Text correction in pen-based computers: an empirical comparison of methods

Abstract

Three methods for correcting text in pen-based computers were compared in an experiment involving 30 subjects. In spite of simulated virtually perfect character recognition, the two methods involving handwriting proved 25~0 slower than the method involving a “virtual keyboard”. There was essentially no difference between the detectable errors: 2 missing between-word spaces, which were to be inserted using a “space” gesture or a space key (on the virtual keyboard); 2 superfluous within-word spaces, to be removed using a “delete” gesture or key; and 2 incorrect letters which were to be overwritten with the correct letter (with the handwriting methods) or deleted prior to insertion of the correct letter (with the virtual keyboard). Apparatus A Philips Advanced Interactive Display (PAID), was used as the pen-based computer. It had a VGA (640 x 480 pixel) 11“ LCD (backlit) display with a stylus attached to the display by a thin cable. Subjects wrote directly on the screen, and immediate feedback was given of the resulting “electronic ink”. With the two handwriting methods, the sentence was displayed in a window comprising 6 rows of 15 boxes (one for each letter, each box measuring 1 x 1 cm); changes to the sentence were reflected in the same window. A Wizard-of-Oz technique was used to simulate essentially perfect character recognition, so as to eliminate the noise that would be introduced into the data by imperfect automatic character recognition: The experimenter worked at a hidden desktop computer whose screen showed the same display as that of the pen-based computer. With the help of specially written software, the experimenter caused the symbols written by the subject to be handled exactly aa if the computer had (correctly) interpreted them; for example, in the delay condition, the results of the subject’s actions were displayed after each delay of 1.5 sees. With the virtual keyboard method, the screen displayed at the top the sentence to be corrected and at the bottom the virtual keyboard, in which each key measured 0.8 x0.8 cm. Subjects The 30 paid subjects (mean age: 25) had no previous experience with pen-based computing, but all had experience with the use of a keyboard. Design Each subject performed 1 correction task with each of the 3 methods (as well as with 5 other methods not discussed here, of which 4 involved handwriting and 1 involved a different type of virtual keyboard). For each subject, the order of using the different methods was randomized, as was the selection of the sentence to be corrected with each method. Procedure Subjects were first given a general introduction and a practice session lasting 20 minutes to acquaint them with all of the variants used. Then each subject performed, with each method, 3 tasks: 2 (not analysed here) that involved only entering a given sentence, followed by 1 text correction task aa described above. RESULTS The time to execute each text correction task was measured between presentation of the sentence and completion of the last correction. The mean times (and standard errors of the means) for completing a correction task are as follows: handwriting with button: 44.4 +.3 sees; handwriting with delay: 42.1 + 3.0 sees; virtual keyboard: 33.6 + 2.7 sees. The times for the two handwriting methods do not differ significantly according to a paired t-test (t(29) = 0.45, p = .65). But each of these mean times is more than 25% higher than the mean time for the virtual keyboard method, both differences being significant (t(29) = 2.24 and t(29) = 2.59 for the differences with the button and the delay methods, respectively, p < .05). The stability of these results is supported by the appearance of the same pattern in the other conditions of our experiment, which included five further methods and involved text entry tasks sa well as correction tasks: Nowhere was there a reliable difference between methods that differed only in their use of a button vs. a delay; and the virtual keyboard was always substantially faster than handwriting.