Grant: Discovery Grants Program - Individual

The learning-by-reading challenge refers to a number of natural language understanding tasks dedicated to the extraction of a formal conceptual model from free texts and to the filtering and integration of this knowledge. In the context of the Semantic Web, this challenge often refers to the ability of learning a domain ontology from free text. This task is particularly challenging when the adopted approach is an open (domain independent) and unsupervised (without any guidance) knowledge extraction. Such a domain independent approach is crucial to the interoperability and reuse of ontology learning methodologies across various applications and domains. This research aims at developing a domain independent methodology and a set of tools for learning the ontology layers (terms, taxonomy, conceptual relations, relation hierarchies, axioms and axiom schemata) from unstructured knowledge sources. In particular, we aim at automatically learning the three last layers, which has been done very rarely in the state of the art.The research will also explore filtering techniques that identify important information and integration techniques that adequately combine knowledge coming from various sources. The interest of the availability of a complete ontology (with all its layers) is that it is essential for agents to reason over semantic data and this will provide effective knowledge retrieval capabilities. The long-term objective of this research is to build learning-by-reading systems by exploring the integration of deep semantic analysis, filtering and knowledge integration techniques. The short-term objective is to build a proof of concept by extracting a domain ontology using a combination of some of these techniques and by focusing on layers that require deep semantic analysis. The major contribution of our proposal is the extraction of domain conceptual models from unstructured sources enabling the indexing of texts and reasoning as well as the exploitation of such conceptual models in various areas such as the Semantic Web and the corporate world.