Elamur Research⌕
☰
home / papers / 2006 / Learning Information Status of Discourse Entities

Learning Information Status of Discourse Entities

Paper · 2006 · Malvina Nissim

Summary

Nissim tackles automatic labeling of the information status of English noun phrases. The task is deciding whether a discourse entity is old (already evoked), mediated (newly mentioned but inferable from context or general knowledge), or new to the hearer. The data are 147 Switchboard telephone dialogues annotated with these three tags, with good inter-annotator agreement (K = .845). The author builds two systems:

  • a hand-written rule cascade that mirrors the annotation guidelines, based mainly on NP type, string overlap and determiner;
  • a C4.5 decision tree (Weka J48) trained on a small set of morpho-syntactic and lexical features taken from the gold treebank.

The learned tree reaches 79.5% accuracy, well above the rules (about 64–66%) and the all-old baseline (47.9%). Merging mediated with new gives a binary task solved at 93.1%. The study matters for two reasons. It was the first learning experiment on information status for English. It also suggests uses in anaphoricity filtering for coreference and in choosing prosody and word order in generation.

Key points

  • Data split: train has 94 dialogues (40,865 NPs), dev has 23 (10,565), and eval has 30 (12,624). The class mix is roughly 48% old, 37% mediated and 15% new.
  • Features:
    • full previous mention
    • mention count (first/second/more)
    • partial previous mention
    • determiner type
    • NP length
    • grammatical role
    • NP type
  • Gold inputs only: syntax comes from the Penn Treebank portion of Switchboard rather than from a parser. Manually annotated coreference was intentionally left out.
  • Per-class results: old entities are found with high precision and recall, and mediated ones with high recall. New entities have fair precision but very poor recall.
  • Learned tree versus rules: the tree's top split is partial previous mention, whereas the rules split first on NP type. NP type alone makes a weak classifier that never predicts new, but removing it still lowers accuracy.

Interesting ideas

  • I1. Retraining after merging different pairs of classes shows which categories behave alike: mediated proved closer to new than to old, which is feedback on the label scheme itself.
  • I2. Partial string overlap with earlier mentions (e.g. "your children" after "my children") is a cheap signal that an entity is mediated through a set relation rather than brand new.
  • I3. The classifier avoids relying on coreference so that its output can later tell a resolver which definite NPs actually need an antecedent.
  • I4. A guideline-based rule cascade serves as an interpretable baseline against which a learned tree is compared, and the learned tree reorders the rules' priorities.
  • I5. The author proposes, as future work, marking an entity as mediated when WordNet or FrameNet links it within some depth to an earlier one.

For Mova

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

  • I1 (learn): Collapse-and-retrain probes could check whether Mova's own label inventories (frame roles, entity states) contain distinctions that the data cannot support.
  • I2 (world): When adding an entity to the text's world, a partial head match with an existing entity could create a set or subset link instead of a fresh, unrelated node.
  • I3 (frames): An explainable old/mediated/new tag computed before coreference could stop the resolver from forcing antecedents onto non-anaphoric definites such as "the sun".
  • I4 (learn): Self-induced repair rules could be checked the same way, comparing a learned decision structure with the hand rules to find which cue the hand rules overweight.
  • I5 (world): The compiled WordNet/FrameNet ontology could supply such bridging links, so that "the engine" after "a car" is attached as a part of an already known entity.

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

🤝National Research CouncilRome · Italy15 works · 15,649 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 worksDiscourse analysis · 466 works

Cite

Cite as Malvina Nissim. 2006. Learning Information Status of Discourse Entities. Proceedings of the 2006 Conference on Empirical Methods in Natural Language Processing.

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

@inproceedings{nissim2006learning,
  title = {{Learning Information Status of Discourse Entities}},
  author = {Malvina Nissim},
  booktitle = {Proceedings of the 2006 Conference on Empirical Methods in Natural Language Processing},
  year = {2006},
  doi = {10.3115/1610075.1610090},
  url = {https://doi.org/10.3115/1610075.1610090}
}

