Elamur Research⌕
☰
home / papers / 1995 / A Uniform Treatment of Pragmatic Inferences in Simple and Complex Utterances and Sequences of Utterances

A Uniform Treatment of Pragmatic Inferences in Simple and Complex Utterances and Sequences of Utterances

Paper · 1995 · Daniel Marcu and Graeme Hirst

Summary

The paper addresses a long-standing difficulty in pragmatics: presuppositions and implicatures are defeasible, but earlier linguistic and computational accounts either fix them permanently once an utterance has been processed or permit cancellation only inside a narrow window. Marcu and Hirst propose stratified logic, a first-order formalism in which every atomic formula has a strength label. The labels are indefeasible (u), infelicitously defeasible (i) or felicitously defeasible (d), and each comes in a true and a false polarity.

Each pragmatic inference type is encoded as a rule stating its necessary trigger conditions. When those conditions hold, the inference is drawn at a defeasible strength. An extended semantic tableau then enumerates the models, and a preference for "optimistic" models selects the readings that remain informative while staying easy to revise. An inference counts as a pragmatic inference only if no stronger information cancels it in any optimistic model.

The algorithm was implemented in Common Lisp with Screamer. It handles simple utterances, complex utterances (the projection problem) and multi-utterance sequences with one uniform mechanism. Its main contribution is that cancellation can happen at any later point in a discourse.

Key points

  • Three strength layers, ordered u < i < d. Each layer is split into truth and falsity, giving a lattice of six values. Separate u-, i- and d-satisfaction relations let a theory be read loosely or strictly, and the strict readings expose any cancellation.
  • Two kinds of defeasibility for "regret". The presupposition of a factive like "regret" is infelicitously defeasible in positive sentences, where cancelling it sounds anomalous. In negated sentences it is felicitously defeasible: "John does not regret… because she did not come" is fine.
  • A dedicated pragmatic quantifier. Pragmatic rules use their own quantifier, which fires only for objects in content that was actually uttered. Lexical semantics uses the ordinary universal quantifier.
  • Projection example. For "Chris is a bachelor or a spinster", the tableau produces 16 models. Only "Chris is an adult" survives as a projected presupposition.
  • Sequence examples.
    • Speech repair: "John is not a bachelor" presupposes John is a male adult. Two utterances later, "He is only five years old" cancels that presupposition.
    • Indirect reply (Green): in the shopping dialogue, the implicature shifts from "no" to "yes" as the answer unfolds.

Interesting ideas

  • I1. Distinguishing inferences that are awkward to retract (i) from those that can be dropped cleanly (d) gives a graded notion of defeasibility, rather than a single default/fact split.
  • I2. Trigger rules should fire only on what was actually said, not on everything entailed, which keeps pragmatic reasoning tied to the utterance surface.
  • I3. An inference is accepted only if it survives in every preferred model, and the system re-evaluates after each new utterance, so nothing is frozen once a sentence ends.
  • I4. Preferring the "optimistic" model, meaning the one that asserts facts at the weakest strength that still covers them, is a principled tie-breaker among competing readings.
  • I5. Repairs and indirect replies serve as natural test cases, because an inference appears and is cancelled several turns later.

For Mova

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

  • I1 (frames): crate prag could tag each presupposition or implicature with an i/d strength, so that a contradiction against an i-level inference is flagged as an anomaly, while a d-level one is quietly withdrawn.
  • I2 (frames): implicature rules in prag could be keyed to nodes in the UD tree or frame produced from the text, and never to world-model consequences, so that each inference traces back to its source words.
  • I3 (world): the world of the text could store presupposed states with a defeasible status and a pointer to their trigger. Later sentences such as corrections or repairs could then retract them, with the retraction recorded as an explainable event.
  • I4 (answer): when a fable or test question has several consistent readings, readers could prefer the one that commits least strongly. They could then report which defeasible assumptions the answer depends on.
  • I5 (learn): build small evaluation sets of dialogue sequences with known generate-then-cancel points, used as negative controls to check that prag retracts stale implicatures.

