Using Stem-Templates to Improve Arabic POS and Gender/Number Tagging

What it is about

The paper builds an end-to-end Arabic pipeline: correction of common spelling confusions (alef forms, ta marbuta and ha, alef maqsura and ya), character-level segmentation with a CRF, POS tagging with a CRF, and gender and number tagging for nouns and adjectives. The key idea is the stem template, the pattern that a root is put into (for example CCAC applied to ktb gives ktAb, "book"). A template often fixes the POS, and some templates fix the gender or the plural type. Templates were found with the Sebawai analyzer, corrected for the cases it got wrong, and reduced to 91 templates observed in the training data. Adding the template as a feature raised POS accuracy from 97.6% to 98.1% on the Penn Arabic Treebank split. For gender and number, a random forest over template, POS, suffix, feminine ending and dictionary tags gave 95.6% and 94.9% on 8,400 nouns and adjectives in 20-fold cross-validation, where each test word was unseen in training. The spelling correction reached 99% on the Treebank.

Ideas

• I1. Use the stem template as a tagging feature for Hebrew: derive the templates (binyanim, nominal patterns, shapes of the stem) from the roots and patterns in MACULA, keep the most frequent ones as a closed set, and use them as features in the tagger. Measure the POS gain on held-out chapters, as the paper did (+0.5 points).
• I2. Keep only templates seen in the gold data: the paper dropped templates never observed, which cut 613 to 91. For Hebrew, the same whitelist, built from MACULA, gives a short, checkable table of noun and verb patterns that tools/hbo can use as a rule.
• I3. A small, readable classifier for gender and number of unseen nouns: features are the pattern, the stem length, the POS, the suffix (feminine -ah, -ot, dual and plural endings), and a dictionary tag. Test with a split by lemma, so no lemma is seen in both parts.

For Mova

I1 and I2 go into The Noun, the Adjective and the Numeral (noun patterns and the segolate and plural classes) and into the tagger features of tools/hbo. I3 is a test for The Noun, the Adjective and the Numeral; the classifier is a small model with its features written out, in line with the transparency rule. A later step for tools/ocrfix is to use the spelling correction on Hebrew OCR editions.

Cite

The paper itself, not this card. BibTeX from the library's record of the work. DOI 10.63317/5m5e2fcsu8jd

@inproceedings{darwish2014using,
  title = {{Using Stem-Templates to Improve Arabic POS and Gender/Number Tagging}},
  author = {Darwish, Kareem and Abdelali, Ahmed and Mubarak, Hamdy},
  booktitle = {Proceedings of the Ninth International Conference on Language Resources and Evaluation (LREC'14)},
  year = {2014},
  doi = {10.63317/5m5e2fcsu8jd},
  url = {https://doi.org/10.63317/5m5e2fcsu8jd}
}

References

10 references, 1 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Sarah Alkuhlani and Nizar Habash. 2012. Identifying broken plurals, irregular gender, and rationality in Arabic text. Proceedings of the 13th Conference of the European Chapter of the Association for Computational Linguistics. find our card
2. Leo Breiman. 2001. Random Forests. Machine Learning. 45(1):5-32. find
3. Kareem Darwish. 2002. Building a Shallow Morphological Analyzer in One Day. ACL Workshop on Computational Approaches to Semitic Languages. 2002. find
4. Mona T. Diab. 2009. Second Generation Tools (AMIRA 2.0): Fast and Robust Tok- enization, POS tagging, and Base Phrase Chunking. 2nd Int. Conf. on Arabic Language Resources and Tools, 2009. find
5. 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 2nd International Conference on Arabic Language Resources and Tools (MEDAR), Cairo, Egypt, 2009 find
6. John Lafferty, Andrew McCallum, and Fernando Pereira. 2001. Conditional random fields: Probabilistic models for segmenting and labeling sequence data. Proceedings of the 18th International Conference on Machine Learning 2001 (ICML 2001), pp. 282-289. find
7. M Moussa, MW Fakhr, Kareem Darwish. 2012. Statistical Denormalization for Arabic Text. Empirical Methods in Natural Language Processing, 228. find
8. H. Mubarak, K. Shaban, and F. Adel. 2009. Lexical and Morphological Statistics of an Arabic POS-Tagged Corpus. In Proceedings of the 9th Conference on Language Engineering ESOLEC2009, 23-24 December 2009, Cairo, Egypt, pp. 147-161. find
9. Y. Lee, K. Papineni, S. Roukos, O. Emam, H. Hassan. 2003. Language Model. Based Arabic Word Segmentation. In the Proceedings of the 41st Annual Meeting of the Association for Computational Linguistics, July 2003, Sapporo, Japan. p. 399 - 406. find
10. Kristina Toutanova, Dan Klein, Christopher Manning, and Yoram Singer. 2003. Feature-Rich Part-of-Speech Tagging with a Cyclic Dependency Network. In Proceedings of HLT-NAACL 2003, pp. 252-259. find

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/lingbulk-2014-darwish-stem-templates-improve-arabic-pos/