Elamur Research⌕
☰
home / papers / 2025 / Colloquial Singaporean English Style Transfer with Fine-Grained Explainable Control

Colloquial Singaporean English Style Transfer with Fine-Grained Explainable Control

Paper · 2025 · Jinggui Liang, Dung Vo, Yap Hong Xian et al.

Summary

The paper addresses style transfer between Colloquial Singaporean English (Singlish) and Standard English. Existing text style transfer methods rarely explain their edits or let users control specific linguistic features. The authors make two contributions. The first is ExpCSEST, a non-parallel corpus of about 140K formal and informal sentences. The text was scraped from the EDMW forum, the Straits Times, and Prime Minister's Office communications. Each sentence carries free-text explanations along six aspects: Syntax, Lexical Borrowing, Pragmatics, Prosody/Phonology, Emoticons/Punctuation and Code-Switching. Annotation began with hand-labelled seed cases and was expanded through in-context prompting of LLMs. The second contribution is MACoE-Style, a multi-agent framework inspired by the "Society of Mind" idea. One LLM agent is assigned to each aspect, and the agents iteratively propose, critique and refine rewrites. Automatic metrics and human judgements show controlled, interpretable transfers in which every change can be traced to a named linguistic aspect. This matters for education, localization and NLP for under-resourced English varieties.

Key points

  • ExpCSEST holds 104,601 informal and 37,676 formal utterances. Informal ones average 23.41 tokens and formal ones 31.10.
  • The taxonomy fixes six aspects (SYN, LEX, PRA, PRO, EMO, COD), but the explanation for each aspect is written freely rather than chosen from a closed label set.
  • Five in-context exemplars per agent worked best. Using 3, 7 or 9 exemplars gave weaker results.
  • When agents ran on a single model, Mistral Large did best, ahead of ChatGPT and Claude. The mixed ChatGPT+Mistral setup outperformed ChatGPT alone.
  • Performance peaked after the first collaboration round, and extra rounds (up to 3) brought no gain.
  • On 200 English-to-Singlish samples, MACoE-Style led in style adaptation. It trailed a single-LLM Direct Prompt with explanation on BARTScore, which suggests a trade-off between style and meaning preservation.

Interesting ideas

  • I1. A closed set of aspect categories with an open-text explanation inside each keeps annotations both countable and expressive, without forcing every nuance into a rigid label.
  • I2. Splitting a task into one specialist per linguistic aspect, with mutual critique, makes each edit attributable to a responsible component.
  • I3. Combining different model families as specialists beat any homogeneous setup except the strongest single model, which hints that diversity among judges adds value.
  • I4. Measuring how much each extra refinement round adds, and finding that the first round already captures most of the gain, is a cheap check against wasted iteration.
  • I5. Evaluating both transfer directions exposed a trade-off between style fidelity and semantic coherence that a one-direction test would have hidden.

For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

  • I1 (tree): UD dialect conversion rules could each be tagged with a fixed aspect class plus a short free-text justification, so conversions stay auditable and grouped.
  • I2 (learn): LLM teacher tiers could be split so that each teacher handles one annotation dimension (e.g., attachments, roles, particles), and disagreements are logged per dimension.
  • I3 (learn): witness parsers for silver data should be deliberately heterogeneous, and agreement should be measured per aspect rather than globally.
  • I4 (learn): self-induced repair rules and self-play loops should report per-iteration gains and stop once a round no longer improves held-out scores.
  • I5 (back): the text → graph → text compressor and generator should be scored separately on meaning preservation and on register or style match, in both directions.

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

🎓Singapore Management UniversitySingapore · Singapore27 works · 2,382 citations🎓Australian National UniversityCanberra · Australia23 works · 6,522 citations🤝DSO National LaboratoriesSingapore · Singapore5 works · 1,381 citations

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 worksComputer Vision and Pattern Recognition · 2,869 works

🧭 Topics 1Research subjects and languages the work is about.English · 2,263 works

Cite

Cite as Jinggui Liang, Dung Vo, Yap Hong Xian, Hai Leong Chieu, Kian Ming A. Chai, Jing Jiang and Lizi Liao. 2025. Colloquial Singaporean English Style Transfer with Fine-Grained Explainable Control. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2025.acl-long.1309

@inproceedings{liang2025colloquial,
  title = {{Colloquial Singaporean English Style Transfer with Fine-Grained Explainable Control}},
  author = {Jinggui Liang, Dung Vo, Yap Hong Xian, Hai Leong Chieu, Kian Ming A. Chai, Jing Jiang and Lizi Liao},
  booktitle = {Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)},
  year = {2025},
  doi = {10.18653/v1/2025.acl-long.1309},
  url = {https://doi.org/10.18653/v1/2025.acl-long.1309}
}

