Game/AI

Regarding your comment:
"I know that Plan Recognition isn't massive in the gaming industry at the moment, but any emerging games/entertainment-related work would be welcomed."

Bot Colony ( www.botcolony.com ) relies on plan recognition to infer the player's goals, and use the knowledge of the player's goals to direct the characters behaviour. The behaviour of goal-driven intelligent robots who fulfill plans has been romanticized in the Bot Colony novel (see page 32, School), but this is actually what has been implemented in the game.
While the player's goals can only be inferred from his actions, which are part of plans, the behaviour of the other characters is entirely goal-driven. When they recognize a certain context, they spring into action and fulfill the plan triggered by that context. See A Generic Memory Module for Events by Dan Tecuci, 2007.

Both the player's and the characters' goals, plans and context in Bot Colony are scripted in Natural Language. We hope to make available the natural-language-based scripting environment to the user community, so that they can extend the game.

The research problem we're currently working on is working our way back, from the knowledge of the mission goals to the current context of the player, so we can anticipate the sub-goals and plans he's likely to select. This is a backward-chaining reasoning problem.

Thanks for your comment Eugene. I have been following the development of Bot Colony, and will be interested to see the technology in action when the game goes live. (when is that?)

You might be interested in the work of my former colleague at the Media Lab, Peter Gorniak, whose thesis was about plan recognition in games to drive behavior and natural language understanding:
http://petergorniak.org/papers/pgorniak_thesis.pdf

My own work is going a similar direction (uses event-recognition to understand natural language), but plans actions and utterances directly from large data-sets of annotated human-human interactions in games.

Jeff