Natural Language Generation

NLG has been viewed as a weaker cousin of NLP.
It's not my area; fortunately it is Elli Maclaren's
Elli is looking at the persuasive power of certain constructs.
Similalry, John Platts is working on story generation.
I've put the NLG component of John's system into GATE, so it should be reasonably easy to get something working.
As NLG is not my thing, this is all a bit tenuous but
current state of NLG for conversation
1. It's relatively easy to get something that is understandable to the user.
2. This may sound a bit odd.
3. Most NLG research is working at a relatively large grain (like Elli's).
4. There hasn't been much work on generating text for conversations
The standard problems are things like repeating a persons full name, or ignoring earlier context.