---
title: "Less is More — paper by Navasardyan et al. (2026)"
description: "The authors argue that this lets resource-poor communities build competitive embedders using only… 16 references, cited by 2 papers here."
authors: ["Zaruhi Navasardyan", "Bagratuni Minsayan", "Spartak Bughdaryan", "Hrant Davtyan"]
year: 2026
venue: "Proceedings of the Second Workshop on Language Models for Low-Resource Languages (LoResLM 2026)"
source: "https://aclanthology.org/2026.loreslm-1.31/"
doi: "10.18653/v1/2026.loreslm-1.31"
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/navasardyan-2026-less-is-more-adapting-text/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/navasardyan-2026-less-is-more-adapting-text/)

[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/5/) / Less is More: Adapting Text Embeddings for Low-Resource Languages with Small Scale Noisy Synthetic Data

# Less is More: Adapting Text Embeddings for Low-Resource Languages with Small Scale Noisy Synthetic Data

**Authors:** Zaruhi Navasardyan, Bagratuni Minsayan, Spartak Bughdaryan and Hrant Davtyan · **Year:** 2026 · **Venue:** [Proceedings of the Second Workshop on Language Models for Low-Resource Languages (LoResLM 2026)](https://md.research.elamur.ai/free/en/venues/loreslm/) ([Workshop on Language Models for Low-Resource Languages 2026](https://md.research.elamur.ai/free/en/venues/loreslm/2026/)) · **Publisher:** [Association for Computational Linguistics](https://md.research.elamur.ai/free/en/publishers/association-for-computational-linguistics/) · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/)

**Links:** [Publisher's page](https://aclanthology.org/2026.loreslm-1.31/) · [PDF](https://aclanthology.org/2026.loreslm-1.31.pdf) · [DOI 10.18653/v1/2026.loreslm-1.31](https://doi.org/10.18653/v1/2026.loreslm-1.31) · [OpenAlex](https://openalex.org/W7140121749)

**Citations:** 1 citations (OpenAlex, 2026-10-14) · 2 in our library, of the 149,439 papers whose bibliographies we read

## Summary

Low-resource languages such as Armenian have few large, clean datasets for training text embedding models, which limits their use in semantic search and RAG. The authors test whether strong multilingual encoders can be adapted cheaply. They translate English Reddit title-body pairs into Armenian with the open-weights Gemma-2-27B-it model, which produces deliberately imperfect output. They then filter the pairs with similarity checks based on multilingual E5 and fine-tune mE5 contrastively.

To measure the effect, they assemble an Armenian benchmark. It combines the MTEB Armenian subset, translated MS MARCO and STS data, and a small hand-built retrieval set.

The main finding is that only 10,000 noisy pairs raise average benchmark scores by roughly 11–12% and retrieval by more than 20% relative. This equals training on 1 million pairs. Better translations, more diverse domains and more data add little. The pattern also holds for Georgian and for EmbeddingGemma as a base model. The authors argue that this lets resource-poor communities build competitive embedders using only public English data and open models.

## Key points

- **Training data:** About 2M pairs were sampled from the sentence-transformers Reddit title-body dataset and machine-translated into Armenian. They were then filtered to around 1M pairs. Pairs were dropped if their English-vs-Armenian pair similarity shifted by more than 0.05, or if the query or passage translation similarity fell to 0.85 or below.
- **Translation noise:** A manual check of 100 pairs found grammar, lexical, literal-translation and entity errors. TER against corrected references was 67.72 for queries and 71.54 for passages.
- **Benchmark contents:**
  - MTEB [hye], with 9 datasets across 5 tasks
  - 185 native Armenian query-document pairs, checked by two reviewers
  - 10K MS MARCO v2.1 validation examples, translated with Gemini 2.0 Flash
  - a 3K translated STS subset
- **Other base model:** EmbeddingGemma improved its retrieval score from 50 to 65 with 10k pairs. It still did not surpass the 10k-tuned mE5, even when trained on 1M examples.
- **Georgian results:** On MTEB [kat], mE5 scored 71.02 at baseline, 76.22 after adaptation and 77.18 after model merging. The model, data and benchmark are released openly.

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

📚 Shelves (5): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 16,667 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 16,194 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 16,039 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 15,891 works · [🗄️ Information Systems](https://md.research.elamur.ai/free/en/shelves/computer-science/information-systems/) · 1,437 works

🧭 Topics (1): [🌱 Low-resource languages](https://md.research.elamur.ai/free/en/topics/low-resource-languages/) · 267 works

## Cite

Cite as: Zaruhi Navasardyan, Bagratuni Minsayan, Spartak Bughdaryan and Hrant Davtyan. 2026. Less is More: Adapting Text Embeddings for Low-Resource Languages with Small Scale Noisy Synthetic Data. Proceedings of the Second Workshop on Language Models for Low-Resource Languages (LoResLM 2026).

