Elamur Research⌕
☰
home / papers / 2015 / CBAS: context based arabic stemmer

CBAS: context based arabic stemmer

Summary

Arabic words are built from roots, usually three to five letters, by applying patterns and adding prefixes or suffixes. Recovering the root (stemming) therefore matters for many Arabic NLP tasks. The problem is ambiguity. Letter mutation, vocalization, missing diacritics and stopwords that ignore derivation rules mean that one surface word can lead back to several candidate roots.

Existing stemmers resolve this in limited ways:

  • Khoja, ISRI and Tashaphyne use fixed, ordered rules.
  • Darwish's approach ranks roots by how often the inferred patterns occur, without looking at the word's surroundings.

CBAS splits the job in two. First it lists every plausible root for a word. Then it chooses among them with distributional semantics: a word–context co-occurrence matrix, weighted with Smoothed Pointwise Mutual Information (SPMI), shows which candidate best matches the context the word appears in.

The authors compared CBAS with other stemmers on a manually annotated set of articles. CBAS reached 81.5% accuracy, at least 9.4% better than the competing systems. The result shows that a word's context can usefully decide between competing morphological analyses.

Key points

  • Candidate roots come from aligning a word to root-pattern templates such as faʿl/fāʿl; prefixes like "al-" are treated as separable additions.
  • The final root is chosen with a distributional semantics model built on a context matrix. The paper discusses three association scores (PMI, PPMI and SPMI) and uses SPMI.
  • Baselines described:
    • Khoja: rule-ordered, and validates roots against a root list.
    • ISRI: no root dictionary, with patterns grouped into sub-groups.
    • Tashaphyne: mainly light stemming.
    • Darwish: learned patterns, with roots ranked by pattern frequency.
  • The paper also contrasts light stemming (stripping only prefixes and suffixes) with full root extraction. It notes that light stemming is competitive in IR, but there is no clear evidence that it beats roots.
  • Evaluation on manually annotated articles gives 81.5% accuracy, at least 9.4% above the other stemmers.

Interesting ideas

  • I1. Separate candidate generation from the decision: list every morphologically valid analysis first, then choose with contextual evidence instead of a fixed rule order. Errors can then be traced to either the generator or the chooser. — For Mova (tokens): English lemma ambiguities such as "left" (leave/left), "found" (find/found) or "saw" (see/saw) could keep every candidate and resolve it with context scores, logging which neighbouring words decided.
  • I2. Use an association score (PMI-family, here a smoothed PMI) between a candidate and its context words, not raw co-occurrence counts, to rank analyses. — For Mova (frames): when a verb fits several VerbNet classes, the class whose typical arguments associate most strongly with the actual arguments could be picked, and the winning association reported as the explanation.
  • I3. Check generated candidates against a closed inventory of valid forms, as Khoja does with its root list, so impossible outputs are rejected before ranking. — For Mova (world): lemma or concept candidates could be filtered against WordNet/Wikidata entries before context ranking, so only real entries reach the knowledge graph.
  • I4. Contrast static, order-dependent rule systems with context-sensitive selection on the same hand-annotated gold set, so the gain from context is measured directly. — For Mova (learn): build a small gold set of ambiguous lemma and sense cases and score rule-only against context-ranked versions of a component, as a targeted ablation.

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: Morphology · Semitic languages

Institutions

🎓Arab Academy for Science, Technology, and Maritime TransportAlexandria · Egypt2 works · 17 citations

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.5121/ijnlc.2015.4301 · arXiv 1611.00027

@misc{eldefrawy2015cbas,
  title = {{CBAS: context based arabic stemmer}},
  author = {El-Defrawy, Mahmoud and El-Sonbaty, Yasser and Belal, Nahla A.},
  year = {2015},
  doi = {10.5121/ijnlc.2015.4301},
  eprint = {1611.00027},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.5121/ijnlc.2015.4301}
}

