TRAnslation: OEM-able, Unified, Integratable Environment and Resources for Operationalization – Traouiero

The Traouiero project aims at allowing the operationalisation of software tools and lingware techniques and resources developed up to now by the GETALP group of the LIG lab as operational prototypes.

In the context of the Web 2.0, the GETALP proposed to replace the traditional paradigm of translation, expensive product paid once delivered, by that of multilingual access, a not very expensive service in the form of subscription, allowing the collaborative increase in the quality of automatically produced “pretranslations”.

A start-up was created around this concept, AXiMAG, a BU (Business Unit) of Floralis (UJF-subsidiary company) since September 2009, and in the process of becoming a SAS. AXiMAG starts to develop the products of the research of the GETALP, and quickly needs “operationalized” and "scaling up" versions and resources.

In addition, the GETALP works on the multimodal MT of phrasebooks, on smartphones, for under-resourced language pairs, and anticipated the possibility of embedded MT in the Ariane-Y project.

This project comprises 3 principal parts, named Ariane++, iMAG++, and LING++. It happens that, for each one of them, we have products or resources resulting from work undertaken for a long time, and others developed in recent PhD theses. There is currently a unique opportunity for operationalising the whole, while calling at the same time upon permanent staff with great expertise, and 3 postdocs very interested by the operationalisation of the software created for their thesis.

The Ariane++ part of the Traouiero project relates to (1) the finalization of the specialized languages (LSPL) of ARIANE and the tests on iPad/iPod and others smartphones, (2) the integration of SECTra_w as a corpus server for MT, (3) the integration of PIVAX as a lexical server for MT, (4) the adaptation to smartphones of the LSPL of Héloïse (ATEF-H, ROBRA-H…) and with the iPad/iPod, (5) the creation of a unified EDL accepting the LSPL of ARIANE-Y (+SysQ-Y) and of Héloïse, (6) the integration and tests on smartphones.

The iMAG++ part relates to (1) the modularization of the iMAG-software and the creation of the iMAG-relay, (2) the organization as agents with control by whiteboard and tasks in infinite loops, (3) the improvement of the TRADOH tool of parameterized call to several MT systems, (4) the creation of SegDoc, a proprietary segmentor-normalisor better than those of Google and others, (5) the specialized automatic generation of empirical MT systems, specialized to the sub-languages of the web sites accessed by iMAGs.

The LING++ part relates to (1) the consolidation of existing “expert” MT systems (Russian-French, English-French, French-English), (2) the reworking of modules of MT systems so that they are usable separately (lemmatisers for Russian, French, English, German; transcripters), (3) scaling up the OMNIA technique of interlingual semantic labelling of spontaneous texts (by interlingual lexemes such as UW++ of UNL), (4) scaling up the current multilingual lexical base related to UW++: from 200K to 1-2M entries (with monolingual and bilingual extraction of multiword expressions).

Each one of these 3 parts will be accompanied by the valorization partner, who will deal with the aspects related with intellectual property, and with valorization. It is important that this project leads to products allowing the open collaborative development of lingware resources, and that the produced software can be used freely for research while giving place to remunerations of associated services by the commercial users.