Elamur Research⌕
☰
home / papers / 2023 / Semantic Parsing and Sense Tagging the Princeton WordNet Gloss Corpus

Semantic Parsing and Sense Tagging the Princeton WordNet Gloss Corpus

Paper · 2023 · Alexandre Rademaker, Abhishek Basu and Rajkiran Veluri

Summary

The Princeton Annotated Gloss Corpus (GlossTag) links words in WordNet definitions and examples to their intended WordNet senses. The 2008 release, however, left about 207 thousand word forms untagged and was never part of the official WordNet 3.0 distribution. This paper presents GlossTag 2022, which continues a manual effort restarted in 2019 to finish that annotation.

The authors first cleaned the data. They checked that each gloss can be rebuilt from its token markup and matches the WordNet 3.0 database, repaired mismatches, split glosses into separate definition and example sentences, and moved quotation attributions into metadata. Next, every sentence was parsed with ERG (English Resource Grammar), an HPSG-based grammar built for wide coverage of English, and sense tags were shifted from surface tokens onto the semantic predicates ERG produces. This makes annotation more consistent, especially for verbs.

They also tested the UKB WSD algorithm as a source of annotator hints and silver data, and analysed what blocks migration to WordNet 3.1. A fully sense-tagged gloss corpus matters because it densely interconnects synsets, pushing WordNet toward a lightweight ontology.

Key points

  • GlossTag 2008 still had 206,711 words (including collocations and multi-word expressions) without sense tags; the 2019 and 2022 releases resume that work under Princeton's open license, at github.com/own-pt/glosstag.
  • Data was converted from XML to JSON-Lines, one object per gloss. Roughly 2100 glosses could not be reconstructed from their token separators and were fixed, and about 758 examples that are attributed quotations had their sources moved to metadata.
  • Sentences are parsed by ERG, and sense annotations are projected from words onto ERG predicates. Tagging every gloss word could raise synset connectivity by about an order of magnitude.
  • Upgrading to WordNet 3.1 is difficult because some sense keys were reused with changed meanings. For example, pluto%1:17:00:: went from planet to asteroid, and new senses such as technology%1:06:00:: may fit old instances better. For now, GlossTag 2022 stays on WordNet 3.0.
  • The authors define "sense stability": a sense counts as stable if its synset keeps the same members across WordNet versions. When a sense moves to a different synset, all senses of its old synset become unstable.

Interesting ideas

  • I1. Sense labels are attached to the semantic predicates produced by a deep grammar rather than to raw tokens, so verb senses line up with the argument structure they occur in.
  • I2. A round-trip check (rebuild the original text from the annotated tokens and separators, then compare it with the source) catches silent tokenization and markup corruption.
  • I3. A coarse-to-fine sense choice first picks a VerbNet class by syntactic behaviour and only then the WordNet sense inside it (e.g., 52 senses of 'make' fall into six classes).
  • I4. Sense stability between resource versions is judged by whether synset membership is preserved, and annotations whose keys changed meaning are flagged for review instead of being copied blindly.
  • I5. An automatic WSD system (UKB) is evaluated on already-tagged data to judge whether it is good enough to give hints and produce intermediate silver releases.

For Mova

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

  • I1 (frames): Mova could anchor WordNet/VerbNet sense choices on frame predicates rather than tokens, so a sense decision is tied to the exact who-did-what-to-whom structure and can be explained through it.
  • I2 (tokens): The en tokenizer could enforce that tokens plus spacing attributes reproduce the input exactly, flagging any gloss or treebank sentence that fails before it enters training.
  • I3 (frames): Verb sense selection in prag could first fix the VerbNet class from the UD tree's valence pattern, which shrinks the candidate senses and gives a readable reason for the final pick.
  • I4 (world): When updating the compiled WordNet ontology, Mova could run a stability diff and quarantine knowledge-graph facts tied to unstable or redefined sense keys.
  • I5 (learn): Disambiguated WordNet glosses could serve both as a gold check for Mova's sense tagging and as a seed for silver data, with the 'tennis problem' style links supplying extra world knowledge.

