Elamur Research⌕
☰
home / papers / 2010 / Disambiguating Compound Nouns for a Dynamic HPSG Treebank of Wall Street Journal Texts

Disambiguating Compound Nouns for a Dynamic HPSG Treebank of Wall Street Journal Texts

Paper · 2010 · Valia Kordoni and Yi Zhang

Summary

The paper looks at noun compounds in English financial news. Their internal bracketing and the hidden relation between their nouns are hard to resolve. This matters when building a deep treebank. The authors describe a "dynamic" HPSG annotation of the WSJ part of the Penn Treebank. The PET parser with the hand-built English Resource Grammar (ERG) produces candidate analyses. Annotators then choose among them by accepting or rejecting individual discriminants, instead of drawing trees by hand. Compounds such as "luxury auto maker" get both left- and right-branching readings. Deverbal compounds also leave the argument-versus-modifier link unresolved. The authors therefore propose two aids. First, they retrain the maximum-entropy parse selection model on newly annotated sections. Second, they add bi-lexical preference features gathered from large automatically parsed corpora, which amounts to a form of self-training. At the time of writing, about half of the WSJ (12 sections) had been annotated. The work shows how a statistical ranker can make rich, grammar-based annotation faster while keeping it independent of the original PTB bracketing.

Key points

  • Tools and resources: ERG, the PET parser, [incr tsdb()] for treebanking, and the DELPH-IN software. Each WSJ sentence keeps up to 500 top readings, found with exact n-best parsing.
  • With discriminant-based annotation, n candidates need about log2 n binary choices, so no more than 9 decisions per sentence. Each sentence ends up marked "gold", "rejected" (a grammar error), or "low confidence" with its partial decisions saved.
  • When the grammar changes, the corpus is reparsed and the stored decisions are replayed. Only sentences left ambiguous after the replay need new manual work.
  • The parse selection model follows Toutanova et al. (2002, 2005). Its features include local subtrees, grandparents and n-grams. Compound bracketing is decided last, after the rest of the sentence is settled.
  • Twelve WSJ sections were done at writing time. The authors chose not to import the NML/JJP NP bracketing of Vadas and Curran (2007), but they plan to measure agreement with it.

Interesting ideas

  • I1. Annotators make yes/no decisions on discriminants rather than picking whole trees, and these decisions are stored, so a reparse with a new grammar can reapply them automatically and keep manual work small.
  • I2. Hard local ambiguities such as compound bracketing are postponed until the surrounding structure is fixed, so the easier decisions narrow the search first.
  • I3. Relation preferences between word pairs are estimated from a large corpus parsed by the system itself and fed back as features, a self-training loop aimed specifically at argument versus modifier inside compounds.
  • I4. The grammar deliberately leaves the deverbal noun's link to its partner underspecified, rather than forcing an early choice, and a separate model or annotator resolves it.
  • I5. The authors keep their annotation independent of an existing resource but still plan to compare against it as an external check.

For Mova

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

  • I1 (learn): Silver-data repairs could be stored as discriminant-like constraints, so they survive parser retraining and only truly new conflicts reach the LLM teacher.
  • I2 (tree): gparse or the perceptron parser could leave NN-compound attachment open and resolve it after clause-level arcs are fixed, using the remaining context.
  • I3 (frames): Counts of noun-noun pairs from parsed text could help decide whether "stock purchase" makes "stock" the theme of the purchase frame or just a modifier.
  • I4 (world): Entities built from compounds could carry an explicitly unresolved role slot that the world builder fills later, keeping each step explainable instead of guessing silently.
  • I5 (learn): A UD-converted or third-party bracketing set could be used only for evaluation of compound structure, avoiding the circularity of training and testing on the same assumptions.

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

🎓Saarland UniversitySaarbrücken · Germany148 works · 18,287 citations🔬German Research Centre for Artificial IntelligenceKaiserslautern · Germany77 works · 29,479 citations

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 works

🧭 Topics 2Research subjects and languages the work is about.Syntax & grammar formalisms · 214 worksTreebanks · 126 works

Cite

Cite as Valia Kordoni and Yi Zhang. 2010. Disambiguating Compound Nouns for a Dynamic HPSG Treebank of Wall Street Journal Texts. Proceedings of the Seventh International Conference on Language Resources and Evaluation (LREC'10).

