ERABAL: Enhancing Role-Playing Agents through Boundary-Aware Learning

TL;DR

ERABAL is a novel framework that improves role-playing agents' consistency by using boundary-aware learning, requiring fewer dialogues and outperforming baseline models in multiple evaluation benchmarks.

Contribution

The paper introduces ERABAL, a boundary-aware learning framework that enhances role-playing agents' consistency with less training data and improved performance.

Findings

ERABAL outperforms baseline models on WikiRoleEval, CharacterEval, and MT-Bench.

ERABAL requires significantly fewer dialogues for training.

The framework is both efficient and effective in maintaining role consistency.

Abstract

Role-playing is an emerging application in the field of Human-Computer Interaction (HCI), primarily implemented through the alignment training of a large language model (LLM) with assigned characters. Despite significant progress, role-playing agents (RPLAs) still struggle with maintaining role-consistency across conversations, particularly when confronted with boundary queries subtly related to character attributes. In this paper, we present ERABAL, a framework aimed at enhancing RPLAs' role-playing capabilities through boundary-aware learning. ERABAL encompasses a generation pipeline for role-specific dialogues and a concomitant methodology for alignment training. Through comprehensive evaluations, we demonstrate that ERABAL is both efficient and effective. By training with significantly fewer dialogues than those used in leading approaches, ERABAL achieves notable improvements across WikiRoleEval, CharacterEval, and the role-playing subset of MT-Bench compared to the generalist baseline models. Our code and datasets will be made publicly available to support further research.