A powerful critique of semantic

networks:
Jerry Fodor’s atomism.

Massimo Piattelli-Palmarini
UofA, Cognitive Science
[email protected]
A rock-bottom notion
• Meaning is intimately related to truth conditions
• Word meanings are concepts
• And concepts “apply to” things
• (Things do, or do not, “fall under” them)
• A concept C is “true of” something, and only of that
something
• That’s its referent (its extension)
• The referent is the “truth-maker” of C
• But this is not enough
• The same referent can be singled out by different
intensions (the morning star/the evening star) (Gottlob
Frege On Sense and Reference 1892/1948)
Meanings
• Meanings are shared, they are public
• They are not the same as the “mental pictures” (or
images, or associations) that a word may elicit in
each individual speaker
• Dog means dog, even if I mentally entertain the
picture a poodle, while you picture a terrier.
• Marriage does not mean different things to a
future bride, her father, a priest, a lawyer etc.
• It’s crucial that our lexical semantics does not
“cut” meanings so thin
• But must cut them thin enough to allow different
intensions to pick out the same extension.
Associations
• This applies also to mental “associations”
• Via a suitable number of links, everything is
associated with everything else
• Cat is commonly associated with mouse, milk
etc.
• These are not “constitutive” of the meaning
of cat
What is constitutive?
• The central issue (Fodor’s issue) here will be:
• Has the association between cat and animal
• A different semantic status than the
association with, say, mouse, milk, meows
etc.?
• Fodor’s answer is: No!
Different strengths of the thesis
• Weakest: The available semantic networks
do not represent lexical meanings
• Moderate: No such network could represent
lexical meanings
• Strong: No kind of network (for instance
inferential networks) can represent lexical
meanings
• Strongest: Lexical meanings have no
external nor internal structure whatsoever
The classic (Aristotelian) theory of
concepts
• There are individually necessary and jointly
sufficient criteria (properties, attributes,
predicates)
• DOG = animal, mammal, domestic, barks,
etc.
• Nothing that “misses” one of these
properties can “fall under” the concept
DOG
• Nothing that has all these properties can
fail to fall under the concept DOG
Problems
• Many counterfactual cases (Chinese
emperors had a race of non-barking dogs)
• Putnam’s Martian robot cats
• Some central properties may not be
accessible to introspection (the typical bird
is “friendly”, while “crunchy” is intimately
associated with acoustic properties, etc.)
An alternative: “family resemblance”
• No property is shared by every member of
the category
• But many members of the category share at
least one property, and frequently many of
them
• Properties are not isolated, but come in
clusters
• p(flies/feathers) >> p(flies/fur)
• p(beak/feathers) >> p(beak/scales)
Problems
• Categories become “hazy”
• Different individuals “mean” different things
by the same word
• Unless
• Some properties are more important (more
diagnostic) than others
• But there is no principled (category-
independent) way of determining which ones
• You may as well “go atomistic”
Another alternative:
• Meanings are micro-structural (the real
criteria are essentialist)
• We have to use manifest attributes, but we
tacitly appeal to internal, essential attributes
which we cannot see (DNA, molecular
composition, design features, etc.)
• We defer to the experts to tell us what these
are
Problems
• We really do not know the meaning of most of the
words we use
• Most meanings are partial, approximate,
defeasible (even for cat, dog, silver etc.)
• Even the experts defer to the continuous progress
of science and scholarship.
• Many meanings have no structural component at
all (uncle, chair, expensive)
• What about verbs? (cut, run, detect etc.)
Inferential Role Semantics (IRS)
• The meaning of a concept is functionally
determined
• By the inferences it licenses, and those it
disallows, and by their strength
• If one is not disposed to assent to all and
only the right inferences, one does not
“have” that concept.
• This also applies to verbs, adjectives,
adverbs
Problems
• Intractable proliferation of inferences
• Again, you have to delimit the crucial ones
• But there is no principled (category
independent) criterion for doing that
• So, maybe atomism?
• That’s Fodor’s choice.
Fodor on concepts
• The ontology of concepts (word meanings)
is intimately tied with the problem of what
it is to have concepts (and meanings), and
to acquire them.
• He is strongly against the idea that the
epistemic problem of possession and
identification is prior to, and ontologically
dominant over, the problem of what
concepts and meanings are.
Contra Wittgenstein and family resemblance
• Concepts are public; they are the sorts of things
that lots of people can, and do, share
• Concept “similarity” (whatever that may mean)
will not do.
• To be viable, it must explain and preserve the
invariance of intentional explanations, but it must
not presuppose a “robust notion of content
identity”.
• No theory of conceptual similarity has been able to
do both.
• Something, at bottom, must be literally shared (for
instance a belief), even if all you want is to
calibrate degrees of similarity.
Identity is what we need
• If you have criteria for literal sharing (for the
identity of concepts and beliefs), then you have a
robust notion of what counts as “public”.
• You do not need similarity.
• There are individually different degrees of intensity
of the same belief.
• There are different mental pictures or associations
for the same concept (say, DOG)
• But concept (and content) identity is always (tacitly)
presupposed.
• If you don’t, than relativism is unstoppable. So, let’s
go for identity.
What is (and what isn’t) constitutive of meaning

