Towards Component-based, Domain-specific, Efficient Question Answering Systems

Andreas Both

Abstract: In the past, the research, as well as industry, brought Question Answering (QA) into daily use. However, there is still an obvious gap between the claim of providing access to any--structured or unstructured--knowledge stored in the world using an interface fitting the demands of regular users. On the one hand side, implementing Question Answering Systems is still hard and time-consuming, on the other hand side, the QA community is still struggling on defining a common ground for collaboration across research fields regarding, for example, realistic benchmarks, maintainability, and broad coverage of knowledge sources. In the talk, challenges of Question Answering will be highlight w.r.t. hybrid QA, domain-specific QA, cross knowledge base QA, etc. Particularly the industry perspective is also presented while aiming at a Question Answering platform which can be assembled from industry components as well as components of the research community. First steps towards this long-term vision are provided by the Qanary framework and similar frameworks aiming at a collaborative approach for the development QA systems which should lead to effective implementations, improved research results as well as a platform economy for QA. Such a platform would industry lead to a tighter collaboration while academics would have the opportunity of accessing precious data for further improvements.