Pattern Recognition

1
Pattern Recognition

• Langston, PSY 4040
• Cognitive Psychology
• Notes 2

2
What is this?
3
What is this?
4
What is this?
5
Whats Out There?

• Were going to explore how you know what it is
  that youre seeing, hearing, etc. Three steps
• Input Get it into the system.
• Identification What is it?
• Recognition What does it mean?

6
Architecture

• Recall our three boxes

7
Architecture

• Recall our three boxes

8
Input (Sensory Store)

• Generic questions
• Capacity?
• Duration?
• Code?
• Forgetting?

9
Sensory Store

• Present some information, get you to report it
  back
• G T F B
• Q Z C R
• K P S N

10
Sensory Store

• How do you report?
• Whole report Tell me everything you saw.
• Partial report Tell me some of it.
• Why make a distinction? If you have a limited
  duration (and it makes sense for this to be
  short), it will fade before you can report it.
• Look at the whole report data

11
Whole report Presentation for 50 ms, 5
observers, more information doesnt lead to more
being reported, the amount stays around 4.5 items.
Sperling (1960, p. 5)
12
Sensory Store

• Sperling (1960)
• Partial report (tone tells you which row, high
  middle, or low).

Sperling (1960, p. 3)
13
Partial report Presentation for 50 ms, 5
observers, they report about 75 of the available
information.
Sperling (1960, p. 5)
14
Sensory Store

• Sperling (1960)
• The duration of the store is less than one
  second.
• After the letters go away, present a tone. Vary
  it from -0.10 seconds to 1.0 second. Return to
  whole report level around 0.25 seconds.

Sperling (1960, p. 9)
15
Sensory Store

• Sperling (1960)
• The code is relatively unprocessed information.
• Forgetting is some form of decay.
• Compare Sperlings (1960) results to your CogLab
  data

16
Sensory Store

• So what? Reading.
• Eyetracking studies (Rayner Sereno, 1994)
• Three regions
• Foveal High resolution, about 8 letters.
• Parafoveal Less acuity, some information
  available, about 12 (more?) letters.
• Peripheral Very low acuity, only gross
  information (e.g., ends of lines).

17
Sensory Store

• So what? Reading.
• Three parts
• Fixations Gather information. Average 200-250
  ms, range from 100-500 ms (partly a function of
  reader and partly a function of processing).
• Saccades Move to new fixation, about 8
  characters, range from 1 to 15.
• Regressions Saccade to earlier part of the text,
  10-15 of saccades.

18
Architecture

• Recall our three boxes

Identification
19
Identification

• Bottom-up vs. top-down

Knowledge, etc.
Representation
Input
20
Three Models

• Template model
• Identify by comparing to a copy of everything
  youve seen.
• Possible support
• Instance theories.
• Perceptual priming.
• Problems
• Too much variability.
• Problems in matching.
• What defines a match?
• Multiple interpretations of the same stimulus.

21
Three Models

• Rotated word

22
Three Models

• E.g.,

23
Three Models

• Feature model
• Identify by breaking into features and analyzing
  those.
• Good feature sets have these properties
• Critical (help you to tell things apart).
• Same with changes in the input environment.
• Unique pattern for every input.
• Reasonably small number of features.

24
Three Models

• Feature model
• Good things about feature models
• Fit nicely with information theory With N
  features, you can classify 2N things.
• 1 feature -gt 2 things.
• 2 features -gt 4 things.
• 8 features -gt 256.
• 20 features -gt 1,048,576.
• Work well with computers.

25
Three Models

• Feature model
• Support
• Confusion matrices.
• Cluster analyses (projector).
• Face recognition.
• Brain organization.

26
Three Models

• Feature model
• Problems
• Hard to get the right set.
• Cant ignore how features combine.

27
Three Models

• Structure model
• Identify by grouping
• Gestalt idea The whole is greater than the sum
  of its parts. Grouping principles
• Proximity Close part of same group.
• Similarity Similar part of same group.
• Continuity Group into continuous forms.

28
Three Models

• Structure model
• Grouping principles
• Closure Prefer closed figures.
• Connectedness Connected part of same group.

29
Three Models

• Lanthier, Risko, Stolz, Besner (2009
  doi10.3758/PBR.16.1.167)
• In addition to features, information about how
  features combine is also important.
• Delete information from midsegments
• Or vertices

30
Three Models

• Lanthier et al. (2009) The kind of information
  deleted mattered

31
Three Models

• Lets try a mini-experiment based on Biederman
  (1987).
• Break into groups A and B.
• You will write down the name of the object that
  you see.

