Please see my Google Scholar page for the most up-to-date list of my publications.

Verb semantics

Currently, one of my foremost interests is verb semantics - how does the human brain process and reason over events, and what are the similarities to how LLMs process and "reason" over events? What is hard about events and verbs and what is easy? I have focused mainly on the two computational semantic tasks called semantic role labeling and semantic proto-role labeling to try to address these (and similar) questions.

Some findings that I've published around these topics:

• A BERT-based model can pretty easily extract positions of predicates, arguments, and 10+ semantic properties per argument in one model pass. [Demo] [Paper]
• LLMs seem to have consistent event role knowledge which aligns with linguistic theory, but it's unclear how to exploit this in prompting for two related tasks, SRL and SPRL. [Paper] [Code]
• Dowty's proto-role properties, when used to label raw text, mirror human perceptions of agency and patiency of various entities. [Paper] [Code]

Improving Wikidata's lexicographic data

For a few years, I've also been involved with a project dedicated to improving the verb and semantic role knowledge in Wikidata using PropBank. We're currently in the process of adding the information contained in 11,300+ PropBank rolesets to Wikidata's English lexemes. The project has taken a few twists and turns and our next publication is currently in progress, but you can follow the breadcrumbs we've left on previous iterations of this effort: [Paper #1] [Paper #2] [Data]