Fog Creek Software
Discussion Board

Voice Recognition = Command Line Prompt?

Technologists are always pontificating on Voice Recognition.  "If only I could tell my computer to turn on the lights".  I think it's overhyped.

VR is a classic example of misidentifying the problem.

The real difficulty in the above example is the interface and connecting your computer to the lights.  If you just replace the VR with a command line, the problem is still difficult.  How do we know WHICH lights to turn on?  Can the user manually overide the computer? How do we give the user feedback if there's an error?  etc., etc.  How do we connect our computer to a high voltage switch. (Yes, I know X10 will do it, but it takes a bit of work.)


Isn't a VR system just a command line interface for people who can't type?

Doesn't it have all the problems that people complained about for 20 years?  ("I can't remember the command word"). PLUS the added difficulty that it might have trouble recognizing your cousin from West Virginia because of how she says "Reyyaad" instead of Red.

Tuesday, December 30, 2003

Hey entrepreneur, careful there... some of us dun lived up 'n them thar hills fer a spell.


Tuesday, December 30, 2003

Haven't we all figured out by now that voice recognition is lame? Don't we all hate those IVR's that make you speak numbers instead of press them? Imagine doing that for your microwave.

What reputable "technologists" have recently expounded on the how great voice recognition is?

For now all the voice recognition I want is the hands free dialing on my cell phone. Even then I only use it when my hands really aren't free.

Tuesday, December 30, 2003

I would love a Star Trek world where I could say "Computer:  Reduce lighting 10%" and it would.

Capn' Kirk
Tuesday, December 30, 2003

The bottleneck with Voice Recognition isn't so much in the technology of understanding what you said.  The bottleneck is in the discourse model.

The trouble is not that computers don't recognize the words you spoke. After all, we can all type, and we don't have computers that do what you want just from typing English instructions to them either.

The reason we want to talk to our computers is not because it's somehow easier than clicking an icon or typing words (even command-line words).

It's because talking is more free-form, more flexible, more interactive. Language can, if both parties understand the topic in the same way, allow much a easier, more fluid, and more natural interface. It can be precise where necessary, and vague where it's clear we share assumptions.

I would say that it's not speaking out loud that's natural (à la voice control interface); it's speaking in a proper conversation that we crave.

Simple voice-command applications have their uses today (like command-and-control for folks with disabilities, or simple voice commands for mobile phones), but they're not going to feel like 'talking to my computer' because it is, exactly like Entrepreneur says, a command-line interface.

Tuesday, December 30, 2003

Heh, reminds me of that old "Voice Recognition Software" joke:

Tuesday, December 30, 2003


Yes, I'm reminded now of something (from How the Mind Works by Pinker, I think):

When people say "speech recognition" they really mean "speech *comprehension*". 

Unfortunately, the layman doesn't recognize that subtle distinction. The reality is that most people have NO CLUE how complex language is.

In fact, I've been discussing, with my wife ( a speech therapist) the theory that you can not LEARN language. You can only be TAUGHT [your first] langauge. I.e., unless you're getting feedback as you learn ("correct, incorrect, try again", etc.) you'll never learn that first language. It's a trial and error (iterative) process.

Tuesday, December 30, 2003

Noam Chomsky would disagree.

Tuesday, December 30, 2003

Dear Entrepreneur,
                              You neither learn nor are taught your mother tongue; you acquire it.

                              You are of course quite correct that without feedback you wouldn't acquire anything, but feedback occurs even in societies where adults never even bother to speak to children.

Stephen Jones
Wednesday, December 31, 2003


Excellent alternative word "acquire". I like that.

My point about not being able to learn without feedback was that I wonder if a computer will EVER learn language. It would need to have a human sit there with it for a few years, giving it some sort of implicit or explicit feedback.

Now, of course, if ONE computer ever learns language, then they'll simply get duplicated, so we have MANY that understand language.

Hmmm.... why is the image of the THE PLANET OF THE APES flashing thru my brain ??

Wednesday, December 31, 2003

Entrepreneur, I think the beginning of your post is largely on target, but the claim about language learning is contradicted by extensive evidence. All normal children acquire language, and even in environments where prevailing language is absent, will typically spontaneously evolve their own. Furthermore, correction by competent speakers (i.e. adults) is not required to achieve proficiency in a language acquired during critical childhood developmental periods, and in fact is simply not provided in certain cultures.

Just flipping open an introductory linguistics textbook I happen to have nearby and looking at the section on first language acquisition yields this: "It is sometimes suggested that parents provide children with direct linguistic training by correcting ill-formed utterances. However, studies of actual interactions between parents and children point in a quite different direction." Papers like this one have some additional information you might find interesting:

John C.
Wednesday, December 31, 2003

Tell me more about about why Noam would disagree.

Perhaps I'll learn more after I read The Language Instinct, by Pinker.

Wednesday, December 31, 2003

John C.,

Thanks for the link. Very interesting. As I said above, this is just a theory I had, so I'm not terribly committed to it being correct.

I've, so far, read only part of the article you cited. However, it seems that humans are somewhat biologically programmed to develop language.

That, too, may suggest that our hardware is preoptimized for language.

I guess my POINT was/is that langauge may not simply be a matter of rules that we learn. I think that we superimpose rules to help us understand language.  I.e., in PHYSICS, there seem to be some underlying rules, that once exposed, give us a deeper understanding of physics. However, langauge seems to be something that has a life of it's own.  We can make some general OBSERVATIONS (plural words usually end in "s", etc.) but these aren't really rules as they are an aspect of langauge that has more or less stabilized. And there is almost ALWAYS an exception to rules in (at least) english. (Such as the s/plural rule).

I'm no expert in this area. Just some thoughts that I had. I'll read the rest of that citation you gave.

Wednesday, December 31, 2003

Language is rule bound; the problem is that the "rules" we think of as adults are not the rules that we use to form correct utterances.

Also bear in mind that every generation recreates the language (if not you would never have language change). There have been many societies where the adults have communicated with each other in pidgin and the children have developed a creole.

Stephen Jones
Wednesday, December 31, 2003

Hmmm... are those language rules EXPLICIT ?

If so, then why is it so difficult to create software the can understand language?

Or, is part of the ability to use/understand language actually in our heads, not just in the rules?

Sunday, January 4, 2004

I'm afraid I don't know what you mean by explicit/implicit?

Stephen Jones
Monday, January 5, 2004

*  Recent Topics

*  Fog Creek Home