--- title: "LoResMT 2026 Shared Task System Description — Panov 2026" description: "This system description reports one participant's entry in the LoResMT 2026 shared task on Turkic languages. 10 references." authors: ["Vladimir Panov"] year: 2026 venue: "Proceedings for the Ninth Workshop on Technologies for Machine Translation of Low Resource Languages (LoResMT 2026)" source: "https://aclanthology.org/2026.loresmt-1.23/" doi: "10.18653/v1/2026.loresmt-1.23" licence: "CC BY 4.0 (ACL Anthology)" licence_url: "https://creativecommons.org/licenses/by/4.0/" summary_licence: "Elamur Cosmic License" canonical: "https://research.elamur.ai/free/en/papers/panov-2026-loresmt-2026-shared-task-system/" --- [Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/panov-2026-loresmt-2026-shared-task-system/) [home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2026](https://md.research.elamur.ai/free/en/papers/2026/page/8/) / LoResMT 2026 Shared Task System Description # LoResMT 2026 Shared Task System Description **Authors:** Vladimir Panov · **Year:** 2026 · **Venue:** Proceedings for the Ninth Workshop on Technologies for Machine Translation of Low Resource Languages (LoResMT 2026) · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/) **Links:** [Publisher's page](https://aclanthology.org/2026.loresmt-1.23/) · [PDF](https://aclanthology.org/2026.loresmt-1.23.pdf) · [DOI 10.18653/v1/2026.loresmt-1.23](https://doi.org/10.18653/v1/2026.loresmt-1.23) · [OpenAlex](https://openalex.org/W7140140793) **Citations:** 0 citations (OpenAlex, 2026-10-14) ## Summary This system description reports one participant's entry in the LoResMT 2026 shared task on Turkic languages. The entry covers only English→Chuvash translation, a pair with very little clean parallel data. Training used two sources: the organizers' automatically aligned English-Chuvash corpus and the GATITOS lexicon data. The much larger Russian-Chuvash corpus and the monolingual Chuvash text were deliberately left out to keep experiments fast. The data was cleaned of quotation marks and leading dashes, pairs with a length ratio beyond 2× were dropped, and duplicates were removed in two passes (MD5 for exact matches, MinHashLSH for near-duplicates). The author fine-tuned a range of models, from umt5-small and the Gemma-3 family up to the dedicated translation LLM tencent/HY-MT1.5-7B. Decoding settings were then tuned with Optuna on FLORES+. The best system was HY-MT1.5-7B with tuned generation parameters, reaching chrF++ 29.67 on the public test set. Decoding tuning alone accounted for a sizeable share of that score. For low-resource MT, the work suggests that model scale, specialization and inference-time tuning can matter as much as the training data. ## Key points - **Data:** about 204k English-Chuvash sentence pairs plus about 4k GATITOS pairs were used for training. FLORES+ dev (997 sentences) was used for evaluation and decoding search. Preprocessing removed roughly 7% of the English-Chuvash corpus. MinHashLSH settings were a Jaccard threshold of 0.7, 128 permutations and shingle size 3. - **Public chrF++ by model:** - umt5-small: 13.03 - gemma-3-270m-it: 14.60 - gemma-3-1b-it: 20.43 - gemma-3-4b-it with QLoRA: 20.43 - HY-MT1.5-7B: 25.42 - HY-MT1.5-7B with tuned