Two-dimensional dialogical patterns extraction

This article addresses the problem of semi-automatic regularity extraction in dialogues in the shape of dialogical patterns. We present a dynamic programming algorithm which runs in O(mA× n × mB) that enables to perform an unsupervised extraction of recurrent patterns from two-dimensional dialogue annotations (with mAand mBthe number of lines in the two considered arrays and n their number of columns). We show how it can be combined to a clustering heuristique in order to extract relevant regularities from an annotated corpus. The parameters of the method and the results obtained (i.e., dialogical patterns and pattern clustering solutions) are evaluated manually by an expert through an experiment.