--- title: "Answering Clarification Questions — paper by Purver et al. (2003)" description: "The paper asks how people interpret clarification requests (CRs) in dialogue and how they respond to… 15 references, cited by 6 papers here." authors: ["Matthew Purver", "Patrick G.T. Healey", "James King", "Jonathan Ginzburg", "Greg J. Mills"] year: 2003 venue: "Proceedings of the Fourth SIGdial Workshop of Discourse and Dialogue" source: "https://aclanthology.org/W03-2103/" licence: "CC BY-NC-SA 3.0 (ACL Anthology, before 2016)" licence_url: "https://creativecommons.org/licenses/by-nc-sa/3.0/" summary_licence: "Elamur Cosmic License" canonical: "https://research.elamur.ai/free/en/papers/purver-2003-answering-clarification-questions/" --- [Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/purver-2003-answering-clarification-questions/) [home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2003](https://md.research.elamur.ai/free/en/papers/2003/) / Answering Clarification Questions # Answering Clarification Questions **Authors:** [Matthew Purver](https://md.research.elamur.ai/free/en/authors/matthew-purver/), [Patrick G.T. Healey](https://md.research.elamur.ai/free/en/authors/patrick-g-t-healey/), James King, [Jonathan Ginzburg](https://md.research.elamur.ai/free/en/authors/jonathan-ginzburg/) and Greg J. Mills · **Year:** 2003 · **Venue:** [Proceedings of the Fourth SIGdial Workshop of Discourse and Dialogue](https://md.research.elamur.ai/free/en/venues/sigdial-workshop-of-discourse-and-dialogue/) ([SIGdial Workshop of Discourse and Dialogue 2003](https://md.research.elamur.ai/free/en/venues/sigdial-workshop-of-discourse-and-dialogue/2003/)) · **Licence:** [CC BY-NC-SA 3.0 (ACL Anthology, before 2016)](https://creativecommons.org/licenses/by-nc-sa/3.0/) **Links:** [Publisher's page](https://aclanthology.org/W03-2103/) · [PDF](https://aclanthology.org/W03-2103.pdf) · [OpenAlex](https://openalex.org/W2578244785) **Citations:** 18 citations (OpenAlex, 2026-10-14) · [6](https://md.research.elamur.ai/free/en/papers/purver-2003-answering-clarification-questions/cited-by/) in our library, of the 149,439 papers whose bibliographies we read ## Summary The paper asks how people interpret clarification requests (CRs) in dialogue and how they respond to them. Both questions matter for dialogue systems, which must handle CRs from users and pose their own. The authors first re-annotated an existing BNC sample of 406 CRs. They marked the kind of response, how far the answer came after the CR, and links between CR form or reading and answer type. They then built a text-chat experiment that silently inserts artificial reprise fragments (RFs) into live conversations between participants. This let them vary the word category of the echoed word (content vs. function) and whether it was a first or a repeated mention. The main findings: - CRs go unanswered surprisingly often. - Answers that do come arrive almost always in the very next turn. - CR form predicts the shape of the answer. - Both part of speech and grounding level reliably shift how an RF is read. - First mentions get answered more often than repeated ones. These results give concrete cues for disambiguating CRs and for choosing how a system should ask its own. ## Key points - The BNC data contain 406 CRs. About 39% look unanswered, but this falls to 17% once two groups are excluded: possible answers transcribed as \, and cases where the asker kept talking. - Of 289 answered CRs, 273 (about 94%) were answered in the next sentence and none later than 3 sentences. Preliminary figures for ordinary questions show only about 70% answered at distance 1, with some reaching distance 6. - CR form predicts answer type: - Sluices get fragment answers and never yes/no. - Conventional CRs ("Pardon?") mostly get full sentences. - RFs get fragments or polar particles. - Constituent-reading CRs are roughly twice as likely to be answered (χ² significant). - In the chat experiment, content-word RFs were read as Clausal 53%, Constituent 29% and Other 18%. Function-word RFs were read almost entirely as Other. - First-mention RFs received a reply 58% of the time, versus 33% for second mentions. The RF Clausal/Constituent split was 87%/6% in the BNC but 48%/34% in the chat setting. ## Interesting ideas - Separate a CR's surface form (taxonomy of reprise fragment, sluice, conventional, etc.) from its reading (clausal, constituent, lexical), because one form can carry several readings. This keeps the ambiguity explicit instead of forcing