Research & Projects
Research & Projects
Properties of Figure and Ground Identified by LMs
I am developing a dataset of sentences attributing properties typical of figures and grounds to nonce words. The dataset will be used to test if LLMs and VLMs are sensitive to the ordering of figures and grounds in prepositional phrases.
Developing a Tagging Scheme for Text Summaries
I am currently working with Professor Amir Zeldes on developing an annotation scheme for tagging what information in a document is present in summaries of the document produced by humans and LLMs. Once the annotation scheme is complete, a portion of the GUM corpus will be tagged to study what types of information from a text are most frequently reflected in the summaries and how human and LLM summaries differ.