15
Natural Language and Logic: Some Difficulties John Barnden Professor of Artificial Intelligence School of Computer Science [email protected]

Natural Language and Logic: Some Difficulties John Barnden Professor of Artificial Intelligence School of Computer Science [email protected]

Embed Size (px)

Citation preview

Page 1: Natural Language and Logic: Some Difficulties John Barnden Professor of Artificial Intelligence School of Computer Science j.a.barnden@cs.bham.ac.uk

Natural Language and Logic:Some Difficulties

John Barnden

Professor of Artificial Intelligence

School of Computer Science

[email protected]

Page 2: Natural Language and Logic: Some Difficulties John Barnden Professor of Artificial Intelligence School of Computer Science j.a.barnden@cs.bham.ac.uk

Substances

Do you like peanut butter? What exactly is it you like?

likes(student123, peanut-butter) ???

.x ( likes(student123, x) is-peanut-butter(x) ) ???

Page 3: Natural Language and Logic: Some Difficulties John Barnden Professor of Artificial Intelligence School of Computer Science j.a.barnden@cs.bham.ac.uk

Object Identity

What is a river? The water is constantly exchanged. And are the banks included?

So what would a logic constant like the-nile stand for???

Lincoln’s ax(e): Repaired bit by bit over years. Is it the same axe at the end?

What would lincolns-ax stand for?

Page 4: Natural Language and Logic: Some Difficulties John Barnden Professor of Artificial Intelligence School of Computer Science j.a.barnden@cs.bham.ac.uk

Subtlety of Common Actions

“Tesco sells pineapples”

What does this mean, exactly?And does it actually matter?

sells(Tesco, pineapples) ?????Simple representation, but what inferences can

be drawn ??

Page 5: Natural Language and Logic: Some Difficulties John Barnden Professor of Artificial Intelligence School of Computer Science j.a.barnden@cs.bham.ac.uk

Subtlety of Prepositions

“There’s a banana in the bowl”The banana need not be within the volume of

the bowl.

“There’s a mirror on my ceiling”The mirror is below the ceiling!

“Vanessa Granola is at her desk”

Page 6: Natural Language and Logic: Some Difficulties John Barnden Professor of Artificial Intelligence School of Computer Science j.a.barnden@cs.bham.ac.uk

Context Often Needed for Precise Meaning:

Some Examples

Pronouns. Ambiguous words.

Prepositional phrase attachment. “Hank saw Vanessa with a telescope”

Did Hank use a telescope? Or did Vanessa have a telescope?

Page 7: Natural Language and Logic: Some Difficulties John Barnden Professor of Artificial Intelligence School of Computer Science j.a.barnden@cs.bham.ac.uk

Quantification: Context-Sensitive

“When Hank arrived everyone laughed” …. .x (is-person(x) laughed(x)) would be wrong

“When Hank arrived everyone sat down to dinner.”

“Hank doesn’t believe anything Vanessa tells him”

Page 8: Natural Language and Logic: Some Difficulties John Barnden Professor of Artificial Intelligence School of Computer Science j.a.barnden@cs.bham.ac.uk

Vague Quantification

Most, a few, several, many, ….

Most.x (is-person(x) anxious(x)) ????

But what inferences could you draw, and how?

Page 9: Natural Language and Logic: Some Difficulties John Barnden Professor of Artificial Intelligence School of Computer Science j.a.barnden@cs.bham.ac.uk

Embedding of Propositions, Situations, etc.

“Vanessa fell over because Hank bumped into her”

In ordinary first-order logic, can’t write things like

cause( bumpinto(H,V), fallover(V) ) because formulas (e.g. bumpinto(H,V) )

can’t be arguments in applications of predicate symbols etc.

Page 10: Natural Language and Logic: Some Difficulties John Barnden Professor of Artificial Intelligence School of Computer Science j.a.barnden@cs.bham.ac.uk

One method: take situations, events, etc. to be objects, just as e.g. people are.

f,b (is-fallover-event(f,V) is-bumpinto-event(b,H,V)

cause(b,f))

Situations, events, etc. are treated as objects in ordinary language anyway.

Page 11: Natural Language and Logic: Some Difficulties John Barnden Professor of Artificial Intelligence School of Computer Science j.a.barnden@cs.bham.ac.uk

Embedding contd.: Another example -- Belief

“Vanessa believes that Hank is lying.”

Can’t write following in ordinary first-order logic if lying is a predicate symbol:

believes(Vanessa, lying(Hank))

Page 12: Natural Language and Logic: Some Difficulties John Barnden Professor of Artificial Intelligence School of Computer Science j.a.barnden@cs.bham.ac.uk

“Honorary” Object Types

fake X, alleged X, imitation X, plastic tree, toy X, model boat, ...

a fake gun is not actually a gun, so it would be bad to write something like

fake (g) is-gun(g)

but it would be nasty to have to use fake-gun(g)

Page 13: Natural Language and Logic: Some Difficulties John Barnden Professor of Artificial Intelligence School of Computer Science j.a.barnden@cs.bham.ac.uk

Figurative Language

Metonymy: “He was listening to Bach”

Metaphor: “The suspicion grabbed me by the back of

my neck.”

Page 14: Natural Language and Logic: Some Difficulties John Barnden Professor of Artificial Intelligence School of Computer Science j.a.barnden@cs.bham.ac.uk

Variety of Types of Difficulty

May need context in order to pin precise meaning down.

Precise meaning may be difficult to pin down even when context fully known.

Precise meaning may be difficult to express in logic, or to do so usefully.

Page 15: Natural Language and Logic: Some Difficulties John Barnden Professor of Artificial Intelligence School of Computer Science j.a.barnden@cs.bham.ac.uk

Final Remarks

This presentation has shown just a selection of the problems of expressing the meaning of natural language utterances in logic.

There are many approaches to the problems, but no-one has a complete solution to all of them and some remain puzzling.

Feel like doing a PhD on the issues??!