UNL-French deconversion as transfer & generation from an interlingua with possible quality enhancement through offline human interaction

What it is about

The UNL-French deconverter generates French from a UNL hypergraph in three stages. First a localisation step adapts the graph to French: a Universal Word with no entry in the French dictionary is replaced by the nearest entry under a pseudo-distance, and ties are broken at random in the default mode. Cultural localisation fills missing information such as gender, number and politeness through heuristics and user profiles. Then the graph is turned into a tree, and classical transfer and generation run in the Ariane-G5 environment, through an abstract structure whose paraphrase is chosen before the surface form. The second half of the paper proposes delaying human interaction until post-editing, and feeding the corrections back into the lexical preferences and into the UNL graph, so that later sentences and other languages improve. No accuracy figures are given.

Ideas

I1. Make the nearest-entry choice deterministic. When a meta concept has no encoder entry, substitute the nearest known concept under a counted distance, such as shared ontology parents or overlap of the seen contexts. Break ties by the BTreeMap order, not at random, and report the share of unseen concepts that get a sensible bare form.
I2. Feed corrections into counts. Keep a log of the encoder choices that a reviewer changed, and let each correction raise the count of the chosen chunk in its context. The correction log is a data file; the encoder does not need retraining.
I3. Log the structure lost in the graph-to-tree step. Count the arcs or attributes dropped when a graph becomes a tree, and report them as their own residual class, so the loss is visible.

For Mova

The nearest-entry substitution is the encoder's back-off in tools/translate/src/profile.rs made explicit. The post-edit feedback is the same mechanism as the author delta, applied to human corrections instead of the standard's errors. The graph-to-tree loss matches the encoder's tree-versus-string problem in open problem 3. Nothing from the paper is measured; the counts come first.

Topics: Machine translation

Cite

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

@inproceedings{serasset1999unl,
  title = {{UNL-French deconversion as transfer \& generation from an interlingua with possible quality enhancement through offline human interaction}},
  author = {Sérasset, Gilles and Boitet, Christian},
  booktitle = {Proceedings of Machine Translation Summit VII},
  year = {1999},
  url = {https://aclanthology.org/1999.mtsummit-1.33/}
}

References

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

Blanc É. & Guillaume P. (1997) Developing MT lingware through Internet: ARIANE and the CASH interface. Proc. Pacific Association for Computational Linguistics 1997 Conference (PACLING'97). Ohme, Japon, 2-5 September 1997, vol. 1/1, pp. 15-22. find
Blanchon H. (1994) Perspectives of DBMT for monolingual authors on the basis of LIDIA-1, an implemented mockup. Proc. 15th International Conference on Computational Linguistics, COLING-94, 5-9 Aug. 1994, vol. 1/2, pp. 115—119. find
Boitet C., Réd. (1982) "DSE-1"— Le point sur ARIANE-78 début 1982. Contrat ADI/CAP-Sogeti/Champollion (3 vol.), GETA, Grenoble, février 1982, 400 p. find
Boitet C. (1994) Dialogue-Based MT and self-explaining documents as an alternative to MAHT and MT of controlled languages. Proc. Machine Translation 10 Years On, 11-14 Nov. 1994, Cranfield University Press, pp. 22.1—9. find
Boitet C. (1997) GETA's MT methodology and its current development towards personal networking communication and speech translation in the context of the UNL and C-STAR projects. Proc. PACLING-97, Ohme, 2-5 September 1997, Meisei University, pp. 23-57. (invited communication) find
Boitet C. & Blanchon H. (1994) Multilingual Dialogue-Based MT for monolingual authors: the LIDIA project and a first mockup. Machine Translation, 9/2, pp. 99—132. find
Boitet C., Guillaume P. & Quézel-Ambrunaz M. (1982) ARIANE-78, an integrated environment for automated translation and human revision. Proc. COLING-82, Prague, July 1982, pp. 19—27. find
Brown R. D. (1989) Augmentation. Machine Translation, 4, pp. 1299-1347. find
Ducrot J.-M. (1982) TITUS IV. In "Information research in Europe. Proc. of the EURIM 5 conf. (Versailles)", P. J. Taylor, ed., ASLIB, London. find
Kay M. (1973) The MIND system. In "Courant Computer Science Symposium 8: Natural Language Processing", R. Rustin, ed., Algorithmics Press, Inc., New York, pp. 155-188. find
Maruyama H., Watanabe H. & Ogino S. (1990) An Interactive Japanese Parser for Machine Translation. Proc. COLING-90, Helsinki, 20-25 août 1990, ACL, vol. 2/3, pp. 257-262. find
Melby A. K., Smith M. R. & Peterson J. (1980) ITS : An Interactive Translation System. Proc. COLING-80, Tokyo, 30/9-4/10/80, pp. 424—429. find
Moneimne W. (1989) (159 p. +annexes) TAO vers I'arabe. Spécification d'une génération standard de I'arabe. Réalisation d'un prototype anglais-arabe à partir d'un analyseur existant. Nouvelle thèse, UJF. find
Nirenburg S. & al. (1989) KBMT-89 Project Report. Center for Machine Translation, Carnegie Mellon University, Pittsburgh, April 1989, 286 p. find
Nyberg E. H. & Mitamura T. (1992) The KANT system: Fast, Accurate, High-Quality Translation in Practical Domains. Proc. COLING-92, Nantes, 23-28 July 92, ACL, vol. 3/4, pp. 1069—1073. find
Quézel-Ambrunaz M. (1990) Ariane-G5 v.3 Le moniteur. GETA, IMAG, juin 1990, 206 p. find
Slocum J. (1984) METAL: the LRC Machine Translation system. In "Machine Translation today: the state of the art (Proc. third Lugano Tutorial, 2-7 April 1984)", M. King, ed., Edinburgh University Press (1987). find
Wehrli E. (1992) The IPS System. Proc. COLING-92, Nantes, 23-28 July 1992, vol. 3/4, pp. 870-874. find

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/serasset-1999-unl-french-deconversion-as-transfer/