Top-down language-to-word inhibition effects in bilingual speakers

Thumbnail Image
Issue Date
Journal Title
Journal ISSN
Volume Title
Multilink is a computational connectionist model inspired by the Bilingual Interactive Activation + (BIA+) model and the Revised Hierarchical Model (RHM) Dijkstra, Wahl, Buytenhuijs, Van Halem, Al-Jibouri, De Korte & Rekké (2019). It is able to simulate recognition, translation, and production of Dutch and English words varying in length and usage frequency for monolinguals and bilinguals. The activation time-course of orthographic, semantic, and phonological word representations in multiple tasks can be mimicked. In this project, we will assess the role of top-down inhibition from language nodes to words by varying relevant parameters in Multilink. By doing so we can investigate whether or not word recognition can be accelerated when we are aware of language information, knowing which language a target word belongs to. This amounts, in fact, to testing BIA+ against BIA, because the existence of language-to-word inhibition is assumed by BIA, but not by BIA+. Cycle times obtained from Multilink simulations are correlated with reaction times (RTs) from multiple datasets: the English Lexicon Project (ELP), Dutch Lexicon Project (DLP), and a dataset for English lexical decision collected by Mulder et al. (2018). Simulations allow us to assess whether there is any evidence for top-down effects from one language on the words of another language. For a broader understanding of processing mechanisms, lateral inhibition between word nodes, bottom-up facilitation from word-to-language nodes, second language proficiency, and neighborhood effects will also be investigated. To obtain well-fitting simulations, the contributions of model architecture, parameters, and lexical content are analyzed and adapted. No significant evidence had been found for top-down language-to-word inhibition. These results are in accordance with the BIA+ model.
Faculteit der Sociale Wetenschappen