Mapping words to actions and events: How do 18-month-olds learn a verb?

1
Mapping words to actions and events How do
18-month-olds learn a verb?

• Mandy J. Maguire, Elizabeth A. Hennon, Kathy
  Hirsh-Pasek, Roberta M. Golinkoff, Carly B.
  Slutzky, and Jenny Sootsman

2
The Charge

• Bloom (1994) The diversity in young vocabularies
  is impressive . . . Whatever principles or
  assumptions are at work for word learning need to
  be considerably more general than those offered
  so far to explain how children learn names for
  objects.

3
Thus. . .

• We need to expand our research interests beyond
  objects to adjectives, pronouns, and verbs.

4
Taking up the charge

• Adjectives (Waxman and colleagues)
• Verbs
• Components of verbs (path, manner, etc.) by
  (Naigles, et al. Pinker, 1984 Akhtar
  Tomasello)
• Cross cultural aspects (Xu Carey, 1996)
• Whorfian Hypothesis (Papafragou, Massey,
  Gleitman 2000 Hohenstein Naigles, 2000)
• Syntactic Approach (Gleitman, Fisher)
• Social aspects of verb learning (Tomasello)

5
Whats Missing

• This research has looked at sophisticated words
  learners, 2 years and up.
• We need to address the very beginning of verb
  learning, the ability to map words to actions.

6
Talk is in 4 parts

• Introduction to labeling actions and the action
  word learning problem
• 3 experiments addressing reference and
  extendibility in action labels
• How we interpret these experiments
• Future directions

7
Verbs are hard

• Gentner (1982) was one of the first to address
  why verbs are such a problem
• defined in diverse ways (motion, instrument,
  results)
• ephemeral events not concrete
• verbs have more definitions than objects
• abstract relatedness as compared to perceptual
  similarity

8
Verbs are really really hard!

• All the reasons that verbs are hard to learn may
  contribute to why they are so hard to study

9
Reference and Extendibility

• Generally principles that apply to more than just
  object words
• Principle of Reference words symbolize, or
  stand for, objects, actions, or events (Hollich,
  Hirsh-Pasek, Golinkoff, 2000)
• Principle of Extendibility most words do not
  refer to a single exemplar as do proper names,
  but to categories of objects, actions, or events
  (Hollich, Hirsh-Pasek, Golinkoff, 2000)

10
Running
11
The goal From the study of object nouns to the
study of first action words

• Use proven methodology
• Bare target
• Novel stimuli

12
Methodology

• Proven Split screen Preferential Looking
  Paradigm
• Bare Intransitive verbs with no influence from
  objects or the objects labels
• Novel Novel aerobics actions

13
The experiments

• Experiment 1 Training with person A and
  extending to person Y
• Experiment 2- Training with A, B, C, and D and
  extending to person Y
• Experiment 3 simplifying the action to the and
  extending to person Y
• Only difference between these will be the
  training

14
Procedure
15
General procedure across experiments

• Subjects 16 participants per study, range 17.92
  - 21.07, M19.36
• Procedure
• Salience trials
• Training phase
• Test trials
• Followed by second novel action pair
• Appropriate counterbalancing

16
Experiment 1 Extending from one exemplar

• If one person performs all of the training can
  infants learn and extend the label to a novel
  person performing the same action?

17
Salience Trials
Look up here! Whats up here? What are they
doing?
18
Training Phase
Look shes blicking! Do you see her blicking?
Watch her blicking!
19
Test Phase
Whos blicking? Do you see her blicking? Watch
her blicking!
20
Predictions

• No salience preference
• If infants learn the action label and are able to
  extend it, they will look longer to the target
  action than the non-target action during the test
  trials than they did for the salience trials

21
Was there a salience preference?

• No

22

• Were they able to learn and extend an action
  label?

23
Results Test Trial Experiment 1

• No

24
Discussion Experiment 1

• Infants dont get it
• Its not that they arent paying attention
• By making it so hard we may have transformed the
  task into a perceptual task in which the infants
  prefer novelty
• Too complex?
• Too few exemplars?

25
Experiment 2 Extension from multiple exemplars

• Salience Trials Same as Experiment 1
• Training - 4 distinct, female actors performing
  the target action consecutively
• Test Trials - Same as in Experiment 1

26
Was there a salience preference?
No
27

• Were they able to learn and extend an action
  label?

28
Results Test Trials Experiment 2

• No, but they are doing better

29
Discussion Experiment 2

• Why arent they getting it?
• Too few exemplars
• Fewer needed for nouns
• Each single actions is too complex
• Too much going on in scene for child to focus

30
Experiment 3 The ultimate simplification

• Point light displays of actions

31
Point light images

• 13 points of light corresponding to the head and
  major joint points of a human.
• Infants as young as 3 months apparently see these
  as human forms (Bertenthal, 1984)
• 3-year-olds can recognize known actions in the
  IPLP and can label them when shown in point light
  (Golinkoff, et al.)

32
Procedure

• Salience Same as Experiment 1 (live action)
• Training Point light displays created from the
  exact video clip used in Experiment 1
• Test trials Same as in Experiment 1 (live
  action)

33
Training
Look shes blicking! Do you see her blicking?
Watch her blicking!
34
Was there a salience preference?

• No

35

• Were they able to learn and extend an action
  label?

36
Results Test Trials Experiment 3

• Yes!

t(15) -2.536, p .005
37
ConclusionsWhat do we know?

• 18-month-olds can learn action labels
• But we need to simplify the visual display
• Simplification allows infants to abstract the
  verbal essence and extend to novel agent
• Verbal essence the semantic component of the
  event that is being encoded by the verb
• Verbal essence is highlighted by the point light
  displays

38
Overall results (Target Non-target)
39
What we dont knowWhy it only emerged in point
light displays

• Simplicity / Focus
• Not what verbs label

40
Hypothesis 1 Simplicity/focus

• Single exemplar may have been complex enough to
  transform the task, moving towards novelty
• Multiple exemplars may begin to allow for
  extraction of the invariant
• Point light is less complex, therefore there are
  fewer options as to what to label

41
Two kinds of simplification

• Remove language to get categorization (Werker)
• Simple actions

42
Hypothesis 2 Not what verbs label

• Verbs may not label very complex actions,
  especially those addressed to small children
• Very few manner verbs in young lexicons (Naigles)
• By using intransitives, the verb is not anchored
  by a ground or goal (Gleitman, Naigles, Fisher)
• These actions may appear unmotivated and without
  clear intent

43
In Conclusion

• Infants can categorize novel actions
• Can map words to these actions (Principle of
  Reference)
• Can extend action labels to new instances
  (Principle of Extendibility)
• Part of what makes verb learning so difficult is
  extracting the verbal essence from complex events

44
Future Directions

• We are looking at the very beginning of verb
  learning, attaching labels to actions
• Future goal Understand how perceptual and
  linguistic factors interact to permit verb
  learning