Memory-Based Morphological Analysis Generation and Part-of-Speech Tagging of Arabic

What it is about

The paper applies memory-based learning (k-nearest neighbours, TiMBL) to two tasks on unvowelised Arabic from the Arabic Treebank 1: generating all vocalised analyses of a word, and tagging the part of speech. Analysis is cast as one classification per letter. Each class encodes the vowels and tags that surround the letter, and a word's analyses are rebuilt by joining the pieces, as in a jigsaw. On 10-fold cross-validation the analyzer reaches precision 53.4%, recall 62.2% and F 57.5 overall, and 32.2 F on unknown words. Overgeneration is the main error. A tagger with a known-word and an unknown-word module reaches 91.9% accuracy (93.1% on known, 73.6% on unknown words). Selecting analyses by the predicted tag gives recall 89.7 and precision 64.0 on all test words, against 97.8 recall with gold tags. The authors state that the tagger's accuracy is the limit of the combination. They also tuned two filters (an n-best cut and a tag-trigram frequency threshold) on the same test fold they report on.

Ideas

• I1. Vocalisation as a table of letter operations: store, for each letter of an unvowelised word, the vowels and tags to its left and right, and join the pieces. This is a finite table, not a model, and every output can be traced back to the pieces that built it. It fits the segmenter tables of tools/hbo (the pre and lex tables) and the unpointed mode.
• I2. Filter overgenerated candidates by two counts: the summed frequency of the table entries used, and a tag-trigram threshold from the training data. Both are plain counts. Tune the cut on a development fold, never on the test fold the paper reports on, which is the flaw in this paper, and keep the held-out part for the once-per-release test.
• I3. Report the upper bound and the real run side by side: once with gold tags, once with predicted tags. The gap (97.8 vs 89.7 recall) tells how much a tagger error costs the selector, and shows where to improve. Do this for tools/hbo before adding any learned tagger.

For Mova

I1 and I2 are candidate additions to tools/hbo (the candidate and filter stage for unpointed text) and are measured against the MACULA-derived Hebrew data. I3 is the reporting pattern for any tagger step there. Arabic is not a Mova language, so the Arabic numbers are context only; the method transfers as a table-and-count design. Relevant to The verb only as an example of where analysis overgenerates.

Topics: Part-of-speech tagging

Cite

The paper itself, not this card. BibTeX from the library's record of the work. DOI 10.3115/1621787.1621789

@inproceedings{marsi2005memory,
  title = {{Memory-Based Morphological Analysis Generation and Part-of-Speech Tagging of Arabic}},
  author = {Marsi, Erwin and Bosch, Antal van den and Soudi, Abdelhadi},
  booktitle = {Proceedings of the ACL Workshop on Computational Approaches to Semitic Languages},
  year = {2005},
  doi = {10.3115/1621787.1621789},
  url = {https://doi.org/10.3115/1621787.1621789}
}

References

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

1. D. W. Aha, D. Kibler, and M. Albert. 1991. Instance-based learning algorithms. Machine Learning, 6:37–66. find our card
2. K. Beesley. 1990. Finite-state description of Arabic morphology. In Proceedings of the Second Cambridge Conference: Bilingual Computing in Arabic and English. find
3. K. Beesley. 1998. Consonant spreading in Arabic stems. In Proceedings of COLING-98. find
4. T. Buckwalter. 2002. Buckwalter Arabic morphological analyzer version 1.0. Technical Report LDC2002L49, Linguistic Data Consortium. available from http://www.ldc.upenn.edu/. ↗
5. V. Cavalli-Sfora, A. Soudi, and M. Teruko. 2000. Arabic morphology generation using a concatenative strategy. In Proceedings of the First Conference of the North-American Chapter of the Association for Computational Linguistics, Seattle, WA, USA. find
6. T. M. Cover and P. E. Hart. 1967. Nearest neighbor pattern classification. Institute of Electrical and Electronics Engineers Transactions on Information Theory, 13:21–27. find
7. W. Daelemans, J. Zavrel, P. Berck, and S. Gillis. 1996. MBT : A memory-based part of speech tagger generator. In E. Ejerhed and I. Dagan, editors, Proceedings of Fourth Workshop on Very Large Corpora, pages 14–27. ACL SIGDAT. find
8. W. Daelemans, A. van den Bosch, and J. Zavrel. 1999. Forgetting exceptions is harmful in language learning. Machine Learning, Special issue on Natural Language Learning, 34:11–41. find
9. W. Daelemans, J. Zavrel, A. van den Bosch, and K. van der Sloot. 2003. MBT: Memory based tagger, version 2.0, reference guide. ILK Technical Report 03-13, Tilburg University. find
10. W. Daelemans, J. Zavrel, K. van der Sloot, and A. van den Bosch. 2004. TiMBL: Tilburg memory based learner, version 5.1, reference guide. ILK Technical Report 04-02, Tilburg University. find
11. E. Daya, D. Roth, and S. Wintner. 2004. Learning Hebrew roots: Machine learning with linguistic constraints. In Proceedings of EMNLP’04, Barcelona, Spain. find our card
12. M. Diab, K. Hacioglu, and D. Jurafsky. 2004. Automatic tagging of arabic text: From raw text to base phrase chunks. In Proceedings of HLT/NAACL-2004. find
13. A. Freeman. 2001. Brill’s POS tagger and a morphology parser for Arabic. In ACL/EACL-2001 Workshop on Arabic Language Processing: Status and Prospects, Toulouse, France. find
14. M. Kay. 1987. Non-concatenative finite-state morphology. In Proceedings of the third Conference of the European Chapter of the Association for Computational Linguistics, pages 2–10, Copenhagen, Denmark. find
15. S. Khoja. 2001. APT: Arabic part-of-speech tagger. In Proceedings of the Student Workshop at NAACL-2001. find
16. G. Kiraz. 1994. Multi-tape two-level morphology: A case study in semitic non-linear morphology. In Proceedings of COLING’94, volume 1, pages 180–186. find our card
17. J. McCarthy. 1981. A prosodic theory of non-concatenative morphology. Linguistic Inquiry, 12:373–418. find
18. A. Soudi. 2002. A Computational Lexeme-based Treatment of Arabic Morphology. Ph.D. thesis, Mohamed V University (Morocco) and Carnegie Mellon University (USA). find
19. A. van den Bosch and W. Daelemans. 1999. Memory-based morphological analysis. In Proceedings of the 37th Annual Meeting of the ACL, pages 285–292, San Francisco, CA. Morgan Kaufmann. find
20. C.J. van Rijsbergen. 1979. Information Retrieval. Buttersworth, London. find

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/aclbulk-2005-marsi-memory-based-morphological-analysis-generation/