Elamur Research⌕
☰
home / papers / 2006 / Improving English Subcategorization Acquisition with Diathesis Alternations as Heuristic Information

Improving English Subcategorization Acquisition with Diathesis Alternations as Heuristic Information

Paper · 2006 · Xiwu Han, Tiejun Zhao and Xingshang Fu

Summary

The paper addresses the filtering stage of automatic subcategorization frame (SCF) acquisition for English verbs. Standard MLE filtering assumes that the frames a verb takes are statistically independent. As a result, it discards rare but valid frames and keeps frequent noise. The authors replace this with a two-threshold filter that uses diathesis alternations, such as the causative-inchoative pair, as heuristic evidence. A frame is accepted outright if it is frequent. A rare frame is accepted if it clears a very low threshold and is an alternation partner of a frame already accepted. This check repeats until the accepted set stops growing.

They tested the filter on Anna Korhonen's public resource: unfiltered hypotheses from Briscoe and Carroll's system for 30 verbs, with BNC sentences and a manual gold standard. Precision rose to 91.18% while recall stayed the same as the MLE baseline. This beats Korhonen's semantically motivated method on precision and relative F-measure. The remaining errors come mostly from hypothesis generation, not filtering.

Key points

  • Data: Korhonen's evaluation resource covers 30 verbs, each with 1000 BNC sentences, spanning 42 SCF types. Hypotheses were produced by Briscoe and Carroll's pipeline: an HMM tagger, a lemmatizer, a probabilistic LR parser trained on a treebank derived from SUSANNE, and a pattern extractor and classifier.
  • Thresholds: θ1 = 0.2 (ten times Korhonen's MLE threshold) and θ2 = 0.002 (one tenth of it). Alternations were used in both directions rather than Korhonen's one-way form.
  • Precision (P) and relative recall (Re_R) by method:
    • Unfiltered: P = 47.85%.
    • MLE (θ = 0.02): P = 67.89%, Re_R = 93.93%.
    • New filter: P = 91.18%, Re_R = 93.93%, Re_F = 92.53.
    • Korhonen's semantic method: P = 87.1%, Re_R = 85.27%.
  • Absolute recall is 32.52%, only about 2 points below no filtering. Absolute figures are not comparable with Korhonen's published numbers because of a discrepancy in her baseline.
  • About 95% of the remaining missed and wrong SCFs trace back to the hypothesis generator, so further gains must come from parsing and extraction.

Interesting ideas

  • I1. A rare candidate earns acceptance only if a structurally related candidate has already been confirmed with strong evidence, so confidence spreads from strong to weak items instead of relying on one global cutoff.
  • I2. The paper separates two recall measures: one against everything the corpus contains, one against what the generator actually proposed. This shows exactly how much loss each pipeline stage causes.
  • I3. The authors treat the independence assumption among a verb's frames as linguistically wrong and encode known alternations as explicit links between frames instead.
  • I4. An error analysis reveals that the bottleneck has moved upstream, which redirects effort to the stage that now dominates the errors.

For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

  • I1 (learn): Self-induced repair rules or silver labels with low support could be kept only when a related, well-attested rule or label already exists, which cuts noise without losing rare true cases.
  • I2 (learn): Evaluations could report frame or parse recall relative to the candidates each stage produced, pinpointing whether errors arise in the tree stage or in later filtering.
  • I3 (world): Compiled verb-class knowledge could store alternation links between VerbNet frames, letting a verb seen only in its transitive use also be licensed for the inchoative where its class permits.
  • I4 (tree): After improving frame filtering, attributing remaining frame errors to UD parse mistakes would guide where parser retraining pays off most.

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

 Institutions

🎓Harbin Institute of TechnologyHarbin · China185 works · 29,140 citations🎓Heilongjiang UniversityHarbin · China2 works · 190 citations

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksArts and Humanities · 17,450 worksLanguage and Linguistics · 17,296 works

🧭 Topics 1Research subjects and languages the work is about.English · 2,379 works

Cite

Cite as Xiwu Han, Tiejun Zhao and Xingshang Fu. 2006. Improving English Subcategorization Acquisition with Diathesis Alternations as Heuristic Information. Proceedings of the COLING/ACL 2006 Main Conference Poster Sessions.

