A Picture Is Worth a Thousand Words, but That`s the

A Picture Is Worth a Thousand Words, but That`s the

A Picture Is Worth a Thousand Words, but That's the Problem: The Role of Syntax in
Vocabulary Acquisition
Author(s): Lila R. Gleitman and Henry Gleitman
Reviewed work(s):
Source: Current Directions in Psychological Science, Vol. 1, No. 1 (Feb., 1992), pp. 31-35
Published by: Sage Publications, Inc. on behalf of Association for Psychological Science
Stable URL: http://www.jstor.org/stable/20182121 .
Accessed: 10/12/2012 11:15
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp
.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.
.
Sage Publications, Inc. and Association for Psychological Science are collaborating with JSTOR to digitize,
preserve and extend access to Current Directions in Psychological Science.
http://www.jstor.org
This content downloaded by the authorized user from 192.168.82.208 on Mon, 10 Dec 2012 11:15:52 AM
All use subject to JSTOR Terms and Conditions
CURRENTDIRECTIONS INPSYCHOLOGICALSCIENCE 31
11 ,\ 68
macaques, journal of Neuroscience,
L. Ungerleider,
190 (1991); D. Boussaoud,
and R. Desimone,
Pathways for motion anal
ysis: Cortical connections of the medial supe
rior temporal and fundus of the superior tem
journal of
poral visual areas in the macaque,
296
462-495
Neurology,
Comparative
(1990).
25. C. Von Der Malsburg,
The Correlation
Theory
of Brain Function (Internal Report 81-2) (Max
Plank Institute for Biophysical Chemistry, De
partment of Neurobiology,
Goettingen, Ger
many, 1981); F. Crick and C. Koch, Towards
a neurobiological
theory of consciousness,
Seminars in the Neurosciences,
2, 263-275
(1990).
26. CM.
Gray, P. K?nig, A.K. Engel, and W.
Singer, Oscillatory responses in cat visual cor
under 2 years of age, the act
For
referred to is not taking place.
example,
'open' is often said when
dren
A Picture IsWorth a Thousand Words,
but That's the Problem: The Role of
Syntax in Vocabulary Acquisition
Lila R. Gleitman
acquire 5 to 10 words a
the 15th month
about
is
the 6th year of life. How
from
day
through
this astonishing
We
discuss
must understand
the type of relation
ship (the short, sharp contact) and
the argument structure.
feat accomplished?
here the mapping
prob
lem for vocabulary
acquisition:
Granted
that infants can entertain
some concept, how do they discover
in the target language
which word
it? For example,
expresses
granted
that they can conceive
of 'dog' and
'open', how do they learn that /dog/
is the label for 'dog' and /open/ for
'open'?
on the
particularly
of
verb
Verbs
problem
acquisition.
nouns
in
in
the
later
than
appear
fant's speech and are also under
We
is opening.3
And
present
nothing
even more often, when something
is
forebear
from
opening,
caregivers
and Henry Gleitman
Children
focus
stood
later. This developmental
pri
nouns is undoubtedly
of
related
ority
to the fact that they typically
label
while
verbs
label
relation
objects,
For
ships among object concepts.1
a
hit
expresses
example,
relationship
between two entities (the arguments
of the verb), the hitter and the one
hit. To understand
hit, then, one
are Pro
Lila and Henry Gleitman
at the Uni
fessors of Psychology
Address
versity of Pennsylvania.
to
L.R.
Gleitman,
correspondence
in Cognitive
Institute for Research
of Pennsylva
Science, University
3401
Walnut
nia,
Street, 4th Floor,
Philadelphia, PA 19104.
tex exhibit
inter-columnar
synchronization
which reflects global stimulus properties, Na
ture, 338, 334-337
(1989).
27. R. Baillargeon, Object permanence
in 3.5- and
4.5-month-old
infants, Developmental
Psy
(1987); R. Baillargeon,
chology, 23, 655-664
Young infants' reasoning about the physical
and spatial properties of a hidden object, Cog
nitive Development,
(1987).
