---
title: "Amharic Abstractive Text Summarization — Zaki 2020"
description: "The paper tackles abstractive summarization for Amharic, a widely spoken African language with almost no ready-made NLP resources."
authors: ["Amr M. Zaki", "Mahmoud I. Khalil", "Hazem M. Abbas"]
year: 2020
venue: "arXiv:2003.13721"
source: "https://arxiv.org/abs/2003.13721"
doi: "10.48550/arxiv.2003.13721"
arxiv: "2003.13721"
licence: "arXiv — non-exclusive license to distribute"
licence_url: "http://arxiv.org/licenses/nonexclusive-distrib/1.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/zaki-2020-amharic-abstractive-text-summarization/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/zaki-2020-amharic-abstractive-text-summarization/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2020](https://md.research.elamur.ai/free/en/papers/2020/page/6/) / Amharic Abstractive Text Summarization

# Amharic Abstractive Text Summarization

**Authors:** Amr M. Zaki, Mahmoud I. Khalil, Hazem M. Abbas · **Year:** 2020 · **Venue:** arXiv:2003.13721 · **Licence:** [arXiv — non-exclusive license to distribute](http://arxiv.org/licenses/nonexclusive-distrib/1.0/)

**Links:** [Publisher's page](https://arxiv.org/abs/2003.13721) · [PDF](https://arxiv.org/pdf/2003.13721) · [DOI 10.48550/arxiv.2003.13721](https://doi.org/10.48550/arxiv.2003.13721) · [arXiv 2003.13721](https://arxiv.org/abs/2003.13721) · [OpenAlex](https://openalex.org/W3014339681)

**Citations:** 6 citations (openalex-snapshot · 2026-10-14) · 1 in our library, of the 140884 papers whose bibliographies we read

## Summary
The paper tackles abstractive summarization for Amharic, a widely spoken African language with almost no ready-made NLP resources. Since no summarization corpus existed, the authors scraped seven Amharic news sites and collected more than 50k articles. They treated each article's headline as its reference summary and kept only the roughly 19k items with longer titles. They also trained their own Word2Vec embeddings for Amharic.

The model is a seq2seq LSTM with attention and pointer-generator copying, trained with Scheduled Sampling. This is a curriculum strategy in which the decoder is gradually fed its own predictions instead of the gold tokens, to reduce exposure bias. On 100 test items the system scored ROUGE-1 of about 20.5, well below English CNN/DailyMail results. The authors attribute the gap mainly to the much smaller and shorter-summary dataset.

The main contribution is practical. They release the dataset and the embeddings, and they show that English-oriented deep architectures can be moved to a low-resource language once embeddings exist.

## Key points
- The corpus was scraped from 7 Amharic news websites. Of more than 50k articles, about 19k with long titles were kept as article/headline pairs.
- The authors trained a custom Word2Vec embedding model for Amharic. Both it and the dataset are released as open source.
- The model is Scheduled Sampling (Bengio et al., 2015) on top of seq2seq + attention + pointer-generator. The code is adapted from the Keneshloo et al. (2019) library, ported to Python 3, and trained on Google Colab.
- Scores on 100 test sentences were BLEU 0.3311, ROUGE-1 F 20.51, ROUGE-2 F 8.59, and ROUGE-L F 14.76. For comparison, the same approach on CNN/DailyMail reaches about 39.53 / 17.28 for ROUGE-1 / ROUGE-2.
- Planned future work is to fine-tune a BERT-style model pretrained on English CNN/DailyMail and transfer it cross-lingually to Amharic.

## Interesting ideas
- **Scheduled Sampling:** training starts with gold history, and the model's own outputs are mixed in more and more as it matures, so training conditions come to resemble inference. — For Mova (tree): the perceptron parser could be trained on transition sequences that increasingly follow its own predicted actions, so it learns to recover from its own earlier attachment errors.
- **Headlines as free supervision:** pairing news articles with their headlines yields summarization data at no annotation cost, which suits any language or domain that lacks curated pairs. — For Mova (back): article/headline pairs could test the text → graph → text compressor, by checking whether the article's meaning graph contains the entities and events the headline names.
- **Filtering scraped data by a simple length criterion:** the authors kept only items with longer titles, a cheap heuristic that removes uninformative pairs before training. — For Mova (learn): silver data could likewise pass through cheap structural filters (length, coverage) before the witness parsers vote, so they spend effort only on plausible items.
- **Learning when to copy versus generate:** the pointer-generator chooses between copying a source token and producing one from the vocabulary, which handles names and out-of-vocabulary words. — For Mova (back): the generator could copy surface forms of names and rare words straight from the source text instead of regenerating them from lemmas, keeping output faithful and explainable.
- **Explaining a gap against a reference result:** the authors account for the score difference from the English benchmark by dataset size and summary length, not just reporting the lower numbers. — For Mova (learn): evaluation reports could pair each score with a comparable reference run and state the data size and test-set size (here only 100 items), so weak numbers are read in context.

