Reproduce “When Harry Meets Superman: The Role of The Interlocutor in Persona-Based Dialogue Generation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Daniela Occhipinti, Marco Guerini, Malvina Nissim
This research systematically investigates how an interlocutor's profile influences persona-based dialogue generation, finding that their biographical information significantly aids in identifying the target speaker's persona. The study reveals that fine-tuning leads to a deeper, more integrated persona representation compared to zero-shot models, which often exhibit superficial 'copy-paste' behaviors, especially when interacting with unfamiliar conversational partners.
Key insights
- The interlocutor's biographical information is a critical cue for identifying the target speaker's persona in generated dialogues, outperforming the impact of just conversational turns.
- Dialogue models demonstrate robust persona adaptation across unfamiliar topics but struggle significantly when faced with unfamiliar interlocutor pairings, indicating limitations in generalizing to unexpected social contexts.
- Fine-tuning models on specific persona-based dialogues promotes a deeper, less trivial integration of persona traits, moving beyond the superficial 'copy-paste' behavior observed in zero-shot generations.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.