Categorial grammars and Lambek calculus found a nice embedding in Linear Logic, and a lot of work have presented proof nets uses for linguistic purposes, with a special look at proof nets for Lambek calculus. But they have mainly explored the syntactic capabilities of proof nets, describing parsing processes. We present here our vision of the generation process based on semantic proof nets. The main features of this proposal consist in the use of proof nets for lambek calculus but also for semantic recipes of lexical entries, so that no lambda-term unification occur with their limitations (undecidability from second order). Based on a graph calculus, the solutions of this process can be expressed as a matrix computation.