My current research interests are as difficult to pin down as my music taste.
I am generally motivated by ideas that 1) improve our understanding of how things (e.g. black box models, natural languages, cognition, the universe...) work, 2) promote safety, inclusivity, and fairness, 3) become elegant solutions for really hard problems that will work well in the long run.
Here are some topics that I have worked on:
Model Interpretability: I am interested in understanding how current NLP models use contextual information to make decisions during language generation (EMNLP'22) and using this information to guide models to process context more efficiently. I am also interested in how neural networks process language on a fundamental level and how machine intelligence compares with human cognition.
Sign Language Processing: I am interested in modeling signed languages from a linguistic perspective and extending existing language technologies to signed languages.
I have argued the importance for the NLP community to include signed languages, both socially and scientifically, and how to get involved (ACL'21), I researched how to translate a signed language into a spoken language (ECCV'20, COLING'20),
how to perform data augmentation for Sign Language Translation (MTSummit21), and how to perform coreference resolution of pronominal pointing signs (EMNLP'21).
Context-aware Machine Translation: I am interested in when context, either on an intra-sentential (within the current sentence), inter-sentential (across multiple sentences), or extra-linguistic (e.g. social, temporal, cultural) level, is required during translation, and how to model these features in machine translation.
My works examine when translation requires context and how well models perform on these translations (arxiv'21), how much context-aware models are actually using context (ACL'21), whether models are using the type of context we expect them to (ACL'21), and how to encourage models to use more/better context.
Please reach out if you'd like to chat or collaborate! I am generally responsive to emails and Twitter messages, I do not check LinkedIn very often, and I prefer that people do not contact me on social media that I have not listed on this website.