Semantic relations at the Machine learning era: where are the (good old) patterns gone? (keynote talk)

International audienceAlthough looking for semantic relations in text has been the topic of a large variety of research works since the early 90's, it is still considered as a difficult task without easy solution. This question is addressed by more and more researchers with very different backgrounds, either in computer science (NLP, semantic web, information extraction, etc.), linguistics, terminology, etc. After various attempts of cross-disciplinary fertilization, it seems that each domain investigates its own solutions and methods without a real integration of results from other disciplines. For instance, the large efforts carried out in terminology and in computational linguistics to evaluate, define, improve pattern-based approaches are not much taken into account in Natural Language Processing, and even less in ontology learning. In a symmetric manner, relation extraction using machine learning has still little impact in linguistic analyses and when designing lexical resources. And Hearst's patterns are regularly the core of new work that little advances the state of the art. I will first draw an overview of the current approaches to identify semantic relations in text, with a special focus on the ways pattern -based solutions have evolved since early works. Then I will present a few works that have evaluated the complementarity of various techniques to support this task. I will finally promote the idea to capitalize better all the experiments and tools developed up to now, in particular by sharing patterns and learning methods, but also but investigating more systematically how existing techniques can be used together in a single platform, and mutually benefit of each other's results