Morphological Disambiguation of Hebrew: A Case Study in Classifier Combination

What it is about

The HAMSAH analyser gives about 2.64 analyses per token on a hand-checked corpus of about 89,000 tokens. The authors train one simple SNoW classifier per attribute (POS, gender, number, person, tense, definiteness, state, prefix length, has-properties) and combine their scores, then apply six hard constraints on pairs of adjacent analyses (for example, a construct-state word must be followed by a nominal phrase). Ten-fold evaluation gives 91.44% accuracy with the hierarchical combination and the constraints, against 90.26% for the simple sum of scores and about 86% for the most-frequent baseline. The error rate falls by 25% against the earlier state of the art. Most remaining errors are in the state feature (19%) and in definiteness of words where the article is assimilated into a preposition (about 16%).

Ideas

• I1. Six hard pair constraints as a finite-state filter on the analysis lattice, applied before any statistical choice: for example, a non-present verb cannot be followed by "shel", and a preposition without a suffix must be followed by a nominal. Test each constraint on the gold-checked data and keep only the ones with zero violations.
• I2. The assimilated definite article (ב, כ, ל before ה) is a large error source: give the analysis lattice an explicit "definite by assimilation" flag and test it as a feature before any learning.
• I3. Order the decisions: first decide the has-properties gate (atomic word or attributed word), then predict only the attributes that apply, and report the change in both accuracy and remaining ambiguity.

For Mova

I1 goes into the disambiguation step of tools/hbo and into Syntax: the clause and the sentence, with the constraint list written as rules. I2 belongs to Pronouns, particles and prepositions (prefixes) and Prepositions, prefixes, conjunctions and particles. I3 is a design choice for the segmenter and tagger of tools/hbo; measure it with the ambiguity and accuracy numbers the paper gives. The paper's numbers are from 2007 and a different corpus, so re-measure before using them.

Cite

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

@inproceedings{shacham2007morphological,
  title = {{Morphological Disambiguation of Hebrew: A Case Study in Classifier Combination}},
  author = {Shacham, Danny and Wintner, Shuly},
  booktitle = {Proceedings of the 2007 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning (EMNLP-CoNLL)},
  year = {2007},
  url = {https://aclanthology.org/D07-1046/}
}

References

14 references, 6 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Meni Adler and Michael Elhadad. 2006. An unsupervised morpheme-based hmm for hebrew morphological disambiguation. In Proceedings of the 21st International Conference on Computational Linguistics and 44th Annual Meeting of the Association for Computational Linguistics, pages 665–672, Sydney, Australia, July. Association for Computational Linguistics. find our card
2. Roy Bar-Haim, Khalil Sima’an, and Yoad Winter. 2005. Choosing an optimal architecture for segmentation and POS-tagging of Modern Hebrew. In Proceedings of the ACL Workshop on Computational Approaches to Semitic Languages, pages 39–46, Ann Arbor, Michigan, June. Association for Computational Linguistics. find our card
3. Yaacov Choueka and Serge Lusignan. 1985. Disambiguation by short context. Computers and the Humanities, 19:147–157. find
4. Ezra Daya, Dan Roth, and Shuly Wintner. 2004. Learning Hebrew roots: Machine learning with linguistic constraints. In Proceedings of EMNLP’04, pages 357– 364, Barcelona, Spain, July. find our card
5. Michael Elhadad, Yael Netzer, David Gabay, and Meni Adler. 2005. Hebrew morphological tagging guidelines. Technical report, Department of Computer Science, Ben Gurion University. find
6. Radu Florian. 2002. Named entity recognition as a house of cards: Classifier stacking. In Proceedings of CoNLL-2002, pages 175–178. Taiwan. find
7. Nizar Habash and Owen Rambow. 2005. Arabic tokenization, part-of-speech tagging and morphological disambiguation in one fell swoop. In Proceedings of the 43rd Annual Meeting of the Association for Computational Linguistics (ACL’05), pages 573–580, Ann Arbor, Michigan, June. Association for Computational Linguistics. find our card
8. Gennadiy Lemberski. 2003. Named entity recognition in Hebrew. Master’s thesis, Department of Computer Science, Ben Gurion University, Beer Sheva, Israel, March. In Hebrew. find
9. Moshe Levinger, Uzzi Ornan, and Alon Itai. 1995. Learning morpho-lexical probabilities from an untagged corpus with an application to Hebrew. Computational Linguistics, 21(3):383–404, September. find our card
10. 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
11. 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
12. Erel Segal. 1997. Morphological analyzer for unvocalized Hebrew words. Unpublished work. find
13. Erel Segal. 1999. Hebrew morphological analyzer for Hebrew undotted texts. Master’s thesis, Technion, Israel Institute of Technology, Haifa, October. In Hebrew. find
14. Shlomo Yona and Shuly Wintner. 2007. A finite-state morphological grammar of Hebrew. Natural Language Engineering. To appear. find our card

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/aclbulk-2007-shacham-morphological-disambiguation-hebrew-case-study/