---
title: "Question Answering Classification for Amharic Social Media"
description: "Amharic has had no public dataset for question classification, the step that predicts the kind of answer… 20 references, cited by 5 papers here."
authors: ["Tadesse Destaw Belay", "Seid Muhie Yimam", "Abinew Ayele", "Chris Biemann"]
year: 2022
venue: "Proceedings of the 1st Annual Meeting of the ELRA/ISCA Special Interest Group on Under-Resourced Languages"
source: "https://aclanthology.org/2022.sigul-1.18/"
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/en/papers/belay-2022-question-answering-classification-amharic-social/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/belay-2022-question-answering-classification-amharic-social/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2022](https://md.research.elamur.ai/free/en/papers/2022/page/7/) / Question Answering Classification for Amharic Social Media Community Based Questions

# Question Answering Classification for Amharic Social Media Community Based Questions

**Authors:** [Tadesse Destaw Belay](https://md.research.elamur.ai/free/en/authors/tadesse-destaw-belay/), [Seid Muhie Yimam](https://md.research.elamur.ai/free/en/authors/seid-muhie-yimam/), [Abinew Ayele](https://md.research.elamur.ai/free/en/authors/abinew-ali-ayele/) and [Chris Biemann](https://md.research.elamur.ai/free/en/authors/chris-biemann/) · **Year:** 2022 · **Venue:** [Proceedings of the 1st Annual Meeting of the ELRA/ISCA Special Interest Group on Under-Resourced Languages](https://md.research.elamur.ai/free/en/venues/sigul/) ([SIGUL Joint Workshop with ELE, EURALI, and DCLRL: Towards Inclusivity and Equality: Language Resources and Technologies for Under-Resourced and Endangered Languages 2022](https://md.research.elamur.ai/free/en/venues/sigul/2022/)) · **Publisher:** [European Language Resources Association (ELRA)](https://md.research.elamur.ai/free/en/publishers/european-language-resources-association-elra/) · **Licence:** [CC BY-NC 4.0 (ELRA, LREC proceedings)](https://creativecommons.org/licenses/by-nc/4.0/)

**Links:** [Publisher's page](https://aclanthology.org/2022.sigul-1.18/) · [PDF](https://aclanthology.org/2022.sigul-1.18.pdf) · [OpenAlex](https://openalex.org/W6891586970)

**Citations:** 0 citations (OpenAlex, 2026-10-14) · [5](https://md.research.elamur.ai/free/en/papers/belay-2022-question-answering-classification-amharic-social/cited-by/) in our library, of the 149,439 papers whose bibliographies we read

## Summary
Amharic has had no public dataset for question classification, the step that predicts the kind of answer a question expects before a QA system searches for it. The authors gathered questions from @AskAnythingEthiopia, a Telegram channel where people post questions through a bot. The bot makes each asker choose one of 20 topical categories, and the authors use those choices as labels. The questions mix Amharic in Fidäl script, Amharic typed in Latin letters, and English. Romanized Amharic turned out to be about three times more frequent than Fidäl, and no embeddings or language models existed for it. To handle this, the authors wrote a rule-based Latin-to-Ethiopic transliteration tool. They then trained deep learning classifiers, with the best reaching an F-score of 57.79 across the 20 classes. A cross-evaluation in which models were tested on the other script showed large performance drops. The dataset, the transliterator and the preprocessing scripts are openly released. This gives the community a first benchmark for Amharic community QA and a practical way to make romanized text usable with existing models.

## Key points
- **Source and scale:** about 83,851 questions were pulled with the Telethon library from @AskAnythingEthiopia. The channel has existed since May 2019 and has over 78k subscribers. The usable Amharic portion is 7,967 questions in Fidäl and 24,446 in Latin script.
- **Labels:** there are 20 categories, such as Politics, Music, Technology and Religion, plus a catch-all "Other". Each label is the one the asker selected in the posting bot.
- **Transliteration tool:** the Latin-to-Fidäl transliterator is released with the amharicprocessor toolkit and the Amharic semantic models repository (uhh-lt/amharicmodels).
- **Cross-script results (AmRoBERTa embeddings):**
  - A model trained on Amharic and tested on transliterated text drops to F1 21.44.
  - A model trained on transliterated text and tested on Amharic reaches 36.92.
  - Merged training gives 42.11 on the Amharic test and 39.39 on the transliterated test.
  - All of these stay well below same-distribution performance.
- **Error analysis:** most errors involve "Other". The model tends to assign a topical label, such as Technology or Politics, even when the question contains no evidence for it.

