Elamur Research⌕
☰
home / papers / 2013 / ASMA: A System for Automatic Segmentation and Morpho-Syntactic Disambiguation of Modern Standard Arabic

ASMA: A System for Automatic Segmentation and Morpho-Syntactic Disambiguation of Modern Standard Arabic

Paper · 2013 · Muhammad Abdul-Mageed, Mona Diab and Sandra Kübler

Summary

Arabic packs clitics, stems and inflectional affixes into a single written word, so POS tagging requires splitting words and disambiguating their morphology together. The leading tools fall short in different ways. AMIRA 2.1 does not split off inflectional affixes. MADA+TOKAN 3.2 depends on a hand-built morphological analyzer and lexicon. ASMA shows that both steps can be done with language-independent learning and no analyzer. Segmentation is framed as character-level IOB labelling, and each resulting morpheme then gets a fine-grained POS tag. Both tasks use the memory-based tagger MBT, with one classifier per task. ASMA was trained and tested on undiacritized newswire from the Penn Arabic Treebank, using two splits that reproduce the competitors' evaluations. It reaches 98.34% segmentation accuracy, 96.26% POS accuracy with a 139-tag set, and 97.59% with the morphology-free CATiB tagset. These results are close to MADA's. ASMA is also cheaper to build, needs almost no training time, and runs fast. This matters for languages without costly analyzers and for large-scale text processing.

Key points

  • Segments are stems, inflectional affixes or clitics. Labels are {B-SEG, I-SEG, O} per character. Morphotactic boundaries are not restored, so outputs like wilaAy are stems rather than valid lexemes.
  • The classifier is MBT, a memory-based tagger. The tuned settings were IB1, weighted overlap similarity, gain ratio feature weighting and k = 1.
  • The data are PATB parts ATB1V4, ATB2V3, ATB3V3.1 and ATB3V3.2, used without diacritics, with an AMIRA-SPLIT and a MADA-SPLIT.
  • On MADA-SPLIT, word-level segmentation accuracy is 98.10%, versus MADA's 98.85% with an analyzer. CATiB tagging on that split reaches 95.92%.
  • Speed is about 5 minutes per 100,000 words on a PowerPC 970 with 2GB of RAM. Frequent errors are NOUN vs NOUN_PROP vs ADJ, passive vs active verbs (because passives are marked only by diacritics), and ADJ.VN vs ADJ.

Interesting ideas

  • I1. Treating word-internal segmentation as per-character B/I/O labelling turns a morphology problem into ordinary sequence tagging, with no rules or lexicon needed.
  • I2. Memory-based learning with k = 1 keeps every training instance, so each decision can be justified by pointing to the nearest stored example, and training costs almost nothing.
  • I3. Running one model at several tagset granularities (139 tags, RTS, CATiB) showed that coarser tags help accuracy, but not in a strictly monotonic way, and that the tagset should match the downstream consumer.
  • I4. Rebuilding each competitor's exact data split, rather than reporting on a single split, makes cross-system comparisons fair.
  • I5. The error analysis tied confusions to properties of the writing system (missing diacritics, no capitalization), not just to the model.

For Mova

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

  • I1 (tokens): Contractions, possessive 's, hyphenated compounds and noisy tokens could be split by a character-level IOB tagger trained on treebank token boundaries, with every split traceable to its label sequence.
  • I2 (tokens): A nearest-neighbour fallback tagger for rare morphology or lemmas would give explainable decisions ("tagged like this stored case") and could be retrained instantly after corrections.
  • I3 (learn): The evaluation suite could report tagger and parser scores at several label granularities, which would show whether errors come from fine morphological features or from core categories.
  • I4 (learn): When benchmarking gparse or the perceptron parser against published UD results, the evaluation should reproduce each paper's split, and any mismatch should be flagged rather than mixing numbers.
  • I5 (tree): Parser error reports could group confusions by linguistic cause, such as -ing gerund vs participle or lowercase proper nouns, so that repair rules target the real source of ambiguity.

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

🎓George Washington UniversityWashington · United States51 works · 8,148 citations🎓Indiana UniversityBloomington · United States41 works · 3,663 citations

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 works

🧭 Topics 2Research subjects and languages the work is about.Semitic languages · 2,908 worksMorphology · 405 works

Cite

Cite as Muhammad Abdul-Mageed, Mona Diab and Sandra Kübler. 2013. ASMA: A System for Automatic Segmentation and Morpho-Syntactic Disambiguation of Modern Standard Arabic. Proceedings of the International Conference Recent Advances in Natural Language Processing RANLP 2013.

