---
title: "Translation as Augmentation — Wu 2026"
description: "The paper addresses the shortage of expert, CEFR-labelled corpora for estimating text difficulty in lower-resource languages, using Finnish as the test…"
authors: ["Yiheng Wu", "Jue Hou", "Roman Yangarber"]
year: 2026
venue: "Proceedings of the 2nd Workshop on Evaluating Text Difficulty in a Multilingual Context (DeTermIt! 2026)"
source: "https://aclanthology.org/2026.determit-1.5/"
doi: "10.63317/3mmqujpgyua5"
licence: "CC BY-NC 4.0 (ELRA, LREC proceedings)"
licence_url: "https://creativecommons.org/licenses/by-nc/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/papers/wu-2026-translation-as-augmentation-effect-translated/"
---
[Elamur Research](https://md.research.elamur.ai/) · [this page in HTML](https://research.elamur.ai/free/papers/wu-2026-translation-as-augmentation-effect-translated/)

[home](https://md.research.elamur.ai/) / [papers](https://md.research.elamur.ai/free/papers/) / [2026](https://md.research.elamur.ai/free/papers/2026/page/5/) / Translation as Augmentation: Effect of Translated Data on Assessment of Difficulty

# Translation as Augmentation: Effect of Translated Data on Assessment of Difficulty

**Authors:** Yiheng Wu, Jue Hou, Roman Yangarber · **Year:** 2026 · **Venue:** Proceedings of the 2nd Workshop on Evaluating Text Difficulty in a Multilingual Context (DeTermIt! 2026) · **Licence:** [CC BY-NC 4.0 (ELRA, LREC proceedings)](https://creativecommons.org/licenses/by-nc/4.0/)

**Links:** [Publisher's page](https://aclanthology.org/2026.determit-1.5/) · [PDF](https://aclanthology.org/2026.determit-1.5.pdf) · [DOI 10.63317/3mmqujpgyua5](https://doi.org/10.63317/3mmqujpgyua5) · [OpenAlex](https://openalex.org/W7166510920)

**Citations:** 0 citations (openalex-snapshot · 2026-10-14)

## Summary
The paper addresses the shortage of expert, CEFR-labelled corpora for estimating text difficulty in lower-resource languages, using Finnish as the test case. The authors combine 4,544 native Finnish texts with 8,321 Russian learner texts that carry CEFR labels and were machine-translated into Finnish. The native texts come from Easy Language sites, L2 textbooks, Helsingin Sanomat, YLE and YLE Selkouutiset. The translated texts come from the RuFoLa and RuAdapt corpora and were produced with an OpusMT model. Labels across 11 fine-grained CEFR classes are placed on a linear 1–6 scale, and a BERT-based regressor predicts a continuous score.

Mixing native and translated data keeps accuracy on native Finnish test texts close to the native-only model, with R² of 96.01% versus 97.30%. The mix also greatly improves results on translated test data. Models trained only on translations transfer poorly to native text. Ablations show that the most useful translated material is mid-level (B1–B2) or added in moderate amounts. The work suggests that translation can stand in for scarce annotations, provided the MT system keeps the original difficulty level intact.

## Key points
- **Dataset:** The combined corpus has 12,865 Finnish texts covering A1–C2. Of these, 8,321 are machine-translated from Russian (marked "MT ← RU"), drawn from RuFoLa and from the Zlatoust and Encyclop. subsets of RuAdapt. Translation used the OpusMT Tatoeba Slavic–Finnish model.
- **Labels:** The scheme uses 11 classes, the six CEFR levels plus intermediate "+" levels. News texts are treated as a coarse level between C1 and C2.
- **Results on native Finnish test data:**
  - Native-only training gave R² 97.30%.
  - Native + MT training gave R² 96.01%.
  - MT-only training gave R² 18.24% (MSE 155.00%).
- **Results on MT test data:** Native + MT training reached R² 94.26%, while native-only training fell to −99.71%.
- **Ablations:**
  - Adding only the B1–B2 translated subset worked best, with R² 96.68%. Adding A1–A2 or C1–C2 translations raised the error.
  - Proportional sampling peaked at 40% of the MT data, with R² 96.20%.
- **Caveat:** The authors stress that the MT model must preserve CEFR level. They checked this only through manual inspection of a sample by native experts.

