User Modeling of Assistive Technology

1
User Modeling of Assistive Technology

• Rich Simpson

2
The Problem

• The most challenging aspect of designing a
  computer access system for a client is predicting
  and accommodating a clients performance in six
  months based on two hours of interaction with
  that client.

3
The Problem

• Clients may only see the clinician once, and that
  visit only lasts for a few hours
• There may be multiple potential solutions
• Each potential solution may have multiple
  configuration options
• The client has little or no experience with
  assistive technology upon which to base decisions

4
The Problem

• Often, the assistive technology thats easiest to
  use at first will be less efficient in the long
  run
• Morse Code vs Row-Column Scanning

5
The Problem

• What we want
• We want to know how well each potential solution
  would work for a client if the client had six
  months to practice
• What we have
• Observations in the clinic
• Assistive Technology Lending Library

6
Keystroke-Level Modeling

• A simple model for the time it takes an expert
  user to perform a task with a given method on an
  interactive computer system.
• Predictive rather than descriptive or explanatory
• Based on intuition rather than observation
• Intended to allow comparisons between two or more
  designs without having to run user trials

7
Keystroke-Level Modeling

• What does expert mean?
• Knows how to do the task
• Doesnt make mistakes
• Consistent time for each action

8
Keystroke-Level Modeling

• Operators
• K - Keystroking
• P - Pointing
• H - Homing
• D - Drawing
• M - Thinking
• R - System Responding

9
Keystroke-Level Modeling

• Keystroking (K)
• Typing speed
• Can range between 0.08 and 1.20 seconds for
  able-bodied adults using a standard keyboard

10
Keystroke-Level Modeling

• Pointing (P)
• Based on Fitts Law

11
Keystroke-Level Modeling

• Mental Operations (M)
• The time to mentally prepare to execute physical
  operators
• In front of the first K of a string
• In front of all Ps that select commands

12
Keystroke-Level Modeling

• An example saving a file
• Move mouse to File menu
• Press mouse button
• Move mouse to Save option
• Press mouse button
• Type in the name of the file
• Press the enter button

13
Keystroke-Level Modeling

• An example saving a file
• Decide what to do (M)
• Move mouse to File menu (P)
• Press mouse button (K)
• Decide what to do (M)
• Move mouse to Save option (P)
• Press mouse button (K)
• Pick a name for the file (M)
• Type in the name of the file (K x length of name)
• Decide what to do (M)
• Press the enter key (K)

14
Keystroke-Level Modeling

• Simplifications
• Fitts Law vs Steering Law
• All movements (P, K) take the same amount of time
• No actions overlap

15
The Problem

• The most challenging aspect of designing a
  computer access system for a client is predicting
  and accommodating a clients performance in six
  months based on two hours of interaction with
  that client.

16
What is Word Prediction?

• Word prediction is used to reduce the number of
  keystrokes required to generate text.
• The computer supplies a list of best guesses
  for the word the user is currently entering, and
  when the word appears it may be selected from the
  list with a single keystroke.

17
What is Word Prediction?

• Word prediction is used to reduce the number of
  keystrokes required to generate text.
• The computer supplies a list of best guesses
  for the word the user is currently entering, and
  when the word appears it may be selected from the
  list with a single keystroke.

18
What is Word Prediction?

• Word prediction is used to reduce the number of
  keystrokes required to generate text.
• The computer supplies a list of best guesses
  for the word the user is currently entering, and
  when the word appears it may be selected from the
  list with a single keystroke.

19
What is Word Prediction?

• Word prediction is used to reduce the number of
  keystrokes required to generate text.
• The computer supplies a list of best guesses
  for the word the user is currently entering, and
  when the word appears it may be selected from the
  list with a single keystroke.

20
What is Word Prediction?

• Word prediction is used to reduce the number of
  keystrokes required to generate text.
• The computer supplies a list of best guesses
  for the word the user is currently entering, and
  when the word appears it may be selected from the
  list with a single keystroke.

21
Why doesnt Word Prediction always increase text
entry rate?

• Word Prediction doesnt necessarily increase the
  speed with which a person can enter text because
  it trades off physical effort for cognitive
  effort.
• The configuration of a word prediction system can
  have a significant effect on a users performance.

22
Configuring Word Prediction

• Show Number of keystrokes entered before list
  appears
• Hide The number of keystrokes entered after list
  appears before it disappears
• Llen Maximum number of words in list
• MWS Minimum number of letters in each word in
  list

23
The Questions

• Will word prediction increase text entry rate for
  a client?
• How should word prediction be configured to
  maximize text entry rate?

24
Koesters Model of Word Prediction

• Search word prediction list
• Decide what key to press
• Press Key
• Repeat

25
Koesters Model of Word Prediction

• Search word prediction list (ts)
• Decide what key to press (d)
• Press Key (tk)
• Repeat

26
Koesters Model of Word Prediction

• Snumber of searches/number of characters
• Knumber of keystrokes/number of characters
• Twp(S)(ts) (K)(tkM)
• So the question is

27
how do these

• Show Number of keystrokes entered before list
  appears
• Hide The number of keystrokes entered after list
  appears before it disappears
• Llen Maximum number of words in list
• MWS Minimum number of letters in each word in
  list

28
influence S, ts, K and tk?

• Number of searches (S)
• When does the list appear? (Show)
• When does the list disappear? (Hide)
• List search time (ts)
• Length of list (Llen)
• Size of words in list (MWS)
• Number of keystrokes (K)
• When does the list appear? (Show)
• When does the list disappear? (Hide)
• Length of list (Llen)
• Size of words in list (MWS)

29
Since you cant set S and K, what good are these
models?
30
Since you cant set S and K, what good are these
models?

• You can measure ts and tk
• Its hard to measure M (which Koester calls d)
• You can simulate user performance over a range of
  values for Show, Hide, Llen and MWS
• The most promising configurations can be compared
  in trials with the client

31
Experimental Validation

• Six subjects with disabilities
• ABA design
• A was a default condition list always
  displayed, six words in list, no minimum number
  of letters
• B was chosen using the model and observations
  during the first A phase
• For three subjects, B was 61 faster than A
• For the other three subjects, B was 20 faster