Petre: Why Looking Isnt Always Seeing

1
Petre Why Looking Isnt Always Seeing

• CptS 538Spring, 2006
• Christopher Hundhausen

2
Who is Marian Petre?

• Professor of Computing at the Open University in
  the UK (the Open University does distance
  education right)
• Did Ph.D. work with Thomas Green
• Interested in expertise (what makes expert
  programmers expert?) and computer science
  education
• Has performed numerous empirical studies of
  expert programming, and is particularly
  interested in the role of mental imagery and
  representations in programming and
  problem-solving
• Nice website can be found at http//mcs.open.ac.uk
  /mp8/Index.htm

3
In a Nutshell

• Graphics are generally regarded to be superior to
  text
• However, in the realm of programming, the program
  does not view a program representation as though
  it were art purposeful perusal is necessary.
• Does the given picture convey the same thousand
  words to all viewers?
• Petres bottom-line As we have learned (in this
  class), the goodness of graphical representation
  depends on the extent to which it facilitates
  efficient (programming) task performancedoes it
  make accessible the right information for the
  task?

4
Petres Key Arguments

• Graphical representations are comprehensible
  because of secondary notation
• Graphical readership is an acquired skill
• Experts read and write diagrams differently from
  novices
• No representation is universally superior we
  need to identify appropriate criteria for
  choosing representational horses for cognitive
  courses.

5
Secondary Notation

• Use of layout and perceptual cues to clarify what
  things mean and to give hints to the reader
• May well distinguish good graphics from bad
  graphics
• Catch-22 Secondary notation is NOT part of the
  formal system!
• Ability to read/write secondary notation
  distinguishes experts and novices

6
Graphics Significantly Slower Than Text

• Effect was uniform across all participants!
• Results were replicated in a study of Petri net
  representations
• Results were task specific (Who would have
  predicted that?)
• Results seemed to be dependent on secondary
  notation issues (e.g., layout)

7
Expert vs. Novice Reading Strategies

• Novices
• Some were rigidadhered to same strategy,
  regardless of task
• Others were chaoticabruptly changed strategy
  midtask
• Had difficulty determining what was relevant
• Experts
• More consistent as a group
• Strategy employed matched task
• Text used to guide reading of graphics
• Able to identify relevant entities and prune
  search space
• Used fingers as markers

8
Expert Writing Strategies

• Conveying expertise through a graphical
  representation does not equate to a mere capture
  ("memory dump") of the expert's knowledge (What
  would Roschelle call this?)
• Rather, an expert diagram skillfully highlights
  some information at the expense of other
  information
• Design conventions support fruitful use of
  secondary notation, but are rarely written down
  because they are hard to specify "in the general
  case." This is why it's hard to automate the
  generation of good graphics

9
Experts Can Distinguish Expert Diagrams from
Novice Diagrams

• Experts apply and break rules in a systematic
  way, allowing them to recognize "an underlying
  system of usage" and ultimately distinguish
  expert from novice diagrams
• Experts may identify expert diagrams as such, but
  may nonetheless rearrange them in order to better
  reflect their priorities

10
Despite Results, Graphics Are Appealing!

• Several (fallacious) characteristics make them
  appealing
• Richness
• Promotion of Gestalt
• "Closeness of mapping"
• Easier to access and comprehend
• More fun (but could be an illusion!)

11
Summary

• Graphical readership is a learned skill
• Knowing where to look and what to look for
• Eye movements and scanning behavior change
• Implications for programming language design
• Novices need more constrained languages that
  minimize possibilities for secondary
  notationtextual
• Graphical languages provide better
  expressiveness, but more trainingmore suitable
  for experts
• Yet the author acknowledges the possibility that
  graphical languages, if well designed, could
  benefit novices

12
Petre and Green Learning to Read Graphics

• CptS 538Spring, 2006
• Christopher Hundhausen

13
Who is Thomas R. Green?

• Eminent cognitive psychologist from the U.K.
• Generally interested in models and frameworks to
  help us better understand and analyze information
  artifacts such as programming languages and user
  interfaces
• Well-known examples of his work include task
  action grammars and cognitive dimensions
  framework, which has been VERY influential within
  the visual languages community

14
How is This Article Related to the Petre Article?

• Check out reference 19 on p. 37 of the Petre
  article. What does it refer to?
• That's right This article, which presents an in
  depth analysis of novice and expert graphics
  reading strategies
• Frankly, there's not too much above and beyond
  the Petre article, which nicely synopsizes this
  work.

