Semi-Supervised Frame-Semantic Parsing for Unknown Predicates

Summary

Frame-semantic parsers trained on FrameNet struggle with predicates that never appear in the lexicon or its annotated sentences. FrameNet lists only about 9,300 annotated target types. Das and Smith address this coverage gap with graph-based semi-supervised learning. They build a large graph whose vertices are lemmatized targets with coarse POS tags. Most vertices come from Lin's dependency-based distributional thesaurus, and a smaller share are targets already annotated in FrameNet 1.5. Edge weights combine distributional similarity with similarity between frame distributions of observed targets. Label propagation then spreads frame distributions from known targets to unknown ones. When the SEMAFOR parser meets an unseen target, it considers only the top-ranked frames from the propagated distribution rather than the whole frame inventory. For unseen targets, this raises frame identification accuracy by 15.7 points and full-parse F1 by 13.7 points over the supervised SEMAFOR baseline. It also makes decoding considerably faster. The method matters because it extends a hand-built semantic resource using unlabeled data rather than other expert lexicons. The authors expect even larger benefits in domains outside news.

Key points

Graph construction: The graph has 61,702 target vertices, roughly six times the number of FrameNet targets. Vertices come from Lin's thesaurus, which was built from a 64-million-word parsed corpus. Each vertex is linked to its 20 nearest neighbours, and adjectives and adverbs were split using Gigaword POS statistics.
Gains on unseen targets: Frame identification accuracy rises by 15.7 points and full frame-semantic parsing F1 by 13.7 points; both gains are significant. On the full test set, relative error drops by 5.4% for frame identification and 1.3% for full parsing, since only 3.2% of test targets are unseen.
Baseline comparison: A self-training baseline did worse than SEMAFOR. A graph built only from the thesaurus (LinGraph) beat both baselines but trailed the combined-similarity graph (FullGraph).
Speed: Restricting candidates to M = 2 frames instead of all 877 cuts frame identification time on unknown targets from 111 s to 19 s. On the full test set, it drops from 131 s to 39 s.
Availability: The extended parser is released as SEMAFOR at ark.cs.cmu.edu.

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

41 papers of our library cite it (of the 140673 whose bibliographies we read); 8 of them have a page here, 33 not yet.

Cite

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

@inproceedings{das2011semi,
  title = {{Semi-Supervised Frame-Semantic Parsing for Unknown Predicates}},
  author = {Das, Dipanjan and Smith, Noah A.},
  booktitle = {Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies},
  year = {2011},
  url = {https://aclanthology.org/P11-1144/}
}

