Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

When Harry Meets Superman: The Role of The Interlocutor in Persona-Based Dialogue Generation

About

Endowing dialogue agents with persona information has proven to significantly improve the consistency and diversity of their generations. While much focus has been placed on aligning dialogues with provided personas, the adaptation to the interlocutor's profile remains largely underexplored. In this work, we investigate three key aspects: (1) a model's ability to align responses with both the provided persona and the interlocutor's; (2) its robustness when dealing with familiar versus unfamiliar interlocutors and topics, and (3) the impact of additional fine-tuning on specific persona-based dialogues. We evaluate dialogues generated with diverse speaker pairings and topics, framing the evaluation as an author identification task and employing both LLM-as-a-judge and human evaluations. By systematically masking or disclosing information about the interlocutor, we assess its impact on dialogue generation. Results show that access to the interlocutor's persona improves the recognition of the target speaker, while masking it does the opposite. Although models generalise well across topics, they struggle with unfamiliar interlocutors. Finally, we found that in zero-shot settings, LLMs often copy biographical details, facilitating identification but trivialising the task.

Daniela Occhipinti, Marco Guerini, Malvina Nissim• 2025

Related benchmarks

TaskDatasetResultRank
Distribution AlignmentPopAlign-Bench Average
Behavioral Alignment Score25.8
9
Distribution AlignmentPopAlign-Bench Public-B
Behav-JS0.232
9
Distribution AlignmentPopAlign-Bench Reddit-A
Behav-JS28.4
9
Distribution AlignmentPopAlign-Bench Public-A
Behav-JS25.4
9
Dialogue Quality CalibrationPopAlign-Bench (test)
Average Score5.44
9
Distribution AlignmentPopAlign-Bench Reddit-B
Behav-JS0.265
9
Showing 6 of 6 rows

Other info

Follow for update