The Language of Thought

1
The Language of Thought

• Philosophy of Mind 2007
• Lecture 6

2
Recap

• Mental content what a thought is about
• Brentano intentionality the mark of the mental
• But mental states supervene on physical states
  how can a physical state be about something?
• Last week - how thoughts get content
  interpretationist, causal/correlational,
  teleological
• Broad and narrow content

3
This week

• This week how are thoughts structured?
• Is thought like a language?
• The language of thought (LoT) and its competitors

4
Propositional attitudes

• We can think of content as having two elements
• The thing were thinking of
• How were thinking of it
• Intentional states - propositional attitudes
• Alf believes that the blancmange is vanilla
• Bert desires that etc

5
Propositional attitudes

• Same objects (blancmange, vanilla)
• Different attitudes (belief, desire)
• Last week how thoughts come to refer to certain
  objects
• This week how thought expresses propositional
  attitudes

6
Propositional attitudes

• So
• Alf believes that the blancmange is vanilla

Intentional content (as last week)
Attitude
Proposition
7
Propositions

• Thinkers might express propositions in English
  (or Japanese, Tagalog, Russian)
• But we can abstract propositional content from
  these natural languages
• How does thought express or represent that
  content?
• As a language of thought (Jerry Fodor)?
• A representational, computational theory

8
Language of thought

• Four things to consider
• How thought represents a proposition
• How thought represents an attitude
• How the brain can process meaning, truth, logical
  relations
• Why must thought be like a language?

9
1. Representing a proposition

• the blancmange is vanilla
• Atoms and molecules (B-M J)
• Semantics of atoms as last week
• Combine to express propositions (truth-evaluable)
  
• As sentences in a language of thought
  mentalese

Proposition
10
1. Representing a proposition

• A simple picture ..
• Thought type brain state type
• Brain state a blancmange, ß vanilla
• Combine tokens of these two brain state types a
  ß
• In a certain syntactic (structural) relationship
• And you get a representation in mentalese of the
  proposition the blancmange is vanilla

11
2. Representing attitudes

• How does Alfs belief that the blancmange is
  vanilla differ from his desire that etc?
• Represent different causal/functional roles
• Belief disposed to eat/reject blancmange
• Desire disposed to seek/make blancmange
• LoT theory having propositions in the
  (metaphorical) belief box or desire box
• Treated in different ways by the brain

12
3. LoT as a computational theory

• LoT is a representational theory the mind
  operates with representations of the world
• Also a computational theory
• brains manipulate representational states
• In ways that respect meaning, content
• Computational theory of mind (CTM)
• How thoughts can be about certain things
• How thoughts can be related in appropriate, or
  rational ways

13
3. LoT as a computational theory

• How? The brain is just a physical object
• Like a rock, or a bowl of porridge
• Or a computer but computer operations appear to
  respect meaning and rationality
• E.g. maths problems

14
3. LoT as a computational theory

• LoT representations have both semantic and
  syntactical properties
• Semantics what an atomic element of LoT means,
  stands for
• Syntax structure
• LoT symbols (a, ß, d) and sentences have semantic
  content (see above last week)
• But the brain responds to the structure of
  symbols and sentences

15
3. LoT as a computational theory

• The brain doesnt know anything
• Responding to structure responding
  appropriately to meaning
• This produces intelligence response to meaning
  of representations
• Compare a vending machine
• Beliefs, desires, etc different sorts of
  computation over these representations.

16
The whole picture?

• So to understand a thought we might
• Examine a brain state
• Establish syntactic components (e.g. a, ß)
• Consult semantic theory, decide what these
  elements represent. E.g. cat friendly
• What function does the state perform?
• If it belongs in the belief box then I believe
  that the cat is friendly, if the desire box
  then I desire that the cat is friendly etc

ß?ddae???µ?pd
?µ?pdßdae???d
17
4. Why must it be a language?

• Central idea
• atomic elements combined in syntactic structures
• combinatorial semantics (Fodor)

18
4 Why must it be a language?

• a) Crucially, thoughts are sentence like
• Compositional - put together from parts
• Alf believes that the blancmange is vanilla
• And so can share parts with other thoughts
• Betty wants to buy some blancmange
• Clare likes vanilla ice-cream
• Reflecting rational relations

19
4 Why must it be a language?

• Compositionality means we can have different
  thoughts about the same things
• The elements of thought can be combined in
  different ways
• Fodor LoT is the only game in town
• The only theory that can explain certain features
  of thought . E.g.

20
4 Why must it be a language?

• b) Systematicity elements of language rearranged
  to express different but related propositions
• a)John loves Mary or b) Mary loves John
• Similar systematic relationship between cognitive
  capacities you cant have (a) without (b) (and
  vice versa)
• Systematicity in inference also

21
4 Why must it be a language?

• c) Productivity and language is productive
• Once we know the elements of language we can
  create entirely new sentences
• Ditto thought we can combine and develop
  thoughts, think new thoughts
• Compositionality parts making wholes

22
4 Why must it be a language?

• d) Similar thoughts have similar effects, based
  on their common elements
• Frank is upstairs
• Frank is working upstairs
• Frank has gone upstairs
• And similar effects might suggest similar causes
  (shared elements of thought)
• (The same is true of sentences )

23
4 Why must it be a language?

