Elamur Research⌕
☰
home / papers / 2003 / A Plethora of Methods for Learning English Countability

A Plethora of Methods for Learning English Countability

Paper · 2003 · Timothy Baldwin and Francis Bond

Summary

The paper addresses deep lexical acquisition: assigning English nouns to countability classes (countable, uncountable, plural only, bipartite) from corpus evidence. Its aim is a general recipe for classifying words with a fixed set of linguistic diagnostics. The authors extract 10 feature clusters (206 unit features) from the written BNC using three independent preprocessors: an fnTBL POS tagger, an fnTBL chunker and the RASP dependency parser. They test two feature representations. The distribution-based one uses relative frequencies computed in several ways. The agreement-based one measures how often the preprocessors agree token by token on each diagnostic. They also compare one multiclass classifier against a suite of binary classifiers, and propose a feature selection method. Gold labels come from COMLEX 3.0 and the ALT-J/E transfer dictionary. The best setup is a distribution-based suite of binary classifiers that pools all three preprocessors. On open data its agreement with hand annotation (about 92.4%) is close to the agreement between the two reference lexicons themselves. This shows that rich lexical properties can be learned reliably from shallow and deep processing output.

Key points

  • Four countability classes are used, and a noun may belong to several. Gold data merges COMLEX 3.0 and ALT-J/E, whose mutual agreement is 93.8%.
  • Nouns were kept only if they appeared at least 10 times as NP heads in the output of all three preprocessors. This gave 20,530 nouns, of which 9,031 have gold labels and form the evaluation set.
  • The 206 unit features expand into 1,284 distribution-based values. The agreement-based representation covers 53 class-indicative diagnostics.
  • With all features, the distribution-based suite (All MEAN) reaches accuracy .937 / F .959 on countable nouns and .954 / F .895 on uncountable nouns.
    • Plain binary-occurrence features do significantly worse.
    • Feature selection gives roughly a 30× speed-up with no significant F-score loss (N > 100 for countable, N > 50 for uncountable).
  • Earlier approaches scored lower: semantic class from ALT-J/E predicts countability 78% of the time, and a Cyc-based method reaches 89.5%.

Interesting ideas

  • I1. Several independent preprocessors (tagger, chunker, parser) supply the same diagnostics, and the token-level agreement between them becomes a feature that separates reliable signal from tool-specific noise.
  • I2. Each raw count is normalised three ways: against the whole corpus, against the word's own frequency, and against the other options in the same feature cluster. This beat simple presence/absence features by a significant margin.
  • I3. Selected top-N features are compared with randomly chosen N features as a control. The gap is significant for all N < 1000, which shows the ranking matters and is not just an effect of having fewer features.
  • I4. Classifier output is judged against the agreement between two independent human-built lexicons, which gives a realistic ceiling instead of treating either lexicon as perfect truth.
  • I5. Countability is treated as a lexical property that shapes which determiners, quantifiers and agreement patterns a noun accepts, and one noun can carry different senses with different classes.

For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

  • I1 (learn): Agreement between witness parsers on a given construction could serve as a per-token confidence score when choosing silver data, rather than only a yes/no filter.
  • I2 (tokens): Lexicon properties such as noun countability or verb transitivity could be induced from parsed corpora using these multi-view relative frequencies.
  • I3 (learn): Any pruning of features or repair rules could be checked against a random-subset negative control of the same size.
  • I4 (learn): Evaluations of compiled knowledge could report the agreement between existing resources (e.g., WordNet vs. other lexicons) as the upper bound.
  • I5 (world): Storing learned countability per noun would let the world model decide whether "papers" means individual countable documents or "paper" means a mass substance, which helps quantity reasoning in maths word problems.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓Stanford UniversityStanford · United States434 works · 314,419 citations🏢NTTTokyo · Japan26 works · 5,083 citations

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 works

🧭 Topics 1Research subjects and languages the work is about.English · 2,263 works

Cite

Cite as Timothy Baldwin and Francis Bond. 2003. A Plethora of Methods for Learning English Countability. Proceedings of the 2003 Conference on Empirical Methods in Natural Language Processing.

