The SUMMA project implementation is arranged in nine work packages and will operate over 36 months. The project started in February 2016.
Work package 1 is concerned with the development of the initial use cases (external and internal monitoring), the detailed user requirements, and the development of a third use case (data journalism). WP2 concerns data collection (both training data and large-scale streams of “as-live” data), data and metadata management, and the definition of an initial ontology.
Work packages 3-4-5 are the main R&D work packages. Work package 3 focuses on shallow stream processing including speech recognition, machine translation, metadata extraction, and story clustering. Work package 4 deals with the extraction of entities and relations in order to automatically construct knowledge bases. Work package 5 is concerned with deep natural language understanding including semantic parsing, summarisation, and sentiment identification.
Work package 6 takes care of the integration into the overall SUMMA platform, including backend and visualisation systems for the use cases. Work package 7 involves field testing and evaluation at the component level, at the platform level, and in prototype deployment at BBC, DW, and user group partners.