cs.CLJun 5, 2026

Style or Content? Evaluating Style Classifiers with Controlled Content Overlap

Authors: Zhuo LiuHaozheng DuXiangxiang XuHangfeng He

Organizations: University of Rochester

Abstract

Style classifiers can use content cues that correlate with style labels in naturally collected data, yet we lack a systematic way to measure this reliance. We study this problem with a controlled content overlap setup built on parallel Bible translations. Specifically, we define the overlap parameter αα as the normalized residual of mutual information between content identity and style label, so that it measures how much content is shared across style classes: from no shared content (α=0α=0) to fully shared content (α=1α=1). Cross-overlap evaluation of RoBERTa-based classifiers shows that low-overlap models degrade when content cues are removed, while high-overlap models transfer more robustly. A cross-style content retrieval probe further shows that content becomes less recoverable as αα increases, with training dynamics showing this removal occurs gradually. Together, these results suggest that controlled overlap provides a simple diagnostic for separating style learning from content shortcuts.

Explore similar work

CardsList
  1. Content-Style Identification via Differential Independence

    May 18, 2026Subash Timilsina, Hoang-Son Nguyen, Sagar Shrestha +1Photorealistic Style TransferJacobian