# Evaluating Bias in Chinese Large Language Models across Tasks and Registers: Task Operationalization Effects and Irony Classification Risks

arXiv: https://arxiv.org/abs/2610.14464 · HTML: https://mlsanity.com/paper/2610.14464/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 21 of its 38 references that are in the map.
- First posted on arXiv 10 Oct 2026.

## Builds on (most cited first, 21 of 21)

- [Training language models to follow instructions with human feedback](https://mlsanity.com/paper/2203.02155/) (2022)
- [Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena](https://mlsanity.com/paper/2306.05685/) (2023)
- [LIMA: Less Is More for Alignment](https://mlsanity.com/paper/2305.11206/) (2023)
- [StereoSet: Measuring stereotypical bias in pretrained language models](https://mlsanity.com/paper/2004.09456/) (2020)
- [Semantics derived automatically from language corpora contain human-like biases](https://mlsanity.com/paper/1608.07187/) (2016)
- [ChatGPT Outperforms Crowd-Workers for Text-Annotation Tasks](https://mlsanity.com/paper/2303.15056/) (2023)
- [CrowS-Pairs: A Challenge Dataset for Measuring Social Biases in Masked Language Models](https://mlsanity.com/paper/2010.00133/) (2020)
- [BBQ: A Hand-Built Bias Benchmark for Question Answering](https://mlsanity.com/paper/2110.08193/) (2021)
- [The Woman Worked as a Babysitter: On Biases in Language Generation](https://mlsanity.com/paper/1909.01326/) (2019)
- [Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training](https://mlsanity.com/paper/2401.05566/) (2024)
- [Justice or Prejudice? Quantifying Biases in LLM-as-a-Judge](https://mlsanity.com/paper/2410.02736/) (2024)
- [Do LLMs exhibit human-like response biases? A case study in survey design](https://mlsanity.com/paper/2311.04076/) (2023)
- [Fundamental Limitations of Alignment in Large Language Models](https://mlsanity.com/paper/2304.11082/) (2023)
- [Benchmarking Cognitive Biases in Large Language Models as Evaluators](https://mlsanity.com/paper/2309.17012/) (2023)
- [Evaluating and Mitigating Discrimination in Language Model Decisions](https://mlsanity.com/paper/2312.03689/) (2023)
- [COLD: A Benchmark for Chinese Offensive Language Detection](https://mlsanity.com/paper/2201.06025/) (2022)
- [Facilitating Fine-grained Detection of Chinese Toxic Language: Hierarchical Taxonomy, Resources, and Benchmarks](https://mlsanity.com/paper/2305.04446/) (2023)
- [ToxiCloakCN: Evaluating Robustness of Offensive Language Detection in Chinese with Cloaking Perturbations](https://mlsanity.com/paper/2406.12223/) (2024)
- [CHBias: Bias Evaluation and Mitigation of Chinese Conversational Language Models](https://mlsanity.com/paper/2305.11262/) (2023)
- [Explicit vs. Implicit: Investigating Social Bias in Large Language Models through Self-Reflection](https://mlsanity.com/paper/2501.02295/) (2025)
- [Irony in Emojis: A Comparative Study of Human and LLM Interpretation](https://mlsanity.com/paper/2501.11241/) (2025)

## Abstract (arXiv)

In Chinese content governance, large language models are increasingly used to identify toxic and discriminatory expressions, yet both task operationalization and ironic expression can alter bias measurement readings. We conducted a preregistered multi-task behavioral audit of multiple Chinese large language models. In the content classification experiment, the core panel's classification error rate rose from 10.9% in the formal register to 23.5% in the yinyang (an echoic-irony register) version, a risk difference of +12.7 percentage points. (95% CI [+9.7, +15.7]). Inter-rater agreement was lower in the yinyang register (majority-vote kappa = 0.299), so the increase contains a component of label uncertainty propagation, but the direction of increase was consistent under both adjudicated and majority-vote reference standards. Explicit irony-aware prompts reduced the intervention panel's error rate by 9.6 percentage points. We operationalized the 29 shared proposition anchors into three task forms: social consensus estimation (Form A), high-stakes decision (Form B), and controlled association rating (Form C); independent manipulation checks found that all 58 A-C comparison units received unanimous agreement, whereas the low A-B convergence was confirmed by calibrated majority adjudication (87.8% supporting the minority judgment); on the irony-probe materials, the interactions of B and C relative to A were +0.577 and +1.121, respectively. Thus, when evaluating large language models for Chinese content governance, the effect of task operationalization on bias readings is an empirical question that requires data to answer, and safety evaluation should incorporate the ironic register as a test condition.