*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: [Linguistics & languages](https://md.research.elamur.ai/free/en/shelves/linguistics/)

Topics: [Summarization](https://md.research.elamur.ai/free/en/topics/summarization/) · [Semitic languages](https://md.research.elamur.ai/free/en/topics/semitic-languages/)

## Cite

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

```bibtex
@misc{zaki2020amharic,
  title = {{Amharic Abstractive Text Summarization}},
  author = {Zaki, Amr M. and Khalil, Mahmoud I. and Abbas, Hazem M.},
  year = {2020},
  doi = {10.48550/arxiv.2003.13721},
  eprint = {2003.13721},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2003.13721}
}
```

- [← previous Transfer Learning for Less-Resourced Semitic Languages Speech Recognition: the Case of Amharic](https://md.research.elamur.ai/free/en/papers/aclbulk-2020-woldemariam-transfer-learning-less-resourced-semitic/)
- [next → Arabic Opinion Mining Using a Hybrid Recommender System Approach](https://md.research.elamur.ai/free/en/papers/harrag-2020-arabic-opinion-mining-using-hybrid/)
- [↑ back to papers of 2020](https://md.research.elamur.ai/free/en/papers/2020/page/6/)

## References

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

1. Dzmitry Bahdanau, Kyunghyun Cho, and Yoshua Bengio. Neural machine translation by jointly learning to align and translate. arXiv preprint arXiv:1409.0473, 2014. [↗](https://arxiv.org/abs/1409.0473) [our card](https://md.research.elamur.ai/free/en/papers/bahdanau-2014-neural-machine-translation-jointly-learning/)
2. Samy Bengio, Oriol Vinyals, Navdeep Jaitly, and Noam Shazeer. Scheduled sampling for sequence prediction with recurrent neural networks. In Proceedings of the 28th International Conference on Neural Information Processing Systems - Volume 1, NIPS15, pp. 11711179, Cambridge, MA, USA, 2015. MIT Press. [find](https://scholar.google.com/scholar?q=Samy+Bengio%2C+Oriol+Vinyals%2C+Navdeep+Jaitly%2C+and+Noam+Shazeer%2E+Scheduled+sampling+for+sequence+prediction+with+recurrent+neural+networks%2E+In+Proceedings+of+the+28th+International+Conference+on+Neural+I)
3. Sumit Chopra, Michael Auli, and Alexander M. Rush. Abstractive sentence summarization with attentive recurrent neural networks. In Proceedings of the 2016 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pp. 93–98, San Diego, California, June 2016. Association for Computational Linguistics. doi: 10.18653/v1/N16-1012. URL https://www.aclweb.org/anthology/N16-1012. [↗](https://doi.org/10.18653/v1/n16-1012) [our card](https://md.research.elamur.ai/free/en/papers/chopra-2016-abstractive-sentence-summarization-attentive-recurrent/)
4. Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. BERT: Pre-training of deep bidirectional transformers for language understanding. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers), pp. 4171–4186, Minneapolis, Minnesota, June 2019. Association for Computational Linguistics. doi: 10.18653/v1/N19-1423. URL https://www.aclweb.org/anthology/N19-1423. [↗](https://doi.org/10.18653/v1/n19-1423)
5. Karl Moritz Hermann, Tomas Kocisky, Edward Grefenstette, Lasse Espeholt, Will Kay, Mustafa Suleyman, and Phil Blunsom. Teaching machines to read and comprehend. In C. Cortes, N. D. Lawrence, D. D. Lee, M. Sugiyama, and R. Garnett (eds.), Advances in Neural Information Processing Systems 28, pp. 1693–1701. Curran Associates, Inc., 2015. URL [http://papers.nips.cc/paper/5945-teaching-machines-to-read-and-comprehend.pdf](http://papers.nips.cc/paper/5945-teaching-machines-to-read-and-comprehend.pdf). [↗](http://papers.nips.cc/paper/5945-teaching-machines-to-read-and-comprehend.pdf) [our card](https://md.research.elamur.ai/free/en/papers/hermann-2015-teaching-machines-read-comprehend/)
