NSF Award
2335702
The FrameNet team has built a super-dictionary of English that humans can read and computers can also use for natural language processing (NLP). Words that describe the same kind of event with the same set of "roles" are grouped together in a SEMANTIC FRAME. Over the past 25 years, FrameNet researchers have defined more than 1,224 frames for 13,687 words and labeled 202,978 examples of them. This project will shift the FrameNet public website from several aging servers to cloud hosting, so that students and NLP researchers will be able to read and download the data for years to come.<br/><br/>The FrameNet website is a vital point of entry into Frame Semantics for thousands of visitors. Both casual visitors and researchers can either search for lexical units to find frames, or search frames to find lexical units. Computational Linguistics research has benefited from the influence of the theory of Fillmore's Frame Semantics, and FrameNet data has been widely used in the development of systems for NLU. For example, recent research has creatively combined statistical and symbolic approaches to NLU, such as using BERT (Devlin et al. 2019) to improve automatic semantic role labeling, or injecting FrameNet labels into neural nets.<br/><br/>FrameNet has had much broader impact than initially envisioned. Instructors use FrameNet to introduce students at all levels to lexical semantics. The thousands of downloads of FrameNet data have sparked various studies outside of traditional NLP, e.g. detection of moral values in tweets, explorations of framing in religious and political texts. Berkeley FrameNet has inspired development of FrameNets in a dozen languages and research on cross-lingual projections of frames and frame shifts when translating. The Berkeley FrameNet project has also hired many students from diverse backgrounds, many of whom have gone on to successful careers in NLP and/or academia.<br/><br/>The URL of the project is https://framenet.icsi.bekeley.edu Because the previous servers for the project are aging and will be retired soon, this URL has been redirected to a duplicate of the FrameNet public website on a server at Universidade Federal de Juiz de Fora in Brazil, thanks to help from colleagues there. The purpose of this project is to recreate the website and database on a cloud-hosting platform in the US, and maintain it there for at least five years. The FrameNet lexical database, including all frames, frame elements, lexical units, and annotations will be retained indefinitely at secure locations worldwide.<br/><br/>This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.
