Automatic Construction of a Grounded Multimedia Ontology of Objects to Illustrate Concepts in a Learning Process Adrian Popescu, Christophe Millet, Pierre - Alain Moëllic, Patrick Hède, Gregory Grefenstette Commissariat à l'Energie Atomique - LIST/LIC2M {adrian.popescu, patrick.hede, gregory.grefenstette}@cea.fr { milletc, moellicp}@zoe.cea.fr Abstract When building an e-learning system, we need a way of organizing the information to be presented to the user. Formal ontologies are one way of resolving this knowledge structuring task. The main use of ontologies in e-learning systems concerns the modelling of the domain of interest and this is equally our approach. Here we present a technique for automatically transforming unstructured data (from text and images) about domains into grounded ontologies that can then be used for e-learning applications. The chosen domain is “tools”, a collection that contains only picturable objects. We implement a grounded ontology that provides exemplars (definitions and images) at each node in the hierarchy. These exemplars can be used to illustrate concepts and ideas presented during learning, or in response to a learner’s questions. One main characteristic of our method is the fully automatic population of large scale ontology. 1. Introduction With the exponential increase of the quantity of data to be manipulated in computer applications, it becomes essential to efficiently organize and manipulate information. The richness of the included information is an essential condition for a good interactive application, but it raises serious problems concerning the management of this information. One successful solution for information organization is the use of formal ontologies [6]. It is undisputed that the hierarchical organization of concepts is proper to the way humans model entities in the world and this fact supports the use of ontologies in interactive computer applications, such as e-learning systems [1], [3], [10]. Our approach is close to that in [1] in that we propose the use of images and text organized as a multimedia ontology as fundamental layer for an e-learning application. The rest of this paper is organized as follows. We continue, in Section 2, by situating our work in the context of the Semantic Web [4] and by briefly discussing the relation between this initiative and e- learning. In Section 3, we discuss current approaches to ontology construction and present the methodology we developed for automatically building multimedia ontologies. We equally describe the way entities are represented in the created ontology. We illustrate, in Section 4, the utility of our work for e-learning by presenting an example of a learning situation. As we speak about an ongoing project, before concluding, we present the perspectives for enlarging and ameliorating both the linguistic and the image parts of the ontology. 2. Semantic Web and e-Learning The Semantic Web (SW) initiative [4] aims at better structuring the data on the Internet. It proposes a stack, with layers going from unique description of entities (URI’s) to a trust level that is, for the moment only theoretically described. From our point of view, the most interesting layer is the ontological one. It provides means for formally describing a domain in the world. Moreover, this level is technologically specified enough to allow the easy creation of applications. There exists an associated language, OWL (Ontology Web Language) as well as easy to use dedicated software, like Protégé [12]. As underlined in [13] the Semantic Web and e- Learning applications are strongly related. Through our approach, we propose a tight relation as we build an ontology, intended to be central in e-learning applications, using OWL component of the SW. This way, the integration of e-Learning applications in the SW framework becomes straightforward, as they are based on the same technology.