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

🏢IBMArmonk · United States154 works · 72,606 citations

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksArts and Humanities · 17,450 worksLanguage and Linguistics · 17,296 works

🧭 Topics 1Research subjects and languages the work is about.Semantic parsing & meaning representation · 193 works

Cite

Cite as Alexandre Rademaker, Abhishek Basu and Rajkiran Veluri. 2023. Semantic Parsing and Sense Tagging the Princeton WordNet Gloss Corpus. Proceedings of the 12th Global Wordnet Conference.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2023.gwc-1.30

@inproceedings{rademaker2023semantic,
  title = {{Semantic Parsing and Sense Tagging the Princeton WordNet Gloss Corpus}},
  author = {Alexandre Rademaker, Abhishek Basu and Rajkiran Veluri},
  booktitle = {Proceedings of the 12th Global Wordnet Conference},
  year = {2023},
  doi = {10.18653/v1/2023.gwc-1.30},
  url = {https://doi.org/10.18653/v1/2023.gwc-1.30}
}

References

31 references, 3 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Eneko Agirre, Oier Lopez De Lacalle, Aitor Soroa, and Informatika Fakultatea. 2009. Knowledge-based wsd and specific domains: Performing better than generic supervised wsd. In IJCAI, pages 1501–1506. https://github.com/own-pt/sensetion.el ↗
  2. Eneko Agirre, Oier López de Lacalle, and Aitor Soroa. 2018. The risk of sub-optimal use of open source nlp software: Ukb is inadvertently state-of-the-art in knowledge-based wsd. find
  3. Eneko Agirre and Aitor Soroa. 2009. Personalizing pagerank for word sense disambiguation. In Proceedings of the 12th Conference of the European Chapter of the Association for Computational Linguistics, pages 33–41. Association for Computational Linguistics. find
  4. Satanjeev Banerjee and Ted Pedersen. 2002. An adapted lesk algorithm for word sense disambiguation using wordnet. In Computational Linguistics and Intelligent Text Processing, pages 136–145, Berlin, Heidelberg. Springer Berlin Heidelberg. find
  5. Pierpaolo Basile, Marco de Gemmis, Anna Lisa Gentile, Pasquale Lops, and Giovanni Semeraro. 2007. Uniba: Jigsaw algorithm for word sense disambiguation. In Proceedings of the Fourth International Workshop on Semantic Evaluations (SemEval-2007), pages 398–401, Prague, Czech Republic. Association for Computational Linguistics. find
  6. Michele Bevilacqua, Tommaso Pasini, Alessandro Raganato, Roberto Navigli, et al. 2021. Recent trends in word sense disambiguation: A survey. In Proceedings of the Thirtieth International Joint Conference on Artificial Intelligence, IJCAI-21. International Joint Conference on Artificial Intelligence, Inc. find
  7. Peter Clark, Christiane Fellbaum, and Jerry Hobbs. 2008a. Using and extending wordnet to support question-answering. In Proceedings of the 4th Global Wordnet Conference, pages 111–119, Hungary. find
  8. Peter Clark, Christiane Fellbaum, Jerry R Hobbs, Phil Harrison, William R Murray, and John Thompson. 2008b. Augmenting wordnet for deep understanding of text. In Semantics in Text Processing. STEP 2008 Conference Proceedings, pages 45–57. find
  9. Ann Copestake. 2002. Implementing typed feature structure grammars, volume 110. CSLI publications Stanford. find
  10. Ann Copestake, Dan Flickinger, Carl Pollard, and Ivan A Sag. 2005. Minimal recursion semantics: An introduction. Research on language and computation, 3(2):281–332. ↗
  11. Christiane Fellbaum, editor. 1998. WordNet: An Electronic Lexical Database (Language, Speech, and Communication). The MIT Press. find
  12. Dan Flickinger. 2000. On building a more efficient grammar by exploiting types. Natural Language Engineering, 6(1):15–28.