decoding: 29.67 - **Decoding gain:** tuning generation settings raised HY-MT1.5-7B from 25.42 to 29.67, a relative gain of about 16–17%. The best configuration used 5 beams, temperature ≈0.28, top_k=77, top_p ≈0.996 and repetition penalty ≈1.036. - **QLoRA vs. full fine-tuning:** QLoRA on gemma-3-4b-it did no better than full fine-tuning of gemma-3-1b-it. With limited data, parameter-efficient tuning may cancel out the advantage of a larger model. - **Tooling:** Transformers (umt5), Unsloth (Gemma-3) and LLaMA-Factory (HY-MT1.5-7B) were used for training. Training ran for 1–3 epochs with an effective batch size of 16, using 8-bit AdamW, bf16 and Flash Attention 2. *Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.* Fields: [Linguistics & languages](https://md.research.elamur.ai/free/en/shelves/linguistics/) · [Machine translation](https://md.research.elamur.ai/free/en/shelves/machine-translation/) ## Cite Cite as: Vladimir Panov. 2026. LoResMT 2026 Shared Task System Description. Proceedings for the Ninth Workshop on Technologies for Machine Translation of Low Resource Languages (LoResMT 2026). The paper itself, not this page. BibTeX from the library's record of the work. ```bibtex @inproceedings{panov2026loresmt, title = {{LoResMT 2026 Shared Task System Description}}, author = {Vladimir Panov}, booktitle = {Proceedings for the Ninth Workshop on Technologies for Machine Translation of Low Resource Languages (LoResMT 2026)}, year = {2026}, doi = {10.18653/v1/2026.loresmt-1.23}, url = {https://doi.org/10.18653/v1/2026.loresmt-1.23} } ``` - [← previous LoRA Fine-Tuning of English–Norwegian NMT for the Oil & Gas Industry](https://md.research.elamur.ai/free/en/papers/yang-2026-lora-fine-tuning-english-norwegian/) - [next → Lost in Translation: Repurposing semantic similarity benchmarks for evaluating lexical-semantic consistency in LLM-based machine translation](https://md.research.elamur.ai/free/en/papers/ye-2026-lost-translation-repurposing-semantic-similarity/) - [↑ back to papers of 2026](https://md.research.elamur.ai/free/en/papers/2026/page/8/) ## References 10 references, 3 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet. 1. Takuya Akiba, Shotaro Sano, Toshihiko Yanase, Takeru Ohta, and Masanori Koyama. 2019. Optuna: A next-generation hyperparameter optimization framework. In Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. [our card](https://md.research.elamur.ai/free/en/papers/akiba-2019-optuna-next-generation-hyperparameter-optimization/) 2. Hyung Won Chung, Xavier Garcia, Adam Roberts, Yi Tay, Orhan Firat, Sharan Narang, and Noah Constant. 2023. Unimax: Fairer and more effective language sampling for large-scale multilingual pretraining. In The Eleventh International Conference on Learning Representations. [find](https://scholar.google.com/scholar?q=Hyung+Won+Chung%2C+Xavier+Garcia%2C+Adam+Roberts%2C+Yi+Tay%2C+Orhan+Firat%2C+Sharan+Narang%2C+and+Noah+Constant%2E+2023%2E+Unimax%3A+Fairer+and+more+effective+language+sampling+for+large%2Dscale+multilingual+pretraining%2E) 3. Yaowei Zheng, Richong Zhang, Junhao Zhang, Yanhan Ye, Zheyan Luo, Zhangchi Feng, and Yongqiang Ma. 2024. Llamafactory: Unified efficient fine-tuning of 100+ language models. