Manual Text Entry: Experiments,
Models, and Systems
Poika Isokoski
Tampere Unit for Computer-Human Interaction
Writing
At School
• I did not understand why the inferior
Multiple writing technologies
• Pen and paper and computers are
Structure of the Work
I: Device Independent Text Input: Rationale and an Example
II: Comparison of two Toucpad-Based Methods for Numeric Entry
III: Evaluation of Multi-device Extension of Quikwriting
IV: Performance of Menu-augmented Soft Keyboards
VI: Combined Model for
Text Entry Rate Development
V: Model for Unistroke writing time
Paper IV: Introduction
• Performance of Menu-Augmented Soft
Keyboards
Paper IV: Model
• Simulator:
– Fitts’ law for pointing ( t = a + b log(A/W+1) ).
– Constant 160 ms for selection (McQueen et al., GI 1995).
– Given a text string the simulator computes how long entering it takes.
• Used the 500 phrase set published by
MacKenzie and Soukoreff (CHI2003).
Paper IV: Model: results
• QWERTY: 26% faster
0
Dvorak QWERTY OPTI II ATOMIK FITALY
W
Paper IV: Model: discussion 1
• Why does it seem to be faster?
• Because it saves stylus travel:
– Without menu:
Paper IV: Model: discussion 2
• Is the model realistic and reliable?
– Maybe.
– It could reflect user’s motor performance fairly accurately.
– It does not know anything about cognitive factors.
Paper IV: Experiment I: task
• Measured the motor performance.
– 12 participants entered repeating patterns:
Paper IV: Experiment I: results
• When the patterns are simple (left),
using the menu is faster
• When the patterns are more complex
(right), the difference is smaller
Paper IV: Experiment I: discussion
• Atomic menu selection is faster than
pointing (on average).
• Deciding whether to point or select
takes time.
• It could be that in real writing situation
the cognitive planning dominates
making using the menu slower than not using it.
Another experiment was needed.
Paper IV: Experiment II: task
• Text transcription task:
– The same software
– 6 participants
– 20 sessions (15 min with the menu and 15 min without it per session)
Paper IV: Experiment II: results
• Text entry rate:
Paper IV: Conclusions
• Using the menu may allow faster text
entry rates, but
– only after a lot of training (>>5 hours) – at the cost of added cognitive load
– the increase in text entry rate may not be that great (model says max 26%).
• The menu does not disturb those users
Summary
• The thesis contributes in four ways:
– Device independent text entry
– Data on text entry method performance – Models of user performance
– A software architecture for text entry