Elamur Research⌕
☰
home / papers / 2007 / A Unified Local and Global Model for Discourse Coherence

A Unified Local and Global Model for Discourse Coherence

Summary

The paper tackles automatic modeling of discourse coherence, which is used for ordering sentences and judging text quality. Local models, such as entity grids, capture how neighboring sentences share nouns but struggle to locate document beginnings, endings and topic shifts. Global HMM content models track topic progression but ignore sentence-to-sentence links. The authors build a single generative model that learns both kinds of evidence jointly. A hidden topic HMM, following Barzilay and Lee (2004), provides global structure. Its emissions for previously mentioned entities come from an entity grid. To make the grid work generatively, they propose a "relaxed entity grid." It drops the assumption that entities behave independently and instead lets known nouns compete for the available syntactic slots. Experiments use the AIRPLANE corpus. The combined model orders sentences about as well as the Soricut and Marcu (2006) mixture model and clearly outperforms both of its parent models on discrimination. This shows that a jointly trained, more principled model can rival a mixture of separately trained components.

Key points

  • Entity grid setup. The grid marks each noun's role in each sentence as S, O, X or absent. Roles are derived from Charniak and Johnson (2005) parser trees, coreference resolution is omitted, and a history horizon of h = 2 is used.
  • Ordering task (Kendall's τ).
    • Combined topic-based model: about .50, comparable to Soricut and Marcu (2006).
    • Barzilay and Lee (2004) HMM: .44.
    • Relaxed entity grid alone: about 0.
  • Discrimination task (original document vs. 20 random permutations, 2000 tests in total):
    • Combined model: 94%.
    • Barzilay and Lee (2004) HMM: 74%.
    • Relaxed entity grid: 87%.
    • Naive generative grid: 81%.
  • Cross-validation on AIRPLANE training data. With the relaxed grid as emission model, the topic-based system reaches τ .54 and 96% discrimination. With the naive grid, it reaches τ .39 and 85%.
  • Feature set. The model relies solely on unigram language models plus features drawn from the entity grid, a strict subset of the features used by Soricut and Marcu. The authors therefore suggest that adding it to such a mixture could improve results further.

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

Fields: Linguistics & languages

Topics: Discourse analysis

Cited by

29 papers of our library cite it (of the 140673 whose bibliographies we read); 3 of them have a page here, 26 not yet.

Cite

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

@inproceedings{elsner2007unified,
  title = {{A Unified Local and Global Model for Discourse Coherence}},
  author = {Elsner, Micha and Austerweil, Joseph and Charniak, Eugene},
  booktitle = {Human Language Technologies 2007: The Conference of the North American Chapter of the Association for Computational Linguistics; Proceedings of the Main Conference},
  year = {2007},
  url = {https://aclanthology.org/N07-1055/}
}

References

16 references, 2 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Ernst Althaus, Nikiforos Karamanis, and Alexander Koller. 2004. Computing locally coherent discourses. In Proceedings of the 42nd ACL, Barcelona. find
  2. Regina Barzilay and Mirella Lapata. 2005. Modeling local coherence: an entity-based approach. In Proceedings of the 43rd Annual Meeting of the Association for Computational Linguistics (ACL’05). find
  3. Regina Barzilay and Lillian Lee. 2004. Catching the drift: Probabilistic content models, with applications to generation and summarization. In HLT-NAACL 2004: Proceedings of the Main Conference, pages 113–120. find
  4. Matthew J. Beal, Zoubin Ghahramani, and Carl Edward Rasmussen. 2001. The infinite Hidden Markov Model. In NIPS, pages 577–584. find
  5. Eugene Charniak and Mark Johnson. 2005. Coarse-to-fine n-best parsing and MaxEnt discriminative reranking. In Proc. of the 2005 Meeting of the Assoc. for Computational Linguistics (ACL), pages 173–180. find our card
  6. Peter Foltz, Walter Kintsch, and Thomas Landauer. 1998. The measurement of textual coherence with latent semantic analysis. Discourse Processes, 25(2&3):285–307. find
  7. Barbara J. Grosz, Aravind K. Joshi, and Scott Weinstein. 1995. Centering: A framework for modeling the local coherence of discourse. Computational Linguistics, 21(2):203–225. find our card
  8. Derrick Higgins, Jill Burstein, Daniel Marcu, and Claudia Gentile. 2004. Evaluating multiple aspects of coherence in student essays. In HLT-NAACL, pages 185– 192. find
  9. Roger Kibble and Richard Power. 2004. Optimising referential coherence in text generation. Computational Linguistics, 30(4):401–416. find
  10. Mirella Lapata and Regina Barzilay. 2005. Automatic evaluation of text coherence: Models and representations. In IJCAI, pages 1085–1090. find
  11. Mirella Lapata. 2003. Probabilistic text structuring: Experiments with sentence ordering. In Proceedings of the annual meeting of ACL, 2003. find
  12. Mirella Lapata. 2006. Automatic evaluation of information ordering: Kendall’s tau. Computational Linguistics, 32(4):1–14. find
  13. E. Miltsakaki and K. Kukich. 2004. Evaluation of text coherence for electronic essay scoring systems. Nat. Lang. Eng., 10(1):25–55. find
  14. Advaith Siddharthan, Ani Nenkova, and Kathleen McKeown. 2004. Syntactic simplification for improving content selection in multi-document summarization. In COLING04, pages 896–902. find
  15. Radu Soricut and Daniel Marcu. 2006. Discourse generation using utility-trained coherence models. In Proceedings of the Association for Computational Linguistics Conference (ACL-2006). find
  16. Y.W. Teh. 2006. A Bayesian interpretation of interpolated Kneser-Ney. Technical Report TRA2/06, National University of Singapore. find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/elsner-2007-unified-local-global-model-discourse/