The proposed disambiguation methodology is based on the manipulation of tree structures. A kind of ambiguity is described with a set of patterns called a beam. A pattern contains variables and descibes a tree structure with constraints on its geometry and labelling. Once a beam has been recognized, a question is prepared. The question items are produced through the manipulation, with a set of basic operators, of the values given to the variables instanciated during the recognition of the beam.
A particular disambiguation module is described with a lingware which is language and analyzer dependant. This lingware is then used as input data to an interactive disambiguation engine so as to describe a particular instance of a running disambiguation module. In this paper we are going to describe the engine and the lingware we have developed at ATR-ITL for English input.