Project Details
Project name
Frame Blending by LLMs
Project Summary
Despite LLMs’ proficiency in various tasks, they struggle with incorporating characteristics like frame blending into sentence generation, a concept well-developed in linguistics and Frame Semantics. The significance of frame semantics lies in understanding words based on conceptual structures, enhancing insights into linguistics, and contributing to fields like economics and politics. The research aims to employ training, fine-tuning, and prompt engineering technologies when using open-source LLMs for generating frame-blending examples. The proposed method involves leveraging FrameNet data for training and fine-tuning LLMs to enhance frame-blending capabilities. The expected outcome is improved frame blending and overall semantic understanding in LLMs, contributing to a deeper exploration of their capabilities and future prospects. Furthermore, a detailed schedule is provided at the end of this proposal.
Mentor
Wenyue Xi, Mark Turner