Breaking the Word Learning Barrier: How children learn their first words. Kathy Hirsh-Pasek Roberta Golinkoff Beth Hennon Mandy Maguire. Outline. I. The Word Learning Problem II. Three Theories of Lexical Acquisition III. The Emergentist Coalition Model IV. Evidence for the Theory
Breaking the Word Learning Barrier:How children learn their first words
The Word Learning Problem
At twelve months, David utters his first word. In just seven or eight months, he will be learning up to nine new words a day and will have over 50 words in his productive vocabulary.
• Segment words from the constant flow of speech (Jusczyk, Cutler, & others).
• Find coherent objects, actions, and events in their environment (Spelke, Baillargeon, & others).
• Map the words to those objects and events in a symbolic way (Markman, Clark, Waxman, Tomesello, & others).
Three theories of Word Learning
Words map to objects, actions, and events
Words map to whole objects
Words map to more than one exemplar
“The typical way children acquire words...is almost completely opposite of the Quinean paradigm. Children do not try and guess what it is that the adult intends to refer to; rather...it is the adult who guesses what the child is focused on and then supplies the appropriate word (pp. 240-241)”
• Single cues not multiple cues
We need to avoid snapshots and find a system which embraces change.
Mutually exclusive or compatible?
Part I I I
An Attempt at Integration
We propose a hybrid-developmental model in which children start with basic word learning principles but the character of these principles changes over time. For example, children might start with a principle of reference -- that a word refers. At first, a word is ASSOCIATED with the most interesting object. Later, word reference is SOCIALLY informed, mapping onto the object the speaker has in mind.
We propose a model of an active child which has the following properties:
Only from the combined action of multiple cues is word learning even possible.
• Domain-general to specific
• Own view to other’s view
• Indexical (signal) to symbolic
Principles are thus the products, not the engines of development.
• Do children use multiple, overlapping cues in word learning?
• Does the weighting change over time?
• In this manner, are word learning principles emergent products?
Part IV: Evidence for the Emergentist Coalition Model
The Principles of Reference and Extendibility
as cases in point.
The Case of Reference
Considered by many to be the most basic of word learning principles, often considered to be a “conceptual primitive.” States that words symbolically “refer” to objects, actions, and events.
Immature Principle of Reference
Attaches label to object, action, or event that is the most interesting in the environment -- perceptual cues dominate mapping. Often wrong!
Mature Principle of Reference
Social cues dominate mapping. Child becomes apprentice to adult for quick and reliable learning.
If present children with interesting and boring objects AND label the boring object (e.g.. Look at it, point to it, handle it)…
Children with an immature principle:
Assume the label maps to the interesting object regardless of what the adult does.
Children with an mature principle:
Assume the label maps to the boring object that the speaker has in mind.
“Where’s the Ball.”
“Do you see the ball? Look at the Ball.”
Children will allocate more attention to the object that “matches” the requested object.
Younger children would map words (associate words) to the most interesting object even in the conflict condition while older children will use social cues to map words to objects and will label the boring object if that is the object labeled by the experimenter.
- Toy labeled
- Side of match
- visual fixation
99 subjects; 33 at each of three ages - 12, 19, 24 mo.
Validation: Familiar Trials within paradigm assess whether children can do the task.
Test of model: Novel Trials in which children are trained to map a label onto either an interesting or a boring unfamiliar toy.
Where infants able to do the task?
Mean Looking time (sec)
Was the interesting toy and boring toy really interesting and boring?
Were the children able to follow
eye gaze in the labeling phase?
Mean Diff Looking time (sec)
What we know:
* We can develop methods that allow us to peer
at cues used in the word learning process
* Principle of Reference appears to be emergent:
There is a changing emphasis on social cues.
What we don't know:
* Whether 12 month olds are capable of learning a
label in our task at all.
The current results leave us with two potential interpretations:
The infants are clueless and cannot label at all
The infants are learning a label BUT are
labeling the interesting object -- the
perceptually salient choice in accordance with our hypothesis.
