We present a method for dimension reduction in multicriteria preference modeling, when taking a bounded distributive lattice as evaluation space. This method aims to reducing the complexity of learning an aggregating model on qualitative data. In this setting, the aggregation model of choice is the Sugeno integral. Learning such a model with qualitative data can thus be turned into an optimization problem with 2n parameters (where n is the number of considered criteria). The reducing method that we propose is based on such qualitative data, and we will give some preliminary application results.