Elamur Research⌕
☰
home / papers / 2019 / Amharic-Arabic Neural Machine Translation

Amharic-Arabic Neural Machine Translation

Summary

The paper tackles machine translation from Amharic into Arabic, a pair with almost no parallel data even though both languages are Semitic and Arabic is widely used for religious teaching in Ethiopia. The authors first built a small sentence-aligned corpus by adapting Quranic material from Tanzile: an existing Arabic monolingual text and its Amharic rendering were reworked into translation pairs. They then trained two attention-based encoder-decoder NMT systems with the open-source OpenNMT toolkit. The two systems are identical apart from the recurrent unit, one using LSTM and the other GRU. Both were compared with Google Translation on the same data. The LSTM variant scored best at 12% BLEU, ahead of GRU at 11% and Google at 6%. The authors also describe why these languages are hard for MT, pointing to root-and-pattern word formation, heavy affixation, missing machine-readable lexicons and the absence of capitalization. The absolute scores are low, but the work provides a first neural baseline and a reusable corpus-construction route for an under-served language pair.

Key points

  • Two NMT models, LSTM-based and GRU-based, use the same attention encoder-decoder setup adapted from OpenNMT.
  • The training data is a small Amharic–Arabic parallel corpus of Quranic text derived from the Tanzile collections.
  • On BLEU, LSTM OpenNMT scored 12%, GRU OpenNMT 11% and Google Translation 6%.
  • The difficulties the authors name include root-and-pattern (interdigitating) morphology and prefix, suffix, infix and circumfix affixation. They also cite a shortage of machine-readable lexicons, no capitalization, and out-of-lexicon items such as proper names carrying affixes.
  • The authors consider the result decent for such a small corpus and expect improvements from more training data.

Interesting ideas

  • I1. A parallel corpus can be assembled cheaply by pairing an existing text with its existing translation when both share a canonical unit, such as numbered verses, so alignment comes almost for free. — For Mova (back): verse- or chapter-numbered classics with established translations could give aligned pairs for testing whether text → graph → text translation through the meta-world preserves meaning unit by unit.
  • I2. The two models are a clean ablation, since only the recurrent cell type changes and the rest of the architecture and data stay fixed, so any score difference can be attributed to that one component. — For Mova (tree): gparse experiments could swap exactly one component at a time (encoder block, attention variant) on a fixed treebank split, so LAS changes are explainable rather than confounded.
  • I3. A widely used commercial system is scored on the same test data as an external reference, which keeps the in-house numbers honest. — For Mova (learn): every evaluation report could include an outside parser or LLM baseline next to the negative controls, so gains are read against a known point and not only against earlier Mova versions.
  • I4. Complex words are analysed as a stem wrapped in layered affixes, with the paper's Arabic example showing a circumfix for person and number, a suffix for the object and stacked prefixes for "will" and "and". This shows how one surface token can carry several separate meanings. — For Mova (tokens): out-of-lexicon English forms, such as affixed proper names or unknown derivations, could be split into a stem plus an explicit affix list before lemma lookup, recording each removed piece as an explainable feature.
  • I5. The authors observe that regular inflection like plural marking makes translation ambiguous, and that productive morphology inflates the vocabulary, which is a reason to model morphology instead of treating each surface form as a separate word. — For Mova (world): number and other inflectional features carried up from morphology could set entity cardinality (one item versus several) in the world model, rather than leaving that to surface-word matching.

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

Topics: Machine translation · Semitic languages

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.1912.13161 · arXiv 1912.13161

@misc{gashaw2019amharic,
  title = {{Amharic-Arabic Neural Machine Translation}},
  author = {Gashaw, Ibrahim and Shashirekha, H L},
  year = {2019},
  doi = {10.48550/arxiv.1912.13161},
  eprint = {1912.13161},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.1912.13161}
}

