Learning Hebrew Roots: Machine Learning with Linguistic Constraints

What it is about

The root of a Hebrew word (usually three consonants) is hard to find, because vowels are unwritten, particles attach without spaces, and weak roots change their shape. The authors train one SNoW classifier per root radical, on letter-position, letter-bigram, prefix and suffix features. Combining the three classifiers directly gives an F-measure of 58.89 on the sequential model. Adding linguistic scoring (a regular root's consonants occur in order, with at most one w or y between two radicals, the suffix must be a legal one, and the root must be in a list) and keeping all candidates within a fixed drop δ = 0.4 of the top score gives F = 84.38 on held-out data (precision 80.90, recall 88.16). Six human subjects reached F = 81.86 on 200 words without context. Regular roots score 94.81 and irregular ones 66.06. The annotated data are 5,242 word types from a newspaper corpus, with 1,043 roots.

Ideas

• I1. A rule-only root candidate filter: list the roots from the root table (2,152 roots in the union of the two sources the paper uses) whose radicals occur in order with only w or y allowed between them, and rank them with the paper's three-class score. It needs no learning, and it would give macula dict the root column it lacks.
• I2. Keep every candidate within a fixed drop of the top score, rather than a fixed top-n. The threshold is deterministic, and it lets a word with one clear root return one root. The paper's own table shows why a fixed top-n loses precision.
• I3. Report F-measure separately for regular and irregular roots, and for the weak paradigms (middle radical w or y, and identical final radicals). Use the paper's human baseline as the reference line. The weak paradigms belong in the test for The weak verbs.

For Mova

I1 goes into macula dict (roots) and is checked against The weak verbs. I2 is a decision rule for root output in tools/hbo. I3 is the evaluation plan for both.

Cite

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

@inproceedings{daya2004learning,
  title = {{Learning Hebrew Roots: Machine Learning with Linguistic Constraints}},
  author = {Daya, Ezra and Roth, Dan and Wintner, Shuly},
  booktitle = {Proceedings of the 2004 Conference on Empirical Methods in Natural Language Processing},
  year = {2004},
  url = {https://aclanthology.org/W04-3246/}
}

References

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

1. Ken Beesley. 1998a. Arabic morphological analysis on the internet. In Proceedings of the 6th International Conference and Exhibition on Multilingual Computing, Cambridge, April. find
2. Kenneth R. Beesley. 1998b. Arabic morphology using only finite-state operations. In Michael Rosner, editor, Proceedings of the Workshop on Computational Approaches to Semitic languages, pages 50–57, Montreal, Quebec, August. COLING-ACL’98. find our card
3. Yaacov Choueka. 1990. MLIM - a system for full, exact, on-line grammatical analysis of Modern Hebrew. In Yehuda Eizenberg, editor, Proceedings of the Annual Conference on Computers in Education, page 63, Tel Aviv, April. In Hebrew. find
4. Kareem Darwish. 2002. Building a shallow Arabic morphological analyzer in one day. In Mike Rosner and Shuly Wintner, editors, Computational Approaches to Semitic Languages, an ACL’02 Workshop, pages 47–54, Philadelphia, PA, July. find
5. Abraham Even-Shoshan. 1993. HaMillon HaX-adash (The New Dictionary). Kiryat Sefer, Jerusalem. In Hebrew. find
6. Y. Even-Zohar and Dan Roth. 2001. A sequential model for multi class classification. In EMNLP-2001, the SIGDAT Conference on Empirical Methods in Natural Language Processing, pages 10–19. find
7. Radu Florian. 2002. Named entity recognition as a house of cards: Classifier stacking. In Proceedings of CoNLL-2002, pages 175–178. Taiwan. find
8. Joseph Kruskal. 1999. An overview of sequence comparison. In David Sankoff and Joseph Kruskal, editors, Time Warps, String Edits and Macromolecules: The Theory and Practice of Sequence Comparison, pages 1–44. CSLI Publications, Stanford, CA. Reprint, with a foreword by John Nerbonne. find
9. John J. McCarthy. 1981. A prosodic theory of non-concatenative morphology. Linguistic Inquiry, 12(3):373–418. find
10. Uzzi Ornan. 2003. The Final Word. University of Haifa Press, Haifa, Israel. In Hebrew. find
11. Vasin Punyakanok and Dan Roth. 2001. The use of classifiers in sequential inference. In NIPS-13; The 2000 Conference on Advances in Neural Information Processing Systems 13, pages 995– 1001. MIT Press. find
12. Dan Roth. 1998. Learning to resolve natural language ambiguities: A unified approach. In Proceedings of AAAI-98 and IAAI-98, pages 806– 813, Madison, Wisconsin. find
13. H. Schütze and Y. Singer. 1994. Part-of-speech tagging using a variable memory markov model. In Proceedings of the 32nd Annual Meeting of the Association for Computational Linguistics. find
14. Joseph Shimron, editor. 2003. Language Processing and Acquisition in Languages of Semitic, Root-Based, Morphology. Number 28 in Language Acquisition and Language Disorders. John Benjamins. find
15. Erik F. Tjong Kim Sang and Fien De Meulder. 2003. Introduction to the CoNLL-2003 shared task: Language-independent named entity recognition. In Walter Daelemans and Miles Osborne, editors, Proceedings of CoNLL-2003, pages 142– 147. Edmonton, Canada. find
16. Yizxaq Zdaqa. 1974. Luxot HaPoal (The Verb Tables). Kiryath Sepher, Jerusalem. In Hebrew. find

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/aclbulk-2004-daya-learning-hebrew-roots-machine-learning/