Generating English Paraphrases From Formal Relational Calculus Expressions

What it is about

The system paraphrases a database query, written as a relational calculus expression, so that a casual user can check how the query was understood. The paper argues that a conceptual model of the database and a rule for choosing a focus matter more than a full syntactic generator (the prototype had only an ad hoc surface step). The focus is the relation that roots a tree over the other relations of the query; a query whose relations are not linked this way is "not paraphrasable". The paraphrase is built recursively from the focus, and the scope of connectives is shown by indentation. The model is written by hand for each database. There are two worked examples and no quantitative evaluation.

Ideas

• I1. world/src/qud.rs: choose a focus for each question as the relation or fact that the question is about, and build the question as a tree over the facts it needs. Refuse a question whose facts are not connected, with the reason, rather than answer it. Measure the refusal rate and the refusal reasons on the school questions and on the bAbI tasks (world/src/babi.rs); a refusal must be right in the same cases as the gold answer is missing.
• I2. Scope by indentation in the explanation output (world/src/qa.rs): when the system explains an answer, show the nested conditions (and, or, not) as an indented list in the order of the query tree. The output format is fixed, so a test can compare it byte for byte.
• I3. A domain table for each school course: for each relation, the verb or phrase that expresses it, with the role of each argument ("to place": customer first, order second). Keep it in world/src/knowledge.rs, and report the share of lesson facts with a phrase for their relation.
• I4. Negative control for the focus rule: choose a random focus among the relations of a question and measure how much worse the paraphrase is judged, by a fixed checklist (all relations mentioned once, connectives in scope, focus first). A focus rule that does not beat the random one is not a gain.

For Mova

Focus and question structure belong to world/src/qud.rs and world/src/qa.rs; the fact tree is in world/src/knowledge.rs. Scope and connectives are in 27. Negation and 36. Discourse relations and coherence. Measures: the refusal rate and reasons for unconnected questions, the share of relations with a phrase in the domain table, and the random-focus control, run twice for identical output.

Cite

The paper itself, not this card. BibTeX from the library's record of the work. DOI 10.3115/991365.991534

@inproceedings{roeck1986generating,
  title = {{Generating English Paraphrases From Formal Relational Calculus Expressions}},
  author = {Roeck, A.N. De and Lowden, B.G.T.},
  booktitle = {Coling 1986 Volume 1: The 11th International Conference on Computational Linguistics},
  year = {1986},
  doi = {10.3115/991365.991534},
  url = {https://doi.org/10.3115/991365.991534}
}

References

11 references, 0 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. B.K.BOGURAEV and K. SPARCK JONES, "A Natural Language Front End to Databases with Evaluative Feedback", in GARANDIN and GELENBE (eds), New Applications of Databases, Academic Press, London, 1984. find
2. [2] W.F.CLOCKSIN and C.S.MELLISH, Programming Prolog, Springer Verlag, Berlin, 1981. in [3] E.F. CODD, "A Database Sublanguage founded on the ReJational Calculus", in Proceedings of the ACM SIGFIDET Workshop on Data Description, Access and Control, 1971. find
3. [4] E.F. COOD, "Relational Completeness of Database Sublanguages in Database Systems", in Courant Co_mmputer Science Series, Vol 6, Prentice I/all, Englewood Cliffs, 1972. find
4. [5] C.G. DATE, An Introduction to Database Systems (2nd edit.), Addison Wesley Publishing Co, Reading (Mass), 1977, [6] R. GRISHMAN, "Response Generation in Question - Answering Systems", in Proceedings of the 17th ACL, La Jolla, 1979. find
5. [7] P.K. H A L V O R S E N , "Semantics for L e x i c a l Functional Grammar", in Linguistic Inquiry, Vol 14, No 4, 1983. find
6. [8] INTERNATIONAL COMPUTERS LTD, Using Querymaster (200 Level), VME 2000, Publication R00260/00, 1983. find
7. [9] R. KAPLAN and J. BRESNAN, "Lexicai Functional Grammar. A Formal System for Grammatical Representation", in BRESNAN (ed), The Mental R~resetltation of Grammatical Relations, MIT Press, Cambridge ( M - , 1982. find
8. [I0] B.G.T. LOWDEN and A.N. DE ROECK, "Generating English Paraphrases from Relational Calculus Expressions", to a p p e a r in B e h a v i o u r and Information Technol0gy. [11] K. McKEOWN, "Paraphrasing using Given and New Information in a Question Answering System", in Proceedings of the 17th ACL, La Jolla, 1983. find
9. [12] E. M U E C K S T E I N , "Q-Trans: Query Language Translation into English", in Proceedings of the 8th IJKAI, Karlsruhe, 1983. find
10. [13] J.C. THOMAS and J.D. GOULD, "A Psychological Study of Query by Example", in Proceedings NCC 44, 1975. find
11. [14] M. WALLACE and V. WEST, "QPROC: A Natural Language Database Enquiry System Inplemented in Prolog", in ICL Technical Journal, November 1983. find

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/roeck-1986-generating-english-paraphrases-formal-relational/