References

40 references, 3 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. P. M. Nadkarni, L. Ohno-Machado, and W. W. Chapman, "Natural language processing: an introduction," Journal of the American Medical Informatics Association, vol. 18, pp. 544-551, 2011. find
  2. J. Hutchins, "The first public demonstration of machine translation: the Georgetown-IBM system, 7th January 1954," noviembre de, 2005. find
  3. N. Chomsky, "Three models for the description of language," Information Theory, IRE Transactions on, vol. 2, pp. 113-124, 1956. find
  4. A. Aho, "R. Sethi, and J. D. Ullman," Compilers: Principles, Techniques, and Tools, 1988. find
  5. D. Klein and C. D. Manning, "Accurate unlexicalized parsing," in Proceedings of the 41st Annual Meeting on Association for Computational Linguistics-Volume 1, 2003, pp. 423-430. find our card
  6. M. Aljlayl and O. Frieder, "On Arabic search: improving the retrieval effectiveness via a light stemming approach," in Proceedings of the eleventh international conference on Information and knowledge management, 2002, pp. 340-347. find
  7. I. A. Al‐Sughaiyer and I. A. Al‐Kharashi, "Arabic morphological analysis techniques: A comprehensive survey," Journal of the American Society for Information Science and Technology, vol. 55, pp. 189-213, 2004. find
  8. M. F. Porter, "Snowball: A language for stemming algorithms," ed, 2001. find
  9. J. Xu, A. Fraser, and R. Weischedel, "Empirical studies in strategies for Arabic retrieval," in Proceedings of the 25th annual international ACM SIGIR conference on Research and development in information retrieval, 2002, pp. 269-274. find
  10. R. Fathalla, Y. El Sonbaty, and M. A. Ismail, "Extraction of Arabic Words form Complex Color Images," in 9th IEEE International Conference on Document Analysis and Recognition (ICDAR 2007), Brazil, pp. 1223-1227. find
  11. C. Akkaya, J. Wiebe, and R. Mihalcea, "Utilizing semantic composition in distributional semantic models for word sense discrimination and word sense disambiguation," in Semantic Computing (ICSC), 2012 IEEE Sixth International Conference on, 2012, pp. 45-51. find
  12. G. Bouma, "Normalized (pointwise) mutual information in collocation extraction," Proceedings of GSCL, pp. 31-40, 2009. find
  13. K. C. Ryding, A reference grammar of modern standard Arabic: Cambridge university press, 2005. find
  14. A. Pasha, M. Al-Badrashiny, M. Diab, A. El Kholy, R. Eskander, N. Habash, et al., "Madamira: A fast, comprehensive tool for morphological analysis and disambiguation of arabic," in Proceedings of the Language Resources and Evaluation Conference (LREC), Reykjavik, Iceland, 2014. find our card
  15. S. M. Oraby, Y. El-Sonbaty, and M. A. El-Nasr, "Exploring the Effects of Word Roots for Arabic Sentiment Analysis," in International Joint Conference on Natural Language Processing, Nagoya, Japan, 2013, pp. 471-479. find
  16. J. B. Lovins, Development of a stemming algorithm: MIT Information Processing Group, Electronic Systems Laboratory, 1968. find
  17. S. Khoja and R. Garside, "Stemming arabic text," Lancaster, UK, Computing Department, Lancaster University, 1999. find
  18. M. S. Husain, "An unsupervised approach to develop stemmer," International Journal on Natural Language Computing, vol. 1, pp. 15-23, 2012. find
  19. D. Harman, "How effective is suffixing?," JASIS, vol. 42, pp. 7-15, 1991. find
  20. I. Smirnov, "Overview of stemming algorithms," Mechanical Translation, vol. 52, 2008. find
  21. Y. Benajiba, M. Diab, and P. Rosso, "Arabic named entity recognition using optimized feature sets," in Proceedings of the Conference on Empirical Methods in Natural Language Processing, 2008, pp. 284-293. find
