2. Contents of Presentation
What Are Categories? (Why do we need them?)
Philosophical views on category representation
Category Representation in Young Infants
Categorization in Infants. (Do they categorize / How do we measure it?)
Information Used to Form Category Representations
Category Formation vs Category Possession (Online processing/ Previous knowledge)
Order of Emergence
Not just for objects. (Spatial and other abstract categorizations)
4. Which Painting Has more Information Content/Why?
Composition VII, Wassily Kandinsky Scuola di Atene(School of Athens), Raffaello
Sanzio
5. Why do we need categories?
World without categories:
What if every entity was unlike any
other?
All representations are new.
World with categories:
Category Representations.
Intellectual functioning is possible.
Making sense of the world through
inter-connected category
representations.
7. Philosophical Background
Wittgenstein
Family Resemblance Theory (1953):
Things which could be thought to be connected
by one essential common feature may in fact be
connected by a series of overlapping similarities,
where no one feature is common to all.
Sluga H., Family Resemlance, Grazer Philosophische Studien 71
(2006)
*Image from Anime Fullmetal Alchemist Brotherhood
8. Philosophical Background
Rosh, Prototype Theory (1978):
Some members of a category are more central
than others.
Furniture? Chair vs Stool
Central members (or their properties determine
the category)
10. Categorization in Infants
Infants prefer novel stimuli
Novel inputs from Novel and Familiar categories
Measured in looking time
Assumption: Infants look at more in novel stimuli
Young infants are capable of representing
variety of complex object categories
Significant to adult humans
Animals, Furnitures
Both general (animal/mammal) and special levels
Leonardo Da Vinci’s flying Machine
11. Information Used to Form Category Representations
How do infants form category representations?
Identify the cues:
Form the category when the cue is present
Not form the category then the information is
not present
12. Experiments
Distinguish cat and dog categories.
(Quinn & Eimas, 1996)
Considerable perceptual overlap
Cats vs Dogs:
Distinguish when full bodies are presented
Distinguish if only head is presented
Can’t distinguish if only body is presented.
Swapped heads and bodies
*Image is from the reference paper Quinn & Eimas, 1996
13. Experiment Result
Head information provides necessary and sufficient information.
Limitations:
Real world movement patterns
Sound information
Cats vs birds/ humans… other cues?
14. Category Formation vs Category Possession
In Infants are category representations:
Online or?
Based on previous experience
Experimental Tasks:
Infants are presumed to construct more CR as more and more examplars from
familiar categories are presented. (Mareschal,French, & Quinn, 2000)
Age: With age, more likely to tap their own knowledge.
Difficult
15. Experiment Mammal vs Furniture
Not likely to observe elephants and
hippopotamuses how do they categorise them?
They rely on solely perceptual processing ?
Picture books, stories
generic furnitures
Unfamiliar animals are more like familiar
animals
Even in such cases infants may rely on pre-
existing information
16. Order of Emergence
Different Levels of inclusiveness /
hierarchy organized systems
Global(Superordinate)level:
Mammals
Intermediate Level: Cat
Specific(Subordinate Level): Siamese Cat
Conventional Wisdom:
Basic level -> Group together
Experimental Evidence:
17. Not Just for Objects
Category representations for more abstract
things
Spatial representations
3 and 10 months
Above/below , left/right
First for specific objects (a under b)
Later more abstract representations
Experience objects in organized spatial
arrangements
18. Implications
Informational enrichment
Develop categories by encountering
Observable static and dynamic attributes can be
detected from exemplars
Young infants correctly parse much of the world
This underlies the process of knowledge acquisition
Mix of an online learning vs previously acquired
knowledge?
19. Later Work (Quinn, et al., 2009)
Head information was necessary and sufficient
Could it arise from preexisting bias mechanism
Attention to head
Or is it because the head is the most diagnostic part.
Head feature flexibly created in online learning
Eye Tracker
Allows micro analyses
Howmuch fixation is on the image
20. Head Bias
If head bias is pre existing
(by their previous
knowledge) It should be
observed in all stages
Also during familiarization
also
If babies are learning online
they would learn to fixate
more on head
This should develop in
familiarization
21. Experiment 1
Randomly into groups
Familiarized with cats or dogs
Shown novel stimuli
Part 1 (Looking Time)
Familiarization, Overall Looking time measured (Habituation)
Preference Test Trials, Novel category preference calculated
from looking time.
Part 2 (Eye fixation)
Familiarization, Head vs body fixation *Image is from the reference paper Quinn, et al., 2009
22. Part 1 (Looking Time)
Familiarization, Overall Looking time measured
No reliable decrement in looking time from
first and second half
They don't consistently habituate when many
different examples are presented
Preference Test Trials, Novel category
preference calculated from looking time.
Novel category score %63.62
reliably above chance p<0.2
Not dependent on weather they were
familiarized with cats or dogs p>0.2
23. Part 2 (Head vs body fixation)
Familiarization, Head vs body fixation
Half of fixations on image
No significant change in 1st and 2nd half
Head looked %45.48 of the time
%17 of surface
Preference Test Trials, Novel category
preference calculated from head vs body
fixations.
Only novel category head was reliably
different from chance
Indicates pre existing biasing mechanism!
24. Experiment 2
Is Head bias because of high contrast of internal
features
If infants are attending because of face bias: They
won’t attend inverted ones
Like Experiment 1 but all images are inverted
25. Part 1 (Looking Time)
Familiarization: Again no difference between
first and second half
Preference Test (Novel category): They attended novel stim.
above chance %59.56, p<0.1. Again not dependent of previous
familiarization p >0.2
They could categorize with inverted stimuli!
Part 2 (Head and Body Fixations)
Familiarization: %40 of fixations were on
stimulus. fixations were not significantly to the
head (to size)
Preference Test (Novel category): Did not significantly look at
head more
The cause is not contrast!
26. General Discussion
Infants use head information to categorize cats vs dogs
Infants use preexisting knowledge
Preexisting does not imply innate, it means previously have the knowledge. (maybe inante)
Possibly Head bias reflects core mechanism that orients infant attention to the
face
But in upright heads
Head bias is not because of its the highest part (other studies)
Bias could broadly assist infants conceptual development
27. Eye tracking
Future work:
Eye fixation with heart rate deceleration (focused attention)
Eye fixation with computational modelling
Eye tracking with event related potentials.
28. References
1. Quinn, Paul C. "Category representation in young infants." Current Directions
in Psychological Science 11.2 (2002): 66-70.
2. Quinn, Paul C., et al. "Time course of visual attention in infant categorization
of cats versus dogs: Evidence for a head bias as revealed through eye
tracking." Child development 80.1 (2009): 151-161.
3. Sluga H., Family Resemlance, Grazer Philosophische Studien 71 (2006)
4. https://en.wikipedia.org/wiki/Prototype_theory
5. Google image search