Elamur Research⌕
☰
home / papers / 2010 / Automatic Discourse Segmentation: Review and Perspectives

Automatic Discourse Segmentation: Review and Perspectives

Paper · 2010 · Iria da Cunha and Juan-Manuel Torres-Moreno

Summary

The paper looks at discourse segmentation, the first step of RST-based discourse parsing. In this step a text is cut into elementary discourse units (EDUs), which become the nuclei and satellites of a rhetorical tree. The authors review two segmenters that are available online: SLSeg for English and a Brazilian Portuguese segmenter built for the DiZer parser. Both rely on a small set of hand-written lexical and syntactic rules that run over the output of a morphosyntactic parser. From these two systems the authors extract a shared six-step recipe: build a corpus, annotate it with RST, manually study which cues mark boundaries, write rules, feed the rules with parser output, and evaluate. They then ask whether the recipe carries over to Spanish, French and African languages. A Spanish segmenter already exists at LIA and follows this recipe. For African languages, the main obstacles are the absence of morphosyntactic parsers and very little research on discourse markers. For every new language, a further problem is that no rival segmenter exists to compare against. The review matters as a practical roadmap for building segmenters in languages with few resources.

Key points

  • SLSeg (Tofiloski, Brooke, Taboada 2009):
    • It applies 12 lexical/syntactic rules for English.
    • It measures how the choice of parser affects segmentation by comparing the Charniak and Stanford parsers.
    • It is evaluated against the Soricut & Marcu segmenter and against baselines, with very good results.
  • The Brazilian Portuguese segmenter (Mazeiro, Pardo et al. 2007):
    • It also uses 12 rules, built on top of the PALAVRAS parser.
    • It follows the Carlson & Marcu EDU manual and is meant for DiZer.
    • It has no explicit evaluation.
  • The Spanish segmenter at LIA:
    • It uses a medical corpus, RSTTool for annotation and Freeling for morphosyntax.
    • Its rules draw on discourse markers, conjunctions, verb forms and punctuation.
    • Because no other Spanish segmenter exists, it can only be compared with baselines.
  • Two core rules:
    • Every EDU must contain a finite verb.
    • A discourse marker such as "because" or "but" splits a sentence only when both sides contain a finite verb.
  • On African languages:
    • Roughly 2000 languages exist, grouped into six families, and most are mainly oral.
    • One example resource is a Somali web corpus of 3 million words.
    • The authors know of no suitable morphosyntactic parsers and little work on discourse markers for these languages.

Interesting ideas

  • I1. An EDU is defined as a stretch of text that has its own finite verb and carries a rhetorical relation; this keeps clause syntax distinct from discourse structure.
  • I2. A discourse marker is treated as a boundary only when both resulting pieces have a finite verb, so "as well to children" stays inside one unit while "but children can use it" becomes its own unit.
  • I3. SLSeg runs the same segmentation rules on two different parsers to see how parser errors spread into the discourse layer.
  • I4. When no competing system exists, the authors fall back on simple baselines as the evaluation reference.
  • I5. One automatic segmentation is handed to all human annotators, so their disagreements come from the analysis itself rather than from different unit boundaries.

For Mova

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

  • I1 (frames): clause units that each hold a finite verb could be read straight off the UD tree, giving each frame a clean span before relations such as Cause or Contrast are attached.
  • I2 (frames): this conjunctive check could stop Mova from building a separate frame for a verbless phrase that merely follows a connective.
  • I3 (learn): Mova's witness parsers could be scored by how stable the downstream segmentation and frames remain, not just by tree accuracy.
  • I4 (learn): trivial segmenters, such as splitting at every sentence or at every comma, could serve as negative controls that any discourse component must clearly beat.
  • I5 (learn): when the LLM teacher tiers or annotators label relations, Mova could fix the segmentation beforehand so that label agreement can be measured on identical spans.

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

🎓Universitat Pompeu FabraBarcelona · Spain36 works · 11,433 citations

📚 Shelves 4 Computer Science · 9,086 works Artificial Intelligence · 8,889 works Arts and Humanities · 9,296 works Language and Linguistics · 9,206 works

🧭 Topics 2 English · 1,373 works Discourse analysis · 394 works

Cite

Cite as Iria da Cunha and Juan-Manuel Torres-Moreno. 2010. Automatic Discourse Segmentation: Review and Perspectives. International Workshop on African Human Language Technologies. 17-20 Jan 2010.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.2005.00468 · arXiv 2005.00468

@inproceedings{cunha2010automatic,
  title = {{Automatic Discourse Segmentation: Review and Perspectives}},
  author = {Iria {da Cunha} and Juan-Manuel Torres-Moreno},
  booktitle = {International Workshop on African Human Language Technologies. 17-20 Jan 2010},
  year = {2010},
  doi = {10.48550/arxiv.2005.00468},
  eprint = {2005.00468},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2005.00468}
}

