Elamur Research⌕
☰
home / papers / 2026 / YNU-HPCC at SemEval-2026 Task 1: Constraint-Aware In-Context Learning for Multilingual Humor Generation

YNU-HPCC at SemEval-2026 Task 1: Constraint-Aware In-Context Learning for Multilingual Humor Generation

Paper · 2026 · Xulong Zhang, Jin Wang and Xuejie Zhang

Summary

The YNU-HPCC team built a system for SemEval-2026 Task 1, which asks models to write jokes starting from a news headline or from a pair of unrelated words, in English, Spanish and Chinese. They compared two strategies.

The first was an RLHF pipeline on Qwen2.5-3B: SFT on a mixed human/synthetic corpus, then PPO. The PPO reward combined an mdeberta-v3-base reward model, regex-based constraint penalties and a KL term. The second was in-context learning with GPT-5.2. A balanced library of strong example jokes was assembled first. For each test item, three random examples from that library were placed in the prompt to keep outputs varied.

The team could not run frequent human evaluations, so they tested internally with two automatic checks. One was constraint satisfaction (length limits and keyword inclusion). The other was blind pairwise win-rates judged by several LLMs. The ICL approach clearly beat both the PPO model and the vanilla baseline. The final submission placed 4th in English and 2nd in both Chinese and Spanish.

The study suggests that a small fine-tuned model has trouble with culture-specific humor in non-native languages. It also shows that following constraints perfectly does not make a text funny.

Key points

  • SFT data ("V13" dataset, 7,800 items):
    • 100 curated Chinese human jokes, upsampled to 500
    • 3,400 items generated by Qwen3-Max
    • 3,400 items generated by DeepSeek-V3.1
    • 500 weak early outputs, kept as negatives
  • Reward-model training: preference pairs were built automatically from quality tiers. The ordering was human > synthetic > filtered > constraint-violating outputs.
  • Method 2 (ICL): a 120-example "Golden Library" per language (60 headline-based, 60 word-pair-based), with three exemplars sampled at random for each prompt.
  • Keyword inclusion: Method 1 averaged 66.67% (only 44% in Chinese), compared with 98.67% for Method 2. The untuned baseline scored 100% on all constraints.
  • Win-rates: under the DeepSeek-V3.2 judge, Method 2 beat Method 1 by 0.94 (EN), 0.965 (ES) and 0.928 (ZH). Against the baseline, Method 2 won at rates above 0.92 according to the Qwen3.5-Plus and Llama 4 Scout judges.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓Yunnan UniversityKunming · China20 works · 40 citations

 Fields, topics and concepts

📚 Shelves 6The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksArts and Humanities · 17,450 worksLanguage and Linguistics · 17,296 worksPsychology · 1,649 worksSocial Psychology · 330 works

🧭 Topics 2Research subjects and languages the work is about.Language models & pretraining · 3,894 worksMultilingual & cross-lingual NLP · 882 works

Cite

Cite as Xulong Zhang, Jin Wang and Xuejie Zhang. 2026. YNU-HPCC at SemEval-2026 Task 1: Constraint-Aware In-Context Learning for Multilingual Humor Generation. Proceedings of the 20th International Workshop on Semantic Evaluation (2026).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.semeval-1.96

@inproceedings{zhang2026ynu,
  title = {{YNU-HPCC at SemEval-2026 Task 1: Constraint-Aware In-Context Learning for Multilingual Humor Generation}},
  author = {Xulong Zhang, Jin Wang and Xuejie Zhang},
  booktitle = {Proceedings of the 20th International Workshop on Semantic Evaluation (2026)},
  year = {2026},
  doi = {10.18653/v1/2026.semeval-1.96},
  url = {https://doi.org/10.18653/v1/2026.semeval-1.96}
}

