Cross-domain coreference modeling in dialogue state tracking with prompt learning

Heng-Da Xu, Xian-Ling Mao, Puhai Yang, Fanshu Sun, Heyan Huang

Abstract: Highlights•Propose a dialogue state tracking model to track cross-domain coreference slots.•Design a domain-specific prompt learning method to predict slots better.•Conduct experiments on the MultiWOZ 2.3 dataset, achieving a new SoTA performance.•Propose a data cleaning algorithm to clean coreference annotations and reduce noise.