Structuring natural language data by learning rewriting rules

Abstract. The discovery of relationships between concepts is a crucial point in ontology learning (OL). In most cases, OL is achieved from a collection of domain-specific texts, describing the concepts of the domain and their relationships. A natural way to represent the description as-sociated to a particular text is to use a structured term (or tree). We present a method for learning transformation rules, rewriting natural language texts into trees, where the input examples are couples (text, tree). The learning process produces an ordered set of rules such that, applying these rules to a text gives the corresponding tree