## Interesting ideas
- Labels that users must pick when they post content can serve as a large, cheap supervision signal, though catch-all classes add noise.
- A train-on-A/test-on-B grid across text variants shows how much a model depends on surface form, and in which direction the transfer fails.
- When no model exists for a noisy surface form, a rule-based mapping to the canonical form lets existing resources be reused instead of training new ones.
- The analysis of "Other" errors shows a classifier choosing a topical class with no supporting evidence in the question.
- Knowing the expected answer category in advance shrinks the space of candidate answers.

## For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

- **I1** (learn): Mine structured metadata such as forum tags, section headers or exercise types as weak silver labels, and track the noisy catch-all buckets separately.
- **I2** (learn): Run the parser and tagger across registers or UD dialect conversions in a swapped grid as a negative control, to check that gains are not just surface-form memorization.
- **I3** (tokens): Normalize informal spellings, leetspeak or nonstandard orthography to canonical English before tagging and lemmatization, so the existing lexicon and morphology still apply.
- **I4** (answer): Give readers an explicit "no evidence" outcome tied to the meaning graph, so they abstain instead of guessing a plausible-looking label.
- **I5** (answer): Derive the expected answer type from the question's UD tree and frame, and use it to restrict which entities and states the reader searches in the world model.

*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

- [Universität Hamburg](https://md.research.elamur.ai/free/en/universities/universitat-hamburg/) — University, Hamburg, Germany · 79 works · 5,422 citations
- [Instituto Politécnico Nacional](https://md.research.elamur.ai/free/en/universities/instituto-politecnico-nacional/) — University, Mexico City, Mexico · 18 works · 4,307 citations

📚 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 (2): [Semitic languages](https://md.research.elamur.ai/free/en/topics/semitic-languages/) · 2,908 works · [❓ Question answering](https://md.research.elamur.ai/free/en/topics/question-answering/) · 441 works

## Cite

Cite as: Tadesse Destaw Belay, Seid Muhie Yimam, Abinew Ayele and Chris Biemann. 2022. Question Answering Classification for Amharic Social Media Community Based Questions. Proceedings of the 1st Annual Meeting of the ELRA/ISCA Special Interest Group on Under-Resourced Languages.

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

```bibtex
@inproceedings{belay2022question,
  title = {{Question Answering Classification for Amharic Social Media Community Based Questions}},
  author = {Tadesse Destaw Belay, Seid Muhie Yimam, Abinew Ayele and Chris Biemann},
  booktitle = {Proceedings of the 1st Annual Meeting of the ELRA/ISCA Special Interest Group on Under-Resourced Languages},
  year = {2022},
  doi = {10.48448/0550-et20},
  url = {https://doi.org/10.48448/0550-et20}
}
```

- [← previous PoliSe: Reinforcing Politeness Using User Sentiment for Customer Care Response Generation](https://md.research.elamur.ai/free/en/papers/firdaus-2022-polise-reinforcing-politeness-user-sentiment/)
- [next → Stanceosaurus: Classifying Stance Towards Multilingual Misinformation](https://md.research.elamur.ai/free/en/papers/zheng-2022-stanceosaurus-classifying-stance-towards-multilingual/)
- [↑ back to papers of 2022](https://md.research.elamur.ai/free/en/papers/2022/page/7/)

