A Labelled Corpus for Prepositional Phrase Attachment

What it is about

The paper describes a training corpus for prepositional phrase (PP) attachment, built from the Penn Treebank II Wall Street Journal text. The authors derive attachment points from the head rules of Collins and Brooks, and they find 117,822 PPs in the treebank and over 124,500 ambiguous PP instances in all, grouped into patterns of potential ambiguity. The canonical pattern (verb group, noun phrase, PP) has 22,816 instances. The corpus adds lexical and phrasal distances, and phrase function tags, which about 45% of the PPs carry. A decision tree (ID3 with gain ratio) runs on the five patterns with stratified 10-fold cross-validation. Earlier best models on the canonical form reach up to about 85% accuracy, and the authors find no significant gain from replacing names and numbers with keywords.

Ideas

• I1. en/src/parse.rs: add a distance feature for each PP attachment candidate, binned as near, medium and far (1–2, 3–6 and 7+ tokens, as the paper suggests). Measure the current parser's attachment accuracy on EWT test split by bin. Add the bin to the perceptron only if the far bin is clearly weaker, and decide on the two-fold check.
• I2. A pattern census in en parse-eval: count the PP ambiguity patterns (verb group with noun phrase and PP, two PPs, adjective phrase and PP) in EWT and report the attachment accuracy for each one. This shows where the canonical verb-or-noun choice is not the main error.
• I3. Test-only benchmark: attachment accuracy of the en chooser on PTB-derived PP instances, with the 85% of earlier models as a reference. The corpus is derived from the Penn Treebank, so it is never used to train anything Mova builds.

For Mova

The code is en/src/parse.rs and the en parse-eval command. The chapter is 22. Phrases: noun, adjective, adverb and prepositional phrases, with 23-complementation.md for the verb-or-noun choice. The corpus is PTB-derived: test only, never training data for the build. Table 2 of the paper gives per-pattern accuracies, but the text extraction is garbled, so the figures should be read from the PDF before they are quoted.

Topics: Treebanks

Cite

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

@inproceedings{mitchell2004labelled,
  title = {{A Labelled Corpus for Prepositional Phrase Attachment}},
  author = {Mitchell, Brian and Gaizauskas, Robert},
  booktitle = {Proceedings of the Fourth International Conference on Language Resources and Evaluation (LREC'04)},
  year = {2004},
  url = {https://aclanthology.org/L04-1443/}
}

References

13 references, 3 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Eric Brill and Philip Resnik. 1994. A Rule-Based Approach to Prepostional Phrase Attachment Disambiguation. In Procs. of COLING-94. find
2. Michael Collins and James Brooks. 1995. Prepositional Phrase Attachment through a Backed-Off Model. In Procs. of the Third Workshop on Very Large Corpora. find our card
3. Thomas G. Dietterich. 1998. Approximate Statistical Tests for Comparing Supervised Classification Learning Algorithms. Neural Computation, 10(7):1895–1923. find
4. Donald Hindle and Mats Rooth. 1993. Structural Ambiguity and Lexical Relations. Computational Linguistics, 19(1):103–120. find
5. David M. Magerman. 1994. Natural Language Parsing as Statistical Pattern Recognition. Ph.D. thesis, Stanford University. find
6. Mitchell P. Marcus, Grace Kim, Mary A. Marcinkiewicz, Robert MacIntyre, Ann Bies, Mark Ferguson, Karen Katz, and Britta Schasberger. 1994. The Penn Treebank: Annotating Predicate Argument Structure. In Procs. of the Human Language Technology Workshop. find our card
7. Paola Merlo, Matthew W. Crocker, and Cathy Berthouzoz. 1997. Attaching Multiple Prepositional Phrases: Generalized Backed-off Estimation. In Procs. of Second Conference on Empirical Methods in Natural Language Processing. find
8. Brian Mitchell and Robert J. Gaizauskas. 2002. A Comparison of Machine Learning Algorithms for Prepositional Phrase Attachment. In Procs. of LREC-2002, Las Palmas de Gran Canaria. find
9. Tom M. Mitchell. 1997. Machine Learning. McGraw-Hill. find
10. Brian Mitchell. 2004. Prepositional Phrase Attachment using Machine Learning Algorithms. Ph.D. thesis, University of Sheffield. find
11. J. Ross Quinlan. 1986. Induction of Decision Trees. Machine Learning, 1(1):81–106. find
12. Adwait Ratnaparkhi, Jeff Reynar, and Salim Roukos. 1994. A Maximum Entropy Model for Prepositional Phrase Attachment. In Procs. of the ARPA Workshop on Human Language Technology. find our card
13. Jakub Zavrel, Walter M. Daelemans, and Jorn Veenstra. 1997. Resolving PP Attachment Ambiguities with Memory-Based Learning. In Procs. of CoNLL 1997. find

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/lingbulk-2004-mitchell-labelled-corpus-prepositional-phrase-attachment/