An Unsupervised Approach to Recognizing Discourse Relations

Summary

Many discourse relations between adjacent clauses have no explicit connective. In the RST corpus of Carlson et al. (2001), only about a quarter of CONTRAST and CAUSE-EXPLANATION-EVIDENCE relations carry a cue phrase. Systems that depend on cue phrases miss the rest, and deep semantic inference is not practical. Marcu and Echihabi work around this with four broad relation classes: CONTRAST, CAUSE-EXPLANATION-EVIDENCE (CEV), CONDITION and ELABORATION. Each class merges labels from several discourse theories. They collected training pairs automatically by matching cue-phrase patterns such as "but", "because", "if" and "for example" in two corpora: Raw (about 1 billion words) and the parsed BLIPP corpus. They then removed the cue phrases from the pairs. Naive Bayes classifiers trained on simple lexical features learn to tell the relations apart even with no marker present. Some binary distinctions reach up to 93% accuracy. On the hand-annotated RST data, the classifiers correctly labelled many unmarked relations. This shows that very large amounts of noisy, automatically labelled text can support discourse parsing, summarization and question answering.

Key points

Training data: generated from Raw (41,147,805 sentences) and BLIPP (1,796,386 sentences parsed with Charniak's parser). Example counts: about 3.88M CONTRAST, 890K CEV, 1.2M CONDITION and 1.84M ELABORATION. There were also 1M random "no relation" pairs drawn from the same text and 1M drawn from different texts.
Relation classes: each is defined as a union of labels from Mann & Thompson (RST), Hobbs, Lascarides & Asher, and Knott & Sanders. For example, CONTRAST covers ANTITHESIS, CONCESSION and OTHERWISE.
Best result: the strongest binary classifiers reach accuracies of up to 93% on relations with the cue phrases stripped out.
Gains on the RST corpus: for adjacent elementary discourse units, the CONTRAST vs. ELABORATION classifier raised correct CONTRAST labelling from at most 61/238 (cue-marked cases only) to 183/238. The CEV vs. ELABORATION classifier raised CEV labelling from 79/307 to 175/307.
ELABORATION: this relation was the hardest to separate from the others, which agrees with Knott et al.'s argument that ELABORATION is poorly defined.

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

131 papers of our library cite it (of the 140673 whose bibliographies we read); 49 of them have a page here, 82 not yet.

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.3115/1073083.1073145

@inproceedings{marcu2002unsupervised,
  title = {{An Unsupervised Approach to Recognizing Discourse Relations}},
  author = {Marcu, Daniel and Echihabi, Abdessamad},
  booktitle = {Proceedings of the 40th Annual Meeting of the Association for Computational Linguistics},
  year = {2002},
  doi = {10.3115/1073083.1073145},
  url = {https://doi.org/10.3115/1073083.1073145}
}

References

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

Michele Banko and Eric Brill. 2001. Scaling to very very large corpora for natural language disambiguation. In Proceedings of the 39th Annual Meeting of the Association for Computational Linguistics (ACL’01), Toulouse, France, July 6–11. find
Lynn Carlson, Daniel Marcu, and Mary Ellen Okurowski. 2001. Building a discourse-tagged corpus in the framework of rhetorical structure theory. In Proceedings of the 2nd SIGDIAL Workshop on Discourse and Dialogue, Eurospeech 2001, Aalborg, Denmark. find our card
Eugene Charniak. 2000. A maximum-entropy-inspired parser. In Proceedings of the First Annual Meeting of the North American Chapter of the Association for Computational Linguistics NAACL–2000, pages 132– 139, Seattle, Washington, April 29 – May 3. find our card
DUC–2002. Proceedings of the Second Document Understanding Conference, Philadelphia, PA, July. find
Christiane Fellbaum, editor. 1998. Wordnet: An Electronic Lexical Database. The MIT Press. find
Michael A.K. Halliday and Ruqaiya Hasan. 1976. Cohesion in English. Longman. find
Jerry R. Hobbs. 1990. Literature and Cognition. CSLI Lecture Notes Number 21. find
Eduard H. Hovy and Elisabeth Maier. 1993. Parsimonious or profligate: How many and which discourse structure relations? Unpublished Manuscript. find
Alistair Knott and Ted J.M. Sanders. 1998. The classification of coherence relations and their linguistic markers: An exploration of two languages. Journal of Pragmatics, 30:135–175. find
Alistair Knott, Jon Oberlander, Mick O’Donnell, and Chris Mellish. 2001. Beyond elaboration: The interaction of relations and focus in coherent text. In T. Sanders, J. Schilperoord, and W. Spooren, editors, Text representation: linguistic and psycholinguistic aspects, pages 181–196. Benjamins. find
Alex Lascarides and Nicholas Asher. 1993. Temporal interpretation, discourse relations, and common sense entailment. Linguistics and Philosophy, 16(5):437– 493. find
William C. Mann and Sandra A. Thompson. 1988. Rhetorical structure theory: Toward a functional theory of text organization. Text, 8(3):243–281. find our card
Christopher Manning and Hinrich Schütze. 1999. Foundations of Statistical Natural Language Processing. The MIT Press. find
Daniel Marcu. 2000. The Theory and Practice of Discourse Parsing and Summarization. The MIT Press. find
James R. Martin. 1992. English Text. System and Structure. John Benjamin Publishing Company. find
Deborah Schiffrin. 1987. Discourse Markers. Cambridge University Press. find
TREC–2001. Proceedings of the Text Retrieval Conference, November. The Question-Answering Track. find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/marcu-2002-unsupervised-approach-recognizing-discourse-relations/

📄 An Unsupervised Approach to Recognizing Discourse Relations