SPEECH VERSUS KEYING IN COMMAND AND CONTROL - EFFECT OF CONCURRENT TASKING

As a result of Poock's influential work in the early 1980s, command an d control is generally believed to be one specific application where s peech input holds great advantages over keyed data entry. However, a r ecent paper (Damper & Wood, 1995 ''Speech versus keying in command and control applications'', International Journal of Human-Computer Studi es, 42, 289-305) has questioned this interpretation of Poock's data be cause the experimental conditions seemed to bias the results against k eyed entry. While Damper and Wood modelled their experiments on Poock' s, however, there were important differences which mean that their con clusions are uncertain. The objective of the work reported here was to determine if the major difference - the omission of concurrent, secon dary tasking from their study - could explain Damper and Wood's observ ed superiority of keying over speech. Simulated command and control ex periments are described in which speech input, abbreviated command key ing and full command keying are compared under dual-task conditions. W e find that speech input is no faster (a nonsignificant 1.23% differen ce) and enormously more error-prone (1038%, highly significant) than a bbreviated keying for the primary data entry task, but allows somewhat more (11.32%, not significant) of a secondary information-transcripti on task to be completed. Full keying has no advantages whatsoever: we believe that this confirms the methodological flaw in Poock's work. If recognizer errors (as opposed to speaker errors) are discounted, howe ver, speech shows a clear superiority over keying. This indicates that speech input has potential for the future - especially for high workl oad situations involving concurrent tasks - if the technology can be d eveloped to the point where most errors are attributable to the speake r rather than to the recognizer. (C) 1996 Academic Press Limited