Elamur Research⌕
☰
home / papers / 2004 / Intentions, Implicatures and Processing of Complex Questions

Intentions, Implicatures and Processing of Complex Questions

Paper · 2004 · Sanda Harabagiu, Steven Maiorano, Alessandro Moschitti et al.

Summary

The paper tackles complex "scenario" questions, where what the user actually wants goes beyond the literal wording. For example, asking what help North Korea got from the USSR/Russia for its missile program carries unstated sub-questions and assumptions. The authors split such a question into intended questions (definitions of the entities, factoid questions about the focus, and questions about information sources). Each intended question then yields implied questions, which are follow-ups about metonymy, timelines, reliability and alternatives.

Two ways of deriving this intentional structure are proposed:

  • Lexico-semantic method: follows WordNet paths from the question words and filters the newly reached concepts against Grice's conversational maxims.
  • Predicate-argument method: parses both the question and the candidate answers into shallow semantic structures, indexes their elements, and coerces evidence, topic, predicate and argument interpretations. It uses a probabilistic model that extends Lapata and Lascarides' work on verbal metonymy.

On a 100-question test, both methods improved answer correctness over a baseline with no coerced intentions. The predicate-argument method helped more. The work matters because it treats pragmatics as something that changes retrieval keywords and answer extraction.

Key points

  • The running example Q1 (North Korea missile assistance) breaks down into 6 intended questions, Q i 1 to Q i 6. Each one spawns implied questions; for instance, "USSR/Russia" may really mean the government, firms or individuals.
  • The question focus comes from the predicate-argument structure and the argument order. The purpose argument ranks highest, so for Q1 the focus is "missile program". All other arguments become definition questions.
  • The topic is modelled along three WordNet-style dimensions: hyponyms or examples, meronyms or components, and function or usage. The topic model also uses topic signatures (Hovy et al., 2002).
  • Coercion builds on the Lapata & Lascarides factorisation P(i)·P(v|i)·P(o|i,v), estimated by maximum likelihood. It is extended from verb→object pairs to any predicate-argument relation, and to the coercion of topic and evidence.
  • The evaluation used 100 questions under three settings: no intentional knowledge, lexico-semantic implicatures, and predicate-argument intentional structures. The extracted text is garbled here, but it appears to report gains of about 8% and about 22% respectively.

Interesting ideas

  • I1. Split a question into explicit intended sub-questions, then into implied follow-up questions linked back by pronouns ("it", "this", "ones"), so that the hidden intent becomes a set of checkable queries.
  • I2. Treat a named entity as a possible metonymy and create an implied question to confirm the reading, e.g. country versus government versus firms.
  • I3. Accept a WordNet path only if the concept it reaches survives a Gricean maxims check, as when "survive the crisis" coerces DANGER, a belief none of the words states.
  • I4. Interpret logical metonymy with the factorisation P(i)P(v|i)P(o|i,v), widened from verb-object pairs to any predicate-argument relation.
  • I5. Pick the question focus by role ranking, with the purpose argument first, and elaborate it along temporal, resultative, causative or manner lines.

For Mova

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

  • I1 (answer): A fable or tale reader could expand a hard question into definition, fact and source sub-questions and answer each one against the world of the text, keeping a visible trace.
  • I2 (world): When an entity like a country appears as the agent of an action, the world model could keep alternative referents with a recorded choice instead of collapsing them silently.
  • I3 (frames): The implicature module in prag could derive presupposed states (e.g. "is in jeopardy") from verb-plus-object combinations, with each inference step explainable as a WordNet path plus the maxim that licensed it.
  • I4 (frames): For "enjoy the movie" or "began the book", prag could fill in the implicit event (watching, reading) as an explicit frame, ranked by counts from silver data.
  • I5 (answer): A deterministic, role-based choice of focus could steer which world entities the answerer searches first, replacing keyword overlap.

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, 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 Sanda Harabagiu, Steven Maiorano, Alessandro Moschitti and Cosmin Bejan. 2004. Intentions, Implicatures and Processing of Complex Questions. Proceedings of the Workshop on Pragmatics of Question Answering at HLT-NAACL 2004.

