Elamur Research⌕
☰
home / papers / 2007 / Linguistically Motivated Large-Scale NLP with C&C and Boxer

Linguistically Motivated Large-Scale NLP with C&C and Boxer

Paper · 2007 · James Curran, Stephen Clark and Johan Bos

Summary

The paper presents C&C, a toolkit centred on a wide-coverage Combinatory Categorial Grammar (CCG) parser, and Boxer, a semantic module built on top of it. The goal is to show that deep, theory-based analysis can run fast enough for very large text collections. The parser's grammar comes from CCGbank, a CCG conversion of the Penn Treebank. A Maximum Entropy supertagger first narrows the lexical categories for each word. CKY chart parsing then combines them, with the best derivation chosen by either a log-linear or a perceptron model. Further Maximum Entropy taggers handle POS tagging, chunking and named entity recognition. Boxer converts the parser's derivations into Discourse Representation Structures (DRSs), which can also be exported as first-order logic. The system processed the full Gigaword corpus, about one billion words, in under five days on 18 processors. Accuracy is competitive on CCGbank and DepBank. The work matters because it brings recovery of long-range dependencies and formal semantics to web-scale text, for example in open-domain question answering.

Key points

  • The grammar has 425 lexical categories extracted from CCGbank, plus a small set of combinatory rules. The parser handles up to 30 sentences per second, and the taggers exceed 100,000 words per second.
  • Parser accuracy:
    • On CCGbank Section 23 predicate-argument dependencies: 84.8% labelled precision and 84.5% labelled recall.
    • On DepBank, using grammatical relations (GRs) in the Briscoe et al. (2006) style: 82.4% precision and 81.2% recall.
  • Perceptron training matches the log-linear models in accuracy and takes a few hours. Unlike the log-linear models, which need a cluster, it runs on one processor.
  • The tagger models and their training data:
    • The POS tagger exceeds 97% per-word accuracy on Penn Treebank Section 23.
    • The chunker is trained on CoNLL shared task data.
    • The NER is trained on MUC data.
  • Boxer implements a first-order fragment of DRT with a Neo-Davidsonian treatment of events and Van der Sandt-style presupposition handling. It outputs Prolog, a flat labelled format, XML, a readable box format, or first-order formulas.

Interesting ideas

  • I1. Adaptive supertagger–parser loop: the tagger starts with very few candidate categories per word and adds more only when the parser cannot build a full analysis. This "cheap first, widen on failure" strategy could speed up any pipeline that combines a pruning model with an expensive search.
  • I2. Multi-tagging with a threshold: the tagger computes per-word tag distributions with forward-backward and keeps several tags when it is uncertain. This lets later components recover from early errors without facing full ambiguity.
  • I3. Porting by annotating lexical categories: because CCG is lexicalized, a parser can be adapted to a new domain, such as questions, by quickly hand-labelling word-level categories rather than full trees. This is attractive for low-resource languages or domains where full treebanking is too costly.
  • I4. printf-like format strings for input and output: a small format language describes how fields are laid out, and extra columns pass through untouched. This makes the tools easy to plug into different corpora and pipelines.
  • I5. Semantics that feeds logic tools: translating DRSs into first-order logic lets off-the-shelf theorem provers or model builders check consistency or informativeness. This offers a route from parsing to reasoning and to checking the plausibility of extracted facts.

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

🎓University of OxfordOxford · United Kingdom52 works · 154,829 citations🎓The University of SydneySydney · Australia31 works · 67,940 citations🎓Sapienza University of RomeRome · Italy36 works · 8,842 citations

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 9,086 worksArtificial Intelligence · 8,889 worksArts and Humanities · 9,296 worksLanguage and Linguistics · 9,206 works

🧭 Topics 1Research subjects and languages the work is about.Syntax & grammar formalisms · 161 works

Cite

Cite as James Curran, Stephen Clark and Johan Bos. 2007. Linguistically Motivated Large-Scale NLP with C&C and Boxer. Proceedings of the 45th Annual Meeting of the Association for Computational Linguistics Companion Volume Proceedings of the Demo and Poster Sessions.

