Transcript Where actions meet words: - College of Liberal Arts

Where actions meet words:
The paradox of early verb learning
Kathy Hirsh-Pasek
Roberta Golinkoff
Temple University
University of Delaware
With support from many students, graduate
and undergraduate, and NSF
• Mandy Maguire
• Beth Hennon
• Shannon Pruden
• Meredith Meyer
• Carolyn Fenter
• Jennifer Sootsman
• Rachel Pulverman
• Sara Salkind
• Khara Pence
• Dede Addy
• Natalie Hansell
Beginning at the beginning…
Language- what’s the big deal?
Language can
start wars
ruin marriages
allow a colloquium presentation
Language allows us to label objects…..
But more importantly….
Language is about relations
The power of language is not in learning the word
“cabbage” and the word “Jim” but in learning how
to express relations between these words.
“Jim ate the cabbage”
“The cabbage attacked Jim”
“Jim, don’t sit the babies in the cabbage!”
And relations are encoded
in…(among other things…)
VERBS
Verbs form the architectural
centerpiece of the sentence.
You just can’t learn language without
learning verbs!
In this talk…
We begin to explore the new frontiers of
verb acquisition by studying how children
learn their first action words.
We will thus use the term “verb” loosely to
refer to action words.
With this caveat in mind…….
We offer a talk in 4 parts:
• The Paradox:Verbs are HARD to learn
• But children have them in their earliest vocabularies
• Theories of verb learning
• Building verbs: A developmental account
• Explaining the paradox: A beginning
Part 1: Verbs are hard to learn
Act I: Demonstration through a typical
motion event
Qui ckTi me™ and a YUV420 codec decompressor ar e needed to see this pi cture.
QuickTime™ and a
decompressor
are needed to see this picture.
What did you see?
•How would you describe it?
•What nouns did you use?
Sliding board? Child? Apartment building?
Ground? Grass?
•What verbs did you use?
You might have used verbs like…
approach
ascend
bend
climb
descend
go
grab
hit
leave
lift
pull
push
run
sit
slide
stand
step
straighten
swing
tuck
The “verb” problem
• A verb encodes only part of what is
happening in a motion event including
(from Talmy, 1985):
– Manner – the way an action is carried
out
– Path – the trajectory of an action with
respect to some reference point
Cross-Linguistic Differences
• Languages differ in terms of the relative
frequencies of different types of verbs
– Path and Motion
• e.g., Spanish, Turkish, Greek
– La mujer salió de la casa (corriendo)
‘The woman exited the house (running)’
– Manner and Motion
• e.g., English, Indonesian, Chinese
– The woman ran out of the house
Sliding Event
approach
ascend
bend
climb
descend
go
grab
hit
leave
li f t
p u ll
p u sh
run
sit
slide
stand
step
straighten
swing
tuck
PATH
MANNER
Gentner (1992, 2001,2003) suggests verbs
are harder to learn than nouns because…
•Verbs more polysemous than nouns
e.g., “run” - 53 entries!; “ball” - 2 entries
•Label relations as compared to perceptual
similarity or function
•Harder to individuate actions than objects and to
to form categories of actions than objects
( What is the invariant in “running” when performed by
Carl Lewis or your grandmother?)
•Ephemeral events: not concrete
e.g., running vs. cup
Act II: Verbs are really hard
A demonstration from Japanese and
English
The rationale
Some have argued that a noun bias is a
product of learning English. In Asian, “verb
final” languages, children have a higher
proportion of verbs in their early
vocabularies relative to nouns. Thus, verbs
might be as easy to learn as nouns in these
languages.
(Tardiff 1996, Gopnik & Choi)
Standard Scene
“見て! Xっている”
“Look ! (She is) X-ing a)”
a) ‘Ｘ-ing’ is a novel verb.
Two Test Scenes
same object, different action
same action, different object
“Xっているのはどっち?”
“In which (movie) is (she) X-ing?”
The facts
• Participants:
– 41, 3-year-olds (M=3;6)
– 40, 5-year-olds (M=5.0)
• Task: Pointing to one of two scenes on
video
Japanese Results
A replication in English
(Meyer, Hirsh-Pasek,Golinkoff, Imai, Haryu)
• Same tapes used in Japanese
• Same ages: 3 (N=55) and 5 yrs. (N=59)
• 3 language conditions:
- Noun (“Find the blick!”)
