Elamur Research⌕
☰
home / papers / 2014 / Croatian Dependency Treebank 2.0: New Annotation Guidelines for Improved Parsing

Croatian Dependency Treebank 2.0: New Annotation Guidelines for Improved Parsing

Summary

The paper describes the second version of the Croatian Dependency Treebank (HOBS). The first edition of this treebank closely followed the analytical layer of annotation of the Prague Dependency Treebank (PDT). The authors note that the Czech model does not fully suit Croatian subordinate clauses. In Croatian grammar, the pronouns and adverbs that introduce a subordinate clause function as conjunctions. HOBS 1.0 treated connective words this way in most subordinate clauses, but annotated relative clauses after the Czech pattern, which produced inconsistency. In HOBS 2.0, the conjunction receives a new tag, Sub (instead of AuxC), marked with the type of the subordinate clause. Adverbial clauses are further subdivided by meaning (time, condition, concession, etc.). The attachment rules have also been unified: the conjunction depends on the predicate of the main clause, and the predicate of the subordinate clause depends on the conjunction and now always carries the tag Pred. The corpus was manually converted to the new scheme. Inter-annotator agreement increased noticeably. The graph-based parser MSTParser performs with statistically significantly higher accuracy on HOBS 2.0 than on HOBS 1.0, even though the new version has more tags. Thus, aligning the annotation scheme with the grammar of a particular language makes manual annotation more consistent and automatic parsing more accurate.

Key points

  • New tags. The conjunction of a subordinate clause is marked with the tag Sub (by analogy with Coord in coordination), with a specification of its type: Atr, Adv, Obj, Pred or Sb. Adverbial clauses have eight further subtypes: place, time, manner, cause, result, purpose, condition and concession.
  • Experiment. MSTParser and tenfold cross-validation with a 9:1 split were used. Significance was tested with approximate randomization. Even the HOBS 2.0 models with the full set (81 tags) significantly outperformed the HOBS 1.0 models with both the basic (27 tags) and the full (70 tags) set. The advantage holds for all three metrics, including LAS and UAS.
  • Key functions (basic sets, LAS). Pred: 65.89 → 80.69 (+14.8), Sb: 68.85 → 73.99 (+5.14), Obj: 62.81 → 70.06. The share of word forms tagged Pred roughly doubled (from 4.76% to over 9%), since predicates of subordinate clauses are now marked explicitly. The LAS for Sub (72.91) is decent relative to overall accuracy, but attachment to the head (UAS 73.89) still needs improvement.
  • Learning curves. Training was done on nine subsets from 500 to 4500 sentences. The learning rate is similar, but the LAS curves for HOBS 2.0 are consistently higher than those for HOBS 1.0. Models with the basic tag set have higher curves than models with the full one.
  • Availability and plans. Both versions of HOBS are available for research via META-SHARE. The authors plan to combine HOBS 1.0, HOBS 2.0 and SETimes.HR to improve parsing. They also want to explore delexicalized parsing on Croatian and Slovene treebanks and to add semantic annotation.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

Fields: Linguistics & languages

Topics: Treebanks

Cited by

3 papers of our library cite it (of the 140673 whose bibliographies we read); 1 of them have a page here, 2 not yet.

Cite

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

@inproceedings{agic2014croatian,
  title = {{Croatian Dependency Treebank 2.0: New Annotation Guidelines for Improved Parsing}},
  author = {Agić, Željko and Berović, Daša and Merkler, Danijela and Tadić, Marko},
  booktitle = {Proceedings of the Ninth International Conference on Language Resources and Evaluation (LREC'14)},
  year = {2014},
  doi = {10.63317/2n7oie4t94xy},
  url = {https://doi.org/10.63317/2n7oie4t94xy}
}