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

@inproceedings{abdulmageed2013asma,
  title = {{ASMA: A System for Automatic Segmentation and Morpho-Syntactic Disambiguation of Modern Standard Arabic}},
  author = {Muhammad Abdul-Mageed, Mona Diab and Sandra Kübler},
  booktitle = {Proceedings of the International Conference Recent Advances in Natural Language Processing RANLP 2013},
  year = {2013},
  url = {https://aclanthology.org/R13-1001/}
}

References

20 references, 10 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. David Aha, Dennis Kibler, and Marc K. Albert. 1991. Instance-based learning algorithms. Machine Learning, 6:37–66. ↗
  2. Tim Buckwalter. 2004. Arabic morphological analyzer version 2.0. Linguistic Data Consortium. find
  3. Walter Daelemans and Antal van den Bosch. 2005. Memory Based Language Processing. Cambridge University Press, Cambridge, UK. ↗
  4. Walter Daelemans, Jakub Zavrel, Peter Berck, and Steven Gillis. 1996. MBT: A memory-based part of speech tagger-generator. In Proceedings of the 4th Workshop on Very Large Corpora, pages 14–27, Copenhagen, Denmark.
  5. Walter Daelemans, Antal van den Bosch, and Jakub Zavrel. 1999. Forgetting exceptions is harmful in language learning. Machine Learning, 34(1-3):11–43. Special Issue on Natural Language Learning. ↗
  6. Mona Diab, Kadri Hacioglu, and Daniel Jurafsky. 2004. Automatic tagging of Arabic text: From raw text to base phrase chunks. In Proceedings of the 5th Meeting of the North American Chapter of the Association for Computational Linguistics and Human Language Technologies Conference (HLT-NAACL), pages 149–152, Boston, MA.
  7. Mona Diab, Kadri Hacioglu, and Dan Jurafsky. 2007. Automatic processing of Modern Standard Arabic text. Arabic Computational Morphology, pages 159–179. ↗
  8. Mona Diab. 2007. Towards an optimal POS tag set for Modern Standard Arabic processing. In Proceedings of Recent Advances in Natural Language Processing (RANLP). find
  9. Mona Diab. 2009. Second generation AMIRA tools for Arabic processing: Fast and robust tokenization, POS tagging, and base phrase chunking. In 2nd International Conference on Arabic Language Resources and Tools, Cairo, Egypt. find
  10. Nizar Habash and Owen Rambow. 2005. Arabic tokenization, part-of-speech tagging and morphological disambiguation in one fell swoop. In Proceedings of the 43rd Annual Meeting of the Association for Computational Linguistics (ACL), pages 573—580, Ann Arbor, MI.
  11. Nizar Habash and Ryan Roth. 2009. CATiB: The Columbia Arabic Treebank. In Proceedings of the ACL-IJCNLP 2009 Conference Short Papers, pages 221–224, Suntec, Singapore. ↗
  12. Nizar Habash, Owen Rambow, and Ryan Roth. 2009. MADA+TOKAN: A toolkit for Arabic tokenization, diacritization, morphological disambiguation, POS tagging, stemming and lemmatization. In Proceedings of the Second International Conference on Arabic Language Resources and Tools, pages 102–109, Cairo, Egypt. find
  13. Jan Hajič. 2000. Morphological tagging: Data vs. dictionaries. In Proceedings of the1st conference of the North American Chapter of the Association for Computational Linguistics, pages 94–101.
  14. Sandra Kübler and Emad Mohamed. 2012. Part of speech tagging for Arabic. Natural Language Engineering, 18(4):521–548. ↗
  15. Mohamed Maamouri, Anne Bies, Tim Buckwalter, and W. Mekki. 2004. The Penn Arabic Treebank: Building a large-scale annotated Arabic corpus. In NEMLAR Conference on Arabic Language Resources and Tools, pages 102–109. find
  16. Emad Mohamed and Sandra Kübler. 2010. Is arabic part of speech tagging feasible without word segmentation? In Human Language Technologies: The 2010 Annual Conference of the North American Chapter of the Association for Computational Linguistics, pages 705–708. Association for Computational Linguistics. ↗
  17. Lance A. Ramshaw and Mitchell P. Marcus. 1995. Text chunking using transformation-based learning. In Proceedings of the ACL 3rd Workshop on Very Large Corpora, pages 82–94, Cambridge, MA.
  18. Ryan Roth, Owen Rambow, Nizar Habash, Mona Diab, and Cynthia Rudin. 2008. Arabic morphological tagging, diacritization, and lemmatization using lexeme models and feature ranking. In Proceedings of ACL-08: HLT, Short Papers, pages 117–120, Columbus, Ohio.
  19. Beatrice Santorini. 1990. Part-of-speech tagging guidelines for the Penn Treebank Project. Department of Computer and Information Science, University of Pennsylvania, 3rd Revision, 2nd Printing. find
  20. Reut Tsarfaty, Djamé Seddah, Yoav Goldberg, Sandra Kübler, Marie Candito, Jennifer Foster, Yannick Versley, Ines Rehbein, and Lamia Tounsi. 2010. Statistical parsing of morphologically rich languages (SPMRL): What, how and whither. In Proceedings of the NAACL Workshop on Statistical Parsing of Morphologically Rich Languages, Los Angeles, CA.

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/mageed-2013-asma-system-automatic-segmentation-morpho/