A FrameNet-Based Semantic Role Labeler for Swedish

Summary

The paper tackles FrameNet-style semantic role labeling (SRL) for Swedish, a language without a sizeable FrameNet-annotated corpus. The authors avoid manual annotation by building the training data automatically. They first trained an English FrameNet labeler and ran it over the English half of Europarl. They then used Giza++ word alignments to map target words and frame element (FE) spans onto the Swedish half, cleaning the projected brackets with a few Swedish-specific heuristics about prepositions, conjunctions and chunk boundaries. No constituent parser existed for Swedish, and dependency subtrees often did not match FrameNet spans. They therefore designed two FE bracketing algorithms that classify chunks as opening or closing an FE while still drawing on dependency-parse features. For evaluation, they manually translated a sample of the FrameNet example corpus into Swedish. The system clearly beats baselines, though it still trails English systems. Its scores roughly equal the English labeler's quality multiplied by transfer accuracy, so the noise in the training data is the main bottleneck. It is presented as the first Swedish SRL system, and the approach should carry over to any language with an English parallel corpus.

Key points

The English labeler was trained on 100,000 FrameNet example sentences covering the 150 most frequent frames, with a further 8,000 used for development. It used the Collins parser and SVM classifiers built with LIBSVM (Gaussian kernel), and scored precision 0.71 and recall 0.65 under strict scoring.
Projection to Swedish used Europarl, Giza++ alignments and the maximal span method. Sentences were kept only when the target word aligned to a Swedish noun, verb or adjective.
On the Swedish test set, classification accuracy on pre-segmented FEs was 0.75. The full task reached precision 0.67, recall 0.47 and F-measure about 0.55.
An expected-quality estimate multiplied the English scores by earlier transfer figures (P 0.84, R 0.81). It gives P 0.60, R 0.53 and F 0.56. The measured precision is notably higher than this estimate, suggesting the SVMs cope well with noisy training data.
Remaining weaknesses include low bracketing recall and weaker Swedish parsers. Europarl's frame distribution is also skewed, for example toward STATEMENT and DISCUSSION.

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

Cited by

18 papers of our library cite it (of the 140673 whose bibliographies we read); 3 of them have a page here, 15 not yet.

Cite

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

@inproceedings{johansson2006framenet,
  title = {{A FrameNet-Based Semantic Role Labeler for Swedish}},
  author = {Johansson, Richard and Nugues, Pierre},
  booktitle = {Proceedings of the COLING/ACL 2006 Main Conference Poster Sessions},
  year = {2006},
  doi = {10.3115/1273073.1273130},
  url = {https://doi.org/10.3115/1273073.1273130}
}

References

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

Collin F. Baker, Charles J. Fillmore, and John B. Lowe. 1998. The Berkeley FrameNet Project. In Proceedings of COLING-ACL’98, pages 86–90, Montréal, Canada. find our card
Chih-Chung Chang and Chih-Jen Lin, 2001. LIBSVM: a library for support vector machines. find
Michael J. Collins. 1999. Head-driven statistical models for natural language parsing. Ph.D. thesis, University of Pennsylvania, Philadelphia. find our card
Charles J. Fillmore. 1976. Frame semantics and the nature of language. Annals of the New York Academy of Sciences: Conference on the Origin and Development of Language, 280:20–32. ↗
Pascale Fung and Benfeng Chen. 2004. BiFrameNet: Bilingual frame semantics resource construction by cross-lingual induction. In Proceedings of COLING-2004. find our card
Daniel Gildea and Daniel Jurafsky. 2002. Automatic labeling of semantic roles. Computational Linguistics, 28(3):245–288. ↗ our card
Richard Johansson and Pierre Nugues. 2005. Using parallel corpora for automatic transfer of FrameNet annotation. In Proceedings of the 1st ROMANCE FrameNet Workshop, Cluj-Napoca, Romania, 26-28 July. find
Philipp Koehn. 2005. Europarl: A parallel corpus for statistical machine translation. In MT Summit 2005. find
Krzysztof Kuchcinski. 2003. Constraints-driven scheduling and resource assignment. ACM Transactions on Design Automation of Electronic Systems, 8(3):355–383. ↗
Ken Litkowski. 2004. Senseval-3 task: Automatic labeling of semantic roles. In Senseval-3: Third International Workshop on the Evaluation of Systems for the Semantic Analysis of Text, pages 9–12. find our card
Joakim Nivre, Johan Hall, and Jens Nilsson. 2004. Memory-based dependency parsing. In Proceedings of CoNLL-2004, pages 49–56. find our card
Franz Josef Och and Hermann Ney. 2003. A systematic comparison of various statistical alignment models. Computational Linguistics, 29(1):19–51. ↗
Sebastian Padó and Mirella Lapata. 2005. Cross-lingual projection of role-semantic information. In Proceedings of HLT/EMNLP 2005. find
Sameer Pradhan, Kadri Hacioglu, Valerie Krugler, Wayne Ward, James Martin, and Dan Jurafsky. 2005a. Support vector learning for semantic argument classification. Machine Learning, 60(1):11– 39. ↗
Sameer Pradhan, Wayne Ward, Kadri Hacioglu, James Martin, and Daniel Jurafsky. 2005b. Semantic role labeling using different syntactic views. In Proceedings of ACL-2005. find
David Yarowsky, Grace Ngai, and Richard Wicentowski. 2001. Inducing multilingual text analysis tools via robust projection across aligned corpora. In Proceedings of HLT 2001. find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/johansson-2006-framenet-based-semantic-role-labeler/

📄 A FrameNet-Based Semantic Role Labeler for Swedish