Okay, let's start. It seems that rainy weather does affect attendance, so good for you to

be here. Thank you. Okay, so we were looking at the big area of natural language processing

or communication as a sub-area of AI and the idea is to kind of, in this course at least,

to follow this weak AI strategy, looking at little things we can do where natural language

processing techniques are actually useful. I tried to convince you, and still am, that

natural language is non-trivial, even if it is essentially effortless to all of us. We

never break a sweat understanding a conversation. And if you actually break a sweat trying to

understand, say, what's on my slides in natural language, it's usually the content that you're

worried about, not the language as such. And that's something we see quite often in AI.

When we try to simulate or realize parts of intelligence in the machine, we find out that

certain things we thought were trivial aren't. If you think about this ongoing big challenge

for robotics, which is artificial soccer, it taught us in the beginning quite a lot

about, well, it's not quite trivial just solving the vision problem there. If you're a robot,

just finding the ball and remembering which goal to play towards is non-trivial. And the

kind of higher level things like planning an attack, should I give the ball to one of

my co-players or go it alone, those are relatively simple. But just finding the ball, even if

and I'm not sure it is actually right now anymore, but even in the beginning where the

ball was the only thing that was A, round, and B, magenta. It was really in a neon magenta,

the ball was colored that way. And it was, by the rule, prohibited to make anything else

in that color. Even then it was a hard problem just finding the ball and locating it and

doing something with it. And that's one of the attractions that I think about AI is that

we get kind of a sense of how difficult certain aspects of intelligence are. And language,

I'm hoping to convince you, isn't. So it's interesting. So we looked at the truth conditions

invoked by things like adjectives. It is not the case that all adjectives are what we call

intersective. Namely, the meaning of blue is given by intersecting the set of all diamonds

with the set of all blue things. Right? Propositional attitudes, knowing, believing, considering

possible, being allowed to, all of those kind of things, kind of wrap around natural language

and modify their meaning. We looked at various forms of ambiguity. It can come from various

things that can be lexical ambiguity, like in bank, where we have words meaning two things.

But it can also be from the grammar, where, for instance, the stuff with the chasing the

gangster in the red sports car is really what this relative clause in the red sports car

is attached to. It's an attachment ambiguity. It can be attached to John, it can be attached

to the gangster and somewhat unintuitively can be attached to the chased. We have quantifier

scope ambiguities. We have intentional time-dependent ambiguities. We have, I don't even know what

it is to call it, it's kind of relevance-based ambiguities. And we looked at an afra, picking

up pieces of a discourse later by he, she and it pronouns, which is massively ambiguous,

interacts interestingly with things like ellipses, where we are not saying everything we mean.

And all of that is not so simple. Are there any questions so far?

Then I'd like to show you a couple more phenomena, just to have a little bit wider scope. The

King of America is rich. True or false, what do you think?

First. False. Okay. So he is poor. The President. Ah. So you are saying there is no King of

America. So indeed, there is not, even if some recent presidents thought that. So we

have the problem, we're talking about an object that doesn't exist. So you have the intuition

that, oh it should be false. But if it were false, then this sentence should be true.

The King of America isn't rich, but we're probably not as happy with that either. So

maybe, and that's the standard idea of this, is that we're not giving it a truth value

because it's ill formed in some ways. At least if you think about America as the United States,

then the United States are a republic. If you're thinking about Canada, we have all

kinds of ambiguities here, but let's think about the US. And then we probably have a

failure to mean here. Which we can heal, interestingly. If we wrap this sentence in a conditional,