2, 179-200
saying 'open': When
returning home
a mother
from work,
rarely greets
her child by saying, "I am opening
the door, Joey," but says, "Hello!"
instead. Thus, the assumed
contin
events
between
and
gency
opening
of 'open' fails to a con
in each direction.
degree
Even when
the speech act is perti
the utterance
siderable
WORD-WORLD
PAIRINGPROCEDURES
solution to the map
is
ping problem
by appeal to princi
in
of
and pragmatic
ples
perception
ference.
Infants are said to pair
occurrences
of, say, the sound
The standard
/open/ with certain observed events
is the verb most
because
/open/
to say when
the
for
caretaker
likely
In Locke's
is opening.
something
show
words,
"People
ordinarily
them the thing of which
they would
have
them have the idea; and then
repeat to them the name that stands
in the end some
for it."2 Though
sort
must
be part of the
of
this
thing
answer to the mapping
issue, this ex
some
un
leaves
planation
mysteries
a
we
few
of
which
sketch
solved,
below.
Imperfections
Word-World
of the
Contingencies
A first problem
inmapping verbs
their real-world contexts
is that
caretaker speech is not a faithful run
on events in view.
ning commentary
onto
In fully a third of verb uses
Copyright ?
1992 American
to chil
Psychological
nent to current events, the listener's
focus of attention may be elsewhere,
as when
the mother
says, "Come
take your nap!" while
the child in
spects
a cat
on
a mat.
In the cases
the
just discussed,
is exceed
contingency
In other cases,
it is
ingly imperfect.
subtle and invisible: Many verbs un
word-world
by 3- and
adequately
encode
abstract
mental
4-year-olds
acts and states that are not straight
such as want,
forwardly observable,
and
know4
hope,
derstood
the Relevant
Abstracting
Event
World
A related problem
is that the
scene that accompanies
utterance of
a verb includes many events, only
one of which
is encoded
by that
verb. Consider the plight of the child
to whom
the mother says, "Do you
want
this ice cream cone?"
The
is speaking, smiling, holding
and waving
the cone, and perhaps
to
it; the cone is observably
pointing
to eat, dripping,
something
good
mother
Society
This content downloaded by the authorized user from 192.168.82.208 on Mon, 10 Dec 2012 11:15:52 AM
All use subject to JSTOR Terms and Conditions
32 VOLUME 1, NUMBER 1, FEBRUARY1992
an object of present desire,
melting,
and so forth. None of these aspects
of the scene is irrelevant to the con
versational
intent, yet only one of
them is correct to map onto the item
want. A picture isworth a thousand
but that is the trouble: A
words,
describe the varying
one
of
any
picture.
aspects
have
investigators
Usually,
solved these problems only by beg
they raise. For ex
ging the questions
are said to ignore
ample, children
thousand words
whose
adult utterances
meanings
are unrelated to the present context.
avoid
But then, how do children
a
different
incorrect)
(and
positing
pairing of the verb heard with some
event
in the environ
that is present
assumes
if
the
learner
is,
ment? That
that
onto
maps
not know
the
does
of some utterance
heard,
attempt a pairing between
speech
and
events,
meaning
she must
current
is hap
and whatever
a
to
such
learner, for in
pening. Woe
a case we mentioned
she
earlier,
that utterance
then try to pair up "Come take
ob
your nap" with the cat-on-mat
false step.
servation, a decidedly
As for abstract verbs, the prob
must
for observational
they pose
are
aside be
waved
often
learning
cause their acquisition
is relatively
lems
late compared with the acquisition
if this lateness
of action verbs?as
answers
of how they
the question
are learned, in the end.
Verb Doublets
Other
attempts
difficulties
these
to circumvent
a prob
involve
in which
procedure
is based on
choice
abilistic
the
the
mapping
most frequent word-world
match,
across situations.
But this approach
faces another problem. Many verbs
in
come
in pairs that vary primarily
on
a
the speaker-perspective
single
action or event, and thus their situ
are virtually al
ational concomitants
ways
get.
the same.
Both
these
Consider
give
describe
verbs
and
the
between
I same transfer of possession
two parties.
If John gives a book to
gets the
necessarily
Mary, Mary
directors
book from John. Movie
an art of distinguishing
be
make
tween such notions. The camera can
zoom
in on the grateful recipient,
the giver out of
screen altogether.
focus
or off
pairing,
taking advan
tage of the clues to interpretation
that reside in the structure of the sen
tence heard.5 The nature of the rela
tionship
expressed
out across
spelled
ture within
the
the word
Using
a
is
than
rather
get
give
linguistic
way of making the same distinction.