References

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

  1. Agić, Ž. and Merkler, D. (2013). Three Syntactic Formalisms for Data-Driven Dependency Parsing of Croatian. LNCS, 8082:560–567. find
  2. Agić, Ž., Merkler, D., and Berović, D. (2013). Parsing Croatian and Serbian by Using Croatian Dependency Treebanks. In Proc. SPMRL, pages 22–33. find
  3. Agić, Ž. (2012). K-Best Spanning Tree Dependency Parsing With Verb Valency Lexicon Reranking. In Proc. COLING, pages 1–12. find
  4. Anić, V. (2003). Veliki rječnik hrvatskoga jezika. find
  5. Barić, E., Lončarić, M., Malić, D., Pavešić, S., Peti, M., Zečević, V., Znika, M., et al. (1995). Hrvatska gramatika. find
  6. Berović, D., Agić, Ž., and Tadić, M. (2012). Croatian Dependency Treebank: Recent Development and Initial Experiments. In Proc. LREC, pages 1902–1906. http://www.xlike.org/ ↗
  7. Böhmová, A., Hajič, J., Hajičová, E., and Hladká, B. (2003). The Prague Dependency Treebank. In Treebanks, pages 103–127. find
  8. Bohnet, B., Nivre, J., Boguslavsky, I., Ginter, R. F. F., and Hajič, J. (2013). Joint Morphological and Syntactic Analysis for Richly Inflected Languages. TACL, 1:415– 428. find
  9. Bohnet, B. (2010). Top Accuracy and Fast Dependency Parsing is not a Contradiction. In Proc. COLING, pages 89–97. find our card
  10. Buchholz, S. and Marsi, E. (2006). CoNLL-X Shared Task on Multilingual Dependency Parsing. In Proc. CoNLL, pages 149–164. find our card
  11. Erjavec, T. (2012). MULTEXT-East: Morphosyntactic Resources for Central and Eastern European Languages. Language Resources and Evaluation, 46(1):131–142. find
  12. Hajič, J., Panevová, J., Buránová, E., Urešová, Z., Bémová, A., Štepánek, J., Pajas, P., and Kárnık, J. (2001). A Manual for Analytic Layer Tagging of the Prague Dependency Treebank. find
  13. Johansson, R. (2013). Training Parsers on Incompatible Treebanks. In Proc. NAACL-HLT, pages 127–137. find
  14. McDonald, R., Pereira, F., Ribarov, K., and Hajič, J. (2005). Non-projective Dependency Parsing Using Spanning Tree Algorithms. In Proc. HLT-EMNLP, pages 523–530. find our card
  15. McDonald, R., Nivre, J., Quirmbach-Brundage, Y., Goldberg, Y., Das, D., Ganchev, K., Hall, K., Petrov, S., Zhang, H., Täckström, O., Bedini, C., Bertomeu Castelló, N., and Lee, J. (2013). Universal Dependency Annotation for Multilingual Parsing. In Proc. ACL, pages 92–97. find
  16. Merkler, D., Agić, Ž., and Agić, A. (2013). Babel Treebank of Public Messages in Croatian. Procedia — Social and Behavioral Sciences, 95:490–497. find
  17. Mikelić Preradović, N., Boras, D., and Kišiček, S. (2009). CROVALLEX: Croatian Verb Valence Lexicon. In Proc. ITI, pages 533–538. find
  18. Mille, S., Burga, A., Ferraro, G., and Wanner, L. (2012). How Does the Granularity of an Annotation Scheme Influence Dependency Parsing Performance? In Proc. COLING, pages 839–852. find
  19. Nivre, J., Hall, J., Kübler, S., McDonald, R., Nilsson, J., Riedel, S., and Yuret, D. (2007a). find
  20. The CoNLL 2007 Shared Task on Dependency Parsing. In Proc. CoNLL Shared Task Session of EMNLP-CoNLL, pages 915–932. find
  21. Nivre, J., Hall, J., Nilsson, J., Chanev, A., Eryigit, G., Kübler, S., Marinov, S., and Marsi, E. (2007b). Malt-Parser: A Language-independent System for Data-driven Dependency Parsing. Natural Language Engineering, 13(2):95–135. find
  22. Pranjković, I. (1986). Koji i što. Jezik, 34:10–16. find
  23. Raguž, D. (1997). Praktična hrvatska gramatika. find
  24. Silić, J., Pranjković, I., and Požgaj-Hadži, V. (2005). Gramatika hrvatskoga jezika: za gimnazije i visoka učilišta. find
  25. Šonje, J. (2000). Rječnik hrvatskoga jezika. find
  26. Tadić, M. (2007). Building the Croatian Dependency Treebank: The Initial Stages. Suvremena lingvistika, 63:85– 92. find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/papers/agic-2014-croatian-dependency-treebank-2-0/