Elamur Research⌕
☰
home / papers / 2002 / Using an Ontology to Determine English Countability

Using an Ontology to Determine English Countability

Paper · 2002 · Francis Bond and Caitlin Vatikiotis-Bateson

Summary

The paper asks how far the count/mass status of English nouns follows from their meaning. This matters for generating English from languages such as Japanese or Chinese, which do not mark number, so the generator must choose plurals and determiners itself. The authors use the common-noun part of the ALT-J/E Japanese-to-English transfer dictionary (71,833 pairs). Each entry is tagged with semantic classes drawn from a 2,710-node ontology (later published as Goi-Taikei) and assigned one of five noun countability preferences (NCPs), which range from fully countable, through strongly and weakly countable, down to uncountable, with plural-only as a fifth category. They predict an entry's NCP from its combination of semantic classes. Countability was never used when building the ontology, which makes this a fair test. About 78% of nouns get the correct countability. The authors estimate that cleaning lexicon errors would raise this towards 85%. The result shows that countability is largely, though not fully, motivated by semantics. Practical uses include guessing countability for unknown words and flagging inconsistent dictionary entries.

Key points

  • Resources: the ALT-J/E transfer dictionary (71,833 Japanese–English noun pairs; 41,285, or 57.4%, are English multi-word expressions) and an is-a/has-a hierarchy of 2,710 semantic classes.
  • Labels: five NCPs from Bond et al. (1994), with codes CO, BC, BU, UC and PT (e.g. knife, cake, beer, furniture, scissors). The subclasses are collapsed: collectives count as fully countable and semi-countables as uncountable.
  • Accuracy and errors:
    • Semantics predicts countability about 78% of the time.
    • Error analysis suggests an upper bound near 85% if lexicon errors were removed.
    • Pair plural-only nouns (e.g. tongs) were almost all wrong, because shape is absent from the ontology. They make up under 3% of entries.
  • Single- vs multi-word entries:
    • Single-word entries turned out harder than multi-word ones. They have more semantic classes per entry (1.38 vs 1.34) and fewer training entries per class combination (3.7 vs 5.1).
    • Training separate models for single- and multi-word entries and then combining them improved the score in that experiment to 72.0%.

Interesting ideas

  • I1. Testing whether a grammatical property is semantically motivated, using an ontology built without reference to that property, yields an honest measure of how motivated the property is.
  • I2. A checker that warns when an entry's semantic class predicts a different label than the one assigned exposes lexicographer errors, misplaced hierarchy nodes and swapped sense labels.
  • I3. Chaining two predictions, first guessing an unknown word's semantic class and then deriving its countability from that class, gives a fallback for unknown words that beats a plain default.
  • I4. Linking countability to individual senses (rabbit as animal = count, as meat = mass) rather than to the whole entry would let countability be fixed during sense disambiguation.
  • I5. Separating subgroups that behave differently within the same semantic class, such as single-word vs multi-word entries, can beat pooled training despite having less data for each subgroup.

For Mova

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

  • I1 (world): We could measure how well WordNet/Wikidata classes predict features such as countability or measure units before trusting them as compiled knowledge.
  • I2 (learn): Build a consistency auditor that flags lexicon or silver-data entries whose class-based prediction disagrees with their stored feature, as a source of repair candidates.
  • I3 (tokens): Morphology could assign a provisional count/mass feature to out-of-vocabulary nouns via their guessed ontology class, keeping the reason traceable.
  • I4 (back): The generator could pick plural forms, "much/many" and classifiers like "a piece of" based on the sense chosen in the meaning graph.
  • I5 (learn): When inducing feature rules from classes, try stratifying by expression type and check gains against a pooled baseline.

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

🏢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 2Research subjects and languages the work is about.English · 2,263 worksKnowledge graphs & ontologies · 192 works

Cite

Cite as Francis Bond and Caitlin Vatikiotis-Bateson. 2002. Using an Ontology to Determine English Countability. COLING 2002: The 19th International Conference on Computational Linguistics.

