In this paper, we introduce two new methods dedicated to phrase based machine translation. Both are based on mining a parallel corpus in order to nd out the couples of linguistic units which are translation of each other. The presented methods do not rely on any alignment in contrast to what is done usually by the statistical machine translation community. Each of them proposes a complete translation table containing translations of single words and phrases. The rst method is inspired from the well-known trigger language model while the second one is inspired from the association rules mining technique. All experiments ar e conducted on a large part of EUROPARL corpus and highlight the utility of both proposed approaches.