It is commonly believed that machine translation, with its sentence-based and strictly formal equivalences, is incompatible with a style of translations that is concerned with capturing and expressing the communicative force of a text. The author argues that this incompatibility is not unsurmountable: there are statable patterns for the realization of these resources which can often be easily incorporated into the MT output at the level of postediting. In the present article this idea is applied to the processing of real MT output and to the analysis of human translation. All the examples are from production texts submitted to SPANAM, the system which has been developed at the Pan American Health Organization.