This article by Alexandra DeLucia et al. addresses the challenge of creating unique personas for dialogue agents, particularly in non-real-world settings such as fantasy. Existing models, mainly trained on PersonaChat, struggle to extract high-quality persona information in unfamiliar settings. The authors propose a post-hoc adaptation of a trained persona extraction model through a natural language inference method. This method, inspired by the literature of dialog natural language inference, aims to extract structured persona information from dialogue. It offers higher-quality persona extraction and requires less human annotation, making it a promising approach for dialogue agents in new domains.
Publication date: 12 Jan 2024
Project Page: https://arxiv.org/abs/2401.06742
Paper: https://arxiv.org/pdf/2401.06742