References

58 references, 19 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Rajiv Bajpai, Danyuan Ho, and Erik Cambria. 2016. Developing a concept-level knowledge base for sentiment analysis in singlish. In CICLing, pages 347– 361.
  2. Dibyanayan Bandyopadhyay and Asif Ekbal. 2023. Unsupervised text style transfer through differentiable back translation and rewards. In PAKDD, pages 210– 221. find
  3. Keith Carlson, Allen Riddell, and Daniel Rockmore. 2021. Unsupervised text style transfer with content embeddings. In Proceedings of the International Conference on Recent Advances in Natural Language Processing (RANLP 2021), pages 226–233. find
  4. Brahim Chaib-draa, Bernard Moulin, René Mandiau, and P. Millot. 1992. Trends in distributed artificial intelligence. Artif. Intell. Rev., pages 35–66. ↗
  5. Xi Chen, Yan Duan, Rein Houthooft, John Schulman, Ilya Sutskever, and Pieter Abbeel. 2016. Infogan: Interpretable representation learning by information maximizing generative adversarial nets. Advances in neural information processing systems, 29. ↗
  6. Siew Yeng Chow and Francis Bond. 2022. Singlish where got rules one? constructing a computational grammar for singlish. In LREC, pages 5243–5250.
  7. Nathalia Moraes do Nascimento, Paulo S. C. Alencar, and Donald D. Cowan. 2023. Gpt-in-the-loop: Adaptive decision-making for multiagent systems. CoRR. find
  8. Yilun Du, Shuang Li, Antonio Torralba, Joshua B. Tenenbaum, and Igor Mordatch. 2024. Improving factuality and reasoning in language models through multiagent debate. In ICML. ↗
  9. Joseph L. Fleiss and Jacob Cohen. 1973. The equivalence of weighted kappa and the intraclass correlation coefficient as measures of reliability. Educational and Psychological Measurement, 33:613 – 619. ↗
  10. Jessica Foo and Shaun Khoo. 2024. Lionguard: Building a contextualized moderation classifier to tackle localized unsafe content. CoRR, abs/2407.10995. ↗
  11. Linus Tze En Foo, Lynnette Hui Xian Ng, and Peter Bell. 2024. Disentangling singlish discourse particles with task-driven representation.
  12. Leon A Gatys, Alexander S Ecker, and Matthias Bethge. 2016. Image style transfer using convolutional neural networks. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 2414– 2423. find
  13. Junxian He, Xinyi Wang, Graham Neubig, and Taylor Berg-Kirkpatrick. 2020. A probabilistic formulation of unsupervised text style transfer. In ICLR. ↗
  14. Sirui Hong, Mingchen Zhuge, Jonathan Chen, Xiawu Zheng, Yuheng Cheng, Jinlin Wang, Ceyao Zhang, Zili Wang, Steven Ka Shing Yau, Zijuan Lin, Liyang Zhou, Chenyu Ran, Lingfeng Xiao, Chenglin Wu, and Jürgen Schmidhuber. 2024. Metagpt: Meta programming for A multi-agent collaborative framework. In ICLR. OpenReview.net. ↗
  15. Huiyuan Lai, Antonio Toral, and Malvina Nissim. 2021. Thank you bart! rewarding pre-trained models improves formality style transfer. In ACL/IJCNLP, pages 484–494. ↗
  16. Guillaume Lample, Sandeep Subramanian, Eric Michael Smith, Ludovic Denoyer, Marc’Aurelio Ranzato, and Y-Lan Boureau. 2019. Multiple-attribute text rewriting. In ICLR. find
  17. Dianqi Li, Yizhe Zhang, Zhe Gan, Yu Cheng, Chris Brockett, Bill Dolan, and Ming-Ting Sun. 2019. Domain adaptive text style transfer. In EMNLP-IJCNLP, pages 3302–3311. ↗
  18. Juncen Li, Robin Jia, He He, and Percy Liang. 2018. Delete, retrieve, generate: a simple approach to sentiment and style transfer. In NAACL-HLT, pages 1865–1874. ↗
  19. Yi Liao, Lidong Bing, Piji Li, Shuming Shi, Wai Lam, and Tong Zhang. 2018. Quase: Sequence editing under quantifiable guidance. In EMNLP, pages 3855– 3864. find
  20. Zhengyuan Liu, Shikang Ni, Ai Ti Aw, and Nancy F. Chen. 2022. Singlish message paraphrasing: A joint task of creole translation and text normalization. In COLING, pages 3924–3936. find