one label. - Use the echoed word's category and its grounding status as cheap, explainable cues for which reading applies. A repeated function word rarely asks what something means, while a newly introduced content word often does. - Treat "answer comes immediately or not at all" as a structural constraint. Clarification answers are far more local than answers to ordinary questions. - Count non-responses as a real outcome and split them into subtypes (truly ignored, asker continued, unclear transcription). This avoids overestimating how often questions fail. - Insert artificial, controlled turns into otherwise natural interaction to study rare phenomena that corpora barely contain. Participants did not notice the manipulation. ## For Mova How these ideas could serve Mova, our small open language model — its files, stages and tests: - **I1** (frames): prag could tag echo-style questions with a form label plus a ranked reading, so that "Paris?" becomes either a check on the whole claim or a request to explain "Paris". - **I2** (world): the world model already tracks which entities have been mentioned, so first-mention vs. repeat status could feed straight into choosing a constituent reading for clarification questions. - **I3** (answer): when resolving an elliptical fragment after a clarification question in a dialogue-style text, readers can restrict the search to the adjacent turn, which narrows the candidates and keeps the reasoning traceable. - **I4** (learn): evaluation of QA on dialogues could include explicit "no answer present" controls with labelled reasons, serving as negative cases the system must not hallucinate answers for. - **I5** (learn): self-play or silver-data pipelines could inject targeted perturbations, such as echoes or synonym/hypernym swaps, into texts to probe and train specific behaviours that natural treebanks rarely contain. *Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.* 📚 Shelves (4): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 18,614 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 17,965 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 17,450 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 17,296 works 🧭 Topics (1): [🗨️ Dialogue systems](https://md.research.elamur.ai/free/en/topics/dialogue-systems/) · 654 works ## Cite Cite as: Matthew Purver, Patrick G.T. Healey, James King, Jonathan Ginzburg and Greg J. Mills. 2003. Answering Clarification Questions. Proceedings of the Fourth SIGdial Workshop of Discourse and Dialogue. The paper itself, not this page. BibTeX from the library's record of the work. ```bibtex @inproceedings{purver2003answering, title = {{Answering Clarification Questions}}, author = {Matthew Purver, Patrick G.T. Healey, James King, Jonathan Ginzburg and Greg J. Mills}, booktitle = {Proceedings of the Fourth SIGdial Workshop of Discourse and Dialogue}, year = {2003}, url = {https://aclanthology.org/W03-2103/} } ``` - [← previous Arabic Syntactic Trees: from Constituency to Dependency](https://md.research.elamur.ai/free/en/papers/zabokrtsky-2003-arabic-syntactic-trees-constituency-dependency/) - [next → A Plethora of Methods for Learning English Countability](https://md.research.elamur.ai/free/en/papers/baldwin-2003-plethora-methods-learning-english-countability/) - [↑ back to papers of 2003](https://md.research.elamur.ai/free/en/papers/2003/) ## References 15 references, 1 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet. 1. J.B. Bavelas, N. Chovil, D. Lawrie, and L. Wade. 1992. Interactive gestures. Discourse Processes, 15:469– 489. [find](https://scholar.google.com/scholar?q=J%2EB%2E+Bavelas%2C+N%2E+Chovil%2C+D%2E+Lawrie%2C+and+L%2E+Wade%2E+1992%2E+Interactive+gestures%2E+Discourse+Processes%2C+15%3A469%E2%80%93+489%2E) 2. Lou Burnard. 2000. Reference Guide for the British National Corpus (World Edition). Oxford University Computing Services. [find](https://scholar.google.com/scholar?q=Lou+Burnard%2E+2000%2E+Reference+Guide+for+the+British+National+Corpus+%28World+Edition%29%2E+Oxford+University+Computing+Services%2E) 3. Herbert H. Clark and Deanna Wilkes-Gibbs. 1986. Referring as a collaborative process. Cognition, 22:1–39. [↗](https://doi.org/10.1016/0010-0277(86)90010-7) 4. Sorin Dusan and James Flanagan. 2002. Adaptive dialog based upon multimodal language acquisition. In Proceedings of the Fourth IEEE International Conference on Multimodal Interfaces, Pittsburgh, October. [find](https://scholar.google.com/scholar?q=Sorin+Dusan+and+James+Flanagan%2E+2002%2E+Adaptive+dialog+based+upon+multimodal+language+acquisition%2E+In+Proceedings+of+the+Fourth+IEEE+International+Conference+on+Multimodal+Interfaces%2C+Pittsburgh%2C+Octob) 5. Christiane Fellbaum, editor. 