*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: [Machine translation](https://md.research.elamur.ai/free/shelves/machine-translation/) · [Natural language processing](https://md.research.elamur.ai/free/shelves/nlp/)

## Cite

The paper itself, not this page. BibTeX from the library's record of the work.

```bibtex
@inproceedings{wu2026translation,
  title = {{Translation as Augmentation: Effect of Translated Data on Assessment of Difficulty}},
  author = {Wu, Yiheng and Hou, Jue and Yangarber, Roman},
  booktitle = {Proceedings of the 2nd Workshop on Evaluating Text Difficulty in a Multilingual Context (DeTermIt! 2026)},
  year = {2026},
  doi = {10.63317/3mmqujpgyua5},
  url = {https://doi.org/10.63317/3mmqujpgyua5}
}
```

- [← previous Transcription Accuracy in the Icelandic Gigaword Corpus: Evaluating Automatic and Manual Annotation](https://md.research.elamur.ai/free/papers/mechler-2026-transcription-accuracy-icelandic-gigaword-corpus/)
- [next → Transparent Semantic Change Detection with Dependency-Based Profiles](https://md.research.elamur.ai/free/papers/tat-2026-transparent-semantic-change-detection-dependency/)
- [↑ back to papers of 2026](https://md.research.elamur.ai/free/papers/2026/page/5/)

