Extraction of named entity relations in textual data is an important challenge in natural language processing. For that purpose, we propose a new data mining approach based on recursive sequence mining. The contribution of this work is twofold. First, we present a method based on a cross-fertilization of sequence mining under constraints and recursive pattern mining to produce a user-manageable set of linguistic information extraction rules. Moreover, unlike most works from the state-of-the-art in natural language processing, our approach does not need syntactic parsing of the sentences neither resource except the training data. Second, we show in practice how to apply the computed rules to detect new relations between named entities, highlighting the interest of hybridization of data mining and natural language processing techniques in the discovery of knowledge. We illustrate our approach with the detection of gene interactions in biomedical literature.