LinGO Redwoods: A Rich and Dynamic Treebank for HPSG

What it is about

Not read: this card is written from what is reliably known about the paper, and every idea is marked to check against the text. It is the journal account of the Redwoods treebank, whose 2002 project note has its own card (The LinGO Redwoods Treebank: Motivation and Preliminary Applications). Instead of drawing trees by hand, annotators pick the correct analysis among all the parses of the English Resource Grammar (ERG) by answering binary discriminants: small properties that split the parse forest. The recorded decisions make the treebank dynamic: when the grammar changes, the sentences are reparsed and the old decisions are replayed, so only the sentences where they no longer decide need human attention. One stored analysis yields several views: the full derivation, a phrase-structure tree, the semantics (MRS) and dependencies. The paper also reports parse-selection experiments with statistical models trained on the treebank.

Ideas

• I1. A dynamic silver store for annot: keep the teacher's decisions as discriminants (attachment of this phrase, label of this edge) rather than whole trees, replay them after every en change, and count how many still decide uniquely. The count is a regression measure for the parser. (to check: how the paper reports the share of decisions that survive an update)
• I2. Several views from one store: export the same analysis as UD, as frames and as the world's facts, and test that each view is a function of the stored analysis (two runs, byte-identical). (to check: which views the paper lists)
• I3. Parse selection with features of the derivation, not only of the surface: rank en's beam alternatives with features from the frames reader, and measure exact-match accuracy by ambiguity bucket. (to check: the features and numbers reported)

For Mova

The code is annot (silver data and gates), en/src/parse.rs (beam), tools/treeagree and treebanks. The chapter is 20. Syntax: the Universal Dependencies frame. The ERG and Redwoods have their own licences; check them in Data and model licences before any use beyond tests.

Topics: Treebanks

Cite

The paper itself, not this card. BibTeX from the library's record of the work. DOI 10.1007/s11168-004-7430-4

@inproceedings{oepen2004lingo,
  title = {{LinGO Redwoods: A Rich and Dynamic Treebank for HPSG}},
  author = {Oepen, Stephan and Flickinger, Dan and Toutanova, Kristina and Manning, Christopher D.},
  booktitle = {Research on Language and Computation 2(4):575–596},
  year = {2004},
  doi = {10.1007/s11168-004-7430-4},
  url = {https://doi.org/10.1007/s11168-004-7430-4}
}

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/enprag-2004-oepen-lingo-redwoods-rich-dynamic-treebank/