Elamur Research⌕
☰
home / papers / 2015 / Joint Arabic Segmentation and Part-Of-Speech Tagging

Joint Arabic Segmentation and Part-Of-Speech Tagging

Paper · 2015 · Shabib AlGahtani and John McNaught

Summary

The paper addresses two tightly linked problems in Modern Standard Arabic: splitting words into segments (clitics such as conjunctions, prepositions and pronouns attached to a stem) and assigning part-of-speech tags to those segments. Pipelines usually segment first and tag afterwards, so segmentation mistakes carry over into tagging. The authors avoid this by treating the character as the unit of analysis. Each character receives one combined label: a position marker (IOB or IOBES) fused with a POS tag from the Arabic Treebank collapsed tagset. A maximum entropy model, decoded with beam search, predicts these labels, and segment tags are then rebuilt from the character sequence. The method uses no lexicon or morphological analyzer, which removes dictionary coverage as a limiting factor and should make it more tolerant of unknown or misspelled words. On ATB parts 1–3, segmentation reaches about 99.4% and tagging roughly 93–95.5%. The authors present this as state-of-the-art for a lexicon-free tagger.

Key points

  • Joint labels are built as 4 position subcategories × 27 collapsed ATB tags (e.g. S-NN, B-NN, I-NN, E-NN), plus a special label for the space between words.
  • Decoding uses beam search of size 5, which also discards illegal label orders such as an Inside label directly after an End label. Segment boundaries are recovered wherever a non-initial character carries B or S.
  • Segmentation accuracy on the unsegmented corpus is 99.4%. With one model trained on all parts, POS accuracy is 95.45 (ATB1), 95.09 (ATB2) and 93.44 (ATB3), higher than models trained separately on each part (94.37 / 93.75 / 93.03).
  • IOBES beats IOB by about 0.2% in tagging. Removing word-level lexical features lowers tagging by 1.29% but segmentation by only 0.23%.
  • Per-class error rates normalized by frequency show NNP, JJ and VBD as the hardest classes. A direct comparison with MADAMIRA was not possible because it uses a different tagset and only part of the test set.

Interesting ideas

  • I1. Fusing a boundary marker with the grammatical tag in a single character label turns segmentation and tagging into one decision, so one stage cannot pass its errors to the next.
  • I2. Constraining the beam with a grammar of admissible label transitions guarantees well-formed output and is cheaper than repairing broken sequences afterwards.
  • I3. Dividing a class's error count by how often that class occurs exposes truly hard classes, which raw error counts hide because frequent tags dominate them.
  • I4. Ablating whole feature families (here, word-identity features) shows which subtask depends on lexical memory and which relies on internal character patterns.
  • I5. Comparing a model trained on the pooled corpus with models trained per subcorpus, on each subcorpus's test split, measures how much extra cross-source data helps.

For Mova

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

  • I1 (tokens): Contractions and clitic-like pieces in English ("can't", "John's", hyphenated compounds) could be split and tagged in one character-level pass inside crate en, with each split point justified by the tag it implies.
  • I2 (tokens): The tagger's decoder could reject impossible label successions outright, and each rejection could be logged as an explicit, explainable constraint.
  • I3 (learn): Evaluation reports could rank tags, dependency relations or frame roles by relative error, which would point repair-rule induction at the categories that are genuinely weak.
  • I4 (learn): Running similar ablations as negative controls would reveal whether a component's success comes from memorized words or from structure that generalizes.
  • I5 (tree): Treebank mixtures for the perceptron parser and gparse could be checked the same way, to decide whether to merge UD treebanks or keep them separate.

