Tilt Text

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