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

@inproceedings{curran2007linguistically,
  title = {{Linguistically Motivated Large-Scale NLP with C\&C and Boxer}},
  author = {James Curran, Stephen Clark and Johan Bos},
  booktitle = {Proceedings of the 45th Annual Meeting of the Association for Computational Linguistics Companion Volume Proceedings of the Demo and Poster Sessions},
  year = {2007},
  doi = {10.3115/1557769.1557781},
  url = {https://doi.org/10.3115/1557769.1557781}
}

References

20 references, 5 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Kisuh Ahn, Johan Bos, James R. Curran, Dave Kor, Malvina Nissim, and Bonnie Webber. 2005. Question answering with QED at TREC-2005. In Proceedings of TREC-2005. find
  2. Patrick Blackburn and Johan Bos. 2005. Representation and Inference for Natural Language. A First Course in Computational Semantics. CSLI. find
  3. Johan Bos. 2005. Towards wide-coverage semantic interpretation. In Proceedings of IWCS-6, pages 42–53, Tilburg, The Netherlands. find
  4. Ted Briscoe, John Carroll, and Rebecca Watson. 2006. The second release of the RASP system. In Proceedings of the Interactive Demo Session of COLING/ACL-06, Sydney.
  5. Stephen Clark and James R. Curran. 2004a. The importance of supertagging for wide-coverage CCG parsing. In Proceedings of COLING-04, pages 282–288, Geneva, Switzerland.
  6. Stephen Clark and James R. Curran. 2004b. Parsing the WSJ using CCG and log-linear models. In Proceedings of ACL-04, pages 104–111, Barcelona, Spain.
  7. Stephen Clark and James R. Curran. 2007a. Formalism-independent parser evaluation with CCG and DepBank. In Proceedings of the 45th Annual Meeting of the ACL, Prague, Czech Republic.
  8. Stephen Clark and James R. Curran. 2007b. Perceptron training for a wide-coverage lexicalized-grammar parser. In Proceedings of the ACL Workshop on Deep Linguistic Processing, Prague, Czech Republic. find
  9. Stephen Clark, Mark Steedman, and James R. Curran. 2004. Object-extraction and question-parsing using CCG. In Proceedings of the EMNLP Conference, pages 111–118, Barcelona, Spain. find
  10. Michael Collins. 2003. Head-driven statistical models for natural language parsing. Computational Linguistics, 29(4):589–637.
  11. James R. Curran and Stephen Clark. 2003a. Investigating GIS and smoothing for maximum entropy taggers. In Proceedings of the 10th Meeting of the EACL, pages 91–98, Budapest, Hungary. find
  12. James R. Curran and Stephen Clark. 2003b. Language independent NER using a maximum entropy tagger. In Proceedings of CoNLL-03, pages 164–167, Edmonton, Canada. find
  13. James R. Curran, Stephen Clark, and David Vadas. 2006. Multi-tagging for lexicalized-grammar parsing. In Proceedings of COLING/ACL-06, pages 697–704, Sydney. find
  14. Julia Hockenmaier. 2003. Data and Models for Statistical Parsing with Combinatory Categorial Grammar. Ph.D. thesis, University of Edinburgh. find
  15. H. Kamp and U. Reyle. 1993. From Discourse to Logic; An Introduction to Modeltheoretic Semantics of Natural Language, Formal Logic and DRT. Kluwer, Dordrecht. find
  16. Ron Kaplan, Stefan Riezler, Tracy H. King, John T. Maxwell III, Alexander Vasserman, and Richard Crouch. 2004. Speed and accuracy in shallow and deep stochastic parsing. In Proceedings of HLT and the 4th Meeting of NAACL, Boston, MA. find
  17. Takuya Matsuzaki, Yusuke Miyao, and Jun’ichi Tsujii. 2007. Efficient HPSG parsing with supertagging and CFG-filtering. In Proceedings of IJCAI-07, Hyderabad, India. find
  18. Guido Minnen, John Carroll, and Darren Pearce. 2001. Applied morphological processing of English. Natural Language Engineering, 7(3):207–223. find
  19. Adwait Ratnaparkhi. 1996. A maximum entropy part-of-speech tagger. In Proceedings of the EMNLP Conference, pages 133–142, Philadelphia, PA. find
  20. Mark Steedman. 2000. The Syntactic Process. The MIT Press, Cambridge, MA. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/curran-2007-linguistically-motivated-large-scale-nlp/