But only for a listener who under
the meanings
of the two
stands
a
zoom
words. Without
lens, how is
the child who
word-world
neither word to
the adult meant 'give'
knows
is
by the verb
the clause struc
the verb occurs.
which
For example,
the fact that hit is a
verb
(one that ex
two-argument
a
a hit
between
presses
relationship
ter and a thing hit) predicts that it
occur
in transitive
structures,
as
such
John hit Bill, whereas
laugh,
a one-argument
in
verb, will occur
will
intransitive
structures,
such as John
because ev
guess whether
or 'get'? This problem
holds
for
lead/follow, and
chase/flee,
buy/sell,
hundreds of other such doublets.
laughs. More generally,
of the verb requires
ery argument
one noun-phrase
in a syn
position
tactic structure, biff heard in a sen
To get around this difficulty, pro
of the word-world
ap
ponents
proach appeal to biases in event rep
in an
For instance,
resentation.
we
more
will
describe
experiment
tence likelohn biffs is unlikely to be
both
and
3-year-olds
a nonsense
with
fully below,
adults presented
verb and shown a scene of a giving
getting event almost always guessed
verb meant
that the nonsense
'give'
and not 'get' or 'take'. This interpre
tive bias is very real, but itwill not
solve the mapping problem because
mothers cannot be relied on to utter
in the presence
of such a
'give'
scene: As it happens, get ismore fre
in maternal
usage than give.
quent
The same bias in event representa
learn
tion that would
help children
in
them
undermine
should
give
an
if
encode
for
get,
they
learning
event as 'give' and hear "get," their
hypothesis must be that "get" means
'give.' Yet the learning function
not characterized
by confusions
words.
such
among
is
A THEORETICAL
ALTERNATIVE:
SENTENCE-WORLDPAIRING
that these problems can
if
learners perform a sen
be solved
rather than a
tence-world
pairing
We
claim
Published
by Cambridge
University
relation such
describing a two-place
as 'hit' even
if perceptual
biases
commend
such a construal. This op
to the
by attention
new
must
the
verb
be
de
structure,
some
the
other
of
aspect
scribing
scene in view: Perhaps John is con
tion eliminated
vulsed with mirth while he hits Bill.
To
narrow
the correct
for
the search-space
must
learners
mapping,
take advantage
of these argument
taking properties of verbs, and the
are expressed
way these arguments
In this sense, the struc
in sentences.
ture of the sentence
that the child
like a mental
hears can function
zoom lens that cues the aspect of the
scene
is describing.
the speaker
verb learning begins
Consequently,
the child shows rudimen
just when
tary sensitivity to syntactic structure
of
and has a working
vocabulary
simple nouns.
In principle,
a systematic
map
semantics
and
between
syntax
ping
could be one of the presuppositions
that learners bring into the learning
situation, part of the innate process
to
it possible
ing system that makes
One
assumption
language.
in
lan
this
implicit
hypothesized
is
that
machinery
guage-learning
some specific
between
mappings
and their syntactic
verb meanings
across
hold universally
expressions
learn
Press
This content downloaded by the authorized user from 192.168.82.208 on Mon, 10 Dec 2012 11:15:52 AM
All use subject to JSTOR Terms and Conditions
CURRENTDIRECTIONS INPSYCHOLOGICALSCIENCE 33
is that very
Another
languages.
can analyze
the
children
young
structure of a sentence
heard and
use this information to derive the ar
structure of the verb. At first
seems too
a procedure
such
glance,
abstract to grant to babies. Yet the
gument
operation of this kind of mechanism
in a variety of
has been documented
as young
with
children
experiments
as 16 months.
