Sponsored by DARPA, Project Moebius was a study of the feasibility of Learning by Reading, i.e. the ability to automatically enhance existing knowledge bases with information obtained from text. It necessitated the collaboration of three disciplines within computer science: Knowledge Representation and Reasoning (KR&R), Machine Learning (ML), and Natural Language Understanding (NLU).
Early Stage
In a nine month effort, The Friedland Group assembled a think-tank of world leading scientists in the three disciplines, and fleshed out a basic approach, specific use scenarios, a notional architecture, primary technical challenges and an evaluation framework.
Intermediate Stage
The Friedland Group led a two year effort
to provide quantitative proof that the Moebius approach was feasible.
The first year showed that concepts and relations extracted from
previously unseen text could be placed correctly in an existing
knowledge base. Year two showed that knowledge acquired from previously
unseen text could quantitatively improve the problem solving performance
of knowledge bases on previously unseen questions.