It's cold here.

Okay, so yesterday we started with communication as a big area of AI.

And as with many AI areas, this is kind of an intersective area.

We've seen that in all of this, Markov discussion procedures and reasoning with uncertainty

were basically in a kind of intersection between AI and say math and control theory and information

theory and all of those kind of things.

So AI has learned a lot there from electrical engineering and so on, where they had already

looked at some of these problems before.

Here, and for communication obviously there's a huge intersection with linguistics.

So the study of language.

They're basically going to look at relatively little of that even though there's in more

involved techniques, there's a bigger intersection with linguistics.

Real kind of stick to the language technology parts which you can do with relatively

little linguistics.

So I tried to convince you that language is important because it's kind of an exercise

ground for AI techniques and of course eventually we want AI agents to be able to talk to us.

And we want to have AI agents to do what students do, namely read documents and

gain knowledge from that, all of those kind of things.

And many of these things are very natural in natural language because that's for humans

that's almost effortless.

I could talk first or logic to you but probably that would make me unhappy and possibly even

more make you unhappy.

Okay, so we're going to first look at a topic called natural language processing.

Remember natural language as opposed to artificial languages are languages that have naturally

evolved by humans to talking to humans typically.

And that means that unlike say programming languages where we can have a normative grammar

which we can implement which will actually say this program is Java or not.

We do not have something like that.

We do not have a normative grammar for say English or Chinese or German.

And though in school they told you that certain things you cannot do in English right

you always have to have the subject before they predicate and then come to the object.

And so, and yes a couple of rules exist.

They're typically in natural languages not as universal as your teacher made you think.

They typically have exceptions.

They typically only adhere to the language you're supposed to write in essays in school.

They typically do not model what people actually say.

One of the things that obviously is missing which is very very important in say spoken

language is the word which is not even in your school dictionary.

But if you look at what people say then typically there's a lot of ums in there at random

places, not quite random but anyway.

And you'll probably see that in a say random lecture I would guess only 60% of the sentences

are even vaguely grammatical.

We start with something, change our minds, oh yes and so on and we have all these kind

of almost completed sentences there.

You seem to have no problem with that.

All of those things are things we have to deal with in natural languages.

So you can think of kind of natural and artificial languages as being kind of cousins.

They kind of do the same thing.

Only that artificial languages are neat and easy and have context free or something like

that, grammars.