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

@inproceedings{kordoni2010disambiguating,
  title = {{Disambiguating Compound Nouns for a Dynamic HPSG Treebank of Wall Street Journal Texts}},
  author = {Valia Kordoni and Yi Zhang},
  booktitle = {Proceedings of the Seventh International Conference on Language Resources and Evaluation (LREC'10)},
  year = {2010},
  doi = {10.63317/2tur3omvhvun},
  url = {https://doi.org/10.63317/2tur3omvhvun}
}

References

17 references, 4 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Sabine Brants, Stefanie Dipper, Silvia Hansen, Wolfgang Lezius, and George Smith. 2002. The tiger treebank. In Proceedings of the workshop on treebanks and linguistic theories, pages 24–41. find
  2. Ulrich Callmeier. 2001. Efficient parsing with large-scale unification grammars. Master’s thesis, Universität des Saarlandes, Saarbrücken, Germany. find
  3. Pamela Downing. 1977. On the creation and use of English compound nouns. Language, 53:4:810–842. ↗
  4. Dan Flickinger. 2002. On building a more efficient grammar by exploiting types. In Stephan Oepen, Dan Flickinger, Jun’ichi Tsujii, and Hans Uszkoreit, editors, Collaborative Language Engineering, pages 1–17. CSLI Publications.
  5. Jan Hajič, Alena Böhmová, Eva Hajičová, and Barbora Vidová-Hladká. 2000. The Prague Dependency Treebank: A Three-Level Annotation Scenario. In A. Abeillé, editor, Treebanks: Building and Using Parsed Corpora, pages 103–127. Amsterdam:Kluwer. find
  6. Mark Johnson and Stefan Riezler. 2000. Exploiting auxiliary distributions in stochastic unification-based grammars. In Proceedings of the 1st NAACL conference, pages 154–161, Seattle, USA. ↗
  7. Judith Levi. 1978. The Syntax and Semantics of Complex Nominals. Academic Press, INC, New York. find
  8. Mitchell P. Marcus, Beatrice Santorini, and Mary Ann Marcinkiewicz. 1993. Building a large annotated corpus of english: The penn treebank. Computational Linguistics, 19(2):313–330. ↗
  9. Stephan Oepen, Kristina Toutanova, Stuart Shieber, Christopher Manning, Dan Flickinger, and Thorsten Brants. 2002. The LinGO Redwoods treebank: motivation and preliminary applications. In Proceedings of COLING 2002: The 17th International Conference on Computational Linguistics: Project Notes, Taipei, Taiwan.
  10. Stephan Oepen. 2001. [incr tsdb()] — competence and performance laboratory. User manual. Technical report, Computational Linguistics, Saarland University, Saarbrücken, Germany. find
  11. Carl J. Pollard and Ivan A. Sag. 1994. Head-Driven Phrase Structure Grammar. University of Chicago Press, Chicago, USA. find
  12. Kristina Toutanova, Christoper D. Manning, Stuart M. Shieber, Dan Flickinger, and Stephan Oepen. 2002. Parse ranking for a rich HPSG grammar. In Proceedings of the 1st Workshop on Treebanks and Linguistic Theories (TLT 2002), pages 253–263, Sozopol, Bulgaria. find
  13. Kristina Toutanova, Christoper D. Manning, Dan Flickinger, and Stephan Oepen. 2005. Stochastic HPSG parse selection using the Redwoods corpus. Journal of Research on Language and Computation, 3(1):83–105. find
  14. David Vadas and James Curran. 2007. Adding noun phrase structure to the penn treebank. In Proceedings of the 45th Annual Meeting of the Association of Computational Linguistics, pages 240–247, Prague, Czech Republic.
  15. David Vadas and James R. Curran. 2008. Parsing noun phrase structure with CCG. In Proceedings of ACL-08: HLT, pages 335–343, Columbus, Ohio, June. Association for Computational Linguistics. find
  16. Gertjan van Noord. 2007. Using self-trained bilexical preferences to improve disambiguation accuracy. In Proceedings of the Tenth International Conference on Parsing Technologies, pages 1–10, Prague, Czech Republic. find
  17. Beatrice Warren. 1978. Semantic Patterns of Noun-Noun Compounds. Acta Universitatis Gothoburgensis, Göteborg. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/kordoni-2010-disambiguating-compound-nouns-dynamic-hpsg/