6. Sepp Hochreiter and Jürgen Schmidhuber. Long short-term memory. Neural Comput., 9(8):17351780, November 1997. ISSN 0899-7667. doi: 10.1162/neco.1997.9.8.1735. URL https://doi.org/10.1162/neco.1997.9.8.1735. [↗](https://doi.org/10.1162/neco.1997.9.8.1735)
7. Yaser Keneshloo, Tian Shi, Naren Ramakrishnan, and Chandan Reddy. Deep reinforcement learning for sequence-to-sequence models. IEEE Transactions on Neural Networks and Learning Systems, PP:1–21, 08 2019. doi: 10.1109/TNNLS.2019.2929141. [↗](https://doi.org/10.1109/tnnls.2019.2929141)
8. Julian Kupiec, Jan O. Pedersen, and Francine Chen. A trainable document summarizer. In SIGIR, 1995. [find](https://scholar.google.com/scholar?q=Julian+Kupiec%2C+Jan+O%2E+Pedersen%2C+and+Francine+Chen%2E+A+trainable+document+summarizer%2E+In+SIGIR%2C+1995%2E)
9. Chin-Yew Lin. ROUGE: A package for automatic evaluation of summaries. In Text Summarization Branches Out, pp. 74–81, Barcelona, Spain, July 2004. Association for Computational Linguistics. URL https://www.aclweb.org/anthology/W04-1013. [↗](https://aclanthology.org/W04-1013/)
10. Tomas Mikolov, G.s Corrado, Kai Chen, and Jeffrey Dean. Efficient estimation of word representations in vector space. pp. 1–12, 01 2013. [find](https://scholar.google.com/scholar?q=Tomas+Mikolov%2C+G%2Es+Corrado%2C+Kai+Chen%2C+and+Jeffrey+Dean%2E+Efficient+estimation+of+word+representations+in+vector+space%2E+pp%2E+1%E2%80%9312%2C+01+2013%2E)
11. Ramesh Nallapati, Bowen Zhou, Cicero dos Santos, Çağlar Gulçehre, and Bing Xiang. Abstractive text summarization using sequence-to-sequence RNNs and beyond. In Proceedings of The 20th SIGNLL Conference on Computational Natural Language Learning, pp. 280–290, Berlin, Germany, August 2016. Association for Computational Linguistics. doi: 10.18653/v1/K16-1028. URL https://www.aclweb.org/anthology/K16-1028. [↗](https://doi.org/10.18653/v1/k16-1028)
12. Chris D Paice. Constructing literature abstracts by computer: techniques and prospects. Information Processing & Management, 26(1):171–186, 1990. [find](https://scholar.google.com/scholar?q=Chris+D+Paice%2E+Constructing+literature+abstracts+by+computer%3A+techniques+and+prospects%2E+Information+Processing+%26+Management%2C+26%281%29%3A171%E2%80%93186%2C+1990%2E)
13. Kishore Papineni, Salim Roukos, Todd Ward, and Wei-Jing Zhu. Bleu: a method for automatic evaluation of machine translation. In Proceedings of the 40th Annual Meeting of the Association for Computational Linguistics, pp. 311–318, Philadelphia, Pennsylvania, USA, July 2002. Association for Computational Linguistics. doi: 10.3115/1073083.1073135. URL https://www.aclweb.org/anthology/P02-1040. [↗](https://doi.org/10.3115/1073083.1073135) [our card](https://md.research.elamur.ai/free/en/papers/papineni-2002-bleu-method-automatic-evaluation-machine/)
14. Marc’Aurelio Ranzato, Sumit Chopra, Michael Auli, and Wojciech Zaremba. Sequence level training with recurrent neural networks. In 4th International Conference on Learning Representations, ICLR 2016, San Juan, Puerto Rico, May 2-4, 2016, Conference Track Proceedings, 2016. URL http://arxiv.org/abs/1511.06732. [↗](https://arxiv.org/abs/1511.06732) [our card](https://md.research.elamur.ai/free/en/papers/ranzato-2015-sequence-level-training-recurrent-neural/)
15. Abigail See, Peter Liu, and Christoper Manning. Get to the point: Summarization with pointer-generator networks. pp. 1073–1083, 01 2017. doi: 10.18653/v1/P17-1099. [↗](https://doi.org/10.18653/v1/p17-1099)