One
investi
such experiment
16- to 18-month-old
gated whether
utter only
isolated
children who
nouns are attending
to the structure
a new verb occurs,
within which
and can assign semantic roles to the
entities
based on their
participant
Two video
structural positions.6
scenes
were
to the in
shown
taped
fants, while half of them heard, "Big
Bird is tickling Cookie Monster,"
and the other half heard, "Cookie
Monster
?stickling Big Bird." The in
fants looked
longest at the video
screen that matched
the syntax. We
can conclude
that even before any
verbs are uttered, infants understand
the semantic
versus direct
implication of subject
(or, at mini
object
mum, of serially first versus second
As we
noun) in English sentences.
if
is in
this
next,
explain
machinery
it supports deductions
about
place,
which action is being described by a
novel verb.
In several experiments,
we asked
whether
dis
young children would
two
inter
between
different
tinguish
pretations of a novel verb based on
the number and positions
of noun
in
construction
used
with
phrases
that verb.7 In one such study, a vid
eotaped scene shown to infants aged
22
to 24 months depicted
two ac
A duck forced
tions simultaneously:
a rabbit to squat by pushing on its
the duck and rabbit
head, while
each wheeled
its free arm in a circle.
While
this scene, half the
watching
say, "The
subjects heard a voice
duck is biffing the bunny"; the other
half heard,
"The duck
and the
bunny are biffing." This video was
then
removed,
and
the voice
said,
"Find biffing now!" Two new videos
Inone, the duck was
then appeared.
to squat (but there
the
rabbit
forcing
was no arm-wheeling).
In the other,
the duck and rabbit were
side by
side wheeling
their arms (but the
duck was
not forcing the rabbit to
Fifteen
of 16 infants looked
squat).
scene
at
the
that matched
the
longest
was
in
which
intro
syntax
biffing
to the intran
duced: Those exposed
sitive sentence watched
the arm
and those exposed
sentence watched
scene. Clearly,
the forcing-to-squat
structure was decisive
the sentence
scene,
wheeling
to the transitive
in cuing which
aspect of the com
scene
was relevant to the
initial
plex
of
/biff/.
interpretation
results held as strongly when
the number rather than the position
of noun phrases was the available
syntactic clue to interpretation (e.g.,
push vs. fall, feed vs. eat).
These
To be sure, under all these pre
sentation
the subjects
conditions,
were guessing
the verb meaning
by
the situational
inspecting
But what they thought was
ing was powerfully?albeit
context.
happen
implicitly
?affected
by the linguistic context.
in event representa
biases
Despite
tion, the learner can and does avoid
errors by respecting the semantic
im
plications of structural formats. For
this simple case, the listener posits
that the subject of the sentence
is the
"doer of the action."
Another experiment
directly pit
ted scene-interpretive
biases against
clues.8
syntactic
olds were
shown
Three-
and
4-year
videotaped
scenes
that could be labeled by different
verbs of the doublet
cussed
variety we
dis
earlier
(e.g., chase/flee, give/
introduced
get). The children were
to a puppet who spoke "puppet lan
guage" and asked to tell the experi
menter
what
the puppet meant as
watched
these scenes. For ex
they
one
in
scene, a rabbit ran
ample,
across the screen from left to right;
as
THE RELATIONSBETWEEN
SENTENCEFRAMEAND
VERBSEMANTICS
a
rabbit was disappearing,
at the left and ran
skunk appeared
across.
If the puppet said, "Look!
the
80% of the subjects
"He's
responded,
chasing him," or
"He's running after him." Without
syntactic clues, then, the scene was
more
likely to be interpreted as de
than fleeing. This
picting chasing
bias was enhanced
(held for well
over 90% of subjects) when
sup
Biffing!"
about
that is,
ported by syntactic evidence,
if the puppet said, "The skunk is biff
ing the rabbit." But this conjecture,
the most natural visual
evidently
cognitive
interpretation of the scene,
was offered by fewer than 30% of
who heard, "The rabbit is
re
biffing the skunk." The modal
to this latter sentence was
sponse
"He's running away from him," or
"He's trying to get away from him."
children
Copyright ?