  21. Guoqing Luo, Yutong Han, Lili Mou, and Mauajama Firdaus. 2023. Prompt-based editing for text style transfer. In Findings of EMNLP, pages 5740–5750. find
  22. Sourabrata Mukherjee, Zdenek Kasner, and Ondrej Dusek. 2023. Balancing the style-content trade-off in sentiment transfer using polarity-aware denoising. CoRR, abs/2312.14708. ↗
  23. Sourabrata Mukherjee, Mateusz Lango, Zdenek Kasner, and Ondrej Dusek. 2024a. A survey of text style transfer: Applications and ethical implications. CoRR, abs/2407.16737. ↗
  24. Zhiqiang Hu, Roy Ka-Wei Lee, Charu C. Aggarwal, and Aston Zhang. 2022. Text style transfer: A review and experimental evaluation. SIGKDD Explor., 24:14– 45. find
  25. Sourabrata Mukherjee, Mateusz Lango, Zdenek Kasner, and Ondrej Dušek. 2024b. A survey of text style transfer: Applications and ethical implications. Preprint, arXiv:2407.16737. ↗
  26. Yufang Huang, Wentao Zhu, Deyi Xiong, Yiye Zhang, Changjian Hu, and Feiyu Xu. 2020. Cycle-consistent adversarial autoencoders for unsupervised text style transfer. In COLING, pages 2213–2223. find
  27. Xing Niu, Sudha Rao, and Marine Carpuat. 2018. Multi-task neural models for translating between styles within and across languages. In COLING, pages 1008–1021. ↗
  28. Harsh Jhamtani, Varun Gangal, Eduard H. Hovy, and Eric Nyberg. 2017. Shakespearizing modern language using copy-enriched sequence-to-sequence models. CoRR, abs/1707.01161. ↗
  29. David Ong and Peerat Limkonchotiwat. 2023. SEA-LION (Southeast Asian languages in one network): A family of Southeast Asian language models. In Proceedings of the 3rd Workshop for Natural Language Processing Open Source Software (NLP-OSS 2023), pages 245–245. find
  30. Vineet John, Lili Mou, Hareesh Bahuleyan, and Olga Vechtomova. 2019. Disentangled representation learning for non-parallel text style transfer. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 424–434.
  31. Kishore Papineni, Salim Roukos, Todd Ward, and Wei-Jing Zhu. 2002. Bleu: a method for automatic evaluation of machine translation. In ACL, pages 311–318.
  32. Nhi Pham, Lachlan Pham, and Adam L. Meyers. 2024. Towards better inclusivity: A diverse tweet corpus of english varieties. CoRR, abs/2401.11487. ↗
  33. Shrimai Prabhumoye, Yulia Tsvetkov, Ruslan Salakhutdinov, and Alan W. Black. 2018. Style transfer through back-translation. In ACL, pages 866–876.
  34. Qwen Team. 2024. Qwen2.5: A party of foundation models. find
  35. Sudha Rao and Joel Tetreault. 2018a. Dear sir or madam, may i introduce the gyafc dataset: Corpus, benchmarks and metrics for formality style transfer. In Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long Papers), pages 129–140.
  36. Sudha Rao and Joel R. Tetreault. 2018b. Dear sir or madam, may I introduce the GYAFC dataset: Corpus, benchmarks and metrics for formality style transfer. In NAACL-HLT, pages 129–140.
  37. Emily Reif, Daphne Ippolito, Ann Yuan, Andy Coenen, Chris Callison-Burch, and Jason Wei. 2022. A recipe for arbitrary text style transfer with large language models. In ACL, pages 837–848. ↗
  38. Arkadiy Saakyan and Smaranda Muresan. 2024. ICLEF: in-context learning with expert feedback for explainable style transfer. In ACL, pages 16141–16163. ↗
  39. Mingyue Shang, Piji Li, Zhenxin Fu, Lidong Bing, Dongyan Zhao, Shuming Shi, and Rui Yan. 2019. Semi-supervised text style transfer: Cross projection in latent space. In EMNLP-IJCNLP, pages 4936– 4945. ↗
  40. Tianxiao Shen, Tao Lei, Regina Barzilay, and Tommi S. Jaakkola. 2017. Style transfer from non-parallel text by cross-alignment. In NeurIPS, pages 6830–6841. ↗