  22. K. Darwish and D. W. Oard, "CLIR Experiments at Maryland for TREC-2002: Evidence combination for Arabic-English retrieval," DTIC Document2003. find
  23. L. S. Larkey and M. E. Connell, "Arabic information retrieval at UMass in TREC-10," DTIC Document2006. find
  24. L. S. Larkey, L. Ballesteros, and M. E. Connell, "Light stemming for Arabic information retrieval," in Arabic computational morphology, ed: Springer, 2007, pp. 221-243. find
  25. L. S. Larkey, L. Ballesteros, and M. E. Connell, "Improving stemming for Arabic information retrieval: light stemming and co-occurrence analysis," in Proceedings of the 25th annual international ACM SIGIR conference on Research and development in information retrieval, 2002, pp. 275-282. find
  26. K. Taghva, R. Elkhoury, and J. Coombs, "Arabic stemming without a root dictionary," in null, 2005, pp. 152-157. find
  27. T. Zerrouki. (2010). Tashaphyne, Arabic light stemmer/segment. find
  28. K. Darwish, "Building a shallow Arabic morphological analyzer in one day," in Proceedings of the ACL-02 workshop on Computational approaches to semitic languages, 2002, pp. 1-8. find
  29. K. R. Beesley, "Arabic morphological analysis on the Internet," in Proceedings of the 6th International Conference and Exhibition on Multi-lingual Computing, 1998. find
  30. O. Smrž, "Elixirfm: implementation of functional arabic morphology," in Proceedings of the 2007 Workshop on Computational Approaches to Semitic Languages: Common Issues and Resources, 2007, pp. 1-8. find our card
  31. T. Buckwalter, "Buckwalter {Arabic} Morphological Analyzer Version 1.0," 2002. find
  32. N. Habash, O. Rambow, and R. Roth, "MADA+ TOKAN: A toolkit for Arabic tokenization, diacritization, morphological disambiguation, POS tagging, stemming and lemmatization," in Proceedings of the 2nd International Conference on Arabic Language Resources and Tools (MEDAR), Cairo, Egypt, 2009, pp. 102-109. find
  33. M. Diab, K. Hacioglu, and D. Jurafsky, "Automated methods for processing arabic text: From tokenization to base phrase chunking," Arabic Computational Morphology: Knowledge-based and Empirical Methods. Kluwer/Springer, 2007. find
  34. S. N. Saleh and Y. El-Sonbaty, "A feature selection algorithm with redundancy reduction for text classification," in Computer and information sciences, 2007. iscis 2007. 22nd international symposium on, 2007, pp. 1-6. find
  35. S. Oraby, Y. El-Sonbaty, and M. A. El-Nasr, "Finding Opinion Strength Using Rule-Based Parsing for Arabic Sentiment Analysis," in Advances in Soft Computing and Its Applications, ed: Springer, 2013, pp. 509-520. find
  36. A. M. Ezzeldin, M. H. Kholief, and Y. El-Sonbaty, "ALQASIM: Arabic language question answer selection in machines," in Information Access Evaluation. Multilinguality, Multimodality, and Visualization, ed: Springer, 2013, pp. 100-103. find
  37. A. M. Ezzeldin, Y. El-Sonbaty, and M. H. Kholief, "Exploring the Effects of Root Expansion, Sentence Splitting and Ontology on Arabic Answer Selection," Natural Language Processing and Cognitive Science: Proceedings 2014, p. 273, 2015. find
  38. M. Abbas, K. Smaïli, and D. Berkani, "Evaluation of Topic Identification Methods on Arabic Corpora," JDIM, vol. 9, pp. 185-192, 2011. find
  39. S. Alansary, M. Nagi, and N. Adly, "Building an International Corpus of Arabic (ICA): progress of compilation stage," in 7th international conference on language engineering, Cairo, Egypt, 2007, pp. 5-6. find
  40. S. Godbole and S. Sarawagi, "Discriminative methods for multi-labeled classification," in Advances in Knowledge Discovery and Data Mining, ed: Springer, 2004, pp. 22-30. find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/el-2015-defrawy-cbas-context-based-arabic-stemmer/