Talking to computers

One of the shared assumptions in computer research is that talking to computers is a really great idea. Such a good idea that speech is regarded as the natural interface between human and computer.

Each company with enough money to spare and enough egoism to believe that it can shape everyone's future now has a 'natural language' research group. Films and TV series set in the future use computers with voice interfaces to show how far technology has advanced from our own primitive day and age. The unwritten assumption is that talking to your house will in the end be as natural as shouting at your relatives.

The roots of this shares delusion lie in the genuine naturalness of spoken communication between humans. Meaning is transferred from person to person so effortlessly that it must be the best way of transferring information from a human to another object.

This view is misguided on many different levels. First people are so good at talking and at understanding what others say because they share a common genetic heritage. Children's brains are hard-wired with a general language structure that their surrounding spoken-word environment. The old view that language is learned by copying parents and other adults has been discredited in recent years, to be replaced by the theory that words are attached to a way that grammar 'emerges', as it were, rather than is taught.

This view of human language, added to shared human experience, shows how people understand each other precisely in a conversation where a transcript would make little sense. Unfinished sentences, in-jokes, catchphrases, hesitation markers like 'er' and 'you know', and words whose meaning is only clear in the context of that one conversation are no bar to human understanding, but baffled early attempts at computer speech recognition.

Recent advances in artificial intelligence address the problem but only in part.

Pioneering linguistic research by scientists has revealed much of the underlying structure of human language, that programmers can now mimic that structure in their software and use statistical and other techniques to make up for the lack of shared experience between operator and machine.

Some of the obvious drawbacks of universal voice control have already been countered. The dreadful prospect of an office full of people talking to their machines has brought about the headset and the throat microphone; these also address the fact that people feel ridiculous talking to something which is non-human. The increasing sophistication of voice-processing and linguistic-analysis tools cuts out the dangers of inaccurate responses to input, preventing the computer from having to respond to every single word uttered, no matter how nonsensical in the overall context.

The fundamental objection to natural language interfaces is that they're about as unnatural as you can get. You might be able to order a computer about in its limited sphere of action, but it'll never laugh at your jokes, make sarcastic comments, volunteer irrelevant but interesting information or do any of the other things that make real human conversation so fascinating. If interaction is limited to didactic instruction from human to computer, why use up valuable processing time performing the immensely difficult task of decoding language correctly? To keep your hands free? For what, precisely?

There's another psychological reason why language control is difficult: the decline in domestic service throughout this century, the absence of military experience from the lives of the last two generations, and the flattering out of business management have all combined to produce a population that's not accustomed to giving crisp orders and expecting them to be obeyed.

Controlling a computer by word power works beast if you imitate a drill sergeant, avoiding all 'could you's ' and 'would you mind's' that most of us use when trying to make someone to do something they'd rather not do. This modern variant of the servant problem opens up the chance of ambiguity and error when interacting with a machine.

It could be said, though, it's just as well as we've forgotten how to give orders. Slaves always have had a" reputation for conspiring against their master's backs.

Дата добавления: 2015-11-16