A "Pivot" XML-Based Architecture for Multilingual, Multiversion Documents: Parallel Monolingual Documents Aligned Through a Central Correspondence Descriptor and Possible Use of UNL

We propose a structure for multilingual, multiversion documents, built on the model of the web-oriented, cooperative lexical multilingual data base PAPILLON: a document is represented by a collection of monolingual XML "volumes" interlinked by a central volume of "interlingual links". Here, the links relate subdocuments (XML trees) corresponding to each other in monolingual "volumes". We are developing a Java application to enable direct editing of a multilingual document through the web, at the level of monolingual volumes as well as through bilingual or trilingual interfaces inspired by those of commercial "translation workbenches". Another goal is easy integration with machine translation and multilingual generation tools. For this, we add a spe- cial UNL volume. In a first stage, we split the UNL-xml document in several monolingual documents, again represented by XML files. Each document con- tains the text in a particular language, plus the corresponding UNL graphs, and can be modified independently. The interface is easy to build, but realigning the documents after a series of such modifications is a very difficult task.