A Pattern-based Analyzer for French in the Context of Spoken Language Translation: First Prototype and Evaluation

In this paper, we describe a first prototype of a pattern-based analyzer developed in the context of a speech-to-speech translation project using a pivot-based approach (the pivot is called IF). The chosen situation involves a French client talking to an Italian travel agent (both in their own language) to organize a stay in the Trentino area. An IF consists of a dialogue act, and a list, possibly empty, of argument values. The analyzer applies a "phrase spotting" mechanism on the output of the speech recognition module. It finds well-formed phrases corresponding to argument values. A dialogue act is then built according to the instantiated arguments and some other features of the input. The current version of the prototype has been involved in an evaluation campaign on an unseen corpus of four dialogues consisting of 235 speech turns. The results are given and commented in the last part of the paper. We think they pave the way for future enhancements to both the coverage an...

• Type: Journal
• Year: 2002
• Where: COLING
• Authors: Hervé Blanchon
• Keywords: COLING 2002, COLING 2008, Dans La Situation, Dialogue Act, Dialogues