Save
CLD Final
Chomsky & Connectionist Theories
Save
Share
Learn
Content
Leaderboard
Learn
Created by
Audrey Henderson
Visit profile
Cards (14)
universal
grammar
- noam
chomsky
-
properties
of language that
all
language's have (subject, verb, object)
- initial
state
of language when you are born
"language
acquisition
device
"
-
domain
specific
and
nature
theory
what
role
does the
environment
have?
- children
must
be
exposed
to language
input
- BUT only need
minimal
exposure
-
environment
sets
parameters
of language
i.e. which order of subject, verb, object, which phonemes does the language use
chomsky's
transformational
/
generative
grammar
- grammar
starts
as a
deep
structure
-
transformations
are made to
change
a sentence for to the surface
structure
EX: if we want to ask a questions, we start by forming a sentence: Jessica is pretty.
Then we change things to make it a questions: is Jessica pretty?
arguments
for
universal
grammar
- children hear
limited
input
- children
learn
to produce
sentences
they have
never
heard
- children are
not
directly
taught
language
- language
develops
very
quickly
and
easily
for kids
poverty
of
stimulus
- the
language
kids hear isn't
perfect
, but kids use language
correctly
critical
periods
- class defining features:
-- period of peak neural plasticity
-- cutoff point for development
-- occurs early in development
-- brief
-- deprivation has permanent and irreversible effects
critical
periods
for
language
- eric
lenneberg
(theorist)
- around
puberty
ability to develop
language
declines
evidence
for
critical
periods
- children raised in
isolation
- deaf children raised by
hearing
parents
- difficulty learning a
second
language
as an
adult
- can never really conduct the "best" experiment to determine whether a critical period for
language
exists
connectionist
theories
- a theory based on
computer
modeling
that is applied to how
humans
develop
language
- children
develop
language
due to the
statistical
regularities
in the
language
they hear
- brain processes this "
statistical info
" and figure out what it means based on
parallel
distributed
processing
parallel
distributed
processing
- brain learns by processing
multiple
pieces
of
information
across
multiple
layers
of
processing
input
- child
hears
input
hidden
layers
(of processing)
- child's brain sends that
input
to different
neurons
to be
processed
- which
units
it gets sent to is based on how they've
processed
similar
info
in the past
- can trace a path from
input
to
hidden units
- path that is more likely to be
activated
is the one that has been most
successful
in the past (how much each path is weighted)
output
- after being
processed
by multiple hidden layers, the child produces
output
(speech/language)
feedback
- during the
language
phase,
feedback
is provided to the
system
- that feedback changes the
weight
for each
pattern
- that feedback changes which
output
will occur