Ellipsis-Dependent Reasoning: a New Challenge for Large Language Models

We propose a novel challenge for large language models: ellipsis-dependent reasoning. We define several structures of paired examples, where an  ellipsis example is matched to its non-ellipsis counterpart, and a question is posed which requires resolution of the ellipsis. Test results show that the best GPT-3 models perform well on non-elliptical examples but struggle with all but the simplest ellipsis structures. I also report on current work with a variety of recent models, including GPT4.

This talk will be given by Daniel Hardt, Copenhagen Business School (CBS). 

The Center for Language Generation and AI at Aarhus University is committed to fostering the interchange of ideas and support for researchers in the area of language generation and AI. By opening our biweekly talks to the public, we aim to create a platform for knowledge sharing and collaboration among researchers, students, and the broader community.

These talks will be conducted virtually, allowing attendees from around the world to participate. Registration details and links to join each talk will be provided on our website. All talk will take place at 13:30 (CEST) on Zoom. The link is: https://aarhusuniversity.zoom.us/j/62020950492

For more information and updates on our biweekly talks, please check our news page or follow us on LinkedIn or Twitter.

For media inquiries and questions, please contact: pascale.moreira@cc.au.dk