- Bare verb (“Blicking!” Where’s blicking?”)
- Rich syntax: Agent/Obj/Verb (“Where is she
blicking?” “Look at her. She is blicking it.”)
English Results
Act III: Verbs are really, really hard
So we simplified the design.
Asked children to learn and extend only
one novel action, no novel object present.
And they still couldn’t do it by age 3 years
THE BOTTOM LINE?
We got a headache!
THE PARADOX:
Verbs are really, really, really hard to learn…
BUT…
They appear in children’s earliest vocabularies
* Choi & Bowerman, 1991; Choi, 1998
* Brown ;Bowerman, deLeon & Choi, 1995
*Fenson et al., 1994; Tardif, 1996, 1999
Part II:Addressing the paradox
Three theories of verb learning
Three theories
• The “Universal Concepts” theories
• The “Language-specific” theories
• The “Hybrid” theories
Universal Concepts Theories
Universal
concepts
PATH
MANNER
CONTAINMENT
CAUSALITY
Language
maps onto
concepts
“The central problem is how do children, from an initially
equivalent base, end up controlling often very differently
structured languages.”
Bowerman & Levinson (2001)
Evidence for Universal Concepts
Theories?
 Languages around the world draw on the same set of concepts
(Talmy, Langacker, etc.)
 Perceptually salient (concrete, individuated) information will
be coded first.
 Developmental data: Bowerman (1974) “ He falled it.” and
Clark (2001) (“y” for inherent properties (he is short) and “ed” for
temporary (he is tired).
Language-Specific Theories
Language
Concepts
 PATH
MANNER
CONTAINMENT
CAUSALITY
Evidence for language-specific
theories?
Words are invitations to form categories (Brown, Balaban &
Waxman; Maguire, Hirsh-Pasek, & Golinkoff ).
The terms, “pour” vs “spill” invite listeners to find distinctions
between these concepts.
 Verbs learned one at a time, then generalized via
common syntax (Tomasello’s Verb Island Hypothesis).
Choi & Bowerman, 1991; Akhtar & Tomasello, 1997); Schlyter, (1990 on
bilingual development in French and German);
GO used with separate senses (Theakston et al.,2002)
Hybrid theories
Language
input
Universal
concepts
Together
determine
Verb meaning
Hybrid Theories
• Natural partitions hypothesis (Gentner & Boroditsky’s, 2001; Gentner,
2003): Abstract universal concepts that are easily individuated across
multiple instances.
• Slobin (2001): Both conceptual primitives and language input work
jointly in the child’s construction of verb meaning.
• Gleitman et al, 1991, Fisher et al., 2002; Naigles: Syntax of language
critical to “zooming in” on verb meaning.
• Hirsh-Pasek & Golinkoff (forthcoming), Emergentist Coalition Model:
children start with universal perceptual/conceptual foundation using
syntactic and social cues to prune language-specific verb meaning.
The Emergent Coalition Model
Linguistic and social cues sculpt universal concepts
in ways consistent with the native language
Universal
perceptual/
conceptual
Linguistic
Social
2nd and 3rd year of life
Predictions
• Infants should be able to discriminate and categorize universal
concepts (e.g., path, manner)
• When action meets words, children should assume that the
word labels the most perceptually salient universal relational
concept (e.g., path over manner)
• Embedding the verb in rich syntax, allows children to map the
verb to the action in language-specific ways
• Attuned to speaker social intent, children should map a verb to
an action in language-specific ways (in progress)
To investigate this we need…
• To find universally available concepts used
differently across languages
– Enter PATH and MANNER
• To find methodologies that can assess verb
comprehension in young children of different ages
– Enter Habituation, Preferential Looking
(IPLP), Preferential Pointing Paradigms
(PPP)
Part 3: Building verbs:
A developmental account
What does it take to learn a verb?
A 3-pronged approach:
1. Nonlinguistic conceptions of actions in events
(finding action; processing actions in ways relevant to language;
forming categories of actions.)
2. What happens when action meets word?
What does it take for a baby to learn a verb? What factors influence
early verb learning?
3. How is children’s verb learning influenced by the
syntax of the target language and by an
understanding of speaker social intent?