1992 American
Psychological
in hand, we can
These findings
look a little more closely at a theory
that can explain them. Our view is
that the several syntactic structures
inwhich particular verbs can appear
determine
certain
verbs' meanings,
related
aspects
taking
Within
aspects
of
the
those
specifically,
to the argument
of the verbs.
properties
con
the coarse-grained
on meaning
imposed
by
straints
these structural
facts, the search
is narrowed
space for verb meaning
to allow observation
to do itswork.
as examples
Consider
the verbs
give, go, think, and explain. Overall,
these verbs are very different in their
yet they share certain se
meanings,
mantic properties. Both give and ex
the transfer of some
describe
two parties, and for
entity between
this reason, they can both appear in
such
structures,
three-noun-phrase
as lohn gives the pencil
to Bill and
lohn explains the facts to Bill. In the
an object goes from
give sentence,
plain
John's to Bill's hands, and in the ex
the facts go from
plain sentence,
Society
This content downloaded by the authorized user from 192.168.82.208 on Mon, 10 Dec 2012 11:15:52 AM
All use subject to JSTOR Terms and Conditions
34 VOLUME 1, NUMBER 1, FEBRUARY1992
John's to Bill's mind. A noun phrase
is required for the source (John), the
goal (Bill), and that which moves be
I without
an endless
coming
categories
In recent work,
of
proliferation
to mind.
we have shown
them (the pencil or the facts).
that describe no such transfer
in this sentence
frame
sound odd
to
the
Bill,
(e.g., lohn goes
pencil
scope of these se
in the
relations
mantic-syntactic
verb lexicon. We asked two sets of
(adult) subjects to partition a set of
lohn thinks the facts to Bill). This first
verbs, one group providing a seman
tic partitioning,
the other a syntactic
are
If
the structures
partitioning.9
tween
Verbs
thus partitions
a
into
transfer set
verbs
the four
a
set
nontransfer
(give, explain) and
(go, think).
semantic
But there is another
structural
distinction
syntactic property that lines up these
four words differently: Although
go
and give pertain to physical actions,
think and explain pertain to mental
is a syntactic
states or acts. There
correlate of this semantic
property
with
appearance
also,
namely,
tensed
sentence
thinks
that Bill
lohn
complements:
is tall and lohn ex
that Bill is tall. Verbs
acts are
physical
plains (toMary)
that describe
in this construction
anomalous
(John
is tall, lohn gives (to
Mary) that Bill is tall). This distinc
tion partitions the four verbs into a
and a
set (think, explain)
mental
goes
that Bill
physical set (go, give).
These
cross-classifications
the overall
constrain
verely
se
verb in
there are hun
terpretation. Though
dreds of mental verbs and hundreds
of
the number
of transfer verbs,
have both
verbs that, like explain,
is small,
these properties
including
items as shout, tell,
such additional
like explain,
and argue. And these,
mental
describe
transfer,
specifi
the
communication.
Thus,
cally,
several clause structures that are fit
ting to a single verb provide conver
as to itsmeaning.
The
gent evidence
manner
communica
of
specific
tion?whether
shouting or whisper
to be derived
or
ing
explaining?has
by inspecting the scene, to be sure,
for such manner distinctions
(as op
distinc
to argument-taking
posed
in
tions) are not formally encoded
structures.
But the
the sentence
structural factors have narrowed the
for the verb meaning
search-space
to work
observation
for
enough
the power
and
then the
predictive of the semantics,
more
two verbs'
ranges
syntactic
the more the verbs should
overlap,
in their construals.
We
overlap
found that the two partitionings were
correlated.
Furthermore,
powerfully
were
found
the same correlations
the same syntactic
(and involved
when
the experiment
properties)
was replicated
in Italian and in He
For example,
relations be
brew.
tween an actor and an event are de
that accept
scribed
by verbs
as
across
sentence
complements
as
That
there
within
well
languages.
such
universal
syntactic
re
is an obvious
semantic mappings
for a learning machinery
quirement
be
that uses these relations as input to
verb learning.
In related work, we have shown
to
that the actual usage of mothers
children
under
age
2?when
verb
the syn
learning begins?provides
tactic information
required by this
Each of the 24 verbs most
model.10
of
used by a sample
frequently
in a distinctive
occurred
mothers
range of structures,
relatedness
among
and the semantic
these verbs was
by the degree of
predicted
in their syntactic ranges.
that
the findings
Finally, given
children as young as 16 months?
