Elamur Research⌕
☰
home / papers / 2015 / Machine Translation Experiments on PADIC: A Parallel Arabic DIalect Corpus

Machine Translation Experiments on PADIC: A Parallel Arabic DIalect Corpus

Paper · 2015 · Karima Meftouh, Salima Harrat, Salma Jamoussi et al.

Summary

Arabic dialects are mostly spoken, lack standard spelling, and have few written resources. This makes data-driven translation between them and Modern Standard Arabic (MSA) difficult. The authors address this gap with PADIC, a multi-dialect parallel corpus created manually rather than harvested from the web, because online dialect text is noisy and inconsistent in script. PADIC aligns, at the sentence level, 6400 sentences in MSA and in five dialects: two Algerian (Annaba, Algiers), one Tunisian (Sfax), Syrian and Palestinian. Spelling was normalized with a simple convention: words that exist in MSA keep MSA orthography, and other words are written as pronounced. The authors trained statistical MT systems for every language pair. For MSA as the target, they interpolated the language model with one built from a large Arabic corpus, and they also compared LM smoothing methods. Closely related dialects translate best, especially the two Algerian ones, and Palestinian is the dialect closest to MSA. Neither smoothing nor LM interpolation produced statistically significant gains. The corpus is a building block toward dialect speech-to-speech translation.

Key points

  • PADIC contains 6400 sentences for each of 6 varieties:

    • MSA
    • ANB (Annaba)
    • ALG (Algiers)
    • TUN (Sfax)
    • SYR (Damascus)
    • PAL (Palestinian)

    The authors describe it as the largest parallel Arabic dialect corpus at the time and the first to cover both Maghreb and Middle-East dialects. The only earlier multi-dialect corpus (Bouamor et al., 2014) had 2000 sentences.

  • Translation quality was reported with BLEU, TER and METEOR.

    • The best pair was ANB↔ALG, which share up to about 60% of their vocabulary.
    • SYR↔PAL also scored well.
    • The weakest pair was Syrian–Algerian, partly because of French borrowings in Algerian.

  • Among the dialects, PAL gives the best translations to and from MSA. Unexpectedly, TUN is closer to MSA than the Algerian dialects and is easier to translate into MSA than SYR.

  • The smoothing comparison was run on English–Arabic training sets S2, S5, S12 and S15, each tested on 20 test sets of 500 sentences. A T-test at α = 0.05 found no significant difference between the methods. For example, Kneser-Ney (KN) BLEU on S2 ranged from 4.25 to 9.56.

  • Future work includes adding Moroccan and rewriting parts of large MSA corpora into dialects.

Interesting ideas

  • I1. Normalize spelling against the standard language. Write a word in standard orthography whenever it exists there, and spell it phonetically only otherwise. This keeps the data consistent and still lets pronunciation be recovered for speech work. The same rule could suit surzhyk or regional Ukrainian varieties.
  • I2. Build the corpus by hand when web text is unusable. Mixed scripts, Latin transliteration with digits, and code-switching made crawling impractical here. Producing translations manually from scratch can be the more reliable path for low-resource varieties.
  • I3. Test whether a technique matters, not just whether one run looks better. The authors repeated experiments over 20 test sets and applied a significance test. This kind of repeated evaluation shows whether a small BLEU difference is real or just noise, which is especially valuable with small corpora.
  • I4. Read MT scores as a measure of language closeness. Translation quality between pairs tracked how related the dialects are. Cross-pair MT scores could serve as a cheap empirical proxy for distance between language varieties.
  • I5. Borrow target-side data from the standard language. Interpolating a small in-domain LM with one trained on a large standard-language corpus is an inexpensive way to use abundant related data. It is worth trying, though in this paper it did not give significant gains.

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

🔬Laboratoire Lorrain de Recherche en Informatique et ses ApplicationsNancy · France23 works · 3,750 citations🎓École Nationale Supérieure d'InformatiqueAlgiers · Algeria4 works · 375 citations

Also on the byline: Laboratoire de Recherche en Informatique.

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 9,086 worksArtificial Intelligence · 8,889 worksArts and Humanities · 9,296 worksLanguage and Linguistics · 9,206 works

🧭 Topics 2Research subjects and languages the work is about.Machine translation · 958 worksSemitic languages · 1,484 works

Cite

Cite as Karima Meftouh, Salima Harrat, Salma Jamoussi, Mourad Abbas and Kamel Smaili. 2015. Machine Translation Experiments on PADIC: A Parallel Arabic DIalect Corpus. Proceedings of the 29th Pacific Asia Conference on Language, Information and Computation.

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

@inproceedings{meftouh2015machine,
  title = {{Machine Translation Experiments on PADIC: A Parallel Arabic DIalect Corpus}},
  author = {Karima Meftouh, Salima Harrat, Salma Jamoussi, Mourad Abbas and Kamel Smaili},
  booktitle = {Proceedings of the 29th Pacific Asia Conference on Language, Information and Computation},
  year = {2015},
  url = {https://aclanthology.org/Y15-1004/}
}

