Elamur Research⌕
☰
home / papers / 2019 / Ambiguity in Explicit Discourse Connectives

Ambiguity in Explicit Discourse Connectives

Summary

The paper argues that explicit discourse connectives can be ambiguous in more ways than usually assumed. Earlier work, notably Pitler and Nenkova (2009), addressed two kinds: whether a token acts as a connective at all (usage) and which relation it signals (sense). The authors survey four further ambiguity types that are well known for other linguistic forms, using examples and counts from the newly released PDTB-3.

The four types are:

  • Part-of-speech: the same word acting as a conjunction or an adverbial.
  • Multi-word expression: a string read as separate words or as one connective.
  • Scope: an ambiguity that changes the relation's sense.
  • Semantic role: which argument of a CONCESSION relation cancels the expected inference.

They also show that the expanded PDTB-3 sense hierarchy creates new sense ambiguities that were absent in the PDTB-2.

The paper is a position and analysis study rather than a new model. Its point is that overlooking these ambiguities causes mislabelled relations, both in manual annotation and in automatic shallow discourse parsing. It calls on researchers to test whether syntax-based disambiguation is sufficient or needs extending.

Key points

  • Corpus sizes: The PDTB-2 has over 43K relations, about 18.4K of them explicit. The PDTB-3 adds roughly 12.5K intra-sentential and 1K inter-sentential relations. Of its ~53,600 tokens, about 57% are unchanged from the PDTB-2, 19% were revised, and 24% are new.
  • New sense, new ambiguity: COMPARISON.SIMILARITY was applied to 135 new tokens and 68 re-annotated ones (41 explicit). This makes connectives such as as well, as if and while newly ambiguous, whereas similarly remains unambiguous.
  • Part-of-speech ambiguity: since, before and however change their sense, sometimes to the reverse temporal or concessive direction, depending on PoS (IN vs. RB, or RB vs. WRB). Correct PoS tagging is enough to resolve this.
  • Multi-word and scope ambiguity: Examples of the multi-word type are only to, but also, but then and or otherwise. For scope, about 9% of over 1,600 relations with a to-clause Arg2 were labelled CONDITION.ARG2-AS-COND, and syntax alone may not resolve such cases.
  • Semantic role ambiguity: This affects postposed although, though, even though and while. Of 324 PDTB-3 tokens, 260 are ARG1-AS-DENIER and 64 are ARG2-AS-DENIER. The 402 preposed tokens are treated as unambiguous.

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 · Part-of-speech tagging

Cited by

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

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/w19-0411

@inproceedings{webber2019ambiguity,
  title = {{Ambiguity in Explicit Discourse Connectives}},
  author = {Webber, Bonnie and Prasad, Rashmi and Lee, Alan},
  booktitle = {Proceedings of the 13th International Conference on Computational Semantics - Long Papers},
  year = {2019},
  doi = {10.18653/v1/w19-0411},
  url = {https://doi.org/10.18653/v1/w19-0411}
}

References

10 references, 5 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Hollenstein, N., N. Schneider, and B. Webber (2016). Inconsistency detection in semantic annotation. In Proceedings, Language Resources and Evaluation Conference (LREC 2016), Potoroz, Slovenia. find
  2. Konstas, I., F. Keller, V. Demberg, and M. Lapata (2014). Incremental semantic role labeling with tree adjoining grammar. In Proceedings, Empirical Methods in Natural Language Processing. find
  3. Pitler, E. and A. Nenkova (2009). Using syntax to disambiguate explicit discourse connectives in text. In ACL-IJCNLP ’09: Proceedings of the 47th Meeting of the Association for Computational Linguistics and the 4th International Joint Conference on Natural Language Processing. find our card
  4. Prasad, R., N. Dinesh, A. Lee, E. Miltsakaki, L. Robaldo, A. Joshi, and B. Webber (2008). The Penn Discourse TreeBank 2.0. In Proceedings, 6th International Conference on Language Resources and Evaluation, Marrakech, Morocco. find our card
  5. Prasad, R., B. Webber, and A. Joshi (2014). Reflections on the Penn Discourse Treebank, comparable corpora and complementary annotation. Computational Linguistics 40(4), 921–950. find our card
  6. The PDTB Research Group (2008). The Penn Discourse TreeBank 2.0 Annotation Manual. Available at http://www.seas.upenn.edu/˜pdtb/, or as part of the download of LDC2008T05. ↗
  7. Webber, B., R. Prasad, A. Lee, and A. Joshi (2019). The Penn Discourse Treebank 3.0 Annotation Manual. Technical report, University of Pennsylvania. Available at https://catalog.ldc.upenn.edu/docs/LDC2019T05/PDTB3-Annotation-Manual.pdf. ↗
  8. Xue, N., H. T. Ng, S. Pradhan, R. Prasad, C. Bryant, and A. Rutherford (2015). The CoNLL-2015 shared task on shallow discourse parsing. In Proc 19 th Conference on Computational Natural Language Learning – Shared Task, Beijing, pp. 1–16. find our card
  9. Xue, N., H. T. Ng, S. Pradhan, A. Rutherford, B. Webber, C. Wang, and H. Wang (2016). find
  10. CoNLL 2016 shared task on multilingual shallow discourse parsing. In Proc 20 th Conference on Computational Natural Language Learning – Shared Task, Berlin, pp. 1–19. find our card

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/papers/webber-2019-ambiguity-explicit-discourse-connectives/