TiltText University of TorontoTiltText 多伦多大学.pptVIP

TiltText: Using Tilt for Text Input to Mobile Phones TiltText: Using Tilt for Text Input to Mobile Phones Daniel Wigdor Ravin Balakrishnan Text Messaging Estimated 500,000,000,000 text messages in 2003 worldwide More popular outside North America Ambiguity Pressing “2” : {2,a,b,c,A,B,C} Solutions MultiTap Language-based disambiguation T9 Letterwise Wordwise Alternate Layouts: MultiTap: ~2.1 KSPC e.g.: {6,6,6,,6,6} = “on” T9: ~1.2 KSPC e.g.: {6,6} = “on”, “no”, “mo”,… T9: Problems Ambiguity persists Inconsistent Eyes-free operation impossible Only English-Like text No numerals Real “texting” impossible(“b4”,”btw”,”lol”,”rotflmao”…) What’s best? Low KSPC Eyes-free Non-language specific 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. TiltType UIST’02 No formal evaluations TiltText: 1 KSPC + Tilt Action Tilt Detection: Key Tilt Difference between press release Slow: 3 consecutive actions keypress, tilt, key-release Pilot study: poor performance 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” 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” Our Prototype Uses Absolute tilt Tilts from board via serial port 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 Results: Overall Speed Overall, TiltText 16% faster (including error correction) Power-law extrapolation Results: Between Participant Data from 1st technique seen by each partic