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 3: System Demonstrations),… [our card](https://md.research.elamur.ai/free/en/papers/zheng-2024-llamafactory-unified-efficient-fine-tuning/) 4. Eric Zhu, Vadim Markovtsev, Aleksey Astafiev, Arham Khan, Chris Ha, Wojciech Łukasiewicz, Adam Foster, Sinusoidal36, Spandan Thakur, Stefano Ortolani, Titusz, Vojtech Letal, Zac Bentley, fpug, hguhlich, long2ice, oisincar, Ron Assa, Senad Ibraimoski, and 8 others. 2024. ekzhu/datasketch: v1.6.5. [find](https://scholar.google.com/scholar?q=Eric+Zhu%2C+Vadim+Markovtsev%2C+Aleksey+Astafiev%2C+Arham+Khan%2C+Chris+Ha%2C+Wojciech+%C5%81ukasiewicz%2C+Adam+Foster%2C+Sinusoidal36%2C+Spandan+Thakur%2C+Stefano+Ortolani%2C+Titusz%2C+Vojtech+Letal%2C+Zac+Bentley%2C+fpug%2C+hguhlic) 5. Michael Han Daniel Han and Unsloth team. 2023. Unsloth. [find](https://scholar.google.com/scholar?q=Michael+Han+Daniel+Han+and+Unsloth+team%2E+2023%2E+Unsloth%2E) 6. Tri Dao. 2024. FlashAttention-2: Faster attention with better parallelism and work partitioning. In International Conference on Learning Representations (ICLR). [find](https://scholar.google.com/scholar?q=Tri+Dao%2E+2024%2E+FlashAttention%2D2%3A+Faster+attention+with+better+parallelism+and+work+partitioning%2E+In+International+Conference+on+Learning+Representations+%28ICLR%29%2E) 7. Alexander Jones, Isaac Caswell, Orhan Firat, and Ishank Saxena. 2023. "GATITOS: Using a New Multilingual Lexicon for Low-resource Machine Translation". In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pages 371–405, Singapore. Association for Computational L… [find](https://scholar.google.com/scholar?q=Alexander+Jones%2C+Isaac+Caswell%2C+Orhan+Firat%2C+and+Ishank+Saxena%2E+2023%2E+%22GATITOS%3A+Using+a+New+Multilingual+Lexicon+for+Low%2Dresource+Machine+Translation%22%2E+In+Proceedings+of+the+2023+Conference+on+Empiric) 8. NLLB Team, Marta R. Costa-jussà, James Cross, Onur Çelebi, Maha Elbayad, Kenneth Heafield, Kevin Heffernan, Elahe Kalbassi, Janice Lam, Daniel Licht, Jean Maillard, Anna Sun, Skyler Wang, Guillaume Wenzek, Al Youngblood, Bapi Akula, Loic Barrault, Gabriel Mejia Gonzalez, Prangthip Hansanti, and 20 o… [find](https://scholar.google.com/scholar?q=NLLB+Team%2C+Marta+R%2E+Costa%2Djuss%C3%A0%2C+James+Cross%2C+Onur+%C3%87elebi%2C+Maha+Elbayad%2C+Kenneth+Heafield%2C+Kevin+Heffernan%2C+Elahe+Kalbassi%2C+Janice+Lam%2C+Daniel+Licht%2C+Jean+Maillard%2C+Anna+Sun%2C+Skyler+Wang%2C+Guillaume+We) 9. Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Clement Delangue, Anthony Moi, Pier-ric Cistac, Tim Rault, Rémi Louf, Morgan Funtowicz, Joe Davison, Sam Shleifer, Patrick von Platen, Clara Ma, Yacine Jernite, Julien Plu, Canwen Xu, Teven Le Scao, Sylvain Gugger, and 3 others. 2020. Transf… [our card](https://md.research.elamur.ai/free/en/papers/wolf-2020-transformers-state-art-natural-language/) 10. Mao Zheng, Zheng Li, Tao Chen, Mingyang Song, and Di Wang. 2025. Hy-mt1.5 technical report. Preprint, arXiv:2512.24092. — [arXiv:2512.24092](https://arxiv.org/abs/2512.24092) [find](https://scholar.google.com/scholar?q=Mao+Zheng%2C+Zheng+Li%2C+Tao+Chen%2C+Mingyang+Song%2C+and+Di+Wang%2E+2025%2E+Hy%2Dmt1%2E5+technical+report%2E+Preprint%2C+arXiv%3A2512%2E24092%2E+%E2%80%94+%5BarXiv%3A2512%2E24092%5D%28https%3A%2F%2Farxiv%2Eorg%2Fabs%2F2512%2E24092%29)