Georges FAFIOTTE & Christian BOITET
Specification of first EMMI Experiments
for the MIDDIM project in the context of
Interpreting Telecommunications
Abstract:This note specifies the first EMMI-based experiments conducted during our research stay at ATR,
in the framework of the MIDDIM project.
In these experiments, we want to test what influence variations in speech recognition ability and
language proficiency of the interpreter, and in domain knowledge of all participants, will have on the
number of disambiguation questions asked by the (automatic) interpreter.
We expect that the records, videos, and transcripts resulting from those experiments will provide
for the refinement of the preliminary user modeling we propose here.
We begin by specifying the envisaged dialogue scenario, and the most relevant characteristics of
the participants. Then we sketch a methodology for EMMI-based simulation situations. Finally, we
propose a first set of experiments.