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

@inproceedings{han2006improving,
  title = {{Improving English Subcategorization Acquisition with Diathesis Alternations as Heuristic Information}},
  author = {Xiwu Han, Tiejun Zhao and Xingshang Fu},
  booktitle = {Proceedings of the COLING/ACL 2006 Main Conference Poster Sessions},
  year = {2006},
  doi = {10.3115/1273073.1273116},
  url = {https://doi.org/10.3115/1273073.1273116}
}

References

17 references, 1 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Boguraev B. K., E. J. Briscoe. Large lexicons for natural language processing utilizing the grammar coding system of the Longman Dictionary of Contemporary English. Computational Linguistics, 1987: 219-240 find
  2. Brent, M., From Grammar to Lexicon: unsupervised learning of lexical syntax, Computational Linguistics 19(3) 1993: 243-262. find
  3. Briscoe, Ted and John Carroll, Automatic extraction of subcategorization from corpora, Proceedings of the 5th ACL Conference on Applied Natural Language Processing, Washington, DC, 1997: 356-363.
  4. Chomsky, Noam, Aspects of the Theory of Syntax, MIT Press, Cambridge, 1965. find
  5. Chrupala, Grzegorz, Acquiring Verb Subcategorization from Spanish Corpora, PhD program “Cognitive Science and Language”, Universitat de Barcelona, 2003: 67-68. find
  6. Ellis, R ． Understanding Second language Acquisition, Oxford University Press.1985 find
  7. Gamallo, P., Agustini, A. and Lopes Gabriel P., Using Co-Composition for Acquiring Syntactic and Semantic Subcategorisation, Proceedings of the Workshop of the ACL Special Interest Group on the Lexicon (SIGLEX), Philadelphia, 2002: 34-41. find
  8. Han, Xiwu, Tiejun Zhao, Haoliang Qi, and Hao Yu, Subcategorization Acquisition and Evaluation for Chinese Verbs, Proceedings of the COLING 2004, 2004: 723-728. find
  9. Korhonen, Anna, Automatic Extraction of Subcategorization Frames from Corpora –Improving Filtering with Diathesis Alternations, 1998. Please refer to http://www.folli.uva.nl/CD/1998/pdf/keller/korhon en.pdf Korhonen, Anna, Subcategorization Acquisition, Dissertation for PhD, Trinity Hall University of Cambridge, 2001. ↗
  10. Korhonen, Anna, Subcategorization Acquisition, Technical Report Number 530, Trinity Hall University of Cambridge, 2002. find
  11. Korhonen, Anna, Yuval Krymolowski, Zvika Marx, Clustering Polysemic Subcategorization Frame Distributions Semantically, Proceedings of the 41st Annual Meeting of the Association for Computational Linguistics, 2003: 64-71. find
  12. Korhonen, Anna. Subcategorization Evaluation Resources. http://www.cl.cam.ac.uk/users/alk23/sub-cat/subcat.html. 2005 ↗
  13. Levin, B., English Verb Classes and Alternations, Chicago University Press, Chicago, 1993. find
  14. McCarthy, D., Lexical Acquisition at the Syntax-Semantics Interface: Diathesis Alternations, Subcategorization Frames and Selectional Preferences, PhD thesis, University of Sussex, 2001. find
  15. Peters, A. The Unit of Language Acquisition, Cambridge University Press. 1983. find
  16. Sarkar, A. and Zeman, D., Automatic Extraction of Subcategorization Frames for Czech, Proceedings of the 19th International Conference on Computational Linguistics, Saarbrucken, Germany, 2000. Please refer to http://www.sfu.ca/~anoop/papers/pdf/coling0final.pdf ↗
  17. Shulte im Walde, Sabine, Inducing German Semantic Verb Classes from Purely Syntactic Subcategorization Information, Proceedings of the 40th Annual Meeting of the Association for Computational Linguistics, 2002: 223-230. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/han-2006-improving-english-subcategorization-acquisition-diathesis/