• e) One thought can cause (lead to) another
• Rational connectedness depends on the information
  carried by the syntactic structure of thought
• After all, computers move from one state to the
  next for just this reason
• So, the central idea atomic elements combined in
  syntactic structures

24
Summary

• LoT as a representational theory
• Mental states are intentional states
• Represent as sentences in LoT - meaningful
  elements combined syntactically
• Belief boxes, desire boxes
• A computational theory
• Thought is processing
• This involves structure, not meaning
• Thought is languagelike because it is
  compositional, systematic, productive, etc

25
The Only Game in Town?

• Philosophy of Mind 2007
• Lecture 6a

26
Recap

• Fodor et al thought must be language-like,
  because thoughts share features of language (an
  IBE)
• So thoughts as sentences in a LoT (mentalese)
• Representational (intentional realism)
• Computational processing based on thought
  symbols and combinatorial semantics

27
Now read on

• This lecture
• Fodors arguments in WTSHTBALoT
• Objections to the Language of Thought
• Alternatives
• Map theory
• Connectionism

28
Fodor WTSHTBALoT

• Aunty accepts thoughts as causally related
  intentional states
• But denies that they have constituent structure
  
• Fodors three arguments against Aunty
• Complex intentions are decomposable into parts
  structure of thought must reflect this
• If you dont accept compositionality, you have no
  story about thought processes
• Systematicity and productivity

29
Objections to representationalism

• Objections to the idea that mental states need
  have intentional content
• May be inspired by behaviourism, more recent
  eliminativism various anti-dualist intuitions
• Do we need to postulate any more than some code
  by which the mind operates?
• Some hints of this in Dennett, Churchlands,
  connectivist writing

30
Objections to representationalism

• Dennett (CFTCC) how to understand
  representation?
• Nothing is intrinsically a representation of
  anything
• The risk of undischarged homunculi
• But doesnt Fodor postulate self-understanding
  representations?

31
Objections to computational theory

• From the other direction
• Can computation, symbol manipulation, explain
  thought?
• Searles Chinese Room
• Dreyfus expert performance cant be captured
  algorithmically. We just see something

32
Specific objections to LoT

• 1) Dennett (BWMR) could symbol tokens
  unambiguously identify beliefs?
• Maybe behaviour is more important (Cf
  interpretationism, Witt. behaviourism)
• Some beliefs are vague, not explicitly formulable
• To avoid regress the brain must store some info
  in more basic form
• Maybe the system structure does the work
• Or self-understanding representations

33
Specific objections to LoT

• 2) The Churchlands (in Lycan) linguistic
  abilities come late in evolution
• So why should we assume that cognitive functions
  take a linguistic form?
• A discontinuity with the cognitive abilities of
  other animals?

34
Specific objections to LoT

• 3) Fodor argues (in TLoT) for concept nativism
  (back-tracked in Concepts)
• All the concepts we will ever need must be innate
  
• We could never learn a concept if we didnt
  already have a LoT symbol corresponding to it
• Can we break concepts down to constituents?
• Very few can be broken down
• Seems implausible

35
Map Theory

• Any representational theory must be structured
• relationships among the namers correspond to
  relationships among the named (B-MJ p179)
• LoT achieves this via syntactic structure
• But why not a map in the head (B-MJ)?

36
Map Theory

• Our beliefs structured like a map (an analogue
  system?)
• Maps and sentences
• Both can lose merely local information
• But maps dont give discrete (single-belief)
  information sentences do
• Cant extract just some information essentially
  rich
• And no minimum representational unit
• Is this more like our beliefs?

37
Map Theory

• Something map-like can represent any empirical
  fact about the world
• LoT reflects the way computers operate but why
  must brains be like computers?
• Evolutionary story maps can evolve by becoming
  more detailed (theyre messy)
• But LoT can seem all or nothing

38
Map Theory

• Can it reflect the features of thought as well as
  LoT?
• Systematicity rational relations between
  similarly structured representations
• B-MJ Yes. Spatial relations in maps are
  similarly systematic
• Productivity elements can be combined in new
  ways
• B-MJ We can add new features to maps, combine
  maps, etc

39
Map Theory

• Similar causal effect differences and
  similarities in representation show up in
  different behaviour
• If a map shows (any) right turn, you turn right!
• One thought leads to another rational
  connectedness between thoughts
• Similarly reflected in the structure of the map
• And maps can evolve as thoughts evolve

40
Connectionism

• But LoT gives us a complete picture of how the
  mind works
• How might the map theory be realised in the
  brain?
• Possibly connectionism can give us a clue
• PDP - parallel distributed processing
• The brain not a computer as the best model of the
  mind

41
Connectionism

• Neural networks large numbers of nodes
  (neurons) with connections between them
  (synapses)
• Connections have different weights, reflecting
  their relative strength
• As the network operates, the weights are adjusted
  to eliminate wrong answers
• Networks learn certain connections as a result

42
Connectionism
43
Why connectionism?

• NB resembles Humes associationism!
• Reflects real neural structure (does this
  matter?)
• Graceful degradation. Degraded, not destroyed -
  not an all or nothing system (better than maps or
  sentences)
• As a result of distributed representation/
  computation (within limits of modularity)

44
What decides the question?

• Which picture of the mind better reflects the
  features of thought?
• Especially compositionality, systematicity,
  productivity
• Fodor connectionism doesnt necessarily reflect
  systematicity
• associative systems dont require that the
  semantic content of thoughts determines their
  relationships

45
Summary

• LoT may be rejected by
• Opponents of representational theories
• Opponents of computational theories
• Implausibility of thought being language-like
  (Dennett, Churchlands, anti-nativists)
• Map theory as an alternative?
• Connectionism as a (related?) alternative?
• Can anything other than LoT explain systematicity?

46
Seminars and Reading

• As module guide (esp BMJ)
• Dennett Brain writing and mind reading
  (Brainstorms)
• Sterelny The Representional Theory of Mind
• Lycan p 277ff for good summary and reading list
• Blackwell Companion articles on connectionism,
  computational models, Jerry Fodor, language of
  thought
• Online Stanford Encyclopedia article on
  Computational theory of mind
• And look ahead to next weeks reading Fodors
  Propositional Attitudes