- Linguistics in Autism
- Utilizing word vectors, we analyze semantic similarity in both typically-developing children and children with autism spectrum disorder (ASD)
- Code repository (private): https://github.com/langcomp/vectoraut
- Using Language Models to Understand Cognition
- Do better language models better predict cognitive processes measured by eye-tracking? Using a neural net-based language model, we use generalized additive models (GAMs) to investigate how language models with lower perplexity can better model eye-tracking data.
The essay below is more of a Research Vision. It was part of my application for the RSG Northwestern Program. For another take on my research, you can also check out my thoughts on Computational Linguistics vs Natural Language Processing.
At the beginning of the 21st-century, the field of linguistics is at a crossroads. On the one hand, formal and theoretical practitioners have made enormous gains in the last fifty years, shaping the scientific understanding of language and cognition. On the other hand, computer scientists and engineers are making similarly significant strides in tackling linguistics tasks by relying solely on the statistical distributions of words and letters, irrespective of any linguistic structures or cognitive insights. As a result, the subfields of linguistics are becoming increasingly Balkanized, and the entire field comes across as out-of-touch or irrelevant to general, practical concerns.
The truth, however, is exactly the opposite: understanding linguistics can provide immense insights into understanding human motivation and can improve the efficacy of many language engineering tasks. To accomplish this, though, linguists need to better communicate with both each other and the general public.
By participating in the RSG Research Communication Training Program, I hope to become a better facilitator to help bridge these gaps. I have always felt comfortable in both quantitative- and humanities-based fields, and understand how to explain each side to the other. I often find myself explaining technical concepts to colleagues and peers by translating (but not diluting) technical information into terms that they find both interesting and understandable. Similarly, I feel comfortable expressing abstract philosophical concepts in more definite terms that can then be converted to computer programs or algorithms.
As noted on the RSG website, “…it is increasingly important to convey the results of…research clearly and concisely to a broad variety of audiences.” I am acutely aware of this, both in general and in my specific field. I look forward to the opportunity to refine my skills in this area, to help advance scientific progress both within and outside of academia.