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

@inproceedings{bond2002using,
  title = {{Using an Ontology to Determine English Countability}},
  author = {Francis Bond and Caitlin Vatikiotis-Bateson},
  booktitle = {COLING 2002: The 19th International Conference on Computational Linguistics},
  year = {2002},
  doi = {10.3115/1072228.1072280},
  url = {https://doi.org/10.3115/1072228.1072280}
}

References

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

  1. Keith Allan. 1980. Nouns and countability. Language, 56(3):541–67. ↗
  2. Francis Bond and Satoru Ikehara. 1996. When and how to disambiguate? — countability in machine translation —. In International Seminar on Multimodal Interactive Disambiguation: MIDDIM-96, pages 29–40, Grenoble. (Reprint of MIDDIM-1996). find
  3. Francis Bond and Kentaro Ogura. 1998. Reference in Japanese-to-English machine translation. Machine Translation, 13(2–3):107–134. find
  4. Francis Bond and Kyonghee Paik. 1997. Classifying correspondence in Japanese and Korean. In 3rd Pacific Association for Computational Linguistics Conference: PACLING-97, pages 58–67. Meisei University, Tokyo, Japan. find
  5. Francis Bond, Kentaro Ogura, and Satoru Ikehara. 1994. Countability and number in Japanese-to-English machine translation. In 15th International Conference on Computational Linguistics: COLING-94, pages 32–38, Kyoto. (http://xxx.lanl.gov/abs/cmp-lg/9511001). ↗
  6. Francis Bond. 2001. Determiners and Number in English contrasted with Japanese — as exemplified in Machine Translation. Ph.D. thesis, University of Queensland, Brisbane, Australia. ↗
  7. Ted Briscoe and Ann Copestake. 1999. Lexical rules in constraint-based grammars. Computational Linguistics, 25(4):487–526. find
  8. Ann Copestake. 1992. The Representation of Lexical Semantic Information. Ph.D. thesis, University of Sussex, Brighton. find
  9. Terumasa Ehara and Hozumi Tanaka. 1993. Kikaihonyaku-ni-okeru shizengengo shori (natural language processing in machine translation). Journal of Information Processing Society of Japan, 34(10):1266–1273. (in Japanese). find
  10. Christine Fellbaum, editor. 1998. WordNet: An Electronic Lexical Database. MIT Press. find
  11. 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 Third Machine Translation Summit: MT Summit III, pages 101–106, Washington DC. (http://xxx.lanl.gov/abs/cmp-lg/9510008). ↗
  12. Satoru Ikehara, Satoshi Shirai, Akio Yokoo, Francis Bond, and Yoshie Omi. 1995. Automatic determination of semantic attributes for user defined words in Japanese-to-English machine translation. Journal of Natural Language Processing, 2(1):3–17. (in Japanese). ↗
  13. Satoru Ikehara, Masahiro Miyazaki, Satoshi Shirai, Akio Yokoo, Hiromi Nakaiwa, Kentaro Ogura, Yoshifumi Ooyama, and Yoshihiko Hayashi. 1997. Goi-Taikei — A Japanese Lexicon. Iwanami Shoten, Tokyo. 5 volumes/CDROM. find
  14. Mutsumi Imai and Dedre Gentner. 1997. A crosslinguistic study of early word meaning: Universal ontology and linguistic influence. Cognition, 62:169–200. find
  15. Ray Jackendoff. 1991. Parts and boundaries. In Beth Levin and Steven Pinker, editors, Lexical and Conceptual Semantics, pages 1–45. Blackwell Publishers, Cambridge, MA & Oxford, UK. ↗
  16. Kazumi Kawamura, Yasuhiro Katagiri, and Masahiro Miyazaki. 1995. Multi-dimensional thesaurus wth various facets,. In IEICE Technical Report NLC94-48, pages 33–40. (in Japanese). find
  17. Melanie Siegel. 1996. Definiteness and number in Japanese to German machine translation. In D. Gibbon, editor, Natural Language Processing and Speech Technology, pages 137–142. Mouton de Gruyter, Berlin. find
  18. Anna Wierzbicka. 1988. The Semantics of Grammar. John Benjamins, Amsterdam. ↗

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/bond-2002-ontology-determine-english-countability/