Mean Difference in Looking Time
(* indicates p<.05)
* The label is making a difference, but only in the coincidental condition.
* 12-month olds are lured by perceptual salience, but they did not attach the label to the interesting object in the conflict condition. Thus, contrary to predictions, they are conservative learners who need consistent cues, both social and perceptual, in order to attach labels to objects.
So where’s Fido?
Is there ever a time when children rely solely on perceptual cues to make world-to-word mappings?
And this is what we found…..
Results of Study 3 with 10-month olds
No differences across conflict and coincidental conditions!
Salience2.80 s. 2.25 s.*
Training 4.04 s. 2.21 s.***
Novels 2.80 s. 1.66 s.***
Glorp 2.06 s. 1.53 s. (ns)
Recovery 2.47 s. 1.34 s. **
At the very beginning of word learning, children label the most interesting object regardless of what the adult is doing.
Our recent SRCD monograph:
Hollich, G., Hirsh-Pasek,K., & Golinkoff, R.M. (2000) Breaking the language barrier: An emergentist coalition model for the origins of word learning.
We know from the principle of reference that children can label a single object, but do they know that most words refer to categories of objects, rather than to single objects, actions and events?
Child operates with proper noun hypothesis; infant
labels unique object and then quickly moves to
extension based on perceptual similarity -- using child’s
point of view.
Child uses speaker’s point of view to extend word to
category member even if no perceptual similarity.
167 children participated; at ages
Data reported from 95 children who learned the label for the original exemplar (ranging from 60% of the 10- and 12-month-olds to 75% of the 24-month-olds).
10-month olds prefer to look at the original exemplar when paired with an exemplar that is identical with the exception of color (p < .05)
12-month-olds show a trend towards the proper noun hypothesis (p = .06)
No evidence for proper noun hypothesis at other ages
At 19- and 24-months
At 10-, 12- and 14- months
These children CAN extend, but extension is fragile and is influenced by context.
If they are presented with this display:
They CAN extend
If they are presented with this display:
They CAN’T extend
Younger children will default to the proper noun hypothesis if given any opportunity to use it.
Older more sophisticated word learners have learned that the default is the opposite: Words label categories.
As in the acquisition of grammar, it is easier to go from a narrow hypothesis to a broad one than the reverse.
As they hear labels used with perceptually similar objects, they learn -- on very few exposures -- to extend and to form categories (Hollich, 1999, Smith 2000)
Yet, all category members do not share perceptual features (e.g., bean bag chairs and dining room chairs are both chairs) Can they extend labels to these categories?
The Social Extension Experiments
Consistent with the theory, we hypothesized that children with an immature principle of extension, who rely on perceptual information will not accept the category label for a perceptually dissimilar object.
Children with a mature principle of extension will use both perceptual and social cues for category extension.
Exploration and labeling
“Eve, it’s a modi “
“Wow, a modi.”
“Oh, a modi”
“Eve, Look , another modi”
“And look at this!”
“Eve, where’s the modi?”
“Do you see the modi?”
“Show me the modi.”
Infants learning their first words are already operating with immature word learning principles that help get language learning off the ground.
These principles are informed by multiple inputs -- attentional, social, and linguistic.
These principles are both conservative and fragile.
The nature of these principles changes over time with attentional cues dominating early and social and linguistic cues becoming increasingly important.
Part V: Conclusions
• Is there evidence that children use multiple cues?
Only a method that allows us to look at competing cues over time will reveal the true complexity of the system.
The emergentist coalition model embraces many of the characteristics of the associationistic, social pragmatic, and constraints/principles views and demonstrates how seemingly opposing perspectives can be united in a word learning theory that looks at developmental change over time.
Thus far we see THAT change occurs. We must now begin to evaluate what motivates the change.
The emergentist coalition model is a proxy for developmental theory--writ large:
NOT searching for parsimony by identifying
a “smoking gun.”
The new wave: the “radical middle”
We need not be constrained by myopic theories of complex phenomena
For only when we look at how multiple inputs interact across time will developmentalists be able to understand complex behaviors and see a whole baby