Xiaojun Bi, Shumin Zhai
2016 Proceedings of the 2016 CHI Conference on Human Factors in Computing Systems - CHI '16  
Figure 1. Keyboard Layouts Optimized for Gesture Typing and Learnability. (a) IJQwerty: one key difference from Qwerty (I and J are swapped). (b) QwertyH1: a key is at most horizontally one key away from its Qwerty position. Compared to Qwerty, 8 pairs of keys get swapped: {E, R}, {Y, U}, {P, O}, {A, S}, {D, F}, {G, H}, {K, L}, and {B, N}. (c) Qwerty1: a key is at most one key away from its Qwerty position in both horizontal and/or vertical directions. Changed keys are classified into 6 sets:
more » ... fied into 6 sets: {Q, and {M, L}. Within each set, a key is moved one of other keys' Qwerty position. Note that colored rectangles are for illustration only. They do not appear on the actual keyboards. ABSTRACT Despite of a significant body of research in optimizing the virtual keyboard layout, none of them has gained large adoption, primarily due to the steep learning curve. To address this learning problem, we introduced three types of Qwerty constraints, Qwerty1, QwertyH1, and One-Swap bounds in layout optimization, and investigated their effects on layout learnability and performance. This bounded optimization process leads to IJQwerty, which has only one pair of keys different from Qwerty. Our theoretical analysis and user study show that IJQwerty improves the accuracy and input speed of gesture typing over Qwerty once a user reaches the expert mode. IJQwerty is also extremely easy to learn. The initial upon-use text entry speed is the same with Qwerty. Given the high performance and learnability, such a layout will more likely gain large adoption than any of previously obtained layouts. Our research also shows the disparity from Qwerty substantially affects layout learning. To minimize the learning effort, a new layout needs to hold a strong resemblance to Qwerty.
doi:10.1145/2858036.2858421 dblp:conf/chi/BiZ16 fatcat:zorgsvzchzfgrg6poi5wrltrfy