ArcANE: Do Role-Playing Language Agents Stay in Character at the Right Time?
Abstract
Role-playing language agents require dynamic character development that evolves through narratives, necessitating benchmarks that evaluate psychological trajectory alignment rather than static factual recall, with ArcANE demonstrating superior performance when character arc information is conditioned into models.
Role-playing language agents (RPLAs) should play characters whose values and behavior evolve as the story progresses, not maintain a fixed persona. Existing benchmarks measure factual recall at a given chapter, not whether responses align with the character's psychological trajectory, especially in scenarios the source text never explores. We introduce ArcANE (Arc-Aware Narrative Evaluation), an automatically constructed benchmark spanning 17 novels and 80 principal characters. A Character Arc segments the narrative into phases along a psychological axis, and each probe poses the same scenario across phases, spanning both situations within the source text and situations beyond it. Across six models and six context modes, conditioning on the Character Arc tops every other context strategy on every model, and the gap is largest on scenarios outside the source text where retrieval has nothing to find. We further fine-tune open-weight models on the same data to obtain ArcANE-8B/32B, which widen the Arc advantage even more on scenarios outside the source text.
Community
Role-playing language agents (RPLAs) should play characters whose values and behavior evolve as the story progresses, not maintain a fixed persona. Existing benchmarks measure factual recall at a given chapter, not whether responses align with the character's psychological trajectory, especially in scenarios the source text never explores. We introduce ArcANE (Arc-Aware Narrative Evaluation), an automatically constructed benchmark spanning 17 novels and 80 principal characters. A Character Arc segments the narrative into phases along a psychological axis, and each probe poses the same scenario across phases, spanning both situations within the source text and situations beyond it. Across six models and six context modes, conditioning on the Character Arc tops every other context strategy on every model, and the gap is largest on scenarios outside the source text where retrieval has nothing to find. We further fine-tune open-weight models on the same data to obtain ArcANE-8B/32B, which widen the Arc advantage even more on scenarios outside the source text.
Really compelling framing. Shifting evaluation from static factual recall to psychological trajectory alignment feels like the right direction for RPLAs, and the out-of-source-text scenarios are a clever stress test where retrieval genuinely can't help.
I'm curious how the Character Arc conditioning holds up on characters with non-linear or ambiguous arcs (like unreliable narrators), where the psychological axis might be harder to segment cleanly. Nice work!
the arc-grounded context architecture is neat, but i'm curious how it handles non-linear character arcs where phases loop back or diverge after long lulls. what happens when arcs loop back or reset after big plot twists, and does that hurt phase fidelity when the text lets the character revisit earlier states? i’d also love an ablation where you remove the per-phase trajectory prompt and compare to a plain temporal cue, to see how much the arc signal actually buys you in out-of-text scenarios. btw the arxivlens breakdown helped me parse the method details, it covers the arc-grounded prompt design nicely: https://arxivlens.com/PaperView/Details/arcane-do-role-playing-language-agents-stay-in-character-at-the-right-time-2734-205d827d
Models citing this paper 0
No model linking this paper
Datasets citing this paper 0
No dataset linking this paper
Spaces citing this paper 0
No Space linking this paper