mlsanityData to 7 Oct 2026

Translation Asymmetry in LLMs as a Data Augmentation Factor: A Case Study for 6 Romansh Language Varieties

arXiv 2603.25489 ↗First posted 26 Mar 2026cs.CLcited by 1

Not on the map. Why?

Abstract (arXiv)

Recent strategies for low-resource machine translation rely on LLMs to generate synthetic data based on text in higher-resource languages. We revisit this idea for Romansh, a language with 6 distinct varieties. LLMs tend to confuse these varieties when translating into Romansh, but they are quite good at translating out of Romansh into a high-resource language such as German. Due to this asymmetry, the direction of data augmentation is a crucial choice. We find that contrary to recent strategies, creating synthetic translations into the higher-resource language is the superior approach, and only this approach allows us to surpass a Gemini 3 Pro baseline on German-Romansh translation (+23 BLEU over Gemini in the lowest-resource variety). A human evaluation confirms that our experiments yield the first model that generates fluent translations in the individual Romansh varieties.

Key facts. Cited by 1 paper in the map (data through 7 Oct 2026). Builds on 22 of its 41 references that are in the map. First posted on arXiv 26 Mar 2026.

Builds on

22 of 41 references.

References · 22

  1. 2018A Call for Clarity in Reporting BLEU Scorescited by 1,790
  2. 2015Improving Neural Machine Translation Models with Monolingual Datacited by 784
  3. 2023xCOMET: Transparent Machine Translation Evaluation through Fine-grained Error Detectioncited by 260
  4. 2023MADLAD-400: A Multilingual And Document-Level Large Audited Datasetcited by 180
  5. 2025FineWeb2: One Pipeline to Scale Them All -- Adapting Pre-Training Data Processing to Every Languagecited by 134
  6. 2025WMT24++: Expanding the Language Coverage of WMT24 to 55 Languages & Dialectscited by 79
  7. 2023Dictionary-based Phrase-level Prompting of Large Language Models for Machine Translationcited by 74
  8. 2019Domain Robustness in Neural Machine Translationcited by 64
  9. 2023Investigating the Translation Performance of a Large Multilingual Language Model: the Case of BLOOMcited by 58
  10. 2024Segment Any Text: A Universal Approach for Robust, Efficient and Adaptable Sentence Segmentationcited by 41
  11. 2019Domain, Translationese and Noise in Synthetic Data for Neural Machine Translationcited by 26
  12. 2019Using Monolingual Data in Neural Machine Translation: a Systematic Studycited by 21
10 more
  1. 2025BOUQuET: dataset, Benchmark and Open initiative for Universal Quality Evaluation in Translationcited by 21
  2. 2024Shortcomings of LLMs for Low-Resource Translation: Retrieval and Understanding are Both the Problemcited by 17
  3. 2025Causal Estimation of Tokenisation Biascited by 16
  4. 2025Scaling Low-Resource MT via Synthetic Data Generation with LLMscited by 14
  5. 2025SwiLTra-Bench: The Swiss Legal Translation Benchmarkcited by 10
  6. 2023Terminology-Aware Translation with Constrained Decoding and Large Language Model Promptingcited by 8
  7. 2024Rule-Based, Neural and LLM Back-Translation: Comparative Insights from a Variant of Ladincited by 5
  8. 2025Expanding the WMT24++ Benchmark with Rumantsch Grischun, Sursilvan, Sutsilvan, Surmiran, Puter, and Valladercited by 4
  9. 2025The Mediomatix Corpus: Parallel Data for Romansh Language Varieties via Comparable Schoolbookscited by 3
  10. 2026Robust Language Identification for Romansh Varietiescited by 2

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 1.

1 paper.

  1. 2026Reinforcement Learning Elicits Contextual Learning of Unseen Language Translationcited by 0