Integrated Morphological and Syntactic Disambiguation for Modern Hebrew
What it is about
Hebrew morphology (prefixed particles, clitics, a fused word that may span several phrases) makes word-level parsing hard. The paper builds the first statistical parser for Modern Hebrew that does not need manual morphological disambiguation. It runs three stages: a morphological analyser proposes segmentations, a HMM tagger assigns POS tags to segments, and a PCFG parser builds trees over the segments. Model I passes only the chosen segmentation to the parser; Model II passes the tagged segments, so that the lowest non-terminals carry the morphology. The treebank has about 3,250 sentences of length 2 to 20. Labeled precision and recall are about 60% for both models, Model II gains up to about 2 points, and the string coverage falls from 99.2% to 95.9%. Scores are measured on character spans, because the constituents do not follow word boundaries.
Ideas
- I1. Let the parser see the tagged segments, not only the chosen segmentation: keep the segmentation and tag choice as a lattice and pass the best tag per segment to a parser. Measure on the same split to see whether labeled scores go up at a cost in coverage.
- I2. Compare tag sets on the parse score, not on the tag score: adding finite and non-finite verb forms lowered tagging accuracy but raised parsing accuracy. Test a finite/non-finite split in the 11. Modern Hebrew in Universal Dependencies tag mapping by parse F.
- I3. Score constituents as character spans (start, end, label) rather than as morpheme or word lists, so that a span across word boundaries is scored correctly. Add this span evaluation to
tools/macula comparefor Hebrew parses.
For Mova
I1 is a design note for the segment stage of tools/hbo (lattice into the parser; gparse takes segment tags). I2 is a test on the tag mapping in 11. Modern Hebrew in Universal Dependencies. I3 goes into the evaluation of tools/macula compare and Syntax: the clause and the sentence. The 2006 system is a baseline only; the data is small, and Mova's own parser is neural-free.
Cite
@inproceedings{tsarfaty2006integrated,
title = {{Integrated Morphological and Syntactic Disambiguation for Modern Hebrew}},
author = {Tsarfaty, Reut},
booktitle = {Proceedings of the COLING/ACL 2006 Student Research Workshop},
year = {2006},
doi = {10.3115/1557856.1557867},
url = {https://doi.org/10.3115/1557856.1557867}
}References
- Meni Adler and Dudi Gabai. 2005. Morphological Analyzer and Disambiguator for Modern Hebrew. Knowledge Center for Processing Hebrew. find
- Roy Bar-Haim. 2005. Part-of-Speech Tagging for Hebrew and Other Semitic Languages. Master’s thesis, Technion, Haifa, Israel. find
- Eugene Charniak, Glenn Carroll, John Adcock, Anthony R. Cassandra, Yoshihiko Gotoh, Jeremy Katz, Michael L. Littman, and John McCann. 1996. Taggers for Parsers. AI, 85(1-2):45–57. find
- Eugene Charniak. 2000. A Maximum-Entropy-Inspired Parser. In Proceedings of NAACL 2000. find
- Michael Collins. 1997. Three Generative, Lexicalised Models for Statistical Parsing. In Proceedings of ACL-EACL 1997. find
- Nizar Habash and Owen Rambow. 2005. Arabic Tokenization, Part-of-Speech Tagging and Morphological Disambiguation in One Fell Swoop. In Proceedings of ACL 2005. find our card
- Helmut Schmid, 2000. LoPar: Design and Implementation. Institute for Computational Linguistics, University of Stuttgart. find
- Erel Segal. 2000. A Probabilistic Morphological Analyzer for Hebrew Undotted Texts. Master’s thesis, Computer Science Department, Technion, Isreal. find
- Khalil Sima’an, Alon Itai, Yoad Winter, Alon Altman, and Noa Nativ. 2001. Building a Tree-Bank for Modern Hebrew Text. In Traitement Automatique des Langues, volume 42, pages 347–380. find
This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/aclbulk-2006-tsarfaty-integrated-morphological-syntactic-disambiguation-modern/