A Factored Functional Dependency Transformation of the English Penn Treebank for Probabilistic Surface Generation

What it is about

The authors convert the English Penn Treebank (about one million words) from phrase structure into a functional dependency corpus. Each word gets over three dozen features: one of 33 surface roles, a head, a factored part of speech, a base form, a grouping of clause, noun phrase or other, and junction data for coordination. Prepositions and complementizers are treated as role markers hanging on the word to their left, not as heads of phrases. The corpus was built for a probabilistic surface realizer, and it is tested by regenerating the sentences of Section 23 from the unordered structure. The result is 98.8% simple string accuracy and 87.1% exact match over 2,416 sentences. The paper also argues from bigram and trigram counts that "came in first" is preferred to "came to first" (counts 21 against 1 for the trigram).

Ideas

• I1. Role markers in en/src/parse.rs: a preposition is a dependent of its left-hand word with a case-style role, and its head is the noun to its right. Try this as a lexicon-driven attachment rule and measure LAS/UAS on UD EWT dev, run twice for determinism.
• I2. Preposition choice by counts: store trigram counts of (verb, preposition, noun) in a BTreeMap table built from the lessons and use it in en/src/lin.rs to pick between "in" and "to" in generated text. A negative control: swap the two counts and the choice must flip.
• I3. Exact-match regeneration as the measure for tools/roundtrip: take a tree, remove the word order, linearize it back, and report string accuracy and exact match, separately for each layer.
• I4. Grouping rules as data: the paper counts how often it nests coordinated noun phrases (6,030) and compound prepositions (3,985). Write these groupings as explicit rules in en/src/parse.rs and log how many times each fires on the school texts, so each change is traceable.
• I5. POS correction table (about 3,255 corrections over 20 kinds, e.g. a closed-class tag fixed by its parent's category): add a small corrections table in en/src/tag.rs, each row with its frequency on the training texts.

For Mova

Prepositions and role markers: 22. Phrases: noun, adjective, adverb and prepositional phrases and 23. Complementation. Coordination nesting: 24. Coordination and subordination. Word order from the tree: 29. Ellipsis, word order and information structure, with en/src/lin.rs and tools/roundtrip. The paper's data (the Treebank) is not ours to build on; use only the ideas, and test with our own UD data.

Topics: Treebanks

Cite

The paper itself, not this card. BibTeX from the library's record of the work. DOI 10.63317/3n5vxgc579gz

@inproceedings{langkildegeary2006factored,
  title = {{A Factored Functional Dependency Transformation of the English Penn Treebank for Probabilistic Surface Generation}},
  author = {Langkilde-Geary, Irene and Betteridge, Justin},
  booktitle = {Proceedings of the Fifth International Conference on Language Resources and Evaluation (LREC'06)},
  year = {2006},
  doi = {10.63317/3n5vxgc579gz},
  url = {https://doi.org/10.63317/3n5vxgc579gz}
}

References

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

1. S. Bangalore and O. Rambow. 2000a. Using tag, a tree model, and a language model for generation. In Proc. of 1st INLG. find
2. A. Cahill, M. McCarthy, J. van Genabith, and A. Way. 2002. Automatic annotation of the penn-treebank with lfg f-structure information. In Proc. LREC Workshop on Linguistic Knowledge Acquisition and Representation - Bootstrapping Annotated Language Data. find
3. E. Charniak. 2000. A maximum-entropy-inspired parser. In Proc. NAACL. find
4. D. Klein and C. Manning. 2003. Accurate unlexicalized parsing. In Proc. ACL. find our card
5. I. Langkilde-Geary. 2002. An empirical verification of coverage and correctness for a general-purpose sentence generator. In Proc. INLG. http://www.isi.edu/licensed-sw/halogen/verifcc.ps. ↗
6. M. Marcus, G. Kim, M. Marcinkiewicz, R. MacIntyre, A. Bies, M. Ferguson, K. Katz, and B. Schasberger. 1994. The Penn treebank: Annotating predicate argument structure. In ARPA Human Language Technology Workshop. find our card
7. S. Oepen, K. Toutanova, S. Shieber, C. Manning, D. Flickinger, and T. Brants. 2002. The lingo redwoods treebank: Motivation and preliminary applications. In Proc. COLING. 0. find our card
8. Rambow, C. Creswell, R. Szekely, H. Taber, and M. Walker. 2002. A dependency treebank for english. In LREC. find our card
9. E. Ringger, M. Gamon, R. Moore, D. Rojas, M. Smets, and S. Corston-Oliver. 2004. Linguistically informed statistical models of constituent structure for ordering in sentence realization. In Proc. Coling. find

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/geary-2006-factored-functional-dependency-transformation-english/