We present components of a system which uses statistical, corpus-based machine learning techniques to perform instantiated goal recognition -- recognition of both a goal schema and its parameter values. We first present new results for our previously reported statistical goal schema recognizer. We then present our goal parameter recognizer. The recognizer is fast (linear in the number of observed actions) and is able to recognize parameter values which were never seen in training. In addition, this is the first system we are aware of that uses statistical methods to recognize goal parameter values.
Nate Blaylock, James F. Allen