In the present article a number of goals of integral MT systems are discussed: 1) Perfect MT, to solve all the problems of translation, from analysis of the source text to generation of the target, on any subject, from and into any language in an efficient way; 2) General interlingual MT, to generate reasonably correct translations in all subject areas from and into 'n' different languages, based on an interlingua as pivot; 3) Restricted interlingual MT, to generate pretty good translations in restricted domains from and into 'n' different languages based on an interlingua as pivot; and 4) General-purpose non-interlingual MT, to produce rought translations in any domain, from and into substets of language combinations, such that they can be used as an information tool or as input for a postediting mode, with research focused on high-priority trouble spots that repeatedly impair the smooth functioning of the system in operational settings.