Okay, let's start.

You've worked more on the LP implementation.

I would like to very briefly recap this idea of a Tableau machine.

Essentially the Tableau machine tries to be the basis essentially of a somewhat cognitive model of natural language understanding in its entirety.

Remember the task we want to solve is essentially we start with natural language.

We transport that into what we often call a quasi logical form.

Logical forms in an under specified representation that can still have various things that need to be resolved like pronouns for an afro, all kinds of things.

And then we go to some kind of a logical form after pragmatics.

This part is compositional.

It's based on a grammar.

A semantic lexicon.

This here is not compositional.

It's based on world knowledge.

We're really there right now.

And that's what the Tableau machine is to basically ideally fully do.

And the idea of the Tableau machine is we have a set of world knowledge.

And that world knowledge you should imagine as being the common knowledge of all people that we're assuming or the group that's around in the situation.

Plus the situation knowledge.

Things the group that is conversing.

Seas for instance.

That's kind of also we think of that as world knowledge.

And then we have say the dialogue or a monologue.

I don't care.

And we have an input sentence.

We saturate the Tableau.

And while in propositional logic, which is what we're really doing right now, we can do that fully.

And semi-decidable logics like first-order logic or higher logic or so.

This might be infinite.

And so we have to somehow decide when we want to terminate, when we want to call it quits.

For instance, when the next input sentence comes along, that might be a good place to stop thinking.

And then we have all these branches in there.

We choose the branch.

That is our current model.

And when the next input sentence comes along, we repeat.

We have all kinds of branches.

We choose one.

And that's the mechanism.

Again, how do we choose the branch?

Open question.

Lots of stuff.

Depends on this.

Also, when do we do which kinds of inferences?

Kind of related to the cutoff question.

We prefer certain inferences as well.

All of those are open.

All of that will have a huge impact on the behavior of this model as a performance model.

Okay.

Anything else I should say?

Any questions?

Yes.