Description de l'outil

Nom de l'outil Ariane-G5
Type d'outil Autre (préciser)
Complément sur le type Environnement de développement et d’exécution de systèmes (linguiciels) de traduction automatique
Mots clés TAO, langages spécialisés pour la programmation linguistique, environnements spécialisés pour la programmation linguistique
URL de l'outil

1. Short description of Ariane-G5/LIDIA

Ariane-G5 is a generator (G) of MT systems based on five (5) rule-based specialized languages for linguistic programming (SLLP). The components of an application written in a SLLP are variable (attribute type) definitions, formats (templates), grammars, and dictionaries. The components are compiled into a packed binary form. The compiled components are used as parameters by the "engine" of the corresponding SLLP (table-driven program).

The structures representing the units of translation are strings of characters (before and after translation) and otherwise trees decorated with complex attributes declared by the linguists and filled by information coming from the dictionaries or computed by the grammars.

As opposed to almost all existing systems, Ariane-G5 presents the advantage that the unit of translation is not restricted to the sentence, but may contain several paragraphs (in practice, up to 1 or 2 standard pages of 250 words or 1400 characters).

Ariane-G5 has a complete interactive interface which manages the lingware components as well as the data base of texts (and corresponding intermediate results) and makes sure that compiled files and intermediate results on texts are consistent with the source lingware at any time. It can also be piloted through the network from the CASH environment written in HyperCard on Macintosh : development can be done from anywhere in the world with a Mac, Hypercard and e-mail (to port it to Toolbook on PC would be easy).

Finally, Ariane-G5 has been complemented by the "LIDIA network", a network of virtual machines which makes it possible to use Ariane-G5 in distant applications using all or only parts of the translation process.

2. Implementation

Ariane-G5 runs under VM/ESA/CMS, on IBM computers with 390 architecture. Since 1993, it is accessible through the Internet. The minimal computer background necessary to use Ariane-G5 consists of learning the elementary commands for beginning and ending a VM/ESA session (login, logout), the XEDIT screen editor, and, for the developers of MT systems, the organization of the interactive monitor and the SLLPs.

The source code of Ariane-G5 proper represents 500,000 source lines in a mixture of assembler, PL360, PLI, EXEC, Pascal/VS, and REXX. This comprises the compilers and "engines" of the 5 languages for linguistic programming, and all the management routines for a specialized data-base containing the "lingware" (dictionaries, grammars) and the "corpuses" (texts in various forms : source, translations, intermediate results...), as well as all messages (30,000 for each dialogue language, currently French and English). The source code of the net-related extensions (LIDIA network, interfaces for e-mail, http, telnet...) represents about 100,000 lines, mostly in REXX and partly in assembler.

Informations complémentaires

Utilisation effective dans l’équipe élargie (même site), à l’extérieur
Disponibilité de l'outil pour la recherche et l'enseignement utilisable chez l’auteur, convention préalable
Disponibilité de l'outil pour une utilisation commerciale utilisable chez l’auteur, convention préalable
Disponibilité des sources sources non disponibles
Systèmes d'exploitation sur lesquels tourne l'outil  
Type d'interface d'utilisation Interface ligne de commande, Interface de programmation (API)
Ressources linguistiques associées disponibles Maquettes pédagogiques (BEX-FEX et FEW-BEX anglais-français) Autres systèmes (russe-français, anglais-divers, français-anglais spécialisé) ? : sur convention


Nom Durian GETA (divers auteurs)
Affiliation GETA (groupe d'étude pour la traduction automatique)


Nom Christian Boitet
Affiliation GETA, CLIPS, IMAG-campus


Ces renseignement ont été saisis le 04/11/1999 par Christian Boitet.