32

• Group A get ready

33

• 1.

34

• 2.

35

• 3.

36

• 4.

37

• 5.

38

• Group B get ready

39

• 1.

40

• 2.

41

• 3.

42

• 4.

43

• 5.

44

• Answers

45

• 1.

46

• 2.

47

• 3.

48

• 4.

49

• 5.

50

• Recoverable stimuli
• The contours have been deleted in regions where
  they can be replaced through collinearity or
  smooth curvature (p. 135).

51

• Non-recoverable stimuli
• The contours have been deleted at regions of
  concavity so that collinearity or smooth
  curvature of the segments bridges the concavity.
  In addition, vertices have been altered, for
  example, from Ys to Ls, and misleading symmetry
  and parallelism have been introduced (p. 135).

52
(No Transcript)
53
(No Transcript)
54
(No Transcript)
55
(No Transcript)
56
(No Transcript)
57
Three Models

• How can gestalt features help explain this?

58
Recognition/Meaning

• Meaning and identification appear to be separate.
• What well do is look at language recognition
  (reading and listening) at several levels
• Letters
• Orthography
• Word superiority
• Speech sounds

59
Recognition/Meaning

• Letters A number of features influence letter
  identification
• Serif vs. sans-serif (f vs. f)
• Weight difference (e vs. e)
• Bias
• X-height
• Spacing (proportional vs. non-proportional)
• Proportions

60
Recognition/Meaning

• Features can influence identification, as in this
  (admittedly not ideal) example

61
Recognition/Meaning

• Find the x
• N N Z N Z N Z N Z
• Z N Z Z N Z Z N N
• N N N Z N X N Z N
• N N Z N Z N Z N Z
• Z N Z Z N Z Z N N

62
Recognition/Meaning

• Find the x
• O O P O P O P O P
• P O P P O P P P O
• O O P P O X P O P
• O O P O P O P O P
• P O P P O P P P O

63
Recognition/Meaning

• Word recognition Additional features
• Word envelope
• Orthography Rules for combining letters
• Avoid doubling letters
• To pronounce
• V C V
• V C C V
• V C

64
Recognition/Meaning

• Word recognition
• Orthography
• To correct a V C pattern, add a dummy e
• Fin, fine, can, cane
• To correct a V C V pattern, double
• Runing, running
• Orthography can help with
• Pronunciation mab, mabing, mabe, mabbing
• Letter expectations

65
Recognition/Meaning

• An example of combining the additional features
  on word recognition Word superiority effect.
• Which should be identified better
• d
• word
• Look at CogLab data for word superiority

66
Recognition/Meaning

• Lanthier et al. (2009) When they put degraded
  letters into words

67
Recognition/Meaning

• A similar finding Huey (1908) finds that words
  can be perceived at distances that are too far
  for the letters within the words to be perceived.
• Note the paradox How do you see the word without
  seeing the letters first?

68
Recognition/Meaning

• Interactive activation A possible explanation
  (McClelland Rumelhart, 1981).

69
Recognition/Meaning
70
Recognition/Meaning
71
Recognition/Meaning
72
Recognition/Meaning
73
Recognition/Meaning

• Putting it together
• Aoccdrnig to a rscheearch at an Elingsh
  uinervtisy, it deosn't mttaer in waht oredr the
  ltteers in a wrod are, the olny iprmoetnt tihng
  is taht frist and lsat ltteer is at the rghit
  pclae. The rset can be a toatl mses and you can
  sitll raed it wouthit porbelm. Tihs is bcuseae we
  do not raed ervey lteter by it slef but the wrod
  as a wlohe. ceehiro. (c.f., http//www.snopes.com/
  language/apocryph/cambridge.asp)

74
Recognition/Meaning

• OK. Lets try unfamiliar text
• Tihs was onriglaily caeertd a lnog tmie ago in
  rposnese to an off-tpioc cmnoemt in a Sosaldht
  aciltre taht I cnat fnid any mroe. Snice tehn,
  its been pssaed aunord a lot. I hvae no ieda how
  or why, but mnay tosdnauhs of polpee cmoe hree
  ecah week to sblarmce wodrs.

75
Recognition/Meaning

• First and last letters
• Gciarcdno ot esreacrh ta na Eilsngh evrsyuinti,
  ti eodsnt rettma ni hwta roedr teh relstte ni a
  rodw rae, eth nyol rpatotinm hitng si atht rftis
  nad satl tetrel si ta teh rihgt ecalp. Eht rets
  anc eb a toalt smse nda yuo nca lltis arde ti
  houtitw opmlrbe.