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

🎓University of ManchesterManchester · United Kingdom87 works · 44,938 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 worksPart-of-speech tagging · 126 works

Cite

Cite as Shabib AlGahtani and John McNaught. 2015. Joint Arabic Segmentation and Part-Of-Speech Tagging. Proceedings of the Second Workshop on Arabic Natural Language Processing.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/w15-3212

@inproceedings{algahtani2015joint,
  title = {{Joint Arabic Segmentation and Part-Of-Speech Tagging}},
  author = {Shabib AlGahtani and John McNaught},
  booktitle = {Proceedings of the Second Workshop on Arabic Natural Language Processing},
  year = {2015},
  doi = {10.18653/v1/w15-3212},
  url = {https://doi.org/10.18653/v1/w15-3212}
}

References

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

  1. AlGahtani, S., W. Black, and J. McNaught. 2009. ‘Arabic Part-of-Speech Tagging Using Transfor-mat ion-Based Learning’. In Proceedings of the Second International Conference on Arabic Language Resources and Tools. Cairo, Egypt.: The MEDAR Consortiu m. <http://www.elda.org/ medar-conference/pdf/43.pdf>. ↗
  2. AlGahtani, S. 2012. ‘Arabic Named Entity Recognition: A Corpus-Based Study’. University of Manchester. <http://www.manchester.ac.uk/escholar/uk-ac-man-scw:158690>. ↗
  3. Beesley, K. 2001. ‘Finite -State Morphological Analysis and Generation of Arabic at Xero x Research: Status and Plans in 2001’. In ACL Workshop on Arabic Language Processing: Status and Perspective, 1:1–8. Toulouse, France. <http://www.xrce.xero x.co m/Research-Develop ment/Publications/2001-0094/%28language%29/eng-GB>. ↗
  4. Diab, M. 2009. ‘Second Generation AMIRA Tools for Arabic Processing: Fast and Robust Tokenization, POS Tagging, and Base Phrase Chunking’. In Proceedings of the MEDAR’09 Conference. Cairo, Egypt. <www.elda.org/ medar-conference/pdf/56.pdf>. find
  5. Diab, M., K. Hacioglu, and D. Ju rafsky. 2004. ‘Automatic Tagging of Arabic Text: Fro m Raw Text to Base Phrase Chunks’. In Proceedings of HLT-NAACL 2004: Short Papers, 149–152. <http://dl.acm.org/citation.cfm?id=1614022> ↗
  6. Diab, Mona, Nizar Habash, Owen Rambow, and Ryan Roth. 2013. ‘LDC Arab ic Treebanks and Associated Corpora: Data Divisions Manual’. arXiv Preprint arXiv:1309.5652. <http://arxiv.org/abs/1309.5652>. ↗
  7. Habash, N., and O. Rambow. 2005. ‘Arabic Tokenization, Part-of-Speech Tagging and Morphological Disambiguation in One Fell Swoop’. In Proceedings of the 43rd Annual Meeting on Association for Computational Linguistics, 573– 580doi:10.3115/1219840.1219911. . <http://dl.acm.org/citation.cfm?id=1219911> ↗
  8. Habash, N., O. Rambow, and R. Roth. 2009. ‘Mada+ Tokan: A Toolkit for Arabic To kenization, Diacrit ization, Morphological Disamb iguation, Pos Tagging, Stemming and Lemmatization’. In Proceedings of the 2nd International Conference on Arabic Language Resources and Tools (MEDAR) . Cairo, Egypt. <www.elda.org/ medar-conference/pdf/24.pdf>. find
  9. Bar-Haim, R., K. Sima’an, and Y. W inter. 2005. ‘Choosing an Optimal Architecture for Seg mentation and POS-Tagging of Modern Hebrew’. In Proceedings of the ACL Workshop on Computational Approaches to Semitic Languages, 39– 46. <http://dl.acm.org/citation.cfm?id=1621796> ↗
  10. Khoja, S. 2001. ‘APT: Arab ic Part-of-Speech Tagger’. In Proceedings of the Student Workshop at NAACL-2001, 20–25. <https://nats-www.informat ik.uni-hamburg.de/intern/proceedings/2001/naacl/srw/pdf/srw -06.pdf>. ↗