## References

22 references, 3 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Sweta Agrawal and Marine Carpuat. 2023. Controlling pre-trained language models for grade-specific text simplification. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pages 12807–12819, Singapore. [↗](https://doi.org/10.48448/fd56-6w47)
2. Wejdan Alkaldi and Diana Inkpen. 2023. Text simplification to specific readability levels. Mathematics, 11(9):2063. [↗](https://doi.org/10.3390/math11092063)
3. Sandra Aluísio, Lucia Specia, Caroline Gasperin, and Carolina Scarton. 2010. Readability assessment for text simplification. In Proceedings of the 5th Workshop on Innovative Use of NLP for Building Educational Applications, Los Angeles, California. [find](https://scholar.google.com/scholar?q=Sandra+Alu%C3%ADsio%2C+Lucia+Specia%2C+Caroline+Gasperin%2C+and+Carolina+Scarton%2E+2010%2E+Readability+assessment+for+text+simplification%2E+In+Proceedings+of+the+5th+Workshop+on+Innovative+Use+of+NLP+for+Building+Ed) [our card](https://md.research.elamur.ai/free/papers/aluisio-2010-readability-assessment-text-simplification/)
4. Fernando Alva-Manchego, Regina Stodden, Joseph Marvin Imperial, Abdullah Barayan, Kai North, and Harish Tayyar Madabushi. 2025. Findings of the TSAR 2025 shared task on readability-controlled text simplification. In Proceedings of the 4th Workshop on Text Simplification, Accessibility, and Readability, Suzhou, China. [find](https://scholar.google.com/scholar?q=Fernando+Alva%2DManchego%2C+Regina+Stodden%2C+Joseph+Marvin+Imperial%2C+Abdullah+Barayan%2C+Kai+North%2C+and+Harish+Tayyar+Madabushi%2E+2025%2E+Findings+of+the+TSAR+2025+shared+task+on+readability%2Dcontrolled+text+sim)
5. Ion Madrazo Azpiazu and Maria Soledad Pera. 2019. Multiattentive recurrent neural network architecture for multilingual readability assessment. Transactions of the Association for Computational Linguistics, 7:421–436. [↗](https://doi.org/10.1162/tacl_a_00278)
6. Kevyn Collins-Thompson and James P. Callan. 2004. A language modeling approach to predicting reading difficulty. In HLT-NAACL 2004: Proceedings of the Human Language Technology Conference of the NAACL, pages 193–200. [find](https://scholar.google.com/scholar?q=Kevyn+Collins%2DThompson+and+James+P%2E+Callan%2E+2004%2E+A+language+modeling+approach+to+predicting+reading+difficulty%2E+In+HLT%2DNAACL+2004%3A+Proceedings+of+the+Human+Language+Technology+Conference+of+the+NAACL)
7. Anna Dmitrieva. 2025. Resources and Tools for Automatic Text Simplification: Cases of Russian and Finnish. Ph.D. thesis. [find](https://scholar.google.com/scholar?q=Anna+Dmitrieva%2E+2025%2E+Resources+and+Tools+for+Automatic+Text+Simplification%3A+Cases+of+Russian+and+Finnish%2E+Ph%2ED%2E+thesis%2E)
8. Anna Dmitrieva and Aleksandra Konovalova. 2023. Creating a parallel Finnish-Easy Finnish dataset from news articles. In Proceedings of the 1st Workshop on Open Community-Driven Machine Translation, pages 21–26, Tampere, Finland. [find](https://scholar.google.com/scholar?q=Anna+Dmitrieva+and+Aleksandra+Konovalova%2E+2023%2E+Creating+a+parallel+Finnish%2DEasy+Finnish+dataset+from+news+articles%2E+In+Proceedings+of+the+1st+Workshop+on+Open+Community%2DDriven+Machine+Translation%2C+pa)
9. Anna Dmitrieva and Jörg Tiedemann. 2021. Creating an aligned Russian text simplification dataset from language learner data. In Proceedings of the 8th Workshop on Balto-Slavic Natural Language Processing, pages 73–79, Kiyv, Ukraine. Association for Computational Linguistics. [find](https://scholar.google.com/scholar?q=Anna+Dmitrieva+and+J%C3%B6rg+Tiedemann%2E+2021%2E+Creating+an+aligned+Russian+text+simplification+dataset+from+language+learner+data%2E+In+Proceedings+of+the+8th+Workshop+on+Balto%2DSlavic+Natural+Language+Process)
10. Caroline Gasperin, Lucia Specia, Tiago F. Pereira, and Sandra M. Aluísio. 2009. Learning when to simplify sentences for natural text simplification. In Proceedings of the Encontro Nacional de Inteligência Artificial (ENIA), Bento Gonçalves, Brazil. [find](https://scholar.google.com/scholar?q=Caroline+Gasperin%2C+Lucia+Specia%2C+Tiago+F%2E+Pereira%2C+and+Sandra+M%2E+Alu%C3%ADsio%2E+2009%2E+Learning+when+to+simplify+sentences+for+natural+text+simplification%2E+In+Proceedings+of+the+Encontro+Nacional+de+Intelig%C3%AA)
11. Aaron Hurst, Adam Lerer, Adam P Goucher, Adam Perelman, Aditya Ramesh, Aidan Clark, AJ Ostrow, Akila Welihinda, Alan Hayes, Alec Radford, et al. 2024. GPT-4o system card. arXiv preprint arXiv:2410.21276. [↗](https://arxiv.org/abs/2410.21276) [our card](https://md.research.elamur.ai/free/papers/openai-2024-gpt-4o-system-card/)