References

32 references, 6 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Mira Ariel. 1990. Accessing Noun Phrase Antecedents. Routledge, London-New York. find
  2. Collin F. Baker, Charles J. Fillmore, and John B. Lowe. 1998. The Berkeley FrameNet project. In Christian Boitet and Pete Whitelock, editors, Proceedings of COLING-ACL, pages 86–90.
  3. Jean Carletta, Shipra Dingare, Malvina Nissim, and Tatiana Nikitina. 2004. Using the NITE XML Toolkit on the Switchboard Corpus to study syntactic choice: a case study. In Proceedings of the 4th International Conference on Language Resources and Evaluation (LREC2004), Lisbon, May 2004. find
  4. Jean Carletta. 1996. Assessing agreement on classification tasks: the kappa statistic. Computational Linguistics, 22(2):249–254. ↗
  5. Herbert H. Clark. 1975. Bridging. In Roger Schank and Bonnie Nash-Webber, editors, Theoretical Issues in Natural Language Processing. The MIT Press, Cambridge, MA. find
  6. Jacob Cohen. 1960. A coefficient of agreement for nominal scales. Educational and Psychological Measurements, 20:37–46. ↗
  7. Miriam Eckert and Michael Strube. 2001. Dialogue acts, synchronising units and anaphora resolution. Journal of Semantics, 17(1):51–89. find
  8. Christiane Fellbaum, editor. 1998. WordNet: An Electronic Lexical Database. The MIT Press, Cambridge, MA. ↗
  9. Talmy Givón. 1983. Introduction. In Talmy Givón, editor, Topic Continuity in Discourse: A Quantitative Cross-language Study. John Benjamins, Amsterdam/Philadelphia. find
  10. J. Godfrey, E. Holliman, and J. McDaniel. 1992. SWITCHBOARD: Telephone speech corpus for research and development. In Proceedings of ICASSP-92, pages 517–520. find
  11. Barbara Grosz, Aravind K. Joshi, and Scott Weinstein. 1995. Centering: a framework for modeling the local coherence of discourse. Computational Linguistics, 21(2):203–225.
  12. Eva Hajičová, Barbara Partee, and Petr Sgall. 1998. Topic-focus articulation, tripartite structures, and semantic content. In Studies in Linguistics and Philosophy, volume 71. Dordrecht. ↗
  13. Eva Hajičová. 1984. Topic and focus. In Petr Sgall, editor, Contributions to Functional Syntax. Semantics and Language Comprehension (LLSEE 16), pages 189–202. John Benjamins, Amsterdam. ↗
  14. M.A.K. Halliday. 1976. Notes on transitivity and theme in English. Part 2. Journal of Linguistics, 3(2):199–244. find
  15. Julia Hirschberg and Christine H. Nakatani. 1996. A prosodic analysis of discourse segments in direction giving monologues. In Proceedings of 34 th Annual Meeting of the Association for Computational Linguistics. find
  16. Ivana Kruijff-Korbayová and Mark Steedman. 2003. Discourse and information structure. Journal of Logic, Language, and Information, 12:249–259. ↗
  17. Knud Lambrecht. 1994. Information structure and sentence form. Topic, focus, and the mental representation of discourse referents. Cambridge University Press, Cambridge. ↗
  18. Sebastian Löbner. 1985. Definites. Journal of Semantics, 4:279–326. find
  19. Mitchell Marcus, Beatrice Santorini, and May Ann Marcinkiewicz. 1993. Building a large annotated corpus of english: The Penn treebank. Computational Linguistics, 19:313–330. ↗
  20. Vincent Ng and Claire Cardie. 2002. Identifying anaphoric and non-anaphoric noun phrases to improve coreference resolution. In Proc of the 19 th International Conference on Computational Linguistics; Taipei, Taiwan, pages 730–736. find
  21. Malvina Nissim, Shipra Dingare, Jean Carletta, and Mark Steedman. 2004. An annotation scheme for information status in dialogue. In Proceedings of the 4th International Conference on Language Resources and Evaluation (LREC2004), Lisbon, May 2004.
  22. Oana Postolache, Ivana Kruijff-Korbayova, and Geert-Jan Kruijff. 2005. Data-driven approaches for information structure identification. In Proceedings of Human Language Technology Conference and Conference on Empirical Methods in Natural Language Processing, pages 9–16, Vancouver, British Columbia, Canada, October. Association for Computational Linguistics. find
  23. Ellen F. Prince. 1981. Toward a taxonomy of given-new information. In Peter Cole, editor, Radical Pragmatics. Academic Press, New York. find
  24. Ellen Prince. 1992. The ZPG letter: subjects, definiteness, and information-status. In Sandra Thompson and William Mann, editors, Discourse description: diverse analyses of a fund raising text, pages 295– 325. John Benjamins, Philadelphia/Amsterdam. find
  25. Mark Steedman. 2000. The Syntactic Process. The MIT Press, Cambridge, MA. ↗
  26. K. Stöber, P. Wagner, Jörg Helbig, S. Köster, D. Stall, M. Thomas, J. Blauert, W. Hess, R. Hoffmann, and H. Mangold. 2000. Speech synthesis using multi-level selection and concatenation of units from large speech corpora. In W. Wahlster, editor, Verbmobil: Foundations of Speech-to-Speech Translation, pages 519–534. Springer-Verlag, Berlin. find
  27. Michael Strube. 1998. Never look back: An alternative to centering. In Proceedings of the 17th International Conference on Computational Linguistics and 36th Annual Meeting of the Association for Computational Linguistics, pages 1251–1257, Montréal, Québec, Canada. ↗
  28. Olga Uryupina. 2003. High-precision identification of discourse new and unique noun phrases. In Proc. of the ACL 2003 Student Workshop, pages 80–86. find
  29. Enric Vallduvı́. 1992. The Informational Component. Garland, New York. find
  30. Renata Vieira and Massimo Poesio. 2000. An empirically-based system for processing definite descriptions. Computational Linguistics, 26(4). ↗
  31. Thomas Wasow. 2002. Postverbal Behavior. CSLI Publications. find
  32. Ian H. Witten and Eibe Frank. 2000. Data Mining: Practical Machine Learning Tools and Techniques with Java Implementations. Morgan Kaufmann, San Diego, CA. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/nissim-2006-learning-information-status-discourse-entities/