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

```bibtex
@inproceedings{navasardyan2026less,
  title = {{Less is More: Adapting Text Embeddings for Low-Resource Languages with Small Scale Noisy Synthetic Data}},
  author = {Zaruhi Navasardyan, Bagratuni Minsayan, Spartak Bughdaryan and Hrant Davtyan},
  booktitle = {Proceedings of the Second Workshop on Language Models for Low-Resource Languages (LoResLM 2026)},
  year = {2026},
  doi = {10.18653/v1/2026.loreslm-1.31},
  url = {https://doi.org/10.18653/v1/2026.loreslm-1.31}
}
```

- [← previous Legitimacy of varieties of English as a non-native language](https://md.research.elamur.ai/free/en/papers/proshina-2026-legitimacy-varieties-english-non-native/)
- [next → Let's Simplify Step by Step: Guiding LLM Towards Multilingual Unsupervised Proficiency-Controlled Sentence Simplification](https://md.research.elamur.ai/free/en/papers/zhang-2026-let-s-simplify-step-step/)
- [↑ back to papers of 2026](https://md.research.elamur.ai/free/en/papers/2026/page/5/)

## References

16 references, 10 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Payal Bajaj, Daniel Campos, Nick Craswell, Li Deng, Jianfeng Gao, Xiaodong Liu, Rangan Majumder, Andrew McNamara, Bhaskar Mitra, Tri Nguyen, Mir Rosenberg, Xia Song, Alina Stoica, Saurabh Tiwary, and Tong Wang. 2018. [Ms marco: A human generated machine reading comprehension dataset](https://md.research.elamur.ai/free/en/papers/bajaj-2016-ms-marco-human-generated/). Preprint, arXiv:…
2. Lucas Bandarkar, Davis Liang, Benjamin Muller, Mikel Artetxe, Satya Narayan Shukla, Donald Husa, Naman Goyal, Abhinandan Krishnan, Luke Zettlemoyer, and Madian Khabsa. 2024. [The belebele benchmark: a parallel reading comprehension dataset in 122 language variants](https://md.research.elamur.ai/free/en/papers/bandarkar-2023-belebele-benchmark-parallel-reading-comprehension/). In Proceedings of the 62nd Annual M…
3. Parishad BehnamGhader, Vaibhav Adlakha, Marius Mosbach, Dzmitry Bahdanau, Nicolas Chapdos, and Siva Reddy. 2024. [Llm2vec: Large language models are secretly powerful text encoders](https://md.research.elamur.ai/free/en/papers/behnamghader-2024-llm2vec-large-language-models-secretly/). arXiv preprint arXiv:2404.05961.
4. Luiz Bonifacio, Hugo Jeronymo, Hugo Q Abonizio, Israel Campiolo, Marzieh Fadaei, Roberto Lotufo, and Rodrigo Nogueira. 2022. Inpars: Data augmentation for information retrieval using large language models. arXiv preprint arXiv:2202.05144. [find](https://scholar.google.com/scholar?q=Luiz+Bonifacio%2C+Hugo+Jeronymo%2C+Hugo+Q+Abonizio%2C+Israel+Campiolo%2C+Marzieh+Fadaei%2C+Roberto+Lotufo%2C+and+Rodrigo+Nogueira%2E+2022%2E+Inpars%3A+Data+augmentation+for+information+retrieval+using+large+language+mo)
5. Alexis Conneau, Guillaume Lample, Marc’Aurelio Ranzato, Ludovic Denoyer, and Hervé Jégou. 2018. Word translation without parallel data. In International Conference on Learning Representations. [find](https://scholar.google.com/scholar?q=Alexis+Conneau%2C+Guillaume+Lample%2C+Marc%E2%80%99Aurelio+Ranzato%2C+Ludovic+Denoyer%2C+and+Herv%C3%A9+J%C3%A9gou%2E+2018%2E+Word+translation+without+parallel+data%2E+In+International+Conference+on+Learning+Representations%2E)
6. Luyu Gao, Yunyi Zhang, Jiawei Han, and Jamie Callan. 2021. [Scaling deep contrastive learning batch size under memory limited setup](https://md.research.elamur.ai/free/en/papers/gao-2021-scaling-deep-contrastive-learning-batch/). In Proceedings of the 6th Workshop on Representation Learning for NLP (RepL4NLP-2021), pages 316–321, Online. Association for Computational Linguistics.