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

🎓University of TorontoToronto · Canada96 works · 142,007 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 1Research subjects and languages the work is about.Implicature · 138 works

Cite

Cite as Daniel Marcu and Graeme Hirst. 1995. A Uniform Treatment of Pragmatic Inferences in Simple and Complex Utterances and Sequences of Utterances. 33rd Annual Meeting of the Association for Computational Linguistics.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.3115/981658.981678 · arXiv cmp-lg/9504017

@inproceedings{marcu1995uniform,
  title = {{A Uniform Treatment of Pragmatic Inferences in Simple and Complex Utterances and Sequences of Utterances}},
  author = {Daniel Marcu and Graeme Hirst},
  booktitle = {33rd Annual Meeting of the Association for Computational Linguistics},
  year = {1995},
  doi = {10.3115/981658.981678},
  eprint = {cmp-lg/9504017},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.3115/981658.981678}
}

References

23 references, 4 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. G. Frege. 1892. 0bet sinn und bedeutung. Zeitschrift fiir Philos. und Philos. Kritik, 100:373-394. reprinted as: On Sense and Nominatum, In Feigl H. and Sellars W., editors, Readings in Philosophical Analysis, pages 85-102, Appleton-Century-Croft, New York, 1947. find
  2. G.J.M. Gazdar. 1979. Pragmatics: Implicature, Presupposition, and Logical Form. Academic Press. find
  3. N. Green and S. Carberry. 1994. A hybrid reasoning model for indirect answers. In Proceedings 3Pnd Annual Meeting of the Association for Computational Linguistics, pages 58-65.
  4. N. Green. 1990. Normal state implicature. In Proceedings 28th Annual Meeting of the Association for Computational Linguistics, pages 89-96. find
  5. N. Green. 1992. Conversational implicatures in indirect replies. In Proceedings 30th Annual Meeting of the Association for Computational Linguistics, pages 64-71.
  6. J.B. Hirschberg. 1985. A theory of scalar implicature. Technical Report MS-CIS-85-56, Department of Computer and Information Science, University of Pennsylvania. Also published by Garland Publishing Inc., 1991. find
  7. G. Hirst, S. McRoy, P. Heeman, P. Edmonds, and D. Horton. 1994. Repairing conversational misunderstandings and non-understandings. Speech Communication, 15:213-229. ↗
  8. G. Hirst. 1991. Existence assumptions in knowledge representation. Artificial Intelligence, 49:199-242. ↗
  9. L. Karttunen and S. Peters. 1979. Conventional implicature. In Oh C.K. and Dinneen D.A, editors, Syntaz and Semantics, Presupposition, volume 11, pages 1-56. Academic Press. find
  10. L. Karttunen. 1974. Presupposition and linguistic context. Theoretical Linguistics, 1:3-44. ↗
  11. P. Kay. 1992. The inheritance of presuppositions. Linguistics £4 Philosophy, 15:333-379. ↗
  12. D. Marcu. 1994. A formalism and an algorithm for computing pragmatic inferences and detecting infelicities. Master's thesis, Dept. of Computer Science, University of Toronto, September. Also published as Technical Report CSRI-309, Computer Systems Research Institute, University of Toronto. ↗
  13. D. Marcu and G. Hirst. 1994. An implemented formalism for computing linguistic presuppositions and existential commitments. In H. Bunt, R. Muskens, and G. Rentier, editors, International Workshop on Computational Semantics, pages 141-150, December.
  14. S. McRoy and G. Hirst. 1993. Abductive explanation of dialogue misunderstandings. In Proceedings, 6th Conference of the European Chapter of the Association for Computational Linguistics, pages 277-286, April. find
  15. R.E. Mercer. 1987. A Default Logic Approach to the Derivation of Natural Language Presuppositions. Ph.D. thesis, Department of Computer Science, University of British Columbia. find
  16. W.V.O. Quine. 1949. Designation and existence. In Feigl H. and Sellars W., editors, Readings in Philosophical Analysis, pages 44-51. Appleton-Century-Croft, New York. find
  17. R. Reiter. 1980. A logic for default reasoning. Artificial Intelligence, 13:81-132. ↗
  18. B. Russell. 1905. On denoting. Mind n.s., 14:479-493. reprinted in: Feigl H. and Sellars W. editors, Readings in Philosophical Analysis, pages 103-115. find
  19. Applcton-Century-Croft, New York, 1949. find
  20. R.A. van der Sandt. 1992. Presupposition projection as anaphora resolution. Journal of Semantics, 9:333-377. ↗
  21. J.M. Siskind and D.A. McAllester. 1993. Screamer: A portable efficient implementation of nondeterministic Common Lisp. Technical Report IRCS-93-03, University of Pennsylvania, Institute for Research in Cognitive Science, July 1. find
  22. R.M. Weischedel. 1979. A new semantic computation while parsing: Presupposition and entailment. In Oh C.K. and Dinneen D.A, editors, Synta~ and Semantics, Presupposition, volume 11, pages 155-182. Academic Press. find
  23. H. Zeevat. 1992. Presupposition and accommodation in update semantics. Journal of Semantics, 9:379-412. ↗

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/marcu-1995-uniform-treatment-pragmatic-inferences-simple/