Towards a Heuristic Categorization of Prepositional Phrases in English with WordNet

What it is about

The paper builds a rule-based system that finds prepositional phrases (PPs) in parsed text and then gives each one a semantic role from seven classes taken from the Penn Treebank: locative, temporal, directional, manner, purpose, extent and beneficiary. It is organised in levels. Level 0 is a small grammar that finds minimal PPs; Level 1 adds the role labels, using the preposition as a first guess and then the WordNet hypernyms of the head noun, which override the guess. The data are 27 Wall Street Journal documents, of which 20 (about 74%) are used to write the rules and 7 (about 26%) to test them. The final grammar reaches recall 87.8%, precision 85.0% and F 86.4% for PP finding. For the categorisation, the total F is 81.5%; directional PPs reach 89.0% and manner PPs only 63.1%.

Ideas

• I1. prag/src/roles.rs: write the role rules over UD trees, with the case marker (case relation) as the first cue and the lemma of the head noun's WordNet hypernym as the second, which overrides the first. Keep the rule table in BTreeMap order and log which rule chose each label, so each choice can be explained.
• I2. Report precision, recall and F for each role class separately, as the paper does, and keep a "left as bare PP" class, so that unsure cases are not forced into a wrong role.
• I3. Use the paper's two-layer design as an ablation: run with the preposition cue only, then with the hypernym layer added. The gain of the second layer is the number to publish.

For Mova

The chapter is 22. Phrases: noun, adjective, adverb and prepositional phrases, with 10. Word classes for prepositions. The code is prag/src/roles.rs for the role rules and en/src/dict.rs for the lexicon of prepositions. The measures: per-class P, R and F on a hand-labelled set of school sentences, a held-out split, and the layer ablation above. The paper's rule-based approach fits Mova; its PP-attachment part is left open there and would be a later step.

Cite

The paper itself, not this card. BibTeX from the library's record of the work. arXiv 1002.1095

@misc{rudzicz2010heuristic,
  title = {{Towards a Heuristic Categorization of Prepositional Phrases in English with WordNet}},
  author = {Rudzicz, Frank and Mokhov, Serguei A.},
  year = {2010},
  eprint = {1002.1095},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1002.1095}
}

References

8 references, 0 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. EAGLES96. Preliminary Recommendations on Semantic Encod- [OW02] T. O’Hara and J. Wiebe. Classifying Preposition Semantic Roles using Class-based Lexical ing Interim Report. 1996. URL: Associations. 2002. NMSU-CS-2002-001 http://www.ilc.cnr.it/EAGLES96 URL: www.cs.nmsu.edu/TechReports /rep/node13.html, Nov 23, 2003. /2002/013.pdf. ↗
3. I. Pratt and N. Francez. On the Seman-tactic Parsing Strategies. University of Con-tics of Temporal Prepositions and Preposition necticut, 1978. PhD thesis. Phrases. University of Manchester, 1997. find
4. N. Francez and M. Steedman. Categori- [Qui85] cal Grammar and the Semantics of Indexical Prepositional Phrases. Human Communication Research Centre, University of Edinburgh, 2003. find
5. R.A. Hudson. Phrases. November 2003. www.phon.ucl.ac.uk/home/dick /tta/phrases/phrases.htm#pp. find
7. (Eds.). Holt Rinehart and Winston Inc„ 1973. 346-65. find
8. Daniel S. Jurafsky and James H. Martin. [Whi90] G. et al Whittemore. Empirical Study of Pre-Speech and Language Processing. Prentice-dictive Powers of Simple Attachment Schemes Hall, Inc., Pearson Higher Education, Upper for Post-Modifier Prepositional Phrases in Pro-Saddle River, New Jersey 07458, 2000. ISBN ceedings of the 28th Annual Meeting of the As-0-13-095069-6. sociation for Computation Linguistics. 1990. pages 23-30. find
9. J. Kimball. Seven Principals of Surface Structure Parsing in Natural Language. MIT Press, 1973. Cognition 2:15-47. 5 Appendix find
10. B. Levin and S. Pinker. Lexical and Concep- 5.1 Appendix A: PP Rule Set tual Semantics. Elsevier Science Publications, The rule set is quite small and simple provided the rest of Amsterdam, 1991. the grammar has a quite a notion of what NP and VP are. C.D. Manning and H. Schutze. Foundations of This basic rule set is used for Level 0 of our work. This Statistical Natural Language Processing. MIT gives us a list of unclassified PPs from a corpus. Later on, Press, 2002. a Perl script run to augment this information with the 7 find

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/rudzicz-2010-towards-heuristic-categorization-prepositional-phrases/