This paper describes a currently running project at IDIAP, aiming at the development of a continuous spoken language recognition system. A particular attention is paid to two important topics under investigation. Firstly, the system characteristics, namely its structure as a multi-agent system and secondly, its search mechanism of a global solution as an optimisation procedure, are studied. As this project is definitely an undergoing research activity, many still important open questions have to be addressed. The major one, concerning a mathematical foundation for the resolution mechanism is covered in a section dedicated to further developments.