Elamur Research⌕
☰
home / papers / 2026 / Development of Burushaski Speech - English Text Translation Dataset

Development of Burushaski Speech - English Text Translation Dataset

Paper · 2026 · Tauqeer Saleem, Abdul Samad, Azkaa Nasir et al.

Summary

Burushaski is a language isolate of northern Pakistan, spoken mainly in the Hunza, Nagar and Yasin valleys. It has an oral tradition, no settled writing norms and almost no speech technology. The authors argue that for such a language the core difficulty is building suitable data, not choosing a model. They therefore propose a speech-first way of creating a Burushaski–English speech translation corpus.

The elicitation design reflects the language's ergative-absolutive alignment, four noun genders and rich verb morphology. It draws on three kinds of material:

  • prompts built from a master sheet of sentence pairs, organised by grammatical category;
  • everyday formulaic expressions;
  • oral narratives and folktales.

Recordings were gathered with a purpose-built mobile app and community-based crowdsourcing. The result is a pilot corpus of about ten hours from 42 speakers.

A fine-tuned Whisper model produced English output that was largely meaningful but had high error rates. Surface metrics also diverged from semantic ones. The work offers a reusable template for other oral, morphologically complex, under-resourced languages.

Key points

  • The authors present this as the first parallel Burushaski speech–English text corpus. After curation it holds roughly 10 hours of audio, contributed by 42 speakers and recorded in both controlled and semi-natural settings.
  • Burushaski has an estimated 90,000–120,000 speakers across three mutually intelligible dialects (Hunza, Nagar, Yasin).
  • Prompt categories were informed by Lahiri (2018) and the Max Planck Field Manuals. They cover:
    • simple and complex sentences;
    • case marking;
    • tense–aspect–mood contrasts;
    • negation.
  • The custom mobile app is built for field conditions with weak infrastructure and low literacy.
  • Baseline Whisper experiments were scored with WER, BLEU and BERTScore. Semantic similarity was relatively good, but lexical substitutions were frequent. Planned future work includes native-speaker human evaluation and CER-based assessment.

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

🎓Habib UniversityKarachi · Pakistan5 works · 4 citations

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksArts and Humanities · 17,450 worksLanguage and Linguistics · 17,296 works

🧭 Topics 1Research subjects and languages the work is about.English · 2,379 works

Cite

Cite as Tauqeer Saleem, Abdul Samad, Azkaa Nasir, Adina Adnan Mansoor, Fatima Faisal and Mahrukh Yousuf. 2026. Development of Burushaski Speech - English Text Translation Dataset. Proceedings of the Second workshop on Challenges in Processing South Asian Languages (CHiPSAL2026).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.63317/3ujn9xefj6ca

@inproceedings{saleem2026development,
  title = {{Development of Burushaski Speech - English Text Translation Dataset}},
  author = {Tauqeer Saleem, Abdul Samad, Azkaa Nasir, Adina Adnan Mansoor, Fatima Faisal and Mahrukh Yousuf},
  booktitle = {Proceedings of the Second workshop on Challenges in Processing South Asian Languages (CHiPSAL2026)},
  year = {2026},
  doi = {10.63317/3ujn9xefj6ca},
  url = {https://doi.org/10.63317/3ujn9xefj6ca}
}

References

6 references, 2 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Babu, A., Wang, C., Tjandra, A., Lakhotia, K., Xu, Q., Goyal, N., Singh, K., von Platen, P., Saraf, Y., Pino, J., Baevski, A., Conneau, A., and Auli, M. (2021). XLS-R: Self-supervised cross-lingual speech representation learning at scale. arXiv:2111.09296.
  2. Baevski, A., Zhou, Y., Mohamed, A., and Auli, M. (2020). wav2vec 2.0: A framework for self-supervised learning of speech representations. In Advances in Neural Information Processing Systems 33 (NeurIPS 2020).
  3. Fathullah, Y., Wu, C., Lakomkin, E., Jia, J., Shangguan, Y., Li, K., Guo, J., Xiong, W., Mahadeokar, J., Kalinli, O., Fuegen, C., and Seltzer, M. L. (2023). AudioChatLlama: Towards general-purpose speech abilities for LLMs. arXiv:2311.06753. find
  4. Finlayson, N., Marsden, E., and Hawkes, R. (2024). Creating and evaluating corpus-informed word lists for adolescent, beginner-to-low-intermediate learners of French, German, and Spanish. find
  5. Language Teaching Research. December 2024. find
  6. Godard, P., Adda, G., Adda-Decker, M., Benjumea, J., Besacier, L., Cooper-Leavitt, J., Kouarata, G.-N., Lamel, L., Maynard, H., Mueller, M., Rialland, A., Schwartz, S., Yvon, F., and Zanon-Boito, M. (2018). A very low resource language speech corpus for computational language documentation experimen… find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/saleem-2026-development-burushaski-speech-english-text/