In this paper we illustrate problems encountered together with proposed solutions when machine translation has to deal with languages with different family origins. The languages chosen are Classical Arabic, Mandarin Chinese and French. The machine translation architectures used are an adapted transfer methodology, and a "global" methodology in seven steps, which is new. What is common between the problems encountered is semantics. If one wishes to have machine translation of high quality there are two issues that must be resolved. The first of these is the semantic aspect of the linguistic units and the second is that the translation calculus has to be adapted to the languages to be processed.