12. Joseph Marvin Imperial and Harish Tayyar Madabushi. 2024. Part of the problem: Toward a finer-grained analysis of LLMs for readability assessment. In Proceedings of the 19th Workshop on Innovative Use of NLP for Building Educational Applications. [find](https://scholar.google.com/scholar?q=Joseph+Marvin+Imperial+and+Harish+Tayyar+Madabushi%2E+2024%2E+Part+of+the+problem%3A+Toward+a+finer%2Dgrained+analysis+of+LLMs+for+readability+assessment%2E+In+Proceedings+of+the+19th+Workshop+on+Innovative+Use)
13. Anisia Katinskaia, Anh-Duc Vu, Jue Hou, Ulla Vanhatalo, Yiheng Wu, and Roman Yangarber. 2025. Estimation of text difficulty in the context of language learning. In BEA: 20th Workshop on Innovative Use of NLP for Building Educational Applications, Vienna, Austria. [find](https://scholar.google.com/scholar?q=Anisia+Katinskaia%2C+Anh%2DDuc+Vu%2C+Jue+Hou%2C+Ulla+Vanhatalo%2C+Yiheng+Wu%2C+and+Roman+Yangarber%2E+2025%2E+Estimation+of+text+difficulty+in+the+context+of+language+learning%2E+In+BEA%3A+20th+Workshop+on+Innovative+Use)
14. J. Peter Kincaid, Robert P. Fishburne, Richard L. Rogers, and Benjamin S. Chissom. 1975. Derivation of new readability formulas (automated readability index, fog count and flesch reading ease formula) for navy enlisted personnel. Technical report, Naval Air Station Memphis (Research Branch Report 8-75). [↗](https://doi.org/10.21236/ada006655)
15. Antonina Laposhina. 2020. A corpus of Russian textbook materials for foreign students as an instrument of an educational content analysis. Russian Language Abroad, 6(283):22–28. [find](https://scholar.google.com/scholar?q=Antonina+Laposhina%2E+2020%2E+A+corpus+of+Russian+textbook+materials+for+foreign+students+as+an+instrument+of+an+educational+content+analysis%2E+Russian+Language+Abroad%2C+6%28283%29%3A22%E2%80%9328%2E)
16. Antonina Laposhina, Tatiana Veselovskaya, Maria Lebedeva, and Olga Kupreshchenko. 2018. Automated text readability assessment for Russian second language learners. In Computational Linguistics and Intellectual Technologies, pages 403–413. [find](https://scholar.google.com/scholar?q=Antonina+Laposhina%2C+Tatiana+Veselovskaya%2C+Maria+Lebedeva%2C+and+Olga+Kupreshchenko%2E+2018%2E+Automated+text+readability+assessment+for+Russian+second+language+learners%2E+In+Computational+Linguistics+and+Int)
17. Matej Martinc, Senja Pollak, and Marko Robnik-Šikonja. 2021. Supervised and unsupervised neural approaches to text readability. Computational Linguistics, 47(1):141–179. [↗](https://doi.org/10.1162/coli_a_00398)
18. Shingo Nahatame and Katsuyoshi Yamaguchi. 2026. Revisiting text readability and processing effort in second language reading: Bayesian analysis of eye-tracking data. Language Learning. [↗](https://doi.org/10.1111/lang.70011)
19. Serge Sharoff. 2022. What neural networks know about linguistic complexity. Russian Journal of Linguistics, 26(2):371–390. [↗](https://doi.org/10.22363/2687-0088-30178)
20. A. Jackson Stenner. 1996. Measuring reading comprehension with the Lexile framework. Technical report, MetaMetrics Inc., Durham, NC. [find](https://scholar.google.com/scholar?q=A%2E+Jackson+Stenner%2E+1996%2E+Measuring+reading+comprehension+with+the+Lexile+framework%2E+Technical+report%2C+MetaMetrics+Inc%2E%2C+Durham%2C+NC%2E)
21. Sowmya Vajjala and Detmar Meurers. 2012. On improving the accuracy of readability classification using insights from second language acquisition. In Proceedings of the 7th Workshop on Building Educational Applications Using NLP (BEA). [find](https://scholar.google.com/scholar?q=Sowmya+Vajjala+and+Detmar+Meurers%2E+2012%2E+On+improving+the+accuracy+of+readability+classification+using+insights+from+second+language+acquisition%2E+In+Proceedings+of+the+7th+Workshop+on+Building+Educati)
22. Kristian Woodsend and Mirella Lapata. 2011. Learning to simplify sentences with quasi-synchronous grammar and integer programming. In Proceedings of Conference on Empirical Methods in Natural Language Processing, Edinburgh, Scotland. [find](https://scholar.google.com/scholar?q=Kristian+Woodsend+and+Mirella+Lapata%2E+2011%2E+Learning+to+simplify+sentences+with+quasi%2Dsynchronous+grammar+and+integer+programming%2E+In+Proceedings+of+Conference+on+Empirical+Methods+in+Natural+Language) [our card](https://md.research.elamur.ai/free/papers/woodsend-2011-learning-simplify-sentences-quasi-synchronous/)