Transcript Words & Meaning: - Department of Psychology

Words & Meaning:
An introduction
Words & Meaning
• Words and meanings are linked, but they are
not identical
• Your textbook gives three reasons why:
– i.) The translation argument
– ii.) The imperfect mapping argument
– iii.) The elasticity argument
– iv.) The partial ignorance argument
Overview
• Intro to words and meanings
• Three classic theories of meaning
• What is meaning made of?
– 3 Feature theories: Classical, Family
Resemblance, Protype
– Can we have meaning without features?
The translation argument
• Many words that exist in other languages
that do not have an equivalent word in
English– i.e. 'schlep', to move a heavy and bulky item
from one place to another.
• Since we have the meaning, but not the
word, in English, a word cannot be identical
to its meaning- we can get one without the
other
The imperfect mapping argument
• If words were the same as meanings, then
there would be one word for every meaning,
and one meaning for every word.
• Of course, there is not: some strings of
letters or phonemes have multiple meanings
(ambiguity), and some meanings have many
words (synonymy).
The elasticity argument
• The meaning of a word is not a fixed thing
but varies with its context of use.
– So 'a tall tale' is not the same thing as 'a tall
man'.
• Since the meaning of a word is not simply a
function of the word, it cannot be that a
word and its meaning are identical.
The partial ignorance argument
• Many of us (especially scrabble players) can
recognize words that we can't give the meanings
of
– Don’t we want to say that in that case the person
‘knows the word’, but not the meaning?
– And if we do, doesn’t that mean that a word is
something other than its meaning?
What is a word?
• Similar arguments apply to a word’s sound or
visual representation
– Many of us know words that we cannot
pronounce, or words that we cannot spell
– We may even know that a word exists for a
meaning, but know neither its spelling or sound
– A word cannot be simply identified with any
one of these aspects, although most words
participate in all three aspects
Open and closed class words
• We want to make a very sharp distinction between two
kinds of words
– Open class words: Words that pick out a thing or an
action or an abstract pattern (like 'peace' or 'justice')
– Closed class words: Words that serve a purpose in
mediating between other words
• Closed class words are function words like articles
'('the', 'a'), conjunctions ('and', 'or', 'because'),
prepositions ('on', 'under') etc. that serve to define
the relationships between words or between their
referents
– Closed class words serve a vital role: to lose them (as in
Broca's aphasia) is largely to lose the ability to
communicate
Complications
• What about words like ‘hey’, ‘um’, or ‘so’?
– They don’t ‘mean’ anything (they have
‘bleached semantics’) but they are still useful
for certain kinds of purposes.
• What about private invented words that have
never been shared with anyone else? Should they
count as words?
Intension/Extension
• We refer to a phrase's definition as its intension: so the
intension of a word like ‘chair’ might be ‘furniture
specifically designed for sitting on’
• we contrast this to the phrase’s extension: the objects that
fall under the intension- in this case, all chairs.
• These have to be kept separate for many reasons, most
notably because knowledge of a phrase’s intension does
not necessarily make its own extension obvious
– e.g. ‘The people I ate dinner with last night’
• Much philosophy of meaning hangs on this distinction
Three ‘classic’ theories of meaning
i.) Meaning as reference
- Two problems
ii.) Meaning as ideational
iii.) Meaning as conventionality
i.) Meaning as reference
• Early philosophy of language tried to build a definition of
meaning from saying that the meaning of a word was
identical to the subset of its intension to which it referredso if I say 'the chair you are sitting on' what I mean is the
chair right there.
• This approach came from mathematics and logic
• Gottlob Frege’s predicate calculus is a form of logic that
analyzes propositions into atomic and quantified
statements
– These propositions, along with related axioms, formed
a kind of calculus that related mathematics and logic
(alas, unsuccessfully as Russell showed).
• Logical atomism: Each symbol (word) had to ‘stand for’ an
object
Problems
• There were many famous philosophical
problems with this:
a.) The difficulty of fixing the referent of a word
b.) Some words have no apparent referent
a.) Fixing the referent of a word
Consider ‘the celestial body which is visible low on the
horizon in the early hours of the morning’
• One can perfectly understand and use this definition
without understanding that its referent is identical to the
referent of the word ‘Venus’- because Venus is in fact the
morning star.
– You might even believe it was true that ‘The morning
star is not the planet Venus’ even though you know the
reference of each term individually- and therefore of
both terms
• Similarly many examples of a word's referent does not
always allow us to fix meaning of the word, due to the
failure of induction
– Quine's ‘gavagai’
b.) Some words have no apparent referent
• Closed-class words or very abstract words or
words which refer to non-existent entities
• Alexius Meinong: Since meaningful things must
refer, non-existent objects must ‘exist’ in logical
space
• Bertrand Russell: ‘Logic must no more admit a
unicorn than zoology can.’
