GDEX: Automatically Finding Good Dictionary Examples in a Corpus

What it is about

The authors built GDEX, a scorer that ranks corpus sentences as candidate dictionary examples for a given collocation. The features are sentence length (10 to 25 words preferred), word frequency (words outside the 17,000 most common are penalised), pronouns and anaphors (penalised), whether the target collocation sits in the main clause, whether the sentence is a whole sentence, and co-occurring "third collocates". Two students picked good examples for 1,000 collocations, and those choices set the feature weights by average rank. In the production run, GDEX gave lexicographers a shortlist of 20 sentences for each of 8,000 collocations for the Macmillan English Dictionary. The team first used the BNC, which was rejected as too old, then UKWaC, a 2-billion-word web corpus, where web lists and junk remained a problem. The authors call the exercise a success in speed.

Ideas

• I1. en/src/dict.rs or the reader of tools/gutenberg: port the sentence-level penalties (length window, rare-word share, pronoun share, whole-sentence test) as a Rust scorer for school reading texts. Measure the share of sentences that pass on a held-out book sample, and check the first two runs are identical.
• I2. Weight fitting with the two-fold check: tune the feature weights on one fold of hand-picked examples and test the rank of the picked sentences on the other fold (en folds). Report the mean rank of the chosen sentence in the shortlist of 20.
• I3. Parse-based feature: the paper's future-work idea is that a sentence the parser cannot analyse is a poor example. Use the en parser's score or its arc count per word as one feature, and check whether it lowers the rate of list-like or junk sentences in a web-style sample.

For Mova

The touch points are en/src/dict.rs for word data, tools/gutenberg for reading lessons, and en/src/parse.rs for the parse-based feature. The chapter is 19. The lexicon: vocabulary, multiword expressions and register, and 37. Register and variation for the register side. The paper's data is CC BY-NC-SA (pre-2016), so only the method is used; any corpus it names is test-only or not used at all. Measure first the length and frequency penalties on the school reading texts against a hand-rated sample.

Cite

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

@inproceedings{kilgarriff2008gdex,
  title = {{GDEX: Automatically Finding Good Dictionary Examples in a Corpus}},
  author = {Kilgarriff, Adam and Husák, Miloš and McAdam, Katy and Rundell, Michael and Rychlý, Pavel},
  booktitle = {Proceedings of the XIII EURALEX International Congress (Barcelona), 2008},
  year = {2008},
  url = {https://euralex.org/category/publications/euralex-2008/}
}

References

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

1. Atkins, S.; Rundell, M. (2008). Oxford Guide to Practical Lexicography. Oxford: Oxford University Press. find
2. Baroni, M.; Bernardini, S. (2004). BootCaT: Bootstrapping Corpora and Terms from the Web. Proceedings, Language Resources and Evaluation Conference. find
3. CEF (2001). Common European Framework of Reference for Languages [on-line]. URL: http://www.coe.int/t/dg4/linguistic/CADRE\_EN.asp ↗
4. Collins-Thompson, K.; Callan, J. (2004). A language-modelling approach to predicting reading difficulty. Proc NAACL-HLT. Boston. find
5. DuBay, W. (2004). The principles of readability. Costa Mesa: Impact Information. find
6. Ferraresi, A. et al. (2008-9). “Introducing and evaluating ukWaC, a very large web-derived corpus of English”. In Proceedings, 4 th WAC workshop, LREC, Marrakech, Morocco. find
7. Flesch, R. (1948). “A new readability yardstick.” Journal of Applied Psychology 32. 221-233. find
8. Kilgarriff, A. (2006). “Collocationality (and how to measure it)”. In Proc EURALEX, Torino, Italy. find
9. Kilgarriff, A.; Grefenstette, G. (2003). Introduction to a Special Issue on Web as Corpus. Computational Linguistics 29 (3). find
10. Kilgarriff, A. et al. (2004). “The Sketch Engine”. In Proc. EURALEX, Lorient, France. find
11. Kotani, K. et al. (2008). “ EFL Learner Reading Time Model for Evaluating Reading Proficiency ”. In Proc CICLING, Haifa, Israel. find
12. Laufer, B. (1992, 2007). “Corpus-based versus lexicographer examples in comprehension and production of new words”. In Proc EURALEX 1992, Helsinki. Reprinted in Fontenelle (ed.) 2007. Practical Lexicography: A Reader. Oxford University Press. find
13. Macmillan English Dictionary for Advanced Learners. 1 st and 2 nd ed.. London: Macmillan, 2002, 2007. find
14. Schwarm, S.; Ostendorf, M. (2005). “Reading Level Assessment using Support Vector Machines and statistical language models”. In Proc ACL, Ann Arbor, USA. find
15. Sinclair, J. et al. (1998). Language independent statistical software for corpus exploration. Computers and the Humanities 31. 229-255. find
16. Someya, Y. (2000). Word level checker: Vocabulary profiling program by AWK. [on-line] URL: http://www.cl.aoyama.ac.jp/english/newSite/wlc/index.html [Access date: 28 March 2008]. ↗

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/lex-2008-kilgarriff-gdex-examples/