Multilinguals do not have several independent language systems. Instead, research has shown that a multilingual’s languages form an integrated system: each language affects processing of the other(s). However, very little is known about the representations and mechanisms underlying this interaction beyond the level of individual words. It cannot be explained by current computational models of sentence comprehension, which deal only with the monolingual case. This project aims at uncovering the architecture of the multilingual sentence processing system and explaining how a multilingual’s languages interact.
A major insight from psycholinguistics is that many language-processing phenomena arise from statistical regularities of language (e.g., word-reading speed closely follows word occurrence probability). Reshaping this view for the multilingual case, the project develops a formalization of how one language’s statistical properties impact the processing of another language. This formalization is then implemented in a number of computational cognitive models that simulate aspects of the language comprehension process and account for the unique properties of multilingualism.
The different models embody different ways in which the multilingual system may operate. The most appropriate model architecture is identified by correlating the models’ fine-grained quantitative predictions with behavioural and neural data from sentence-reading studies that compare: monolinguals and bilinguals; native and non-native speakers; and non-natives with different levels of second language exposure end proficiency. Thus, by integrating computational simulations and psycholinguistic experiments, the project provides novel insights that increase, and potentially transform, our understanding of multilingualism; and can drive innovations in foreign language teaching and training.