References

37 references, 9 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

C. Baker, M. Ellsworth, and K. Erk. 2007. SemEval-2007 Task 19: frame semantic structure extraction. In Proc. of SemEval. find our card
C. A. Bejan. 2009. Learning Event Structures From Text. Ph.D. thesis, The University of Texas at Dallas. find
Y. Bengio, O. Delalleau, and N. Le Roux. 2006. Label propagation and quadratic criterion. In Semi-Supervised Learning. MIT Press. find
H. C. Boas. 2002. Bilingual FrameNet dictionaries for machine translation. In Proc. of LREC. find our card
D. Das and S. Petrov. 2011. Unsupervised part-of-speech tagging with bilingual graph-based projections. In Proc. of ACL-HLT. find
D. Das, N. Schneider, D. Chen, and N. A. Smith. 2010a. Probabilistic frame-semantic parsing. In Proc. of NAACL-HLT. find our card
D. Das, N. Schneider, D. Chen, and N. A. Smith. 2010b. SEMAFOR 1.0: A probabilistic frame-semantic parser. Technical Report CMU-LTI-10-001, Carnegie Mellon University. find
K. Erk and S. Padó. 2006. Shalmaneser - a toolchain for shallow semantic parsing. In Proc. of LREC. find
C. Fellbaum, editor. 1998. WordNet: an electronic lexical database. MIT Press, Cambridge, MA. find
C. J. Fillmore, C. R. Johnson, and M. R.L. Petruck. 2003. Background to FrameNet. International Journal of Lexicography, 16(3). find
C. J. Fillmore. 1982. Frame semantics. In Linguistics in the Morning Calm, pages 111–137. Hanshin Publishing Co., Seoul, South Korea. find
M. Fleischman, N. Kwon, and E. Hovy. 2003. Maximum entropy models for FrameNet classification. In Proc. of EMNLP. find our card
P. Fung and B. Chen. 2004. BiFrameNet: bilingual frame semantics resource construction by cross-lingual induction. In Proc. of COLING. find our card
H. Fürstenau and M. Lapata. 2009. Semi-supervised semantic role labeling. In Proc. of EACL. find
D. Gildea and D. Jurafsky. 2002. Automatic labeling of semantic roles. Computational Linguistics, 28(3). find our card
A.-M. Giuglea and A. Moschitti. 2006. Shallow semantic parsing based on FrameNet, VerbNet and Prop-Bank. In Proc. of ECAI 2006. find
D. Graff. 2003. English Gigaword. Linguistic Data Consortium. find
R. Johansson and P. Nugues. 2007. LTH: semantic structure extraction using nonprojective dependency trees. In Proc. of SemEval. find
D. Lin. 1993. Principle-based parsing without overgeneration. In Proc. of ACL. find
D. Lin. 1994. Principar–an efficient, broadcoverage, principle-based parser. In Proc. of COLING. find
D. Lin. 1998. Automatic retrieval and clustering of similar words. In Proc. of COLING-ACL. find
D. C. Liu and J. Nocedal. 1989. On the limited memory bfgs method for large scale optimization. Math. Programming, 45(3). find
Y. Matsubayashi, N. Okazaki, and J. Tsujii. 2009. A comparative study on generalization of semantic roles in FrameNet. In Proc. of ACL-IJCNLP. find our card
R. McDonald, K. Crammer, and F. Pereira. 2005. Online large-margin training of dependency parsers. In Proc. of ACL. find our card
Z.-Y. Niu, D.-H. Ji, and C. L. Tan. 2005. Word sense disambiguation using label propagation based semi-supervised learning. In Proc. of ACL. find
S. Padó and M. Lapata. 2005. Cross-linguistic projection of role-semantic information. In Proc. of HLT-EMNLP. find
M. Pennacchiotti, D. De Cao, R. Basili, D. Croce, and M. Roth. 2008. Automatic induction of FrameNet lexical units. In Proc. of EMNLP. find our card
S. P. Ponzetto and M. Strube. 2007. Deriving a large scale taxonomy from wikipedia. In Proc. of AAAI. find
A. Ratnaparkhi. 1996. A maximum entropy model for part-of-speech tagging. In Proc. of EMNLP. find
L. Shi and R. Mihalcea. 2004. An algorithm for open text semantic parsing. In Proc. of Workshop on Robust Methods in Analysis of Natural Language Data. find
L. Shi and R. Mihalcea. 2005. Putting pieces together: combining FrameNet, VerbNet and WordNet for robust semantic parsing. In Computational Linguistics and Intelligent Text Processing: Proc. of CICLing 2005. Springer-Verlag. find
R. Snow, D. Jurafsky, and A. Y. Ng. 2006. Semantic taxonomy induction from heterogenous evidence. In Proc. of COLING-ACL. find
A. Subramanya and J. A. Bilmes. 2009. Entropic graph regularization in non-parametric semi-supervised classification. In Proc. of NIPS. find
A. Subramanya, S. Petrov, and F. Pereira. 2010. Efficient Graph-based Semi-Supervised Learning of Structured Tagging Models. In Proc. of EMNLP. find
C. A. Thompson, R. Levy, and C. D. Manning. 2003. A generative model for semantic role labeling. In Proc. of ECML. find
S. Tonelli and C. Giuliano. 2009. Wikipedia as frame information repository. In Proc. of EMNLP. find
X. Zhu, Z. Ghahramani, and J. D. Lafferty. 2003. Semi-supervised learning using gaussian fields and harmonic functions. In Proc. of ICML. find

↑ back to top

📄 Semi-Supervised Frame-Semantic Parsing for Unknown Predicates