Nonlinguistic conceptions
of action
Study: Can infants discriminate proposed
universal concepts in nonlinguistic events?
Method:
Habituation
Participants:
18, 7 month-olds : 40, 14-17 months
Vocabulary for older children:
Half above mean; half below (on MacArthur)
Question: Can infants dishabituate to new events that
change the MANNER and/or the PATH of an event?
(Pulverman, Sootsman, Golinkoff & Hirsh-Pasek, 2002)
Enter Starry
The Habituation study…..
(Pulverman)
QuickTime™ and a V ideo decompressor are needed to see this picture.
Stimuli
9 computer-animated motion events
3 Manners
3 Paths
flapping
over
spinning
under
bending
past
NO LANGUAGE ACCOMPANIED THESE EVENTS
Procedure and design
• Habituated to one of 9 stimulus events
• Trials ended after 2-second look away or 30
seconds, whichever came first
• Within subjects design
• IV= test conditions
DV= looking time
Test Events, an eg.
• Control Event (Habituated to )
– Flapping Over
• Path Change Event
– Flapping Under
• Manner Change Event
– Spinning Over
• Both Change Event
– Bending Past
(order counterbalanced across children)
Drumroll Please:Major Findings
1) 7 & 14-mo olds discriminate universal action
components (MANNER and PATH) in ongoing events.
2) HIGH VOCABULARY CHILDREN, pay more attention to MANNER
changes than PATH changes -- consistent with prominence of MANNER in
English.
-Do all high vocabulary children pay attention to MANNER? OR do
children learning Spanish attend to PATH? (Pulverman et al.).
What happens when
actions meet words?
Predictions
• Children should assume that the word labels the most
perceptually salient universal relational concept (e.g.,
PATH over MANNER)
• A common label should focus attention on languagespecific components in this case to MANNER (Gentner)
• Children who already know some relational terms should
be able to use syntactic support to learn novel verbs (Jones
& Smith; Slobin)
Three experiments
• Experiment 1: Do 2-year-olds assume a novel verb label
refers to the PATH of the action?
• Experiment 2: Can 2-year-olds use multiple exemplars of
events to guide verb learning?
• Experiment 3: Can 2-year-olds use multiple exemplars and
syntactic cues to guide verb learning?
Study: Maguire (2003):Will infants attach a
word to the most salient, individuated aspect
of an event?
Method:
Preferential Looking to one of two video events
Participants:
Age - 16 children at each of 2 ages (2 & 2.5)
High ( >95%) and average relational vocabulary
(prepositions, verbs, on MacArthur,) 7 high
relation, 18 average
Will children take a novel verb as a label for
the PATH or MANNER of an event?
Question:
Path = most salient; but
English tends to label MANNER not PATH
Intermodal Preferential Looking Paradigm
Design
•
•
•
•
Introduction: Introduce Starry
Salience trials: Test salience of test trials
Training: Teach a novel action label
Test trials: Does the child take the verb to
mean the PATH or the MANNER of the
event?
Design
Introduction:
Meet Starry. Starry is fun!
Salience:
Look Starry is blicking! Watch Starry blicking!
Training: Spin over
Look Starry is blicking! Watch Starry blicking!
Design cont. TEST TRIALS
Test trial 1:
Bend over; spin past
Where is Starry blicking?
Test trial 2:
Where is Starry blicking?
Test trial 3:
Mutual exclusivity
Where is Starry hirshing?
Test trial 4:
Recovery
Where is Starry blicking?
Results
• No age differences
• Children with more relational words looked
significantly longer to the PATH even though
English tends to have MANNER verbs
• English-speaking 5-year-olds and adults all chose
MANNER as the referent for “blicking.”
Study: Can we make toddlers approach verb learning
like English speaking adults?
Method:
Preferential Looking
Participants:
30 children, 2 ages (2 and 2.5 years)
Do multiple instances of same MANNER
(spinning) across different PATHS (around, under) bias
children to assume that a novel verb labels MANNER?
Question:
NOTE: the only difference in this study is in the training
video which now shows 4 different PATHS, one MANNER.
The training audio remains, “See, Starry blicking.”
Results: Nothing!
• No age effects
• No vocabulary effects
Significance: Seeing multiple instances is not
enough to sway young language learners to a
MANNER bias for verb meaning
Will anything cause children to interpret
a novel verb as labeling the MANNER
of an event?