## References

20 references, 5 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Abedissa, T. (2013). Amharic question answering for definitional, biographical and description questions. Unpublished Master’s Thesis, Computer Science Department, Addis Ababa University, Addis Ababa, Ethiopia. [find](https://scholar.google.com/scholar?q=Abedissa%2C+T%2E+%282013%29%2E+Amharic+question+answering+for+definitional%2C+biographical+and+description+questions%2E+Unpublished+Master%E2%80%99s+Thesis%2C+Computer+Science+Department%2C+Addis+Ababa+University%2C+Addis+Ababa%2C)
2. Akbik, A., Blythe, D., and Vollgraf, R. (2018). [Contextual string embeddings for sequence labeling](https://md.research.elamur.ai/free/en/papers/akbik-2018-contextual-string-embeddings-sequence-labeling/). In Proceedings of the 27th international conference on computational linguistics, pages 1638–1649, New Mexico, USA.
3. Amha, A. (2009). On loans and additions to the Fidäl (Ethiopic) writing system. In The Idea of Writing, pages 179–196. Brill. [↗](https://doi.org/10.1163/ej.9789004174467.i-396.52)
4. Belay, T. D., Ayele, A. A., Gelaye, G., Yimam, S. M., and Biemann, C. (2021). Impacts of homophone normalization on semantic models for Amharic. In 2021 International Conference on Information and Communication Technology for Development for Africa (ICT4DA), pages 101– 106, Bahir Dar, Ethiopia. IEEE. [find](https://scholar.google.com/scholar?q=Belay%2C+T%2E+D%2E%2C+Ayele%2C+A%2E+A%2E%2C+Gelaye%2C+G%2E%2C+Yimam%2C+S%2E+M%2E%2C+and+Biemann%2C+C%2E+%282021%29%2E+Impacts+of+homophone+normalization+on+semantic+models+for+Amharic%2E+In+2021+International+Conference+on+Information+and+Com)
5. Conneau, A., Khandelwal, K., Goyal, N., Chaudhary, V., Wenzek, G., Guzmán, F., Grave, E., Ott, M., Zettlemoyer, L., and Stoyanov, V. (2019). Unsupervised cross-lingual representation learning at scale. arXiv 2019 preprint arXiv:1911.02116, page 8440–8451. [↗](https://arxiv.org/abs/1911.02116)
6. Habtamu, S. (2021). Amharic Question Classification System Using Deep Learning Approach. 9 [https://github.com/uhh-lt/amharicmodels:](https://github.com/uhh-lt/amharicmodels:) The dataset are released under a permissive license Unpublished master thesis, Addis Ababa University. [↗](https://github.com/uhh-lt/amharicmodels:)
7. Lei, T., Shi, Z., Liu, D., Yang, L., and Zhu, F. (2018). A novel cnn-based method for question classification in intelligent question answering. In Proceedings of the 2018 International Conference on Algorithms, Computing and Artificial Intelligence, pages 1–6, Sanya China. [find](https://scholar.google.com/scholar?q=Lei%2C+T%2E%2C+Shi%2C+Z%2E%2C+Liu%2C+D%2E%2C+Yang%2C+L%2E%2C+and+Zhu%2C+F%2E+%282018%29%2E+A+novel+cnn%2Dbased+method+for+question+classification+in+intelligent+question+answering%2E+In+Proceedings+of+the+2018+International+Conference+on+)
8. Li, X. and Roth, D. (2002). [Learning question classifiers](https://md.research.elamur.ai/free/en/papers/li-2002-learning-question-classifiers/). In COLING 2002: The 19th International Conference on Computational Linguistics, page 1–7, Taipei, Taiwan.
9. Li, X. and Roth, D. (2006). Learning question classifiers: the role of semantic information. Natural Language Engineering, 12(3):229–249. [↗](https://doi.org/10.1017/s1351324905003955)
10. Liu, Y., Ott, M., Goyal, N., Du, J., Joshi, M., Chen, D., Levy, O., Lewis, M., Zettlemoyer, L., and Stoyanov, V. (2019). [RoBERTa: A robustly optimized BERT pretraining approach](https://md.research.elamur.ai/free/en/papers/liu-2019-roberta-robustly-optimized-bert-pretraining/). arXiv preprint arXiv:1907.11692. [↗](https://arxiv.org/abs/1907.11692)