and adult controls as well?actually
closely
overlap
make use of the structural evidence
to derive verb construals,
Lederer,
is using
with
the present authors,
a
to
broader
idea
adult subjects
get
information potential of both
and structural supports
situational
for verb learning. To model a word
world
(in which,
pairing procedure
the learner has no access to struc
I tural information but has many ex
of the
Published by Cambridge
University
to scenes
some
in which
posures
one group was
verb
is uttered),
scenes
shown dozens of videotaped
(but with no audio) in each of which
a mother was uttering some single
verb to her child. This procedure
was
repeated for the 24 most fre
in these mothers'
verbs
quent
not
The
could
speech.
subjects
come within
of
distance
calling
what
the
verbs
mystery
guessing
might be (guessing right inonly 7%
scenes
of trials). The set of observed
the verb
underdetermined
vastly
construals.
In a second
condition,
subjects
told all the other content words
in each sentence with
that occurred
each mystery verb (but not their syn
were
tactic positioning
and without
the
Itmight seem that this co
video).
occurrence
information would
help;
for example,
that such
knowing
nouns
as candy,
ice cream,
and
some
with
verb
hamburger occurred
'eat.' But
might suggest that itmeans
this information
improved subjects'
performance
only to 13% correct,
and (as in the first condition)
their
false guesses were wildly
unrelated
to the target verbs (e.g., for the target
included go, catch,
want,
guesses
and eat). This pathetic performance
hardly mirrors the learning charac
teristics of real children.
In a third condition,
subjects
were shown the dozens of structures
in which
verb oc
each mystery
curred in the mothers'
speech, again
without
the videotaped
context, and
to
with all content nouns converted
nonsense.
tests for
This procedure
the
informativeness
ranges.
of
syntactic
These
subjects
correctly
the target verbs over 50% of
guessed
the time.
their false
Moreover,
into
the
guesses
right semantic
(e.g., for the target
neighborhoods
false
want,
guesses were other men
tal verbs, such as expect and hope).
to several structures
Thus, exposure
more
to
is
informative than exposure
fell
several
scenes.
A final group of subjects received
the structural information, as just de
Press
This content downloaded by the authorized user from 192.168.82.208 on Mon, 10 Dec 2012 11:15:52 AM
All use subject to JSTOR Terms and Conditions
CURRENTDIRECTIONS INPSYCHOLOGICAL
SCIENCE 35
scribed, but the nouns were the real
ones
that the mothers
had used.
These
the target
subjects guessed
over 80% of the
verbs correctly
time, though they saw no scenes.
This result reveals that the nouns co
occurring with a verb can add sig
nificant information about itsmean
(and,
ing if the syntactic positioning
the semantic
role) of these
hence,
nouns
is known. Knowing
that ice
cream
and hamburger
occurred
in the mother's
"somewhere"
utter
ance
is
(as in the second condition)
not too informative: After all, the
sentence might be "Ice cream ruins
your appetite" or "The hamburger
fell on the floor." But knowing that
as the di
these "edibles" occurred
rect object of the mystery
verb is a
'eat.'
good clue that itmight mean
We conclude
that structural infor
mation
is a requirement
for efficient
onto
and its mapping
structure
that provides
tional data source.
conceptual
this addi
work was
sup
Acknowledgments?This
from
ported by a grant to L.R. Gleitman
Steven and Marcia
aid we
Roth, whose
with gratitude.
acknowledge
Notes
1. D. Gentner, Why nouns are learned before
Vol. 2. Lan
verbs, in Language Development:
guage, Thought and Culture, S. Kuczak, Ed.
(Erlbaum, Hillsdale, NJ, 1981).
2. J. Locke, An Essay Concerning Human Under
1964;
standing (Meridian Books, Cleveland,
in 1690), Book
original work published
3.IX.9.
3. R. Beckwith, E. Tinker, and L. Bloom, The ac
quisition of nonbasic sentences, paper pre
sented at the Boston Child Language Confer
ence, Boston (1989, October).
4. One particularly startling example of children's
in this regard is that see was the
competence
first verb uttered by a congenially
blind
2-year-old; B. Landau and L. Gleitman, Lan
guage and Experience: Evidence From the
verb
learning. The frame ranges pro
In
vide strong cues to interpretation.
the presence of this structural infor
the complement
selection
mation,
(the syntactically
nouns)
positioned
In
further
clues.
provides significant
set
the
of
scenarios
taken
contrast,
even taken in combina
alone?or
tion with
of
(asyntactic) knowledge
all co-occurring
too
words?leaves
to
much latitude
allow verb identifi
cation.
in all, our work suggests that
cannot be extracted
verbs' meanings
that pairs single
by a procedure
All
to their observational
contin
gencies. This is because verbs do not
as a rule directly encode actions and
events.
