My research centers around the development of learner language, both that of second language learners and novice members of a professional field. My methods are mostly data-driven, using quantitative modeling to analyze the relationship between the language output of learners and variables from their environment and background.
Methods of investigation: Corpus linguistics, computational linguistics, quantitative modeling.
My dissertation focuses on the disciplinary writing development of college students, aiming to find out the developmental trajectories of student writing and the gap between students and experts. I use multi-dimensional analysis (MDA) to find the dimensions that characterize the variation in writings of L1 vs. L2, students vs. experts, and various disciplines. The findings will further the understanding of the writing gap and the need for improvement that should be addressed for purposes beyond higher education
Work in progress.I have worked on three projects at the intersection of these three subfields, all related to native language identification (NLI), a machine learning task that intends to identify one's native language based on one's L2 production. The task is based on the idea that cross-linguistic transfers leave "fingerprints" on the L2 production, thus making it possible to detect one's native language. I approach the task not from an engineering perspective -- focusing on achieving high accuracy -- but rather with the goal of gaining linguistic insights from the model results.
Using NLI to Identify Potential Collocation Transfer in L2 English
We discuss the potential of using the NLI task to automatically identify collocation transfer. Using training data in English, the model findings implied the potential to apply the method to understudied L2 languages to efficiently capture potential collocation transfers. The study also focuses on the interpretation of NLI results, revealing collocation transfer patterns.
https://doi.org/doi: 10.18653/v1/2025.bea-1.49Studying Cross-linguistic Structural Transfer in Second Language Learning
This is a study of large-scale cross-linguistic L2 data (274 L1-L2 pairs), where we revealed the systematic influence of L1 on L2 production, manifested through the aspect of morphosyntax. We also show that certain morphosyntactic features (e.g., verb aspect, person of auxiliary, etc.) are more predictive of the native languages than others (e.g., main constituent order, number of auxiliary).
https://escholarship.org/uc/item/2g87d428Typological influence on L2 and L2
We take one step further to understand the impact of L1 typology on L2 production. Does L2 production from two typologically similar L1s share similar features? How is the production pattern impacted by the typological difference between L1s and L2s?
Work in progress.Noun phrases have been identified as part of the causes behind the literacy crisis of U.S. Could there be something from the English language itself that causes further processing difficulty? Based on psycholinguistic theories of embodied simulation (Bergen, 2015) and chunking in verbal working memory (Cowan et al., 2012), I hypothesize that the head direction in English and abstract head nouns in mathematics together put extra load on working memory, a load that would not exist in right-headed languages. The hypothesis suggests that re-writing noun phrases with concrete items at phrase-beginning help to alleviate reading challenge experienced in learners of math.
Work in progress.