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

@inproceedings{harabagiu2004intentions,
  title = {{Intentions, Implicatures and Processing of Complex Questions}},
  author = {Sanda Harabagiu, Steven Maiorano, Alessandro Moschitti and Cosmin Bejan},
  booktitle = {Proceedings of the Workshop on Pragmatics of Question Answering at HLT-NAACL 2004},
  year = {2004},
  url = {https://aclanthology.org/W04-2505/}
}

References

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

  1. John Chen and Owen Rambow. 2003. Use of deep linguistic features for the recognition and labeling of semantic arguments. In Proceedings of the 2003 Conference on Empirical Methods in Natural Language Processing. Abdessamad Echihabi and Daniel Marcu. noisy-channel approach to question answering. In Proceedings of the 41st Annual Meeting of the ACL, Sapporo, Japan. find
  2. Chin-Yew Lin E.H. Hovy, U. Hermjakob and Deepak Ravichandran. 2002. Using knowledge to facilitate pinpointing of factoid answers. In Proceedings of the 19th International Conference on Computational Linguistics (COLING 2002). find
  3. Daniel Gildea and Julia Hockenmaier. 2003. Identifying semantic roles using combinatory categorial grammar. In Proceedings of the 2003 Conference on Empirical Methods in Natural Language Processing. ↗
  4. Daniel Gildea and Daniel Jurafsky. 2002. Automatic labeling of semantic roles. Computational Linguistic, 28(3):254–288. ↗
  5. Daniel Gildea and Martha Palmer. 2002. The necessity of parsing for predicate argument recognition. In Proceedings of the 40th Annual Conference of the Association for Computational Linguistics (ACL-02), Philadelphia, PA.
  6. King-Shy Goh, Edward Chang, and Kwang-Ting Cheng. 2001. SVM binary classifier ensembles for image classification. In Proceedings of the tenth international conference on Information and knowledge management, pages 395–402. find
  7. Paul H. Grice. 1975a. Logic and conversation. In P. Cole and New York J.L. Morgan (ed.), Academic Press, editors, Syntax and Sematics Vol.3:Speech Acts, pages 41– 58. ↗
  8. Paul J. Grice. 1975b. Syntax and Semantics Vol.3:Speech Acts. P. Cole and J. Morgan, editors. find
  9. Kadri Hacioglu, Sameer Pradhan, Wayne Ward, Jim Martin, and Dan Jurafsky. 2003. Shallow semantic parsing using support vector machines. Technical report.
  10. Sanda Harabagiu, Marius Paşca, and Steven Maiorano. 2000. Experiments with open-domain textual question answering. In Proceedings of the 18th International Conference on Computational Linguistics (COLING-2000), pages 292–298, Saarbrucken, Germany,. find
  11. Sanda M. Harabagiu, Dan I. Moldovan, Marius Pasca, Rada Mihalcea, Mihai Surdeanu, Razvan C. Bunescu, Roxana Girju, Vasile Rus, and Paul Morarescu. 2001. The role of lexico-semantic feedback in open-domain textual question-answering. In Meeting of the ACL, pages 274–281. find
  12. S. Harabagiu, D. Moldovan, C. Clark, M. Bodwen, J. Williams, and J. Bensley. 2003. Answer mining by combining extraction techniques with abductive reasoning. In Notebook of the Twelveth Text REtrieval Converence (TREC-2003), pages 46–53. find
  13. A. Ittycheriah, M. Franz, W. Zhu, and A. Ratnaparkhi. 2000. IBM’s statistical question answering system. In Proceedings of the 9th Text REtrieval Conference, Gaithersburg, MD. 2003. find
  14. A T. Joachims. 1999. T. Joachims, Making large-Scale SVM Learning Practical. In B. Schölkopf and C. Burges and A. Smola (ed.), MIT-Press., editor, Advances in Kernel Methods - Support Vector Learning. find
  15. Maria Lapata and Alex Lascarides. 2003. A probabilistic account of logical metonymy. Computational Linguistics, 29:2:263–317. ↗
  16. Wendy Lehnert. 1978. The process of question answering. In Lawrence Erlbaum Assoc., Hillsdale. find
  17. Sameer Pradhan, Kadri Hacioglu, Wayne Ward, James H. Martin, and Daniel Jurafsky. 2003. Semantic role parsing: Adding semantic structure to unstructured text. In Proceedings of the International Conference on Data Mining (ICDM-2003). find
  18. Dan Moldovan Sanda Harabagiu and Takashi Yukawa. 1996. Testing gricean constraints on a wordnet-based coherence evaluation system. In Working Notes of the AAAI-96 Spring Symposium on Computational Implicature, Stanford, CA. find
  19. Mihai Surdeanu, Sanda Harabagiu, John Williams, and Paul Aarseth. 2003. Using predicate-argument structures for information extraction. In Proceedings of the 41th Annual Conference of the Association for Computational Linguistics (ACL-03), pages 8–15.
  20. V. Vapnik. 1995. The Nature of Statistical Learning Theory. Springer. ↗

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/harabagiu-2004-intentions-implicatures-processing-complex-questions/