Approaches to enlarge bilingual corpora of example sentences to more languages in memoriam

In a multilingual richly structured lexical data base such as Papillon, examples, citations, definitions and glosses expressed in each language have to be translated into all other languages and stored into the data base. Storing can be achieved in a simple and «!seamless!» way by introducing «!auxiliary» lexies and axies for these «!free language elements!». Translating all these elements into all languages is a necessary contribution to the Papillon project. In the TraCorpEx project, we pursue a more restricted goal: to enlarge bilingual corpora of parallel utterances to more languages, because some large such corpora have become freely available and one of them, the "Tanaka corpus" of Japanese-English sentence pairs, has been proven by J.!Breen to be useful as a source of examples while consulting JEDict. Hence, they can be also called "corpora of example sentences". The TraCorpEx methodology has 3 parts. (1) As such corpora almost always contain English, use several MT engines from English to new language L and an automatic comparaison scheme to get a "best suggestion" (or none) for translating each example in L. (2) Produce the "reference" translations in L by using a web-oriented, mutualization-based translation workbench implementing the «!Montaigne!» architecture to enable cooperative work of (preferably volunteer) translators. (3) If no MT engine is available for E-L, produce UNL graphs for the examples (manually, then interactively) and send them to a UNL-L deconverter. In the future, a «!coedition!» technique, still at the prototyping stage, could be used to improve the UNL graphs a posteriori and transparently from any language, and get improved translations in all target languages. Mots-clés!: Papillon multilingual data base, N-N translation of dictionary information, Montaigne architecture, aligned bilingual corpora, example sentences, multilingual sentence memory (MSM).