SEL1007: The Nature of Language

SEL1007: The Nature of Language

Computation, mind, and language: the history of 20th Century linguistics

1

The plan for today

A bit of history: the classical mind-body problem

Computers as a solution Language and the theory of

computers

Descartes and the scientific study of language and mind

“I think, therefore I am”

Invented the Cartesian coordinate system and analytic geometry

Formulated the ‘mind/body problem’

René Descartes (1596-1650)

The ‘mind/body problem’

The world (and the animal kingdom) are basically big machines

But human beings are different Human behavior is neither completely

deterministic nor completely random

In other words, human beings have free will

The ‘mind/body problem’

Language is an important facet of this

“it is quite remarkable that there are no men so dull-witted and stupid…that they are incapable of arranging various words together and forming an utterance from them in order to make their thoughts understood; whereas there is no other animal, no matter how perfect and well endowed it may be, that can do the same.”

-Discourse on Method

The ‘mind/body problem’

So what “causes” free will? Descartes’ (perfectly scientific) response: substance dualism There’s two kinds of ‘stuff’ in the universe

But modern science not so keen on substance dualism

A more modern solution:

=

Why is this helpful for the scientific study of language? Computers provide an acceptable

metaphor. Mental operations (like thought, or

language) aren’t some mystical incomprehensible thing. It’s ‘just like’ what a computer is doing

The hardware/software distinction People had a theory of how

computers worked

So, how does a computer work exactly?

Key member of Bletchley Park team that broke the Nazi “Enigma” code

His formulation of ‘what a computer is’ underlies most of modern computer science and computer technology

Computers transform strings of symbols into other strings via an algorithm

Alan Turing (1912-1954)

Some fundamental concepts

Symbols (and symbol systems) Any physical thing which, by agreement,

represents something else

= USA

The relevant symbol system

Another symbol: p (represents /p/) The symbol system: the Roman alphabet

More fundamental concepts

Strings A series of symbols taken from a

particular symbol system abcde, aakkklubss, powerpointsucks,

banana

More fundamental concepts

Algorithms A sequence of instructions to perform

particular tasks in a particular order

An algorithm for getting from the School Office to the Student Union Go through the double doors to the landing Go down the stairs to the ground floor Exit the Percy Building from the main entrance Walk down the quad Walk under the arches Cross the road Walk 10 metres straight ahead Turn rightNote: Each step is explicit and the steps are in a particular order

Another kind of algorithm: recipes

OK, so what do computers do?

Computation = string transformation

String 1 = (2+2)/3; String 2 = 1.333333 String 1 = ‘the car’; String 2 = ‘el coche’

The computer transforms one string into another by following the algorithm

An example of a ‘Turing Machine’

Doing “1+1=2” with a Turing Machine

Doing “1+1=2” with a Turing Machine

Doing “1+1=2” with a Turing Machine

Language as string transformation: a phrase-

structure grammar A grammar = an algorithm for

producing and understanding language

‘phrase-structure’ = sequences of words are structured as/consist of phrases.

A phrase-structure grammar of (a very small part of)

EnglishS -> NP VP Det -> theNP -> N VP -> V NPNP -> Det N V -> bitesN -> man V -> catchesN -> dogN -> cat

Language Generation

S -> NP VP

Two restrictions on rewriting Only rewrite one symbol at a time Only the leftmost symbol can be

rewritten

S -> Det N VPS -> the N VPS-> the man VP

Language Generation

the man V NPthe man bites NPthe man bites Det Nthe man bites the Nthe man bites the dog

Et voila!

Understanding language using a phrase-structure

grammar

The man bites the dog

Det N V Det N

NP NP

VP

S

But….It’s important to keep two questions separate

The technologyquestion

The natural worldquestion

≠[from http://www.learnfrenchlab.com/how-to-speak-french.html]

[from http://motherboard.vice.com/2010/8/5/eight-sci-fi-robots-that-prove-that-robots-aren-t-going-to-enslave-humanity]

Not doing too badly with the first one (but see YouTube, etc.)

Little more of a problem with the second one Phrase-structure grammars don’t have

the right mathematical properties for natural language

Most successful language parsers require some degree of initial ‘training’ (via a corpus of pre-parsed sentences). Children don’t.

Also, the ‘symbol grounding’ problem Computers manipulate symbols,

but they don’t understand them(imagine trying to learn Chinese

from a Chinese dictionary of Chinese) 吃飯 = 把嘴裡的食物

If the mind is just a computer, then there’s a problem. Human beings must be doing something more.

Next Time

A different 20th century reaction:

Run away! Run away!