References

31 references, 4 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. R. Mitkov, The Oxford handbook of computational linguistics. Oxford University Press, 2004. find
  2. B. Maylath, “Current trends in translation,” Commun. Lang. Work, vol. 2, no. 2, pp. 41–50, 2013. find
  3. J. Oladosu, A. Esan, I. Adeyanju, B. Adegoke, O. Olaniyan, and B. Omodunbi, “Approaches to Machine Translation: A Review,” FUOYE J. Eng. Technol., vol. 1, pp. 120–126, 2016. find
  4. MemoQ, “5 Translation Technology Trends to Watch Out for in 2019,” 2019. [Online]. Available: https://slator.com/press-releases/5-translation-technology-trends-to-watch-out-in-2019/. ↗
  5. Y. LeCun, Y. Bengio, and G. Hinton, “Deep learning,” Nature, vol. 521, no. 7553, p. 436, 2015. find
  6. Y. Wu et al., “Google’s neural machine translation system: Bridging the gap between human and machine translation,” arXiv Prepr. arXiv1609.08144, 2016. ↗
  7. S. Islam, A. Paul, B. S. Purkayastha, and I. Hussain, “CONSTRUCTION OF ENGLISH-BODO PARALLEL TEXT CORPUS FOR STATISTICAL MACHINE TRANSLATION,” Int. J. Nat. Lang. Comput. Vol, vol. 7, 2018. find
  8. H. L. Shashirekha and I. Gashaw, “DICTIONARY BASED AMHARIC-ARABIC CROSS LANGUAGE INFORMATION RETRIEVAL,” Comput. Sci. \& Inf. Technol., vol. 6, pp. 49–60, 2016. find
  9. P. Koehn and R. Knowles, “Six challenges for neural machine translation,” arXiv Prepr. arXiv1706.03872, 2017. ↗
  10. Gezmu Andargachew Mekonnen Andreas Nurnberger and T. B. Bati, “A Parallel Corpus for Amharic-English Machine Translation,” 2018. find
  11. S. M. Strassel and J. Tracey, “LORELEI Language Packs: Data, Tools, and Resources for Technology Development in Low Resource Languages.,” in LREC, 2016. find our card
  12. M. M. Sakre, M. M. Kouta, and A. M. N. Allam, “automated construction of Arabic-English parallel corpus,” J. Adv. Comput. Sci., vol. 3, 2009. find
  13. A. A.-S. Ahmad, B. Hammo, and S. Yagi, “ENGLISH-ARABIC POLITICAL PARALLEL CORPUS: CONSTRUCTION, ANALYSIS AND A CASE STUDY IN TRANSLATION STRATEGIES,” Jordanian J. Comput. Inf. Technol., vol. 3, no. 3, 2017. find
  14. G. Inoue, N. Habash, Y. Matsumoto, and H. Aoyama, “A Parallel Corpus of Arabic-Japanese News Articles,” in Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC-2018), 2018. find
  15. H. M. Alotaibi, “Arabic-English Parallel Corpus: A New Resource for Translation Training and Language Teaching,” Arab World English J. Vol., vol. 8, 2017. find
  16. P. Resnik and N. A. Smith, “The web as a parallel corpus,” Comput. Linguist., vol. 29, no. 3, pp. 349–380, 2003. find our card
  17. K. Cho et al., “Learning phrase representations using RNN encoder-decoder for statistical machine translation,” arXiv Prepr. arXiv1406.1078, vol. 3, 2014. ↗
  18. D. Bahdanau, K. Cho, and Y. Bengio, “Neural machine translation by jointly learning to align and translate,” arXiv Prepr. arXiv1409.0473, vol. 7, 2014. ↗ our card
  19. A. Almahairi, K. Cho, N. Habash, and A. Courville, “First result on Arabic neural machine translation,” arXiv Prepr. arXiv1606.02680, vol. 1, 2016. ↗
  20. M. Wordofa, “Semantic Indexing and Document Clustering for Amharic Information Retrieval,” AAU, 2013. find
  21. M. El-Haj, “Multi-document arabic text summarisation,” University of Essex, 2012. find
  22. L. Rura, W. Vandeweghe, and M. Montero Perez, “Designing a parallel corpus as a multifunctional translator’s aid,” in XVIII FIT World Congress= XVIIIe Congrès mondial de la FIT, 2008. find
  23. J. Tiedemann, “Parallel Data, Tools and Interfaces in OPUS.,” in Lrec, 2012, vol. 2012, pp. 2214–2218. find
  24. M.-T. Luong, H. Pham, and C. D. Manning, “Effective approaches to attention-based neural machine translation,” arXiv Prepr. arXiv1508.04025, 2015. ↗
  25. G. Klein, Y. Kim, Y. Deng, J. Senellart, and A. M. Rush, “Opennmt: Open-source toolkit for neural machine translation,” arXiv Prepr. arXiv1701.02810, 2017. ↗ our card
  26. N. Donges, “Recurrent Neural Networks and LSTM,” 2018. [Online]. https://towardsdatascience.com/recurrent-neural-networks-and-lstm-4b601dd822a5. ↗
  27. S. Hochreiter and J. Schmidhuber, “Long short-term memory,” Neural Comput., vol. 9, no. 8, pp. 1735–1780, 1997. find
  28. J. Chung, C. Gulcehre, K. Cho, and Y. Bengio, “Empirical evaluation of gated recurrent neural networks on sequence modeling,” arXiv Prepr. arXiv1412.3555, 2014. ↗
  29. A. Graves, “Generating sequences with recurrent neural networks,” arXiv Prepr. arXiv1308.0850, 2013. ↗
  30. S. Du, T. Li, X. Gong, Z. Yu, Y. Huang, and S.-J. Horng, “A hybrid method for traffic flow forecasting using multimodal deep learning,” arXiv Prepr. arXiv1803.02099, 2018. ↗
  31. K. Wolk and K. Marasek, “Neural-based machine translation for medical text domain. Based on European Medicines Agency leaflet texts,” Procedia Comput. Sci., vol. 64, pp. 2–9, 2015. Authors Ibrahim Gashaw Kassa is a Ph.D. candidate at Mangalore University Karnataka State, India, since 2016. He graduated in 2006 in Information System from Addis Ababa University, Ethiopia. In 2014, he obtained his master’s degree in Information Technology from the University of Gondar, Ethiopia., and he serves as a lecturer at the University of Gondar from 2009 to May 2016. His research interest is in Cross-Language Information Retrieval, Machine translation Artificial Intelligence Natural Language Processing Available: Dr. H L Shashirekha is a Professor in the Department of Computer Science, Mangalore University, Mangalore, Karnataka State, India. She completed her M.Sc. in Computer Science in 1992 and Ph.D. in 2010 from University of Mysore. She is a member of Board of Studies and Board of Examiners (PG) in Computer Science, Mangalore University. She has several papers in International Conferences and published several papers in International Journals and Conference Proceedings. Her area of research includes Text Mining and Natural Language Processing. find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/gashaw-2019-amharic-arabic-neural-machine-translation/