From Texts to Ontologies: Non-taxonomical Relation Extraction

We propose a methodological approach which aims at creating and acquiring patterns from texts in order to enrich lightweight ontologies with non taxonomical relations. Our approach relies on an existing ontology, a morphosyntactically tagged corpus and an evolutive pattern bank. Our goal is twofold : (i)the morpho-syntactic patterns stored in the base are used to identify new relations between the concepts of the ontology (ii) new patterns are dened according to the syntactic contexts in the corpus of the terms that are concept labels. These patterns are added to the base and can be reused on a new corpus. At the end of the process, the initial ontology is enriched with new relations extracted from the corpus. The originality of our approach is that we guide the pattern denition process by suggesting abstract forms of patterns according to the context of co-occurring terms.