C-VARC: A Large-Scale Chinese Value Rule Corpus for Value Alignment of Large Language Models

TL;DR

This work addresses the Western bias in existing AI value benchmarks by introducing a culturally adaptive Chinese value framework and the Chinese Value Rule Corpus (C-VARC). C-VARC comprises over 250k value rules organized in a three-dimension hierarchy (national, societal, personal) with 12 core and 50 derived values, augmented by a rule-extraction and annotation pipeline that yields high-quality, scalable resources. The authors demonstrate that C-VARC-guided scenario generation improves thematic relevance and diversity, and that Chinese LLMs preferentially align with C-VARC compared to Western benchmarks, supported by human judgments. Additionally, they develop a rule-based method to automatically generate 404k moral dilemmas and analyze cross-model alignment across 17 LLMs, underscoring C-VARC’s utility for scalable value-alignment evaluation in Chinese contexts and its potential as a broadly applicable benchmark for culturally grounded AI ethics.

Abstract

Ensuring that Large Language Models (LLMs) align with mainstream human values and ethical norms is crucial for the safe and sustainable development of AI. Current value evaluation and alignment are constrained by Western cultural bias and incomplete domestic frameworks reliant on non-native rules; furthermore, the lack of scalable, rule-driven scenario generation methods makes evaluations costly and inadequate across diverse cultural contexts. To address these challenges, we propose a hierarchical value framework grounded in core Chinese values, encompassing three main dimensions, 12 core values, and 50 derived values. Based on this framework, we construct a large-scale Chinese Value Rule Corpus (C-VARC) containing over 250,000 value rules enhanced and expanded through human annotation. Experimental results demonstrate that scenarios guided by C-VARC exhibit clearer value boundaries and greater content diversity compared to those produced through direct generation. In the evaluation across six sensitive themes (e.g., surrogacy, suicide), seven mainstream LLMs preferred C-VARC generated options in over 70.5% of cases, while five Chinese human annotators showed an 87.5% alignment with C-VARC, confirming its universality, cultural relevance, and strong alignment with Chinese values. Additionally, we construct 400,000 rule-based moral dilemma scenarios that objectively capture nuanced distinctions in conflicting value prioritization across 17 LLMs. Our work establishes a culturally-adaptive benchmarking framework for comprehensive value evaluation and alignment, representing Chinese characteristics.

Figures (16)

• Figure 1: Examples of Western benchmark values conflicting with Chinese values. These cases, manually curated from SC101forbes2020social and MICziems2022moral, illustrate but do not exhaust the broader spectrum of ethical divergence.
• Figure 2: The Chinese value framework. The framework proposed in this paper is based on the socialist core values, detailing three dimensions and 12 core values, and further expanding to include 50 derived values. The comparison process with main Chinese benchmarks is demonstrated in Appendix \ref{['A.1']}.
• Figure 3: The overall process of constructing C-VARC. The blue box represents the data filtering and selection process of rules from SC101forbes2020social and MICziems2022moral, while the orange box depicts the process of constructing rules based on the Chinese cultural context. A detailed description of the construction process can be found in Appendix \ref{['A']}.
• Figure 4: Distribution of data across the three value levels in the C-VARC.
• Figure 5: The t-SNE visualization of generated scenarios. (a) presents the dimensionality-reduced visualization of 1,200 scenarios directly generated by the LLM; (b) shows the dimensionality-reduced visualization of 1,200 scenarios generated with rule guidance.