Christian BOITET
On the design of MIDDIM-DB,
a data base of ambiguities and
disambiguation methods
Abstract:Interactive disambiguation will be necessary in future Machine Translation and Machine Interpretation
systems to be used in non-restricted contexts by the general public. In order to determine which
ambiguities end users could and should help the system disambiguate, and which (possibly multimodal)
disambiguation methods are most appropriate, we have started to build MIDDIM-DB, a data base of real
ambiguities and simulated or implemented disambiguation methods.
A first version has already been prototyped in HyperCard, and has shown the need for some design
improvements. This paper elaborates on the motivations to build such a data base, and presents the main
aspects of the design of a second version. An important byproduct of this enterprise is an exact and
formal definition of what an ambiguity is, so that an ambiguity becomes a computational object.