Research in targeted model editing, interpretability, and concept erasure from neural networks.Joint post-doc with David Bau (Northeastern) and Yonatan Belinkov (Technion – Israel Institute of Technology).

Research in multilingual NLP, machine translation, and syntactic evaluation of neural language models.

Artificial Intelligence System Design and Development, Spring 2020.

Research in improving few-shot intent classification.

Research in multilingual neural machine translation, entity linking, and word alignment. Also improved the speed-accuracy trade-off in our MT system by implementing convolutional MT models. Work funded by DARPA as part of the AIDA (Active Interpretation of Disparate Alternatives) project.

Various research projects in Machine Learning, Natural Language Processing, and Corpus Linguistics. Affiliations included the Institutional Research and Advanced Analytics team, Department of Linguistics, the Computer Vision Lab (Dept. of Computer Science).Projects included machine-learning-based student grade prediction; neural machine translation for low-resource language varieties; endangered language preservation through the creation of educational resources; and diachronic analyses of the linguistic expression of evolving social attitudes toward gender and sexuality. Presented research at the Illinois Language and Linguistics Society Conference, the Toronto Undergraduate Linguistics Conference, and the National Conference on Undergraduate Research.

Worked with graduate students and faculty in the Statistical Social Language Analysis (SLANG) Lab to improve an entity-event extraction system that, given news articles, automatically extracts the names of people who have been killed by police in the United States. Funded by the National Science Foundation as part of the Research Experiences for Undergraduates program.

Worked with graduate students, postdocs, and faculty at the Center for Language & Speech Processing (CLSP) to create phrase-based translation systems for rare languages, following the specifications of DARPA's Low-Resource Languages for Emergent Incidents (LORELEI) program. Responsibilities included building a morphological generator for Crimean Tatar, extracting parallel training data by writing web data parsers, writing tokenizers for multilingual corpora, and creating a lemma-based English-Uyghur translation model (increased BLEU by over 60%).