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

Text Messaging TiltText : Using Tilt for Text Input • Estimated 500,000,000,000 text messages in 2003 worldwide to Mobile Phones • 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 8 Tilt as input TiltText : 1 KSPC + Tilt Action Q • Add a tilt sensor to device • inexpensive accelerometers eg: {7} = … • Hinckley et al.UIST’00 • Tilt for text input: • Sazawal et al. Unigesture MobileHCI ‘02 • Partridge et al. TiltTypeUIST’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 • floating origin • Implemented on Motorola i95 in Java • Tilts from board via serial port • 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 16 blocks of 20 phrases each 14 in 2 sessions 12 10 WPM • Same phrases for both techniques 8 TiltText 6 • Technique order between participant MultiTap 4 • Measured time & accuracy 2 0 • Participants told to correct mistakes 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 Block 15 16 Power-law extrapolation Results: Between Participant Data from 1 st technique seen by each participant • • TiltText still faster 16 16 0.2134 y = 7.6837x 2 = 0.9263 14 14 R 12 12 0.1184 y = 8.0297x 10 10 WPM WPM 2 = 0.8963 R 8 8 TiltText TiltText 6 MultiTap 6 MultiTap 4 4 2 2 0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 1 3 5 7 9 11 13 15 17 19 21 23 25 Block Block 17 18 3

Results: Error Rate Error Rate: By Letter • TiltText error rate higher than MultiTap • Error rates much higher for some letters 20 40 Error Rate Percentage 18 Error Rate Percentage 35 16 30 14 TiltText 25 12 MultiTap 10 20 8 15 6 10 4 5 2 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 j k l m n o p q r s t u v w x y z Block Correct Letter 19 20 Error Rate: Tilt Direction Conclusions • Direction significantly effects error rate • Implemented TiltText • Creeping posture • Three distinct approaches for tilt • Formal study conducted 40 Error Rate Percentage 35 • TiltText faster despite errors 30 25 20 15 10 5 0 Left Forward Right Back Correct Tilt Direction 21 22 Future Work Acknowledgements • Theoretical TiltText speed • Michael McGuffin • Richard Watson • KSPC is not the whole story • DGP Lab members • Study participants • Implement relative-tilt system • Microsoft Research • Deeper analysis of error causes • Longer study • Optimizing letter/key assignments 23 24 4

25 5