References

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

  1. Grice, P. (1975). “Logic and conversation”. In Cole, P.; Morgan, J. L. (eds.). Syntax and Semantics 3: Speech Acts. New York: Academic Press. 4158. ↗
  2. Halliday, M. A. K.; Hasan, R. (1976). Cohesion in English. Essex: Longman. find
  3. Anscombre, J. C.; Ducrot, O. (1983). L’argumentation dans la langue. Bruselas: Mardaga. find
  4. Sperber, D.; Wilson, D. (1986). Relevance. Oxford: Blackwell. find
  5. Mann, W. C.; Thompson, S. A. (1988). “Rhetorical structure theory: Toward a functional theory of text organization”. Text 8 (3). 243281. ↗ our card
  6. Marcu, D. (2000a). The Theory and Practice of Discourse Parsing Summarization. Massachusetts: Institute of Technology. find
  7. Marcu, D. (2000b). “The Rhetorical Parsing of Unrestricted Texts: A Surfacebased Approach”. Computational Linguistics 26 (3). 395448. find
  8. Sumita, K.; Ono, K.; Chino, T.; Ukita, T.; Amano, S. (1992). “A discourse structure analyzer for Japanese text”. Proceedings of the International Conference on Fifth Generation Computer Systems. 11331140. find
  9. Pardo, T.A.S.; Nunes, M.G.V.; Rino, L.H.M. (2004). “DiZer: An Automatic Discourse Analyzer for Brazilian Portuguese”. Lecture Notes in Artificial Intelligence. 224234. find
  10. Pardo, T.A.S.; Nunes, M.G.V. (2008). “On the Development and Evaluation of a Brazilian Portuguese Discourse Parser”. Journal of Theoretical and Applied Computing. 15 (2). 4364. ↗
  11. Tofiloski, M.; Brooke, J.; Taboada, M. (2009). “A Syntactic and LexicalBased Discourse Segmenter”. In Proceedings of the 47th Annual Meeting of the Association for Computational Linguistics. Singapur. find
  12. Mazeiro, E.; Pardo, T.A.S.; Volpe, M.G. (2007). “Identificação automática de segmentos discursivos: o uso do parser PALAVRAS”. Série de Relatórios do Núcleo Interinstitucional de Lingüística Computacional (NILC). São Carlos, São Paulo. find
  13. Carlson, L.; Marcu, D. (2001). Discourse Tagging Reference Manual. ISI Technical Report ISITR545. Los Angeles: University of Southern California. find
  14. da Cunha, I.; Iruskieta, M. (2009). “La influencia del anotador y las técnicas de traducción en el desarrollo de árboles retóricos. Un estudio en español y euskera”. 7th Brazilian Symposium in Information and Human Language Technology (STIL 2009). São Paolo, Brazil. find
  15. Charniak, E. (2000). “A MaximumEntropy Inspired Parser”. 1 st Meeting of the North American Chapter of the ACL (NAACL). 132139. Seattle, WA. find
  16. Klein, D.; Manning, C.D. (2003). “Fast Exact Inference with a Factored Model for Natural Language Parsing”. Advances in NIPS 15 (NIPS 2002). 310. Cambridge, MA: MIT Press. find
  17. Soricut, R.; Marcu, D. (2003). “Sentence Level Discourse Parsing Using Syntactic and Lexical Information”. Human Language Technology conference (HLTNAACL). 149156. Edmonton, Canada. our card
  18. Bick, E. (2000). “The Parsing System Palavras – Automatic Grammatical Analysis of Portuguese in a Constraint Grammar Framework”. Aarhus: Aarhus University Press. find
  19. O’Donnell, M. (2000). “RSTTOOL 2.4 – A Markup Tool for Rhetorical Structure Theory”. Proceedings of the International Natural Language Generation Conference. 253256. our card
  20. Atserias, J.; Comelles, E.; Mayor, A. (2005). “TXALA un analizador libre de dependencias para el castellano”. Procesamiento del Lenguaje Natural 35. 455456. find
  21. Laenzlinger, C,; Wehrli, É. (1991). “FIPS : Un analyseur interactif pour le français”. TA Informations 32(2). 3549. find
  22. Wehrli, É. (2007). “Fips, a “Deep” Linguistic Multilingual Parser”. ACL 2007 Workshop on Deep Linguistic Processing. 120127. Prague, Czech Republic. find
  23. Abdillahi, N.; Nocera, P.; TorresMoreno, JM (2006). “Boîte à outils TAL pour des langues peu informatisées : le cas du Somali”. JADT 2006 : 8es Journées internationales d’Analyse statistique des Données Textuelles. 697705. find
  24. Vaufreydaz D.; Akbar M.; Roullard J. (1999). “Asru’99”. In Internet documents: a rich source for spoken language modelling. 177280. Keystone Colorado, USA. find
  25. Ghani R.; Jones, R.; Mladenic D. (2000). Mining the web to create minority language corpora. Berlin. find
  26. Purvis, T. (2008). “Variation in the Use of Hesitation Markers in Akan Discourse”. 37th of the New Ways of Analyzing Variation conference. Houston, Texas. find
  27. DikiKidiri, M. (2008). Le vocabulaire scientifique dans les langues africaines. Karthala: Paris. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/cunha-2020-automatic-discourse-segmentation-review-perspectives/