The paper titled ‘Rethinking Response Evaluation from Interlocutor’s Eye for Open-Domain Dialogue Systems’ focuses on the need for more accurate evaluation methods for open-domain dialogue systems. The authors argue that current automatic evaluation methods may not accurately approximate an interlocutor’s judgments. They conducted experiments using the Hazumi dataset and Twitter conversations, finding that interlocutor awareness is crucial for automatic response evaluation. The study also highlights the challenges in evaluating generated responses compared to human responses.
Publication date: 4 Jan 2024
Project Page: https://arxiv.org/abs/2401.02256v1
Paper: https://arxiv.org/pdf/2401.02256