  41. Tianxiao Shen, Jonas Mueller, Regina Barzilay, and Tommi S. Jaakkola. 2020. Educating text autoencoders: Latent representation guidance via denoising. In ICML, pages 8719–8729. ↗
  42. Ishika Singh, David Traum, and Jesse Thomason. 2024. Twostep: Multi-agent task planning using classical planners and large language models. CoRR, abs/2403.17246. ↗
  43. William Strunk. 2017. The elements of style : fourth edition. find
  44. Hongmin Wang, Yue Zhang, GuangYong Leonard Chan, Jie Yang, and Hai Leong Chieu. 2017. Universal dependencies parsing for colloquial singaporean english. In ACL, pages 1732–1744.
  45. Lei Wang, Chen Ma, Xueyang Feng, Zeyu Zhang, Hao Yang, Jingsen Zhang, Zhiyuan Chen, Jiakai Tang, Xu Chen, Yankai Lin, Wayne Xin Zhao, Zhewei Wei, and Jirong Wen. 2024a. A survey on large language model based autonomous agents. Frontiers Comput. Sci., page 186345. ↗
  46. Peng Wang, Xiaobin Wang, Chao Lou, Shengyu Mao, Pengjun Xie, and Yong Jiang. 2024b. Effective demonstration annotation for in-context learning via language model-based determinantal point process. In EMNLP, pages 1266–1280. find
  47. Ruixuan Xiao, Yiwen Dong, Junbo Zhao, Runze Wu, Minmin Lin, Gang Chen, and Haobo Wang. 2023. Freeal: Towards human-free active learning in the era of large language models. In EMNLP, pages 14520–14535. find
  48. Peng Xu, Jackie Chi Kit Cheung, and Yanshuai Cao. 2020. On variational learning of controllable representations for text without supervision. In ICML, pages 10534–10543. ↗
  49. Wei Xu, Alan Ritter, William B Dolan, Ralph Grishman, and Colin Cherry. 2012. Paraphrasing for style. In Proceedings of COLING 2012, pages 2899–2914.
  50. Weizhe Yuan, Graham Neubig, and Pengfei Liu. 2021. Bartscore: Evaluating generated text as text generation. In NeurIPS, pages 27263–27277.
  51. Stanley Yunick. 1995. The step-tongue: Children’s english in singapore. World Englishes, 14(2):304– 307. ↗
  52. Chiyu Zhang, Honglong Cai, Yuezhang Li, Yuexin Wu, Le Hou, and Muhammad Abdul-Mageed. 2024a. Distilling text style transfer with self-explanation from llms. In NAACL: Human Language Technologies: Student Research Workshop, pages 200–211. ↗
  53. Hongxin Zhang, Weihua Du, Jiaming Shan, Qinhong Zhou, Yilun Du, Joshua B. Tenenbaum, Tianmin Shu, and Chuang Gan. 2024b. Building cooperative embodied agents modularly with large language models. In ICLR. ↗
  54. Ruoyu Zhang, Yanzeng Li, Yongliang Ma, Ming Zhou, and Lei Zou. 2023. Llmaaa: Making large language models as active annotators. In Findings of EMNLP, pages 13088–13103. find
  55. Tianyi Zhang, Varsha Kishore, Felix Wu, Kilian Q. Weinberger, and Yoav Artzi. 2020. Bertscore: Evaluating text generation with BERT. In ICLR. ↗
  56. Wenxuan Zhang, Hou Pong Chan, Yiran Zhao, Mahani Aljunied, Jianyu Wang, Chaoqun Liu, Yue Deng, Zhiqiang Hu, Weiwen Xu, Yew Ken Chia, Xin Li, and Lidong Bing. 2024c. Seallms 3: Open foundation and chat multilingual large language models for southeast asian languages. CoRR, abs/2407.19672. ↗
  57. Junbo Jake Zhao, Yoon Kim, Kelly Zhang, Alexander M. Rush, and Yann LeCun. 2018. Adversarially regularized autoencoders. In ICML, pages 5897–5906. ↗
  58. Mingchen Zhuge, Haozhe Liu, Francesco Faccio, Dylan R. Ashley, Róbert Csordás, Anand Gopalakrishnan, Abdullah Hamdi, Hasan Abed Al Kader Hammoud, Vincent Herrmann, Kazuki Irie, Louis Kirsch, Bing Li, Guohao Li, Shuming Liu, Jinjie Mai, Piotr Piekos, Aditya A. Ramesh, Imanol Schlag, Weimin Shi, Aleksandar Stanic, Wenyi Wang, Yuhui Wang, Mengmeng Xu, Deng-Ping Fan, Bernard Ghanem, and Jürgen Schmidhuber. 2023. Mindstorms in natural language-based societies of mind. CoRR, abs/2305.17066. ↗

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/liang-2025-colloquial-singaporean-english-style-transfer/