11. May, R. and Steinberg, A. (2004). Al, building a question classifier for a TREC-style question answering system. AL: The Stanford Natural Language Processing Group, Final Projects. [find](https://scholar.google.com/scholar?q=May%2C+R%2E+and+Steinberg%2C+A%2E+%282004%29%2E+Al%2C+building+a+question+classifier+for+a+TREC%2Dstyle+question+answering+system%2E+AL%3A+The+Stanford+Natural+Language+Processing+Group%2C+Final+Projects%2E)
12. Metzler, D. and Croft, W. B. (2005). Analysis of statistical question classification for fact-based questions. Information Retrieval, 8(3):481–504. [↗](https://doi.org/10.1007/s10791-005-6995-3)
13. Nega, A., Chekol, W., and Kumlachew, A. (2016). Question classification in amharic question answering system: Machine learning approach. International Journal of Advanced Studies in Computers, Science and Engineering, 5(10):14–21. [find](https://scholar.google.com/scholar?q=Nega%2C+A%2E%2C+Chekol%2C+W%2E%2C+and+Kumlachew%2C+A%2E+%282016%29%2E+Question+classification+in+amharic+question+answering+system%3A+Machine+learning+approach%2E+International+Journal+of+Advanced+Studies+in+Computers%2C+Science)
14. Sangodiah, A., Muniandy, M., and Heng, L. E. (2015). Question classification using statistical approach: A complete review. Journal of Theoretical & Applied Information Technology, 71(3):386–395. [find](https://scholar.google.com/scholar?q=Sangodiah%2C+A%2E%2C+Muniandy%2C+M%2E%2C+and+Heng%2C+L%2E+E%2E+%282015%29%2E+Question+classification+using+statistical+approach%3A+A+complete+review%2E+Journal+of+Theoretical+%26+Applied+Information+Technology%2C+71%283%29%3A386%E2%80%93395%2E)
15. Taffa, T. A. and Libsie, M. (2019). Amharic question answering for biography, definition, and description questions. In Proceedings of the 2019 Workshop on Widening NLP, pages 110–113, Florence, Italy. Association for Computational Linguistics. [find](https://scholar.google.com/scholar?q=Taffa%2C+T%2E+A%2E+and+Libsie%2C+M%2E+%282019%29%2E+Amharic+question+answering+for+biography%2C+definition%2C+and+description+questions%2E+In+Proceedings+of+the+2019+Workshop+on+Widening+NLP%2C+pages+110%E2%80%93113%2C+Florence%2C+Italy)
16. Tedla, T. (2015). [amLite: Amharic Transliteration Using Key Map Dictionary](https://md.research.elamur.ai/free/en/papers/tedla-2015-amlite-amharic-transliteration-using-key/). arXiv preprint arXiv:1509.04811. [↗](https://arxiv.org/abs/1509.04811)
17. Van-Tu, N. and Anh-Cuong, L. (2016). Improving question classification by feature extraction and selection. Indian Journal of Science and Technology, 9(17):1–8. [find](https://scholar.google.com/scholar?q=Van%2DTu%2C+N%2E+and+Anh%2DCuong%2C+L%2E+%282016%29%2E+Improving+question+classification+by+feature+extraction+and+selection%2E+Indian+Journal+of+Science+and+Technology%2C+9%2817%29%3A1%E2%80%938%2E)
18. Yang, Y., Liu, J., and Liaozheng, Y. (2018). Chinese question classification based on deep learning. In Advanced Multimedia and Ubiquitous Engineering, pages 315–320. Springer. [find](https://scholar.google.com/scholar?q=Yang%2C+Y%2E%2C+Liu%2C+J%2E%2C+and+Liaozheng%2C+Y%2E+%282018%29%2E+Chinese+question+classification+based+on+deep+learning%2E+In+Advanced+Multimedia+and+Ubiquitous+Engineering%2C+pages+315%E2%80%93320%2E+Springer%2E)
19. Yimam, S. M. and Libsie, M. (2009). TETEYEQ: Amharic question answering for factoid questions. IE-IR-LRL, 3(4):17–25. [find](https://scholar.google.com/scholar?q=Yimam%2C+S%2E+M%2E+and+Libsie%2C+M%2E+%282009%29%2E+TETEYEQ%3A+Amharic+question+answering+for+factoid+questions%2E+IE%2DIR%2DLRL%2C+3%284%29%3A17%E2%80%9325%2E)
20. Yimam, S. M., Ayele, A. A., Venkatesh, G., Gashaw, I., and Biemann, C. (2021). [Introducing various semantic models for Amharic: Experimentation and evaluation with multiple tasks and datasets](https://md.research.elamur.ai/free/en/papers/yimam-2020-introducing-various-semantic-models-amharic/). Future Internet, 13(11). [↗](https://doi.org/10.3390/fi13110275)

*Updated 1 November 2026*