1 T9 : Problems Whats best? Ambiguity persists Low KSPC - - PDF document

1

TiltText: Using Tilt for Text Input to Mobile Phones

Daniel Wigdor & Ravin Balakrishnan

2

Text Messaging

• Estimated 500,000,000,000 text messages

in 2003 worldwide

• More popular outside North America

3

Ambiguity

• Pressing “2” : {2,a,b,c,A,B,C}

4

Solutions

• MultiTap
• Language-based disambiguation
• T9
• Letterwise
• Wordwise
• Alternate Layouts:

5

MultiTap: ~2.1 KSPC

e.g.: {6,6,6,>,6,6} = “on”

6

T9: ~1.2 KSPC

e.g.: {6,6} = “on”, “no”, “mo”,…

2

7

T9: Problems

• Ambiguity persists
• Inconsistent
• Eyes-free operation impossible
• Only English-Like text
• No numerals
• Real “texting” impossible

(“b4”,”btw”,”lol”,”rotflmao”…)

8

What’s best?

• Low KSPC
• Eyes-free
• Non-language specific

9

Tilt as input

• Add a tilt sensor to device
• inexpensive accelerometers
• Hinckley et al.UIST’00
• Tilt for text input:
• Sazawal et al. Unigesture MobileHCI ‘02
• Partridge et al. TiltTypeUIST’02
• No formal evaluations

10

TiltText: 1 KSPC + Tilt Action

eg: {7} = …

P Q R S

11

Tilt Detection: Key Tilt

• Difference between press & release
• Slow: 3 consecutive actions
• keypress, tilt, key-release
• Pilot study: poor performance

12

Tilt Detection: Absolute

• Relative to a fixed origin
• Keypress & tilt actions concurrent
• Consecutive same-tilt: savings
• Consecutive opposite-tilt: extra cost
• High error-rate: “creeping posture”

3

13

Tilt Detection: Relative

• Most recent tilting gesture
• floating origin
• Maintains advantages of Absolute tilt
• Saves work on consecutive same tilts &

consecutive opposite tilts

• No “creeping posture”

14

Our Prototype

• Uses Absolute tilt
• Implemented on Motorola i95 in Java
• Tilts from board via serial port

15

The Study

• Repeated-measures design

10 participants 2 techniques (MultiTap & TiltText) 16 blocks of 20 phrases each in 2 sessions

• Same phrases for both techniques
• Technique order between participant
• Measured time & accuracy
• Participants told to correct mistakes

16

Results: Overall Speed

• Overall, TiltText 16% faster (including error correction)

2 4 6 8 10 12 14 16 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 TiltText MultiTap

Block WPM

17

Power-law extrapolation

y = 7.6837x

0.2134

R

2 = 0.9263

y = 8.0297x

0.1184

R

2 = 0.8963

2 4 6 8 10 12 14 16 1 3 5 7 9 11 13 15 17 19 21 23 25 TiltText MultiTap

WPM Block

18

Results: Between Participant

• Data from 1 st technique seen by each participant
• TiltText still faster

2 4 6 8 10 12 14 16 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 TiltText MultiTap

Block WPM

4

19

Results: Error Rate

• TiltText error rate higher than

MultiTap

Error Rate Percentage Block

2 4 6 8 10 12 14 16 18 20 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 TiltText MultiTap

20

Error Rate: By Letter

• Error rates much higher for some letters

Correct Letter

5 10 15 20 25 30 35 40 a b c d e f g h i j k l m n o p q r s t u v w x y z

Error Rate Percentage

21

Error Rate: Tilt Direction

• Direction significantly effects error rate
• Creeping posture

5 10 15 20 25 30 35 40 Left Forward Right Back

Error Rate Percentage Correct Tilt Direction

22

Conclusions

• Implemented TiltText
• Three distinct approaches for tilt
• Formal study conducted
• TiltText faster despite errors

23

Future Work

• Theoretical TiltText speed
• KSPC is not the whole story
• Implement relative-tilt system
• Deeper analysis of error causes
• Longer study
• Optimizing letter/key assignments

24

Acknowledgements

• Michael McGuffin
• Richard Watson
• DGP Lab members
• Study participants
• Microsoft Research

5

25