  11. Kiraz, G. 2002. ‘Co mputational Nonlinear Morphology: With Emphasis on Semitic Languages’. Computational Linguistics 28 (4): 576– 58doi:10.1162/coli.2002.28.4.576. . <http://dx.doi.o rg/10.1162/coli.2002.28.4.576>. ↗
  12. Kudo, T., and Y. Matsumoto. 2001. ‘Chunking with Support Vector Machines’. In Proceedings of the Second Meeting of the North American Chapter of the Association for Computational Linguistics on Language Technologies, 1–8. <http://dx.doi.org/10.3115/ 1073336.1073361>. ↗
  13. Maamouri, M., A. Bies, and S. Kulick. 2006. ‘Diacritization: A Challenge to Arabic Treebank Annotation and Parsing’. In Proceedings of the Conference of the Machine Translation SIG of the British Computer Society. <http://papers.ldc.upenn.edu/NLTSG/Diacrit izat io nArabicTreebank.pdf>. ↗
  14. Mansour, S., K. Sima’an, and Y. Winter. 2007. ‘Smoothing a Lexicon-Based Pos Tagger for Arabic and Hebrew’. In Proceedings of the 2007 Workshop on Computational Approaches to Semitic Languages: Common Issues and Resources, 97– 103. <http://dl.acm.org/ft_gateway.cfm?id=1654593&ty pe=pdf&CFID=95957381&CFTOKEN=73775663 >. ↗
  15. Pasha, A, M. Al-Badrashiny, M. Diab, A. El Kholy, R. Eskander, N. Habash, M. Pooleery, O. Rambow, and R. Roth. 2014. ‘Madamira: A Fast, Co mpre-hensive Tool for Morphological Analysis and Disamb iguation of Arabic’. In Proceedings of the Language Resources and Evaluation Conference (LREC), Reykjavik, Iceland. <http://www.lrec-conf.org/proceedings/lrec2014/pdf/593_Paper.pdf> ↗
  16. Qian, X, and Y. Liu. 2012. ‘Joint Chinese Word Segmentation, POS Tagging and Parsing’. In Proceedings of the 2012 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning, 501– 511. <http://dl.acm.org/citation.cfm?id=2391007>. ↗
  17. Sawalha, M., and E. S. Atwell. 2010. ‘Fine-Grain Morphological Analyzer and Part-of-Speech Tagger for Arab ic Text ’. In Proceedings of the Seventh Conference on International Language Resources and Evaluation ((LREC’10)), 1258–1265. Valleta, Malta. <http://www.lrec-conf.org/proceedings/lrec2010/pdf/282_Paper.pdf> ↗
  18. Shaalan, K., A. Allam, and A. Go mah. 2003. ‘Towards Automatic Spell Checking fo r Arabic’. In Proceedings of the 4th Conference on Language Engineering ((ELSE ’03)), 240– 247. Cairo, Egypt: Egyptian Society of Language Engineering. <http://www.claes.sci.eg/NARIMS_upload/CLAE SFILES/3847.pdf>. ↗
  19. Shen, L., G. Satta, and A. Joshi. 2007. ‘Gu ided Learning for Bidirectional Sequence Classificat ion’. In Proceedings of the 45th Annual Meeting of the Association of Computational Linguistics, 45:760– 767. Prague, Czech Republic. <http://acl.ldc.upenn.edu/P/P07/P07-1096.pdf>. ↗
  20. Toutanova, K., D. Klein, C.D. Manning, and Y. Singer. 2003. ‘Feature-Rich Part-of-Speech Tagging with a Cyclic Dependency Network’. In Proceedings of the 2003 Conference of the North American Chapter of the Association for Computational Linguistics on Human Language Technology (NAACL ’03), 1:173– 180. Ed monton, Canada. <http://dx.doi.org/10.3115/ 1073445.1073478>. ↗
  21. Tsuruoka, Y., Y. Tateishi, J.D. Kim, T. Ohta, J. McNaught, S. Ananiadou, and J. Tsujii. 2005. ‘Developing a Robust Part-of-Speech Tagger for Biomed ical Text’. Advances in Informatics 3746. Lecture Notes in Co mputer Science: 382– 392. <http://dx.doi.org/10.1007/ 11573036_ 36>. ↗

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/algahtani-2015-joint-arabic-segmentation-part-speech/