1998. WordNet: An Electronic Lexical Database. MIT Press. [find](https://scholar.google.com/scholar?q=Christiane+Fellbaum%2C+editor%2E+1998%2E+WordNet%3A+An+Electronic+Lexical+Database%2E+MIT+Press%2E) 6. Simon Garrod and Gwyneth Doherty. 1994. Conversation, co-ordination and convention: an empirical investigation of how groups establish linguistic conventions. Cognition, 53:181–215. [↗](https://doi.org/10.1016/0010-0277(94)90048-5) 7. Jonathan Ginzburg and Robin Cooper. 2001. Resolving ellipsis in clarification. In Proceedings of the 39th Meeting of the ACL, pages 236–243. Association for Computational Linguistics, July. Jonathan Ginzburg and Robin Cooper. forthcoming. Clarification, ellipsis, and the nature of contextual updates. Linguistics and Philosophy. [find](https://scholar.google.com/scholar?q=Jonathan+Ginzburg+and+Robin+Cooper%2E+2001%2E+Resolving+ellipsis+in+clarification%2E+In+Proceedings+of+the+39th+Meeting+of+the+ACL%2C+pages+236%E2%80%93243%2E+Association+for+Computational+Linguistics%2C+July%2E+Jonathan+G) 8. Beth Ann Hockey, Deborah Rossen-Knill, Beverly Spejewski, Matthew Stone, and Stephen Isard. 1997. Can you predict answers to Yes/No questions? Yes, No and Stuff. In Proceedings of Eurospeech ’97. [find](https://scholar.google.com/scholar?q=Beth+Ann+Hockey%2C+Deborah+Rossen%2DKnill%2C+Beverly+Spejewski%2C+Matthew+Stone%2C+and+Stephen+Isard%2E+1997%2E+Can+you+predict+answers+to+Yes%2FNo+questions%3F+Yes%2C+No+and+Stuff%2E+In+Proceedings+of+Eurospeech+%E2%80%9997%2E) 9. Adam Kilgarriff. 1997. Putting frequencies in the dictionary. International Journal of Lexicography, 10(2):135–155. [↗](https://doi.org/10.1093/ijl/10.2.135) 10. Kevin Knight. 1996. Learning word meanings by instruction. In Proceedings of the Thirteenth National Conference on Artifical Intelligence, pages 447–454. AAAI/IAAI. [find](https://scholar.google.com/scholar?q=Kevin+Knight%2E+1996%2E+Learning+word+meanings+by+instruction%2E+In+Proceedings+of+the+Thirteenth+National+Conference+on+Artifical+Intelligence%2C+pages+447%E2%80%93454%2E+AAAI%2FIAAI%2E) 11. A. Mikheev. 1997. [Automatic rule induction for unknown word guessing](https://md.research.elamur.ai/free/en/papers/mikheev-1997-automatic-rule-induction-unknown-word/). Computational Linguistics, 23(3):405–423. 12. Matthew Purver, Jonathan Ginzburg, and Patrick Healey. 2001. On the means for clarification in dialogue. In Proceedings of the 2nd ACL SIGdial Workshop on Discourse and Dialogue, pages 116–125. Association for Computational Linguistics, September. [find](https://scholar.google.com/scholar?q=Matthew+Purver%2C+Jonathan+Ginzburg%2C+and+Patrick+Healey%2E+2001%2E+On+the+means+for+clarification+in+dialogue%2E+In+Proceedings+of+the+2nd+ACL+SIGdial+Workshop+on+Discourse+and+Dialogue%2C+pages+116%E2%80%93125%2E+Associ) 13. Matthew Purver, Jonathan Ginzburg, and Patrick Healey. 2002. On the means for clarification in dialogue. In R. Smith and J. van Kuppevelt, editors, Current and New Directions in Discourse & Dialogue. Kluwer Academic Publishers. [↗](https://doi.org/10.1007/978-94-010-0019-2_11) 14. Matthew Purver. 2002. Processing unknown words in a dialogue system. In Proceedings of the 3rd ACL SIGdial Workshop on Discourse and Dialogue, pages 174– 183. Association for Computational Linguistics, July. [find](https://scholar.google.com/scholar?q=Matthew+Purver%2E+2002%2E+Processing+unknown+words+in+a+dialogue+system%2E+In+Proceedings+of+the+3rd+ACL+SIGdial+Workshop+on+Discourse+and+Dialogue%2C+pages+174%E2%80%93+183%2E+Association+for+Computational+Linguistics) 15. E. Schegloff. 1987. Some sources of misunderstanding in talk-in-interaction. Linguistics, 25:201–218. Figure 1: Chattool Client Interface Subject A’s View A: Obviously the relatives were coming around like they do to see me A: yeah Figure 2: Story Telling Task Excerpt, Noun Clarification, Subjects 1 & 2 Subject A’s View A: so we agree B: agree? A: yeah to chuck out Susie derkins B: uh huh A: yes Figure 3: Balloon Task Excerpt, Verb Clarification, Subjects 3 & 4 Subject B’s View B: Obviously the relatives were coming around like they do to see me → A: relatives? B: Yeah just unts and uncles → A: ah B: yeah Probe Block Ack Subject B’s View B: so we agree ← Probe Block ← Ack B: yes [↗](https://doi.org/10.1515/ling.1987.25.1.201) *Updated 1 November 2026*