Reference Resolution

There are a host of problems that confront conversational agents.
One problem that is particularly sticky for agents in shared environments is reference resolution.
The problem of reference resolution in documents is difficult, but this is compounded by the dynamic environment.
In the example scenario, the enemy was referenced. Which enemy?
In a document, the enemy would have been referred to before, but in a dynamic environment it may not have been spoken about. It might not even be present.
There are a host of types of reference that need to be resolved including pronominal reference, NP references, and even abstract entities.
To solve this problem, it will be necessary to wed sensory data, linguistic data, and behavioural data.
On the positive side, conversation can also be used to help resolve the reference? (Which enemy, the red one on the left or the one on the right?)
In addition to an algorithm for reference resolution, we are hoping to develop a test bed.
This test bed could be used to compare other reference resolution stratagies.