Analysis into a formal task-oriented pivot without clear abstract - semantics is best handled as "usual" translation

During the development of a multilingual international demo implemented by the CSTAR consortium, five of the six partners have adopted a task-oriented (task and domain specific) pivot architecture. To add French to the system, we have developed, among other components, an analyzer which converts written utterances, coming from a speech recognition system, into the pivot language known as IF (Interchange Format). Perhaps paradoxically, the natural character of the relevant utterances and the lack of formal semantics in the resulting IF structures has led us to construct this analyzer as if we were translating between two poorly defined natural languages. We will describe how we have used the Ariane-G5 environment while adopting a technique inspired both by the example-based machine translation paradigm and by older, "semantic" machine translation approaches.