Study: Adding rich syntax
Adding sentences to last study
Method:
Preferential Looking
Participants:
30 children, 2 ages (2 and 2.5 years)
Question: Can toddlers use syntax + multiple instances
to discern that the verb is labeling MANNER?
NOTE: The only difference in this and the last study is in
the training which now adds syntactic information, “Look,
Starry is blicking around (under, past) the ball!”
Results
• No age effects
• Large vocabulary effects! The higher
relational vocabulary children now assume
that the word labels the MANNER.
Discussion:
• As predicted: Children map verbs onto the most salient
universal action concepts (PATH over MANNER, as
suggested in the perception studies)
• What helps English-speaking children move from a
reliance on PATH to a language-specific reliance on
MANNER? Syntax and/or multiple instances.
For children with higher relational vocabularies sentences
that block the PATH interpretation, yield adult-like
performance.
Part 4: Given the data, how do
we begin to explain the paradox?
Verbs ARE hard to learn(at least
in the lab), but children have
them in their earliest vocabularies
Musings that need your input!
Imagine that verb learning occurs on a
developmental continuum? (Gentner, 2001)
Perceptually based
Specific context
Social intent is clear
Rich language input
Extension limited
Contextually impoverished
Reduced language input
Social intent ambiguous
Extension even metaphorical
18 mo.
4 yrs.
Evidence suggests that Early Verb Learning
is context bound and used in situations where all the cues for
verb usage (perceptual,linguistic, and social) overlap
-Behrend, Forbes & Farrar studies - Young verb learners
very conservative in their extensions
-Tomasello “verb island hypothesis” One verb at a time.
This is exemplified in Chinese early verb use.
Tardiff: Children begin with more focused narrow verbs,
associated with particular and consistent routines or social contexts
(e.g., hug, kiss )
LATER Verb Learning is of two forms.
2.5 - 3 years - children can learn novel verbs and extend
them with syntactic (or perhaps social) support (Maguire,
2003; Fischer, Naigles) BUT only in limited, related contexts
(e.g, , you can substitute the agent but not the instrument)
 5-7 years - rapid extension to new situations is observed,
the meaning of the verb is “lifted” from its originally learned
context and become truly relational.
Why the paradox?
• While toddlers look like they have some verbs
early, when pressed in laboratory research, their
limitations become clear.
• Only older, late preschool children can represent
the abstract relations between language and
events. These children are no longer bound to
context. E.g., Hammering can even be done
without a hammer.
This raises the question: Why can’t the
youngest children learn verbs that are context
free? The learning problem is not with…
• …discriminating or categorizing relations in non-linguistic
events (studies by Pulverman et al., Pruden et al.).
• …forming mappings between language and aspects of the
real world -- lots of work with nouns (e.g., Hollich et al.)
• … or even with recognizing which element in the sentence
is the verb (Golinkoff et al., sensitivity to /ing/ 16-18 months)
Rather, the problem might be that verb
learning requires the ability to abstract
relations across multiple domains.
As Gentner (2003) argued…
As similarity comparisons evolve from being perceptual and
context bound to becoming increasingly sensitive to common
relational structure, children show an increasing capacity to
reason at the level of abstract commonalities and rules.
(p. 201)
Verb learning might not be just about
verbs, it might be about the ability to
reason about relationships.
Indeed, there are suggestive parallels in
the mastery of verb learning and in…
• Number development
Huttenlocher
• The development of analogy
(Ratterman & Gentner, 1998, 2002; Lowenstein & Gentner)
• Even in relational noun development
Island (Keil)
Passenger (Hall &Waxman)
We don’t want to stretch this too far. Yet, we do want
to leave you with some interesting final thoughts…
1. Adult-like verb learning is really, really, really, hard
for children.
2. And it is hard even though, children seem to have
the prerequisites that should enable them to use
verbs productively
3. We believe that part of the problem is that to learn a
verb, children have to coordinate information about
relations across contexts and domains ( language,
social & perceptual)
4. And that this coordination across relations might
prove a stumbling block in language learning and in
other aspects of cognition
Perhaps Piaget got the picture when
he suggested that young children
had difficulty dealing with relations
prior to age 5.