Elamur Research⌕
☰
home / papers / 2019 / Automatic Propbank Generation for Turkish

Automatic Propbank Generation for Turkish

Summary

The authors automatically build a Turkish PropBank, i.e., a corpus annotated with predicate-argument structure for the semantic role labeling task. Manual annotation of such a resource is expensive and requires linguists. They therefore transfer semantic roles from English. They take English–Turkish parallel sentences from the translated Penn Treebank (the corpus of Ak et al., 2018) and add annotations from the first version of English PropBank to them. The predicate in a Turkish sentence is identified through morphological analysis: the word must have a verbal root and a verbal last inflectional group. The English roles are then passed to it, and the argument labels are aligned with the Turkish annotation scheme. For sentences without a tree correspondence to the English ones, words are aligned via WordNet or with the statistical models IBM Model 1 and 2. Turkish roots are found with an FSM morphological analyzer, and words that remain unlabeled are labeled by rules based on case suffixes. Evaluated against manual annotation, the tree-based method achieves 64% accuracy, and about 75% among transferred labels alone. This approach quickly yields a draft PropBank for low-resource languages, leaving annotators only to proofread it.

Key points

  • Data: the original English PropBank (Palmer et al., 2004, LDC) and Turkish translations of Penn Treebank sentences. English annotations could be added to 6 060 of 9 558 sentences. For comparison, the manually annotated Turkish PropBank (Ak et al., 2018) was used: 9 560 sentences, framesets for 1 330 verbs and 1 914 senses.
  • Transfer via parallel trees achieves 64% accuracy relative to manual annotation. Counting only the transferred annotations, it is approximately 75%.
  • WordNet-based alignment has low coverage: only ~14.59% of annotations are transferred correctly. However, of the 5 457 transferred roles, 4 255 are correct, i.e., about 79%.
  • IBM Model 1 and 2, together with a boosting step and morphological rules, yield about 60% correct annotations, with accuracies of 59.44% and 59.86%.
  • The rules rely on case: a noun without a suffix becomes ARG0, the accusative or dative case gives ARG1, the locative or ablative ARGM-LOC, and the instrumental ARG2. Dates and times receive ARGM-TMP, and numerical expressions ARGM-EXT.

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: Semantic role labeling

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.26615/978-954-452-056-4_005

@inproceedings{ak2019automatic,
  title = {{Automatic Propbank Generation for Turkish}},
  author = {AK, Koray and Yıldız, Olcay Taner},
  booktitle = {Proceedings of the International Conference on Recent Advances in Natural Language Processing (RANLP 2019)},
  year = {2019},
  doi = {10.26615/978-954-452-056-4\_005},
  url = {https://doi.org/10.26615/978-954-452-056-4_005}
}

References

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

  1. Meyers A., R. Reeves, C. Macleod, R. Szekely, V. Zielinska, B. Young, and R. Grishman. 2004. The nombank project: An interim report. In HLT-NAACL 2004 Workshop: Frontiers in Corpus Annotation. Association for Computational Linguistics, Boston, Massachusetts, USA, pages 24–31. find our card
  2. M. P. Marcus, M. A. Marcinkiewicz, and B. Santorini. 1993. Building a large annotated corpus of english: The penn treebank. Computational linguistics 19(2):313–330. find our card
  3. K. Ak, O. T. Yıldız, V. Esgel, and C. Toprak. 2018. Construction of a Turkish proposition bank. Turkish Journal of Electrical Engineering and Computer Science 26:570 – 581. find
  4. M. Palmer, D. Gildea, and P. Kingsbury. 2005. The proposition bank: An annotated corpus of semantic roles. Comput. Linguist. 31(1):71–106. find our card
  5. M. Palmer, P. Kingsbury, O. Babko-Malaya, S. Cotton, and B. Snyder. 2004. Proposition bank i. Philadelphia: Linguistic Data Consortium. LDC2004T14. find
  6. A. Akbik, L. Chiticariu, M. Danilevsky, Y. Li, S. Vaithyanathan, and H. Zhu. 2015. Generating high quality proposition banks for multilingual semantic role labeling. In ACL (1). The Association for Computer Linguistics, pages 397–407. find our card
  7. L. Van der Plas, M. Apidianaki, and C. Chen. 2014. Global methods for cross-lingual semantic role and predicate labelling. In COLING. ACL, pages 1279– 1290. find
  8. C. Bonial, J. Bonn, K. Conger, J. D. Hwang, and M. Palmer. 2014. Propbank: Semantics of new predicate types. In Proceedings of the Ninth International Conference on Language Resources and Evaluation (LREC-2014). European Language Resources Association (ELRA). find our card
  9. L. Van der Plas, P. Merlo, and J. Henderson. 2011. Scaling up automatic cross-lingual semantic role annotation. In ACL (Short Papers). The Association for Computer Linguistics, pages 299–304. find
  10. T. Zhuang and C. Zong. 2010. Joint inference for bilingual semantic role labeling. In Proceedings of the 2010 Conference on Empirical Methods in Natural Language Processing. Association for Computational Linguistics, Stroudsburg, PA, USA, EMNLP ’10, pages 304–314. find
  11. P. F. Brown, V. J. D. Pietra, S. A. D. Pietra, and R. L. Mercer. 1993. The mathematics of statistical machine translation: Parameter estimation. Comput. Linguist. 19(2):263–311. find
  12. G. G. Şahin. 2016a. Framing of verbs for turkish propbank. In TurCLing 2016 in conj. with 17th International Conference on Intelligent Text Processing and Computational Linguistics (CICLING 2016). find
  13. G. G. Şahin. 2016b. Verb sense annotation for turkish propbank via crowdsourcing. In 17th International Conference on Intelligent Text Processing and Computational Linguistics (CICLING 2016). find
  14. G. G. Şahin and E. Adalı. 2017. Annotation of semantic roles for the Turkish proposition bank. Language Resources and Evaluation . find
  15. R. Ehsani, E. Solak, and O. T. Yıldız. 2018. Constructing a wordnet for Turkish using manual and automatic annotation. ACM Transactions on Asian Low-Resource Language Information Processing 17(3). find
  16. C. J. Fillmore, J. Ruppenhofer, and Collin F. Baker. 2004. FrameNet and Representing the Link between Semantic and Syntactic Relations, Institute of Linguistics, Academia Sinica, Taipei, pages 19–62. Language and Linguistics Monographs Series B. find
  17. P. Kingsbury and M. Palmer. 2002. From treebank to propbank. In LREC. European Language Resources Association. find our card

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/ak-2019-automatic-propbank-generation-turkish/