Theory of descriptions
• Descriptions are not names, but logical syllogisms
• A logical syllogism can be sensible (well-defined)
but false
– Thereby we can salvage expressions that refer to nonexistent entities without having to postulate their
existence
– In this view language obscures what is actual the case
– Later we shall see that Wittgenstein will have much to
say about this, in two different but related ways
ii.) Meaning as ideational
• Perhaps word Venus does not refer to the celestial body
itself, but to your own idea of the planet.
• There are two problems:
– i.) If words refers to something in your head, then we
have to say that the referent of a word must share the
same properties as things in your head
• The word 'light' refers to something in your head, so
it must be true that 'the word light refers to
something that it is dark'- which is mad.
– ii.) Failure to keep public reference
• If you are talking about something inside you and I
am talking about something inside me, then how do
we ever agree on what we mean?
• Meaning must be fixed publicly in some way.
ii.) Meaning as conventional
• Perhaps meaning is a public construction
– we keep reference public by the way we act- we debug
our internal references by our actions
– If you have one view of what I mean by 'Venus' and I
have another, then we will fix them when our
behaviours collide in some way.
– One implication is that we can mean different things by
words so long as our meaning doesn't have any
practical effects that keep us out of synchrony with
someone else with whom we cannot be desynchronized
What is meaning made of?
• However meaning may be fixed, we still want to
know what it is that is fixed.
• What is meaning made of? How it is organized?
Feature theories
• One common answer is that meaning is made of features
(or dimensions)
– So a cat has features like 'is an animal', 'is a mammal',
'can be domesticated’
– This is popular in AI, since it gives meaning a clear tree
structure allowing for inheritance
– For example, if you know that a cat is a mammal, then
you automatically know that it bears it young live, it
suckles its young, it has fur etc. etc.
How do features combine?
• The 'classical view' says that a concept is defined
by necessary and sufficient features- i.e. by some
features that it must have (the necessary ones), and
by a minimal set of features that are all it needs
(the sufficient ones).
• This is a cold logical view, which assumes that
there is a discernible, specifiable logic underlying
not just 'hard' concepts like 'triangle', but soft ones
like love, peace, and justice too.
How do features combine?
• The 'family resemblance' view says that there is no
necessary or jointly sufficient features, but there is a
greater or lesser agreement of a concept with its set of
possible features
– This allows us to say that an individual can be a more
or less good exemplar of a concept- so, a robin is a
better example of a bird (because it has more 'bird'
features) than a penguin.
– It also allows us to allow things to degrade without
changing category membership- so a canary is almost
always yellow, but it doesn't have to cease being a
canary if we paint it purple
Prototypes
• The best exemplar is a prototype
• Evidence shows that it has some psychological reality:
people will respond faster to the sentence 'a robin is a bird'
than they will to 'a penguin is a bird'.
• There is evidence that even categories which are defined
classically, with necessary and sufficient conditions, show
family resemblance structure
– People systematically judge some numbers as being
'more odd' than others (and respond faster to statements
about them), even though the only requirement for
oddness is not being divisible by 2.
– Same for 'a mother is a woman' (faster) than 'waitress is
a woman' (slower)
How do features combine?
• The ‘knowledge-based’ view is an extension of feature
theories
• One problem with features is knowing where the end:
when do features stop having features so that the system
can 'bottom out' and return an answer.
– In some cases this is obvious (i.e. when the search is on
for a specific feature), but in others it is not
• How do we calculate when we have 'enough'
features to make a judgment such as 'A penguin is
a bird'.
• How do we build idiosyncratic feature sets like
‘Things my grandmother never lived to see that I
own’?
How do features combine?
• Knowledge based approaches say that features are
built in contexts of 'theories'- that concepts cohere
based on a deeper coherence of explanations
which those features together
– Of course, this opens up the same questions we
are trying to answer, but at a different level:
what is a theory made up? What are its
components?
Psychological essentialism
• One theory is ‘Psychological essentialism’: people act as
if things have essences that could be defined, and
sometimes they know what those essences are
– So, we have a theory that says that water is made of 2
hydrogen atoms and 1 oxygen atom, and anything elseno matter how much it looks or acts like water- is not
water.
– most of us accept this: contra family resemblance
theory, some features just are necessary
Psychological contextualism
• ’Psychological contexualism' says that theories of
essentialism can change with context
– Some features just are necessary in some contexts
– We might accept artificial orange juice as orange juice
if we were thirsty, but might want to argue in chemistry
or philosophy class that a liquid containing no matter
from oranges was not orange juice.
Meaning without features?
• Features are defined generally enough to be
all-inclusive: any difference may be defined
as a feature
• Later we will see other approaches that
define features in unexpected ways: as
vectors in co-occurrence space, or as
undefinable variety [i.e. maybe features
have no features!]