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

@inproceedings{baldwin2003plethora,
  title = {{A Plethora of Methods for Learning English Countability}},
  author = {Timothy Baldwin and Francis Bond},
  booktitle = {Proceedings of the 2003 Conference on Empirical Methods in Natural Language Processing},
  year = {2003},
  doi = {10.3115/1119355.1119365},
  url = {https://doi.org/10.3115/1119355.1119365}
}

References

16 references, 5 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Timothy Baldwin and Francis Bond. 2003. Learning the countability of English nouns from corpus data. In Proc. of the 41st Annual Meeting of the ACL, Sapporo, Japan. (to appear).
  2. Francis Bond and Caitlin Vatikiotis-Bateson. 2002. Using an ontology to determine English countability. In Proc. of the 19th International Conference on Computational Linguistics (COLING 2002), Taipei, Taiwan.
  3. Ted Briscoe and John Carroll. 2002a. High precision extraction of grammatical relations. In Proc. of the 19th International Conference on Computational Linguistics (COLING 2002), pages 134–140, Taipei, Taiwan. find
  4. Ted Briscoe and John Carroll. 2002b. Robust accurate statistical annotation of general text. In Proc. of the 3rd International Conference on Language Resources and Evaluation (LREC 2002), pages 1499–1504, Las Palmas, Canary Islands.
  5. Lou Burnard. 2000. User Reference Guide for the British National Corpus. Technical report, Oxford University Computing Services. find
  6. Jean Carletta. 1996. Assessing agreement on classification tasks: the kappa statistic. Computational Linguistics, 22(2):249–254. ↗
  7. Walter Daelemans, Jakub Zavrel, Ko van der Sloot, and Antal van den Bosch. 2002. TiMBL: Tilburg memory based learner, version 4.2, reference guide. ILK technical report 02-01. find
  8. Ralph Grishman, Catherine Macleod, and Adam Myers, 1998. COMLEX Syntax Reference Manual. Proteus Project, NYU. (http://nlp.cs.nyu.edu/comlex/refman.ps). ↗
  9. Satoru Ikehara, Satoshi Shirai, Akio Yokoo, and Hiromi Nakaiwa. 1991. Toward an MT system without pre-editing – effects of new methods in ALT-J/E–. In Proc. of the Third Machine Translation Summit (MT Summit III), pages 101– 106, Washington DC, USA. find
  10. Huan Liu and Hiroshi Motoda. 1988. Feature Extraction, Construction and Selection: A Data Mining Perspective. Kluwer Academic Publishers. find
  11. Guido Minnen, John Carroll, and Darren Pearce. 2001. Applied morphological processing of English. Natural Language Engineering, 7(3):207–23. ↗
  12. Grace Ngai and Radu Florian. 2001. Transformation-based learning in the fast lane. In Proc. of the 2nd Annual Meeting of the North American Chapter of Association for Computational Linguistics (NAACL2001), pages 40–7, Pittsburgh, USA. find
  13. Tom O’Hara, Nancy Salay, Michael Witbrock, Dave Schneider, Bjoern Aldag, Stefano Bertolo, Kathy Panton, Fritz Lehmann, Matt Smith, David Baxter, Jon Curtis, and Peter Wagner. 2003. Inducing criteria for mass noun lexical mappings using the Cyc KB and its extension to WordNet. In Proc. of the Fifth International Workshop on Computational Semantics (IWCS-5), Tilburg, the Netherlands. find
  14. Lane O.B. Schwartz. 2002. Corpus-based acquisition of head noun countability features. Master’s thesis, Cambridge University, Cambridge, UK. find
  15. Erik F. Tjong Kim Sang and Sabine Buchholz. 2000. Introduction to the CoNLL-2000 shared task: Chunking. In Proc. of the 4th Conference on Computational Natural Language Learning (CoNLL-2000), Lisbon, Portugal.
  16. Allan P. White and Wei Zhong Liu. 1994. Bias in information-based measures in decision tree induction. Machine Learning, 15(3):321–9. ↗

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/baldwin-2003-plethora-methods-learning-english-countability/