• There are metaphysical connections between

meanings, because properties are often connected
with other properties, presuppose other properties,
entail other properties etc.
• You want to have counter-factual supporting
criteria for new and possible applications of the
concept.
• No such connection, presupposition or entailment,
however, is constitutive of the meaning of a
concept.
What is primary
• It's not language use, it's not capacities or abilities
that are primary.
• If having concepts were having capacities (to
recognize, to sort, or to draw sound inferences),
then concepts would not be mental particulars,
they would not be things at all,
• and therefore a fortiori they would not be mental
things.
• But they are!
Fodor contra Inferential Role Semantics

• First: Buying IRS would lead to circularity:

“I can’t both tell a computational story
about what inference is and tell an
inferential story about what content is.”
(Concepts, 1998. p.13)
• Second: “an inferential role semantics has
holistic implications that are both
unavoidable and intolerable” (ibid.) (see
also his book with LePore)
Fodor contra Inferential Role Semantics
• Third: atomism. “Satisfying the metaphysically
necessary conditions for having one concept never
requires satisfying the metaphysically necessary
conditions for having any other concept.” (p.14)
• No inference can be constitutive of the meaning of
a concept.
• It is metaphysically conceivable that a mind can
exist that possesses only the concept (the
meaning) DOG, and nothing else (no “cat”, and,
most of all, no “animal” concept).
• Frogs and sticklebacks are neat instantiations
Fodor contra Inferential Role Semantics