If they did, grunting
and
pointing could substitute for elabo
rate human
In
systems.
language
acts and states
stead, verbs encode
words
of the world
and of the mind under
stances
(and invisible)
particular
toward
these
adopted
by the
is
speaker. A further data source
therefore required to rein in the hun
dreds of salient interpretive choices
made available
and
by perception
as
to
inference
the
pragmatic
speak
er's
intent.
appreciation
It is the
of
infant's natural
structure
syntactic
Blind Child (Harvard University Press, Cam
bridge, MA, 1985).
see L. Gleitman, Struc
5. For a fuller discussion,
tural sources of verb learning, Language Ac
1, 3-55 (1990).
quisition,
6. K. Hirsh-Pasek
and R. Golinkoff,
Language
comprehension: A new look at old themes, in
and
Behavioral
Biological
Aspects of Lan
N. Krasnegor, D. Rum
guage Acquisition,
baugh, M. Studdert-Kennedy, and R. Schiefel
busch, Eds. (Erlbaum, Hillsdale, NJ, in press).
7. L. Naigles, Children use syntax to learn verb
journal of Child Language,
17,
meanings,
357-374
H.
(1990). See also L. Naigles,
L.
and
Gleitman,
Gleitman, Children acquire
word meaning components from syntactic ev
idence, in Linguistic and Conceptual Devel
opment, E. Dromi, Ed. (Ablex, New York, in
press).
8. C. Fisher, G. Hall, S. Rakowitz, and L. Gleit
man, When it is better to receive than to give:
Syntactic and conceptual
supports for verb
learning, unpublished manuscript, University
of Pennsylvania, Philadelphia
(1991).
9. C. Fisher, H. Gleitman, and L. Gleitman, On
the semantic content of subcategorization
frames, Cognitive Psychology,
23, 331-392
(1991); H. Geyer, L. Gleitman, and H. Gleit
man, Semantic/syntactic
linkages in the He
brew verb lexicon, unpublished manuscript,
of Pennsylvania,
University
Philadelphia
(1991).
10. A. Lederer, L. Gleitman, and H. Gleitman,
In
put to a deductive verfracquisition procedure,
Lingua (to appear).
Neural Foundations of
Visual Motion Perception
J. Anthony Movshon
and William
The detection and analysis of mo
tion is one of the fundamental
tasks
of vision, because practically every
thing of interest in the visual world
moves. Although motion analysis of
a high order is evident
in such sim
T. Newsome
linked to a variety of mo
the
tasks,
including
the motion
of complex
patterns, the detection of target mo
tion relative to the background,
and
the generation of signals for smooth
has been
tion-related
analysis of
ple visual systems as the fly's, it is
only in primates that a well-defined
anatomical division of the central vi
sual pathways can be seen to be spe
pursuit eye movement.2
cialized
for the analysis of motion.1
The best defined area in this path
is an extrastriate area known as
way
Unlike other areas of the
MT(orV5).
is Professor in
). Anthony Movshon
the Center for Neural Science and
the Department
at
of Psychology
New York University and an Inves
tigator of the Howard
Hughes
Medical
Institute. William
T. New
some is Associate
in the
Professor
of Neurobiology
at
Department
Stanford University School of Med
icine. Address correspondence
to
Center
for
J. Anthony Movshon,
Neural Science, New York Univer
sity, New York, NY 10003.
extrastriate visual cortex,
monkey's
almost all neurons
inMT are direc
tion
selective,
meaning
best
respond
that they
to motion
typically
a given range of directions,
within
and respond not at all (orwith inhi
in the opposite di
bition) to motion
rection.
Copyright ?
The activity
1992 American
of MT
neurons
Psychological
Society
This content downloaded by the authorized user from 192.168.82.208 on Mon, 10 Dec 2012 11:15:52 AM
All use subject to JSTOR Terms and Conditions