Re: Dialogue management

Andy Paterson wrote:
> 
> Hi,
>  Apologies if this is off-topic...
>  I've been working on a dialogue management component for an NLP
> engine. Part of it is responsible for processing concept graphs
> generated by a conceptual analyser, and, in the context of the current
> dialogue, generating an appropriate response/executing an appropriate
> action.
> At the moment, I've been using an object-oriented scripting language
> to drive a slot-based (requirements driven) system. This means writing
> out, by hand, the mappings from concept graph to what slots and
> required to be filled. This is fine for the small domains we are
> currently working with, but I'm worried that when we scale up to
> larger domains, the scripting approach will be impossible to manage.
> Are there any alternatives for generating these mappings, or am I
> going about this in totally the wrong way?
[snip of pseudocode]

I'm not sure if you're going about it in the best way, but I don't know
what else to suggest. At some point there has to be a mapping between
the language model (your concept graphs) and the domain/task model,
where your slots live. I would recommend using a good declarative representation
for your ontology and object models, but I'm biased. (I'm working on a new
Trindikit-inspired toolkit, so I have my own ideas about how dialogue
managers get built. The jury is still out as far as evaluation goes.)

However you organize your slots, you should make sure they're compatible
with the rest of the dialogue management: referent resolution, focus,
grounding, etc. (Depending on the complexity of your system you might not
care much about those aspects.)

--
Carl Burke
[EMAIL PROTECTED]