It is widely accepted in linguistics that hierarchical structures serve as the underlying syntactic structure of natural language. What happens if language is linearized? That is, for a given sentence, untransform” the hierarchical form used to mark a sentence of a particular type, and convert it to a “linearized” form by reordering it and adding artificial markers at some fixed location to signal the syntactic form. No natural, human language works this way, and adults attempting to learn such artificial languages do not use the same brain regions as when they engage with normal, human language syntax. In this project, we aim to test the language acquisition capabilities of deep neural networks and currently popular transformer systems with such artificial languages.

I’m participating in SuperUROP because I’m excited to work on the intersection of two of my interests, computer science and linguistics. I want to apply the classes I’ve taken on machine learning and linguistics to an interesting research question. I hope to get hands-on experience with natural language processing and gain a better understanding of the steps of the research lifecycle.