Text Messaging TiltText: Using Tilt for Text Input to Mobile Phones
• Estimated 500,000,000,000 text messages in 2003 worldwide • More popular outside North America
Daniel Wigdor & Ravin Balakrishnan
2
Ambiguity
Solutions
• Pressing “2” : {2,a,b,c,A,B,C}
• MultiTap • Language -based disambiguation • • •
T9 Letterwise Wordwise
• Alternate Layouts:
3
4
MultiTap: ~2.1 KSPC
T9: ~1.2 KSPC
e.g.: {6,6,6,>,6,6} = “on”
e.g.: {6,6} = “on”, “no”, “mo”,…
5
6
1
T9: Problems
What’s best?
• Ambiguity persists
• Low KSPC
• Inconsistent
• Eyes-free
• Eyes-free operation impossible • Non-language specific
• Only English -Like text • No numerals • Real “texting” impossible (“b4”,”btw”,”lol”,”rotflmao”…) 7
Tilt as input
8
TiltText: 1 KSPC + Tilt Action Q
• Add a tilt sensor to device • •
inexpensive accelerometers Hinckley et al. UIST’00
eg: {7} = …
• Tilt for text input: • •
Sazawal et al. Unigesture MobileHCI ‘02 Partridge et al. TiltType UIST’02
P
R
• No formal evaluations
S 9
10
Tilt Detection: Key Tilt
Tilt Detection: Absolute
• Difference between press & release
• Relative to a fixed origin
• Slow: 3 consecutive actions
• Keypress & tilt actions concurrent
•
keypress, tilt, key- release
• Consecutive same -tilt: savings
• Pilot study: poor performance
• Consecutive opposite -tilt: extra cost • High error-rate: “creeping posture”
11
12
2
Tilt Detection: Relative
Our Prototype
• Most recent tilting gesture
• Uses Absolute tilt • Implemented on Motorola i95 in Java • Tilts from board via serial port
•
floating origin
• Maintains advantages of Absolute tilt • Saves work on consecutive same tilts & consecutive opposite tilts • No “creeping posture”
13
14
The Study
Results: Overall Speed
• Repeated-measures design
•
Overall, TiltText 16% faster (including error correction)
10 participants 2 techniques (MultiTap & TiltText) 16 blocks of 20 phrases each in 2 sessions
14 12
WPM
• • • •
16
Same phrases for both techniques Technique order between participant Measured time & accuracy Participants told to correct mistakes
10 8 TiltText MultiTap
6 4 2 0 1
2
3
4
5
6
7
8
9
10 11 12
13 14 15
16
Block 15
Power-law extrapolation
16
Results: Between Participant • •
16
Data from 1 st technique seen by each participant TiltText still faster 16
0.2134
y = 7.6837x 2
14
R = 0.9263
14
12 0.1184
10
y = 8.0297x
10
WPM
WPM
12 2
R = 0.8963
8
TiltText MultiTap
6
8
TiltText
6
MultiTap
4
4
2
2
0
0 1
3
5
7
9
11
13
15
17
19
21
23
1
25
Block
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
Block 17
18
3
Results: Error Rate
Error Rate: By Letter
•
•
TiltText error rate higher than MultiTap
Error rates much higher for some letters
40
18
Error Rate Percentage
Error Rate Percentage
20
16 14 TiltText
12
MultiTap
10 8 6 4 2
35 30 25 20 15 10 5 0
0 1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
a b c d e f g h i
Block
j k l m n o p q r s t u v w x y z
Correct Letter 19
20
Error Rate: Tilt Direction
Conclusions
• •
• • • •
Direction significantly effects error rate Creeping posture Error Rate Percentage
40 35 30 25
Implemented TiltText Three distinct approaches for tilt Formal study conducted TiltText faster despite errors
20 15 10 5 0 Left
Forward
Right
Back
Correct Tilt Direction
21
22
Future Work
Acknowledgements
• Theoretical TiltText speed
• • • • •
• KSPC is not the whole story
• • • •
Implement relative -tilt system Deeper analysis of error causes Longer study Optimizing letter/key assignments
23
Michael McGuffin Richard Watson DGP Lab members Study participants Microsoft Research
24
4
25
5