References

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

  1. Houda Bouamor, Nizar Habash, and Kemal Oflazer. 2014. A Multidialectal Parallel Corpus of Arabic. In Proceedings of the Language Resources and Evaluation Conference, LREC-2014, pages 1240–1245. David Chiang, Mona Diab, Nizar Habash, Owen Ram-WB Max E[X] 8.97 6.43 11.32 7.99 14.13 9.35 14.27 9.74 σ 2 2.23 2.92 4.32 4.72 σ XY 2.33 3.16 4.45 4.75 Min 4.1 5.18 5.95 6.19 p-value 0.33 0.35 0.36 0.39 bow, and Safiullah Shareef. 2006. Parsing Arabic Dialects. In Proceedings of the European Chapter of ACL (EACL).
  2. Heba Elfardy and Mona Diab. 2013. Sentence Level Dialect Identification in Arabic. In ACL (2), pages 456– 461.
  3. Nizar Habash and Owen Rambow. 2006. Magead: A Morphological Analyzer and Generator for the Arabic Dialects. In Proceedings of the 21st International Conference on Computational Linguistics and 44th Annual Meeting of the Association for Computational Linguistics, pages 681–688. ↗
  4. Nizar Habash, Ryan Roth, Owen Rambow, Ramy Eskander, and Nadi Tomeh. 2013. Morphological analysis and disambiguation for dialectal arabic. In Proceedings of NAACL-HLT, pages 426–432, Atlanta, Georgia.
  5. Salima Harrat, Mourad Abbas, Karima Meftouh, and Kamel Smaili. 2013. Diacritics restoration for arabic dialect texts. In Proceedings of 14th Annual Conference of the International Communication Association (Interspeech), pages 125–132. find
  6. Salima Harrat, Karima Meftouh, Mourad Abbas, and Kamel Smaili. 2014. Building resources for algerian arabic dialects. In Proceedings of 15th Annual Conference of the International Communication Association (Interspeech), pages 2123–2127. find
  7. Salima Harrat, Karima Meftouh, Mourad Abbas, Salma Jamoussi, and Kamel Smaili. 2015. Cross-dialectal arabic processing. In Computational Linguistics and Intelligent Text Processing, 16th International Conference, CICLing 2015 proceeding, part 1, pages 620– 632, April. find
  8. Hanaa Kilany, H. Gadalla, Howaida Arram, A. Yacoub, Alaa El-Habashi, and C. McLemore. 2002. Egyptian Colloquial Arabic Lexicon. In LDC catalog number LDC99L22. ↗
  9. Katrin Kirchhoff, Jeff Bilmes, Sourin Das, Nicolae Duta, Melissa Egan, Gang Ji, Feng He, John Hopkins, Daben Liu, Mohamed Noamany, Pat Schone, Richard Schwartz, and Dimitra Vergyri. 2003. Novel Approaches to Arabic Speech Recognition: Report from the 2002 Johns-Hopkins Summer Workshop. In Proc. IEEE Intl. Conf. on Acoustics, Speech, and Signal Processing, pages 344–347. Philipp Koehn, Hieu. find
  10. Hoang, Alexandra Birch, Chris Callison-Burch, Marcello Federico, Nicola Bertoldi, Brooke Cowan, Wade Shen, Christine Moran, Richard Zens, Chris Dyer, Ondrej Bojar, Alexandra Constantin, and Evan Herbst. 2007. Moses: Open Source Toolkit for Statistical Machine Translation. Proceedings of the Annual Meeting of the Association for Computational Linguistics, demonstation session, pages 177–180. find
  11. Mohamed Maamouri, Ann Bies, Tim Buckwalter, and Hubert Jin. 2004. Arabic treebank: Part 3 v 1.0. In Linguistic Data Consortium. ↗
  12. Franz Josef Och and Hermann Ney. 2003. A Systematic Comparison of Various Statistical Alignment Models. Computational Linguistics, Volume 29, No 1, pages 19–51. ↗
  13. Arfath Pasha, Mohamed Al-Badrashiny, Ramy Kholy Ahmed El Eskander, Mona Diab, Nizar Habash, Manoj Pooleery, Owen Rambow, and Ryan Roth. 2014. Madamira: A fast, comprehensive tool for morphological analysis and disambiguation of arabic. In LREC 2014, Language Resources and Evaluation Conference, Reykjavik, Iceland.
  14. Tachicart Ridouane and Bouzoubaa Karim. 2014. A hybrid approach to translate moroccan arabic dialect. In SITA’14, 9th International Conference on Intelligent Systems. Wael. Salloum and Nizar. find
  15. Habash. 2012. Elissa: A dialectal to standard arabic machine translation system. In Coling’2012, 24th International Conference on Computational Linguistics, pages 385–392. find
  16. Wael Salloum and Nizar Habash. 2013. Dialectal Arabic to English Machine Translation: Pivoting through Modern Standard Arabic. In Proceedings of the 2013 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, NAACL HLT 13, pages 348–358. Hassan. find
  17. Sawaf. 2010. Arabic dialect handling in hybrid machine translation. In AMTA’2010, 9th Conf. of the Association for Machine Translation in the Americas. find
  18. Inguna Skadiņa, Ahmet Aker, Voula Giouli, Dan Tufis, Robert Gaizauskas, Madara Mieriņa, and Nikos Mastropavlos. 2010. A Collection of Comparable Corpora for Under-resourced Languages. In Proceedings of the 2010 Conference on Human Language Technologies – The Baltic Perspective: Proceedings of the Fourth International Conference Baltic HLT 2010, pages 161– 168. ↗
  19. Andreas Stolcke. 2002. Srilm – an Extensible Language Modeling Toolkit. In ICSLP, pages 901–904, Denver, USA. find
  20. Rabih Zbib, Erika Malchiodi, Devlin Jacob, David Stallard, Spyros Matsoukas, Richard Schwartz, John Makhoul, Omar Zaidan, and Chris Callison-Burch. 2012. Machine Translation of Arabic Dialects. In Proceedings of the 2012 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, NAACL HLT 12, pages 49–59.

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/meftouh-2015-machine-translation-experiments-padic-parallel/