In group decision-making problems with linguistic comparison matrices, different decision makers may use linguistic term sets with different granularity and different semantic to represent their preferences. A new approach to deal with this issue was presented, which uniform the linguistic term sets with multi-granularity and multi-semantic into the normalized representations of two-tuple linguistic. It was approved that by this approach there is no information loss and the linguistic comparison matrix can keep its properties after unification. Then, based on some aggregation operators of 2-tuple linguistic, different decision makers' preference information was aggregated into the group preference information and most desirable alternative is selected. An example was presented to validate the proposed method.