For this study proposal, I conducted a brief literature review on multilingualism and its linguistic and cognitive impacts. I found that multilingualism fosters cognitive flexibility, enhances metalinguistic awareness, and supports the positive transfer of linguistic skills across languages.

I then proposed a computational model to explore this more thoroughly and quantitatively. Specifically, I aimed to investigate (1) whether knowing multiple languages facilitates the acquisition of new languages and (2) whether the similarity between known and new languages affects this process. I hypothesized that knowing multiple languages will positively impact future language acquisition, especially for similar languages.

The proposed model will leverage Recurrent Neural Networks (RNNs), using cross-entropy loss to assess next-word prediction, sentence comprehension, and sentence production. It will be trained unsupervised on a multilingual corpus comprising texts in various languages. The training process will center around two elements: training loss and generalization tests. For training loss, the model will optimize for next-word prediction and sentence comprehension; for generalization tests, held-out data from unseen languages will be used to evaluate the model's ability to generalize learning across different linguistic contexts.

The model’s performance will be based on its learning efficiency and its sentence comprehension and production. Learning efficiency will be based on the rate of language acquisition and quantified by training loss over time, while its performance on sentence comprehension and production will be compared with both monolingual models and human benchmarks.

The proposed study's implications could illuminate multilingualism’s cognitive effects, inform language teaching methodologies, and contribute to the development of more effective multilingual educational systems. As multilingualism becomes more prevalent, it is crucial to understand this phenomenon’s cognitive effects.