Legal ontologies have proved their increasingly substantial role in representing, processing and retrieving legal information. By using the knowledge modeled by such ontologies in form of concepts and relations, it is possible to reason over the semantic content of legal documents. Supporting (semi-) automatically the development of ontologies from text is commonly referred to as ontology learning from text. The learning process includes learning of the concepts that will form the ontology and learning of the semantic relations among them.In this paper, we present a new approach for expliciting the semantic relations between Arabic compound nouns concepts. The originality of this work is twofold. Firstly, the technique of inferring relations is based on exploiting the internal structure of the compounds using a defined set of domain-and language-independent rules according to their different structures, on the one hand, and on studying prepositions semantics specifying the inferred relations applying a gamification mechanism that collects human votes, on the other hand. Secondly, relying on the compounds set described by both binary (structural positions in which there are written) and relational attributes (the deduced relations), we used a “Relational Concept Analysis” (RCA) technique, as an adaptation of “Formal Concept Analysis” (FCA), for the construction of interconnected lattices that we transformed into ontological concepts and relations which can be either taxonomic or transversal.Experiments carried out on Arabic legal dataset showed that the proposed approach reached encouraging performance through achieving high precision and recall scores. This performance affects positively the retrieval results of legal documents based on a powerful ontology, which presents our main objective.
Read full abstract