Greg Detre
14:00 Tuesday, September 10, 2002
Prof. Deb Roy
http://courses.media.mit.edu/2002fall/mas962
he finds the most convenient way of thinking about meaning/language is in terms of words, i.e. mental lexicon
infinite number of qualities/events/relations in the world for words to latch onto
words as equivalence of kinds across time � they reflect our objectification of the world
they classify, they blur the differences between things we want to lump together
they highlight the essential features of referents
though what you highlight depends on your purpose
2 approaches
cognitive science: reverse engineering � full-blown system which we can only chip at
computational: analysis by synthesis � not sure what/how to build up
words point/attach/refer to things
doesn�t incorporate mental models
Odgen & Richards (1932) � triangle of:
thought/reference
referent
symbol
imputed relation between symbol + referent � mediates
causation � built-in innately?
correlation, temporal ordering � anything more???
or perhaps we learn motor skills and their effect on the world, and so derive our notion of causation from our notion of our own agency
difference between being able to employ causality in reasoning, and being able to use and understand the word/concept �causality� itself
surprisingly, a large number of people at an Evolution of Language conference at Harvard didn�t believe that the evolution of language was driven by communication
they thought that it grew out of the need for coordination of a distributed control system within an agent, i.e. bits of the brain talking to each other
Minsky: the most efficient way to build a system may be parallel, but serialising may be necessary for simplification (possibly inaccurate paraphrase???)
hard to get away from the notion of truth when talking about meaning
Aristotle � defined truth as the correspondence between what is said and the actual state of affirs (intentionality)
\ the meaning of a statement is anchored to the conditions in the world which make it true
binary truth values � considered a formal truth-preserving calculus
�dog� = corresponds to all the entities in the world (or in possible worlds) that are dogs � i.e. meaning is extension
but how do we store that in our heads?
and how can we know partial definitions? e.g. vitamins or beech vs elm (Putnam)
word meaning = necessary and sufficient binary features
e.g. bachelor = unmarried/human/male/adult
difficult to nail down a set of primitives
maybe you can only have sufficient in a large number of cases, or probabilistic
Witt � what is a game? no single primitive of a game. perhaps �game� is itself a primitive?
taken together, a set/family of resemblance defines the space
what�s a pile of sand? what�s a shoulder?
what primitives do you choose from? what do the primitives themselves mean?
the variable name shouldn�t be thought of as capturing anything of the meaning, e.g. DOG vs XYZ
levels of description
functional vs descriptive
how much implicit knowledge is needed to use the representation?
Jackendoff (2002) � activity in the neural assemblies in the brain is meaning
it�s not clear what is meant to talk of external objects in the world anyway
what do you need to know to use dictionary definitions?
circularity (e.g. �push� in Webster)
relation to:
world
other words
cultural constructions
personal history/state
can use relations to shed light on underlying features of meaning
fast car, fast road, fast food
maybe you can�t break apart these lexical phrases
what�s the core word meaning that stays constant?
relational/structural, e.g. Wordnet
externally grounded, e.g. computer vision, robotics
na� theoretic
maybe change time to 4.30
what do I mean by levels??? layers???
what languages are not word-based??? Turkish??? even those languages have morphemes
�they thought that it grew out of the need for coordination of a distributed control system within an agent, i.e. bits of the brain talking to each other???� � what do I make of this as an idea for the selective pressure on language???
prototypes???
dictionaries???