• “Much of the life of the mind consists in

applying concepts to things” (p.24).
Concepts have their satisfaction conditions
essentially, but it does not follow that “the
confirmation conditions of a concept are
among its essential properties”. (p.25)
• Satisfaction conditions are metaphysical,
while confirmation conditions are
epistemic.
Fodor contra Inferential Role Semantics
• Confirmation may well be (and sometimes is) a
holistic enterprise (mobilizing relevant clues and
inferential skills from everything you know),
• though concept-satisfaction is not holistic.
• Dispositions to draw inferences, to sort things
competently, to make connections etc. are the
consequence of knowing the meaning of concepts
• Not the other way around.
Fodor’s atomism
• A word means the property that the concept it
expresses is locked to.
• Mental states and processes are typically species of
relations to mental representations, of which latter
concepts are typically the parts.
• Thoughts are mental representations analogous to
closed sentences, while concepts (their constituents)
are mental representation analogous to the
corresponding open ones.
Fodor contra iconicity
• "The idea that there are mental representations
is the idea that there are Ideas minus the idea
that Ideas are images" (Concepts 1998, p.8, see
also Hume Variations, 2003)
• Thought is a lot like language, and concepts are
a lot like mental lexical entries (if primitive), or
mental phrasal constituents (if composed).
• This is perfectly OK. We want both of them
(thought and language) to be productive and
systematic.
Fodor’s atomism
• DOG means dog
• PUT means put
• KEEP means keep
• etc. (a Tarskian theory)
• What you have on the right are mental
particulars (atomistic mental representations
of qualities).
• No decomposition
No decomposition
• Kill means KILL
• Not “cause-to-become-not-alive”
• Nor anything like that.
• John poisons Bill on Tuesday and Bill dies
on Saturday.
• Did John kill Bill?
• When did he kill him?
No decomposition
• Sue tells John on Monday that their affair is
over. The poor desperate John commits
suicide on Wednesday.
• Did Sue “kill” John?
• She surely caused him to become not alive.
• Endless counterexamples like this one.
Contra internal structures
• Pace Pustejowski, Jackendoff, Talmy etc.
• It’s not the case that, say
• Melt = <CAUSE, change-of-state, SOLID to
LIQUID>
• This is not a semantic decomposition
• You should not “explain” the perfectly clear with
the totally unclear
• What is the meaning of CAUSE?
• Do we understand “change-of-state”?
• Let’s not confuse physical relations in the world
with conceptual semantics.
Word meanings
• The meaning of a word is:
• its reference
• Plus its Mode of Presentation (MOP)
• This solves Frege’s “evening star”, “morning star” puzzle
• And the “water” H2O puzzle
• Same extension, but different MOPs
• MOPs are mental particulars
• MOPs are "entertained" or "grasped".
• There are many (innumerably many) ways of thinking
about water. This does not mean that the concept WATER
has innumerably many meanings (there are not
innumerably many concepts of WATER).
More about MOPs
• "MOPs are mental objects and referents aren't"
• [...] "Mental objects are ipso facto available to be
proximal causes of mental processes; and it’s
plausible that at least some mental objects are
distinguished by the kinds of mental processes that
they cause; i.e. they are functionally distinguished.
• Suppose that MOPs are in fact so distinguished.
• Then it’s hardly surprising that there is only one
way a mind can entertain each MOP; since, on this
ontological assumption, functionally equivalent
MOPs are ipso facto identical" (ibid p.19)
More about MOPs
• A causal connection with actual tokenings
of the real thing in the real world is
necessary.
• But (pace Skinner et al.) it is not sufficient.
• Tokenings must be “presented” adequately
to the mind.
• Modes of Presentation, plus actual causal
encounters with the extensions, are
necessary and sufficient.
A division of labor
• Satisfaction criteria, identity criteria and
conceptual connections are all metaphysical in
nature (they must explain how mental
representations connect with properties and
objects in the world).
• Confirmation criteria are then evaluated
epistemically (and may well be, indeed, holistic).
• The psychologist’s job is to study how the mind
gets access to, and then manipulates, these
metaphysically necessary relations.
• No less, and no more.
Turing meets Frege
• The convergence of Turing's story (that mental
processes are computations on symbols sensitive
to their contents)
• and Frege's story (that some individuating
component - for Fodor, a mental representation -
has to combine with the way the world is to
determine reference)
• "is about the nicest thing that ever happened to
cognitive science" (Concepts p. 22)
Turing meets Frege
• "Wherever mental states with the same
satisfaction conditions have different
intentional objects (like, for example,
wanting to swallow the Morning Star and
wanting to swallow the Evening Star) there
must be corresponding differences among
the mental representations that get tokened
in the course of having them." (ibid)
Fodor’s story
• “My story is: The laws that govern mental
processes are intentional, hence sensitive to
semantic properties. But their
implementation is syntactic. It would be a
mystery how syntactic processes could
implement semantic regularities, but Turing
showed us how to do so. Proving, thereby,
that he was very clever. Anyhow, that’s the
line I take in `Elm', and I haven't yet been
disconvinced of it”.
Fodor’s story
• For some concepts (RED, DOG, DOORKNOB
etc.) we are directly connected with their
extensions, via personal experience. For other
concepts (METAL, HYDROGEN, NEUTRINO
etc.) the connection is indirect, inherited along a
chain, by deference to other persons (the
experts, books, eyewitnesses etc.)
• The point is that their connection to the
extension is direct (no indefinite regress).
Fodor’s story
• Objects, sets, events, situations etc. connect causally, and
nomologically, to the mind via a suitably abstract power
to convey information.
• “Concepts are categories, and are routinely employed as
such”. (p. 24)
• Things in the world “fall under them”.
• “The thesis that concepts are mental particulars is
intended to imply that having a concept is constituted by
having a mental particular, and hence to exclude the
thesis that having a concept is, in any interesting sense,
constituted by having mental traits or capacities” (p.3)
In essence
• “understanding what a thing is, is invariably prior to
understanding how we know what it is" (p. 5)
• The metaphysics of meaning is primary
• The epistemology of meaning is derived.
• “What bestows content on mental representations is
something about their causal-cum-nomological
relations to the things that fall under them: for
example, what bestows upon a mental representation
the content dog is something about its tokenings being
caused by dogs”