The MiMo2 system is based on a fundamental distinction between `possible' and `best' translation (Landsbergen ). Linguistically, a source text can have many possible translations, i.e. many target language texts that are equivalent in meaning. In practice, some interpretations will be more plausible than others, and some translations of some interpretations will often be preferred over others on various grounds, e.g. style. The MiMo2 system attempts to capture the notion `linguistically possible translation', as will be clarified below. This leads to a system that produces plausible as well as implausible translations. In the future, the selection of the more plausible ones could be based on either interactivity or more machine intelligence.
The MiMo2 system differs from Landsbergen's work in two respects. First, it uses a transfer model, where the Rosetta system is based on an interlingua approach. Second, its rule formalism and linguistic theory are based on unification grammars and a lexicalist approach, where the Rosetta system uses M-grammars, a computationally tractable version of Montague grammar.