  13. Dan Flickinger. 2011. Accuracy v. robustness in grammar engineering. In Emily M. Bender and Jennifer E. Arnold, editors, Language from a Cognitive Perspective: Grammar, Usage and Processing, pages 31–50. CSLI Publications, Stanford, CA. find
  14. Dan Flickinger, Yi Zhang, and Valia Kordoni. 2012. Deepbank. a dynamically annotated treebank of the wall street journal. In Proceedings of the 11th International Workshop on Treebanks and Linguistic Theories, pages 85–96. find
  15. Michael Wayne Goodman. 2019. A python library for deep linguistic resources. In 2019 Pacific Neighborhood Consortium Annual Conference and Joint Meetings (PNC), pages 1–7. IEEE. find
  16. Sanda M. Harabagiu, George A. Miller, and Dan I. Moldovan. 1999. Wordnet 2: a morphologically and semantically enhanced resource. In Proceedings of SIGLEX99: Standardizing Lexical Resources, pages 1–8. find
  17. Michael C McCord. 2004. Word sense disambiguation in a slot grammar framework. Technical Report RC23397, IBM. find
  18. John Philip McCrae, Alexandre Rademaker, Ewa Rudnicka, and Francis Bond. 2020. find
  19. English WordNet 2020: Improving and extending a WordNet for English using an open-source methodology. In Proceedings of the LREC 2020 Workshop on Multimodal Wordnets (MMW2020), pages 14– 19, Marseille, France. The European Language Resources Association (ELRA). find
  20. Rada Mihalcea and Dan I. Moldovan. 2001. extended wordnet: progress report. In Proceedings of NAACL Workshop on Word-Net and Other Lexical Resources, pages 95–100. find
  21. George A Miller. 1993. The association of ideas. The General Psychologist, 29:69– 74. find
  22. George A Miller and Christiane Fellbaum. 2007. WordNet then and now. Language Resources and Evaluation, 41(2):209–214. ↗
  23. George A Miller, Claudia Leacock, Randee Tengi, and Ross T Bunker. 1993. A semantic concordance. In Proceedings of the workshop on Human Language Technology, pages 303–308. Association for Computational Linguistics.
  24. Dan Moldovan and Adrian Novischi. 2004. Word sense disambiguation of wordnet glosses. Computer Speech & Language, 18(3):301–317. ↗
  25. Ian Niles and Adam Pease. 2003. Linking lexicons and ontologies: Mapping wordnet to the suggested upper merged ontology. In Ike, pages 412–416. find
  26. Stephan Oepen. 2001. [incr tsdb()] — competence and performance laboratory. User manual. Technical report, Computational Linguistics, Saarland University, Saarbrücken, Germany. In preparation. find
  27. Stephan Oepen, Kristina Toutanova, Stuart M Shieber, Christopher D Manning, Dan Flickinger, and Thorsten Brants. 2002. The lingo redwoods treebank: Motivation and preliminary applications. In COLING 2002: The 17th International Conference on Computational Linguistics: Project Notes.
  28. Woodley Packard. 2015. Full forest treebanking. Ph.D. thesis, University of Washington. find
  29. Adam Pease and Andrew Cheung. 2018. Toward a semantic concordancer. In Proceedings of the 9th Global Wordnet Conference, pages 97–104, Nanyang Technological University (NTU), Singapore. Global Wordnet Association. find
  30. Alexandre Rademaker, Bruno Cuconato, Alessandra Cid, Alexandre Tessarollo, and Henrique Andrade. 2019. Completing the Princeton annotated gloss corpus project. In Proceedings of the 10th Global Wordnet Conference, pages 378–386, Wroclaw, Poland. Global Wordnet Association. find
  31. Karin Kipper Schuler. 2005. VerbNet: A broad-coverage, comprehensive verb lexicon. Ph.D. thesis, University of Pennsylvania. ↗

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/rademaker-2023-semantic-parsing-sense-tagging-princeton/