References

18 references, 9 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Tom B. Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, Sandhini Agarwal, Ariel Herbert-Voss, Gretchen Krueger, Tom Henighan, Rewon Child, Aditya Ramesh, Daniel M. Ziegler, Jeffrey Wu, Clemens Winter,… — Language Models are Few-Shot Learners
  2. Santiago Castro, Luis Chiruzzo, Santiago Góngora, Salar Rahili, Naihao Deng, Ignacio Sastre, Victoria Amoroso, Guillermo Rey, Aiala Rosá, Guillermo Moncecchi, J. A. Meaney, Juan José Prada, and Rada Mihalcea. 2026. SemEval-2026 Task 1: MWA-HAHA, Models Write Automatic Humor And Humans Annotate. In P… find
  3. Shen Chen, Jin Wang, and Xuejie Zhang. 2025. YNU-HPCC at SemEval-2025 task3: Leveraging zero-shot learning for halluciantion detection. In Proceedings of the 19th International Workshop on Semantic Evaluation (SemEval-2025), pages 28–33, Vienna, Austria. Association for Computational Linguistics. find
  4. Ameet Deshpande, Vishvak Murahari, Tanmay Rajpurohit, Ashwin Kalyan, and Karthik Narasimhan. 2023. Toxicity in chatgpt: Analyzing persona-assigned language models. Preprint, arXiv:2304.05335.
  5. Ari Holtzman, Jan Buys, Li Du, Maxwell Forbes, and Yejin Choi. 2020. The curious case of neural text degeneration. Preprint, arXiv:1904.09751.
  6. Nabil Hossain, John Krumm, and Michael Gamon. 2019. “president vows to cut <taxes> hair”: Dataset and analysis of creative text editing for humorous headlines. In Proceedings of the 2019 Conference of the North, pages 133–142. Association for Computational Linguistics. find
  7. Sunghwan Kim, Dongjin Kang, Taeyoon Kwon, Hyungjoo Chae, Dongha Lee, and Jinyoung Yeo. 2025. Rethinking reward model evaluation through the lens of reward overoptimization. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 13252… find
  8. Rada Mihalcea and Carlo Strapparava. 2005. Making computers laugh: Investigations in automatic humor recognition. In Proceedings of Human Language Technology Conference and Conference on Empirical Methods in Natural Language Processing, pages 531–538, Vancouver, British Columbia, Canada. Association…
  9. Long Ouyang, Jeff Wu, Xu Jiang, Diogo Almeida, Carroll L. Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, John Schulman, Jacob Hilton, Fraser Kelton, Luke Miller, Maddie Simens, Amanda Askell, Peter Welinder, Paul Christiano, Jan Leike, and Ryan Lowe. 2022. Train…
  10. Joseph Tomasulo, Jin Wang, and Xuejie Zhang. 2020. Ynu-hpcc at semeval-2020 task 7: Using an ensemble bigru model to evaluate the humor of edited news titles. In Proceedings of the Fourteenth Workshop on Semantic Evaluation, pages 871–875. International Committee for Computational Linguistics. find
  11. Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, Dan Bikel, Lukas Blecher, Cristian Canton Ferrer, Moya Chen, Guillem Cucurull, David Esiobu, Jude Fernandes, Jeremy Fu, Wenyin Fu, and 49 others… — Llama 2: Open Foundation and Fine-Tuned Chat Models
  12. Lewis Tunstall, Edward Beeching, Nathan Lambert, Nazneen Rajani, Kashif Rasul, Younes Belkada, Shengyi Huang, Leandro von Werra, Clémentine Fourrier, Nathan Habib, Nathan Sarrazin, Omar Sanseviero, Alexander M. Rush, and Thomas Wolf. 2023. Zephyr: Direct distillation of lm alignment. Preprint, arXiv… find
  13. Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Brian Ichter, Fei Xia, Ed Chi, Quoc Le, and Denny Zhou. 2023. Chain-of-thought prompting elicits reasoning in large language models. Preprint, arXiv:2201.11903.
  14. Diyi Yang, Alon Lavie, Chris Dyer, and Eduard Hovy. 2015. Humor recognition and humor anchor extraction. In Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing, pages 2367–2376. Association for Computational Linguistics.
  15. Lianmin Zheng, Wei-Lin Chiang, Ying Sheng, Siyuan Zhuang, Zhanghao Wu, Yonghao Zhuang, Zi Lin, Zhuohan Li, Dacheng Li, Eric P. Xing, Hao Zhang, Joseph E. Gonzalez, and Ion Stoica. 2023. Judging llm-as-a-judge with mt-bench and chatbot arena. Preprint, arXiv:2306.05685.
  16. Shanshan Zhong, Zhongzhan Huang, Shanghua Gao, Wushao Wen, Liang Lin, Marinka Zitnik, and Pan Zhou. 2024. Let’s think outside the box: Exploring leap-of-thought in large language models with creative humor generation. Preprint, arXiv:2312.02439. — [arXiv:2312.02439](https://arxiv.org/abs/2312.02439) find
  17. Chunting Zhou, Pengfei Liu, Puxin Xu, Srini Iyer, Jiao Sun, Yuning Mao, Xuezhe Ma, Avia Efrat, Ping Yu, Lili Yu, Susan Zhang, Gargi Ghosh, Mike Lewis, Luke Zettlemoyer, and Omer Levy. 2023. Lima: Less is more for alignment. Preprint, arXiv:2305.11206. find
  18. Daniel M. Ziegler, Nisan Stiennon, Jeffrey Wu, Tom B. Brown, Alec Radford, Dario Amodei, Paul Christiano, and Geoffrey Irving. 2020. Fine-tuning language models from human preferences. Preprint, arXiv:1909.08593. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/zhang-2026-ynu-hpcc-semeval-2026-task/