76
Recognition/Meaning

• And the unfamiliar text
• Isht aws inialrylog redtcea a lnog miet goa ni
  opersesn ot na ffo-tiocp nmmotce ni a Lshsatod
  tacrlei ttah I ctan fidn yan meor. Ncise nthe,
  tsi eenb daspes dornau a lto. I haev on edia ohw
  ro wyh, ubt nyma dostansuh fo eeplop coem heer
  hcae ekwe ot cresmbla rwods.

77
Recognition/Meaning

• Clearly, the first letters claim has some
  validity. I would suggest that it gives two
  sources of constraint that still makes reading
  relatively easy
• Some of the letter order information is preserved
  in every word (word envelope).
• Words of three letters or less are intact. Given
  the role of function words, thats a big deal.

78
Recognition/Meaning

• Relate this to the interactive activation model.
  How might it account for your being able to read
  scrambled words? Or, why might it have a problem
  accounting for this?

79
Recognition/Meaning

• What about these?
• Tihs txet uess olny wdors taht are shrot.
• Hilpapy aoutrhs sitll issnit on fwinollog
  dullfready oslotebe snellpig.
• Psleae ntoe taht any uaeicipnntatd hmras to
  paiitprtancs or asrvede eetvns must be reroeptd
  to the oficfe of cnpioalmce.
• Rsaerech tseihs taht cltionlecg cmlteope reday
  yuor hvae you sbuimt mnaes fisihend you dtaa are
  and to.

80
Recognition/Meaning

• What about these?
• sbalermcd
• saremlbcd
•  
• unirevisty
• ustveniriy
• ueiinrstvy

81
Recognition/Meaning

• What about these?
• This list of words painters, pertains, pantries,
  in loco parentis.
• a. anhtrsicit, piacvrote, lsngoievns, chlrucanbe

82
Recognition/Meaning

• What about these?
• This list of words painters, pertains, pantries,
  in loco parentis.
• a. anhtrsicit, piacvrote, lsngoievns, chlrucanbe
• b. antichisrt, proaictve, lonivgness, ccrnuhable

83
Recognition/Meaning

• What about this?
• 4)V4\C3D l3epeAlt i whEn J00 4lK L1K3
  t-15. t0 u\d3r_at_\D jOo \/u5 be lEET. 1f
  J00 4r3 NO lEe jOO C_at_\N0T 5p3Alt 0r ReAd
  -I5.

84
Recognition/Meaning

• What about speech?
• Phone Sound you can make (4,096)
• 896 used
• about 100 account for almost all languages
• Phoneme Group of phones treated as one sound by
  a language
• Morpheme Smallest unit that conveys meaning
• Listeme An entry in your mental dictionary

85
Recognition/Meaning

• What about speech?
• Features matter. For example

86
Recognition/Meaning

• What about speech?
• Two ways to describe speech
• Articulatory phonetics Based on how you make a
  sound. A speech sound is
• Air voicing manner place
• http//www.sil.org/mexico/ling/glosario/E005ci-Pla
  cesArt.htm
• http//www.ic.arizona.edu/lsp/Phonetics/Consonant
  sI/Phonetics2d.html

87
Recognition/Meaning

• What about speech?
• Two ways to describe speech
• Acoustic phonetics Based on the actual, physical
  sound wave that is produced.

88
Recognition/Meaning

• What about speech?
• Acoustic phonetics uses the spectrogram.

89
Recognition/Meaning

• What about speech? A couple of issues to provide
  an example
• Parallel transmission

90
Recognition/Meaning

• What about speech? A couple of issues to provide
  an example
• Context conditioned variation

91
Recognition/Meaning

• What about speech? An example of a top-down
  influence on speech perception
• McGurk effect http//psiexp.ss.uci.edu/research/t
  eachingP140C/demos/McGurk_large.mov

92
Recognition/Meaning

• Other context effects in recognition Palmer
  (1975) Objects are recognized in an appropriate
  scene more than in an inappropriate context.

93
Meaning

• A word means
• What it refers to (what it stands for)?
• The image it calls up?
• What else?

94
Meaning

• A word means A two-part model
• Propositions Ideas
• Models derived from them
• E.g., the star is to the left of the circle

95
Grounding

• Chinese-room problem How are symbols grounded?
• Embodied cognition?
• Other ideas?

96
Examples
97
Examples
98
Examples
99
Examples
100
End of Pattern Recognition Show