Humans solve the problems of interpreting multimedia documentation by both exploiting structural regularities in the data, as well as making use of common sense and specialized knowledge that explicitly represents the meaning of data. In spite of this simple observation, current approaches to processing content over the Internet are based either on a statistical approach, which exploits the regularities of the content encoded in a statistical model, or, alternatively, by logical approaches, by exploiting logical knowledge encoded via logical theories (such as ontologies). In both cases this unilateral approach leads to a limitation in the performance or in the quality of the results. We believe that combining statistical knowledge with logical knowledge in a unique system would improve the efficiency and the effectiveness of content management applications. Copilosk aims at investigating how statistical and logical knowledge can be combined and exploited in content and knowledge management, by proposing or extending a theoretical paradigm, test it in three use cases in the area of machine translation, content extraction from text, and image object recognition, and in case of positive results applying it to a real case scenario.

Data and Knowledge management

Human Language Technology

Technology for Vision