7. Liang Wang, Nan Yang, Xiaolong Huang, Binxing Jiao, Linjun Yang, Daxin Jiang, Rangan Majumder, and Furu Wei. 2023. [Improving text embeddings with large language models](https://md.research.elamur.ai/free/en/papers/wang-2024-improving-text-embeddings-large-language/). arXiv preprint arXiv:2401.00368.
8. Odunayo Jude Ogundepo, Akintunde Oladipo, Mofetoluwa Adeyemi, Kelechi Ogueji, and Jimmy Lin. 2022. AfriTeVA: Extending ?small data? pretraining approaches to sequence-to-sequence models. In Proceedings of the Third Workshop on Deep Learning for Low-Resource Natural Language Processing, pages 126–135… [find](https://scholar.google.com/scholar?q=Odunayo+Jude+Ogundepo%2C+Akintunde+Oladipo%2C+Mofetoluwa+Adeyemi%2C+Kelechi+Ogueji%2C+and+Jimmy+Lin%2E+2022%2E+AfriTeVA%3A+Extending+%3Fsmall+data%3F+pretraining+approaches+to+sequence%2Dto%2Dsequence+models%2E+In+Proceeding)
9. David Ifeoluwa Adelani Kosei Uemura, Miaoran Zhang. 2025. [Afrimteb and afrie5: Benchmarking and adapting text embedding models for african languages](https://md.research.elamur.ai/free/en/papers/uemura-2026-afrimteb-afrie5-benchmarking-adapting-text/). arXiv preprint arXiv:2510.23896.
10. Philip May. 2021. Machine translated multilingual sts benchmark dataset. [find](https://scholar.google.com/scholar?q=Philip+May%2E+2021%2E+Machine+translated+multilingual+sts+benchmark+dataset%2E)
11. Niklas Muennighoff, Nouamane Tazi, Loïc Magne, and Nils Reimers. 2023. [Mteb: Massive text embedding benchmark](https://md.research.elamur.ai/free/en/papers/muennighoff-2022-mteb-massive-text-embedding-benchmark/). Preprint, arXiv:2210.07316.
12. Kelechi Ogueji, Yuxin Zhu, and Jimmy Lin. 2021. Small data? no problem! exploring the viability of pretraining multilingual language models for low-resourced languages. In Proceedings of the 1st Workshop on Multilingual Representation Learning, pages 116–126. Association for Computational Linguistic… [find](https://scholar.google.com/scholar?q=Kelechi+Ogueji%2C+Yuxin+Zhu%2C+and+Jimmy+Lin%2E+2021%2E+Small+data%3F+no+problem%21+exploring+the+viability+of+pretraining+multilingual+language+models+for+low%2Dresourced+languages%2E+In+Proceedings+of+the+1st+Works)
13. Nils Reimers and Iryna Gurevych. 2019. [Sentence-bert: Sentence embeddings using siamese bert-networks](https://md.research.elamur.ai/free/en/papers/reimers-2019-sentence-bert-sentence-embeddings-siamese/). In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing. Association for Computational Linguistics.
14. Matthew Snover, Bonnie Dorr, Rich Schwartz, Linnea Micciulla, and John Makhoul. 2006. [A study of translation edit rate with targeted human annotation](https://md.research.elamur.ai/free/en/papers/snover-2006-study-translation-edit-rate-targeted/). In Proceedings of the 7th Conference of the Association for Machine Translation in the Americas: Technical Papers, pages 223–231, Cambridge, Massachu…
15. Gemma Team and 1 others. 2024. Gemma: Open models based on gemini research and technology. arXiv preprint arXiv:2403.08295. [find](https://scholar.google.com/scholar?q=Gemma+Team+and+1+others%2E+2024%2E+Gemma%3A+Open+models+based+on+gemini+research+and+technology%2E+arXiv+preprint+arXiv%3A2403%2E08295%2E)
16. Liang Wang, Nan Yang, Xiaolong Huang, Binxing Jiao, Linjun Yang, Daxin Jiang, Rangan Majumder, and Furu Wei. 2022. [Text embeddings by weakly-supervised contrastive pre-training](https://md.research.elamur.ai/free/en/papers/wang-2022-text-embeddings-weakly-supervised-contrastive/). arXiv preprint arXiv:2212.03533.

*Updated 31 October 2026*