects depended on whether or not stimuli belonged to the target language(s). In particular, pseudowords marked to the target language were more difficult to reject while items that were marked to the non-target language were easier to reject than unmarked items. This demonstrates the importance of manipulating task context to understand the source of markedness effects. The findings suggest that markedness does not affect word recognition per se, but rather response selection processes.
Overall, the thesis contributes to the knowledge of different organisational levels in the brain of balanced bilingual speakers, by showing that neural networks sensitive to language membership exist within shared language processing regions. In addition, the thesis adds to the current understanding of factors that drive language selection despite the assumed integrated lexicon and neural overlap, emphasising the importance of both linguistic features and context cues. Additionally, this provides support for theoretical models of bilingual reading such as the Bilingual Interactive Activation plus model (BIA+).