15
Programmer as Active Reader

• "We propose a model of the programmer as an
  active reader one who uses a graphical system
  with specific tasks or goals in mind" (p. 56)
• Is this what Larkin, Simon, and Casner have in
  mind when they emphasize the task specificity of
  graphics?

16
The Match-Mismatch Hypo

• "No particular notation is universally best
  rather, each notational structure makes some
  kinds of information accessible at the expense of
  other types" (p. 58)
• ? "Performance is best when the form of the
  information required by the question matches the
  form of the representation" (p. 58)

17
The Experiment

• Tested specific match-mismatch hypo
• Nested conditional notation supports "forwards"
  questions (Provide output, given input)
• Declarative notation supports "backwards"
  questions (Provide input, given output)
• 2 x 2 x 2 mixed-factor design crossed three
  independent variables
• notation type (sequential, circumstantial)
• notation medium (text, graphical)
• Subject type (novice (n-5), expert (n6))

18
Statistical Results

• Notation type had significant effect
• Notation medium had stronger signficant effect
• Subject type did not have an effect (!)
• Effect of notation medium existed across subject
  type
• Details presented elsewhere (Darn!)

19
Follow-up Analysis of Strategies

• Experimenter observed participants and took notes
  as the participants worked through tasks,
  thinking aloud. (This isn't satisfying to me. How
  about you?)
• Petre well summarized the results in the first
  article
• Expert strategies were more uniform and
  consistent
• Novice strategies were more chaotic and rigid
• This article offers a few new observations
• Experts used text to guide their reading of
  graphics, unlike novices.
• Experts matched inspection strategy to the
  question
• Novices and experts concentrated on different
  graphical details. Experts took advantage of
  secondary notation. Novices gravitated toward
  what was most visible.

20
Discussion Questions

• 1. According to the article, what are the key
  ways in which experts and novices differ with
  respect to their strategies for reading graphics?
• 2. This paper focuses on the role of 'secondary
  notation' in graphics comprehension tasks. What
  is the relationship between 'secondary notation'
  and 'semantic primitives', as defined in the
  paper we read last week?
• 3. Earlier this semester, we read articles by
  Larkin, Simon, and Casner that claimed that good
  graphics enable the viewer to substitute
  perceptual operations for more costly logical
  operations. Does this paper's take on 'good
  graphics' support or refute this view?
• 4. The tasks used in Petre's empirical studies
  required participants to answer questions
  however, they did not explicitly test the manner
  in which participants interpreted each component
  of the graphics they used. In your view, does the
  semantic mapping technique discussed last week
  provide a superior means of testing readership
  skills?
• 5. What is the danger in assuming that a
  graphical representation will be superior because
  it can provide a closer mapping to the underlying
  domain that it is representing?

21
Discussion Questions (cont.)

• 6. Recall that 'graphical constraining' was a
  central advantage that Scaife and Rogers claimed
  for external representations. Likewise, Petre
  uses the word 'constraint' in her discussion of
  'secondary notation'. Is she talking about the
  same thing? Why or why not?
• 7. Petre claims that "formatting a graphical
  representation into something comprehensible is
  not currently reliably possible by algorithm." In
  light of your reading of the Casner article, do
  you agree with her?
• 8. In both articles, the authors claim that,
  while there are rules of thumb by which experts
  abide, "heuristics that resolve the conflicts
  within a particular context are buried deep in
  experience and expertise and are not easily
  externalized" (p. 69). Do you believe that such
  heuristics could be written down in a manual, so
  that a novice could read the manual and become an
  expert?

22
Appendix Notations