Editing by voice and the role of sequential symbol systems for improved human-to-computer information rates

Composing text on the computer is usually heavily dependent on editing, such as moving text around, correcting spacing, and inserting punctuation characters. Dictation systems, based on automatic speech recognition, are not known for their efficiency as an editing tool - something that significantly reduces their potential for freeing the user from the keyboard. Speech recognition has almost invariably been tied to natural language, but we point out that this approach is inherently disadvantageous in important ways. Instead, given the evidence that humans routinely become experts at sequencing signs not related to natural language, we propose that editing command languages should rely on symbolizations similar to that of the keyboard. We introduce ShortTalk, an editing language whose command sequences are primitive symbol combinations that are not confusable with dictation. We argue that ShortTalk by construction may solve common editing situations much more efficiently than by use of keyboard and mouse. Our experimental results for ShortTalk indicate that an average information rate of about 16 bit/s for editing commands is achievable. Thus, editing by speech may be more efficient than by non-verbal means.