Towards an Online Automatic Translation System of Amazigh Documents based on UNL Graphs

Abstract

The use of IT tools with Tamazight Berber is an absolute requisite for giving this language full citizenship on the Web in particular and in the digital world in general. Thus, the need to create Tamazight digital documents is becoming increasingly urgent. Granted, Unicode now includes the full Tifinagh character set, but the question that remains is how to implement information interchange between Tamazight and the languages of the Web. We contend that the best and simplest way to share concurrent revisions of the same text in multiple languages is to coedit text in natural language and then to render it into Interlingua (IL) for dissemination. This method allows the participants to (i) translate the text, with modifications if any, from L0 into IL, and subsequently to (ii) regenerate the text in L1 ... Ln starting from the same IL representation.

Generators will never be perfect. Therefore, manual editing should always be allowed, because the automatically generated IL form may be irremediably inexpressive, or may altogether be unavailable due to lack of relevant data in the knowledge base. Since Universal Networking Language (UNL) graphs seem to be the best tool for the job, human participants should use a UNL editor for manual translation and revising. We propose that the collaborative manual translations be done on the Web with the help of the bank of multilingual utterances compiled by C. Boitet's team (GETA, CLIPS, IMAG à Grenoble, France). The resulting Tamazight utterances should then be integrated in a UNLXML document. At a later stage it will be necessary to build a UNLTamazight deconverter and a TamazightUNL enconverter, relying on the knowledge base built on top of the objects that have been tagged so far.