31st August 2005, 12:38 AM
Voice recognition has come a long way, this is true.
It is also true that key phrases can be recognized.
The illusion of a conversation may be present, but in what sense are you claiming? If it is a transparent illusion, one that the user will always be fully aware of, then yes.
If it is a fully convincing illusion, then no, we are not at that level. Human conversations have a level of adaptability and randomness that a program can't yet account for, and likely never will until a fundamental shift in computer design takes place.
For example, if the phrase "hand me the" is in the voice recognition software, and the object you want to refer to is also there, then yes, you have successfully communicated. However, what if you say "go get that orange thing by the cat"? This demands that it knows the phrase "go get that", the ability to recognize the word orange as a word that only narrows the list of potential things the user can be referring to, and then the ability to, from there, recognize that in this context, the word orange is also not referring to the fruit by the same word. It would have to recognize that orange is describing the word thing. From there, it would have to realize that "by the cat" also describes the word thing, in that it means location, not "buy the cat".
It starts out somewhat simple, but it quickly develops into a very messy program, at least if you want to go beyond that and get all possible variants of that command as a 10 year old would be able to understand it.
LL was basically saying that conversation on that level is simply beyond us quite completely at this point.
It is also true that key phrases can be recognized.
The illusion of a conversation may be present, but in what sense are you claiming? If it is a transparent illusion, one that the user will always be fully aware of, then yes.
If it is a fully convincing illusion, then no, we are not at that level. Human conversations have a level of adaptability and randomness that a program can't yet account for, and likely never will until a fundamental shift in computer design takes place.
For example, if the phrase "hand me the" is in the voice recognition software, and the object you want to refer to is also there, then yes, you have successfully communicated. However, what if you say "go get that orange thing by the cat"? This demands that it knows the phrase "go get that", the ability to recognize the word orange as a word that only narrows the list of potential things the user can be referring to, and then the ability to, from there, recognize that in this context, the word orange is also not referring to the fruit by the same word. It would have to recognize that orange is describing the word thing. From there, it would have to realize that "by the cat" also describes the word thing, in that it means location, not "buy the cat".
It starts out somewhat simple, but it quickly develops into a very messy program, at least if you want to go beyond that and get all possible variants of that command as a 10 year old would be able to understand it.
LL was basically saying that conversation on that level is simply beyond us quite completely at this point.
"On two occasions, I have been asked [by members of Parliament], 'Pray, Mr. Babbage, if you put into the machine wrong figures, will the right answers come out?' I am not able to rightly apprehend the kind of confusion of ideas that could provoke such a question." ~ Charles Babbage (1791-1871)