Easy-First Dependency Parsing of Modern Hebrew

What it is about

The authors apply their easy-first, non-directional dependency parser to the Hebrew dependency treebank, a small treebank of a language with rich morphology and free word order. The parser makes the easiest attachments first and each decision can see the structure built so far. With the English feature set (lexical preposition features removed) it matches a first-order MST parser on gold tags and beats MaltParser in every condition. Adding noun-adjective agreement features (a binary gender flag and a binary number flag for each candidate pair in a small window) raises unlabeled accuracy on the development section from 83.8 to 84.2 with gold morphology and from 75.5 to 76.2 with automatic morphology, on a par with second-order MST. For the MST parsers the same morphology features hurt when morphology is automatic. Most remaining errors are semantic (coordination, PP attachment, main-verb hierarchy).

Ideas

• I1. Agreement as a local attachment feature: when the parser considers attaching an adjective to a noun, or a verb to a subject, add gender and number match flags for the two heads. The same check can be run as a rule over the trees we build from treebanks/he/ to count agreement violations before any learning.
• I2. Report every parser result twice, with gold and with automatic morphology. The paper's main signal is that the agreement gain survives automatic morphology while the MST gain does not; Mova's gparse runs on tags from tools/hbo and from the Modern Hebrew analyser, so the same two-column table is the honest test.
• I3. Keep the feature set for lexical items small on small treebanks: the authors chose not to lexicalise prepositions because of the treebank size and the productive preposition system. A test for gparse on he: lexicalised versus class-only prepositions.

For Mova

I1 belongs in Nouns, adjectives, pronouns and numerals (noun-adjective agreement) and in 05. The verb: system, binyanim, tense and use for subject-verb agreement. I2 is a measurement for gparse on the treebanks/he/ data. I3 is a feature-set choice for gparse, to be tested before it is written into the grammar.

Topics: Dependency parsing · Treebanks

Cite

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

@inproceedings{goldberg2010easy,
  title = {{Easy-First Dependency Parsing of Modern Hebrew}},
  author = {Goldberg, Yoav and Elhadad, Michael},
  booktitle = {Proceedings of the NAACL HLT 2010 First Workshop on Statistical Parsing of Morphologically-Rich Languages},
  year = {2010},
  url = {https://aclanthology.org/W10-1412/}
}

References

7 references, 2 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Meni Adler. 2007. Hebrew Morphological Disambiguation: An Unsupervised Stochastic Word-based Approach. Ph.D. thesis, Ben-Gurion University of the Negev, Beer-Sheva, Israel. find
2. Yoav Goldberg and Michael Elhadad. 2009. Hebrew Dependency Parsing: Initial Results. In Proc. of IWPT. find our card
3. Yoav Goldberg and Michael Elhadad. 2010. An efficient algorithm for easy-first non-directional dependency parsing. In Proc. of NAACL. find
4. Yoav Goldberg, Meni Adler, and Michael Elhadad. 2008. EM Can find pretty good HMM POS-Taggers (when given a good start). In Proc. of ACL. find
5. Sandra Kübler, Ryan T. McDonald, and Joakim Nivre. 2009. Dependency Parsing. Synthesis Lectures on Human Language Technologies. Morgan & Claypool Publishers. find
6. Ryan McDonald, Koby Crammer, and Fernando Pereira. 2005. Online large-margin training of dependency parsers. In Proc of ACL. find our card
7. Joakim Nivre, Johan Hall, and Jens Nillson. 2006. Malt-Parser: A data-driven parser-generator for dependency parsing. In Proc. of LREC. find

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/aclbulk-2010-goldberg-easy-first-dependency-parsing-modern/