Talking to a machine
© Copyright 1994-2002, Rishab Aiyer Ghosh. All rights reserved.
Electric Dreams #5
21/March/1994

Intro: Though technology advances, we still cannot communicate easily with computers.

You might have noticed that people very familiar with computers tend to personify them. My computer doesn't want to print this proposal. My modem feels neglected, it won't start up. This sort of thing is quite common, and if you think of it, quite natural. After all, we communicate best with our human friends and coworkers; as computers begin to play a major role, at least as our coworkers, if not friends, we try to communicate with them as if they were human.

How would we like to talk to our electronic colleagues? Currently, most of our communication is very formal. We write a letter to the PC when we want anything done. A short letter perhaps, but then, you wouldn't find it very convenient communicating with all your friends in short notes. Of course, the now ubiquitious graphical user interface, or GUI, that first spread through the Apple Macintosh, does reduce somewhat our dependence on words for communication. Instead, you use a mouse to point at icons that are supposed to remind you of what they mean. Often you use organized hierarchical menus instead of graphic icons. However, while GUIs may be a little less intimidating than the one-line prompt of command-line interfaces, they are not particularly friendly. You still have to remember that, for no particular reason, to print a letter you need to go to the File menu. To check the spelling in a proposal you have to figure out whether you want a Tool or a Utility. Most of all, you have to drastically change the way you behave to use a computer, while it should be the computer that adapts to your method of working. What is needed is an interface between us and computers that justifies our frequent personification of them. A natural interface that lets one concentrate on the message being communicated, rather than the elaborate procedures surrounding it.

Humans talk a lot. Our favourite means of communication is speech. It is convenient and informal. To those one talks to frequently, it is usually clear enough -- we don't have to communicate in short notes all the time. It is also easier than typing, or playing with GUIs and mice. We will soon communicate with our personal computers and digital assistants as much, if not more, than with other humans. (This is of course in terms of information transmitted for the purpose of getting something done; not everyone will prefer electronic to human company!) It would be nice if computers could understand our conversations in context just as our friends do. And if we can be informal with our computers, we can definitely talk to them.

Understanding messages in context is the least of the problems in electronic speech recognition. For a long time, personal computers have not been powerful enough to perform the complex operations involved in separating the sound components of speech, far less identify and understand words and phrases. (Incidentally, these components are separated by hair follicles in the human ear!) Now technology is advanced enough to provide some speech recognition even on palmtop computers; it's just the inertia of interface designers that prevents it from appearing in wide use.

Many technical problems still remain. Most systems require training for a particular individual's speech (which is not necessarily a bad thing -- your computer should recognize your voice), and find spoken sentences, without uniform pauses between words, difficult. But the first GUI systems were not perfect either, and some desktop computers have finally appeared with a speech interface. Not surprisingly, these are also from Apple.

It may be a long time before I can scold my recalcitrant digital secretary, but it won't be long before I can tell it to print that proposal. Of course, it might still not want to!




  • Electric Dreams Index
  • dxm.org Homepage