"Carl Gundel" <[email protected]> writes: > LOL! I love your example. :-) > > I used to work at a company working on natural language processing (in > Smalltalk no less). We had more than a dozen doctorate linguists and > computational linguists working at LingoMotors. Here's just one single and > overwhelming example of a challenge to overcome. A perfectly grammatical > sentence in a human language can have many valid parse trees, and realize > that this isn't a design fault of the parser. Then you have to pick the one > that the speaker intended. This is no mean feat. > > So, first correctly recognize all the spoken words (hard enough), being sure > to know where the sentence boundaries are (also hard), then parse them > correctly into the possible correct senses (much harder), and then finally > decide based on expert knowledge and context that may not be present which > sense is the correct one (really, really difficult). > > Natural language wins? Not anytime soon. > > -Carl Gundel
My intuition, based on a very limited course on speech recognition at University and my own heavy bias towards programming languages, is that 'serious' use of speech commands will end up evolving some terse, phonetic, unambiguous vocal programming language. It would resemble speech in the same way that a bash session resembles an email chain. There are probably languages like this in the wild already. My reasoning is by analogy with text-based programming. Even Excel users are used to saying "SUM(A15:B20) / 1.5" rather than "the sum of the range from A fifteen to B twenty all divided by one and a half". I don't imagine the equivalent we'll be saying "sum open-paren dollar one five colon..." but something more phonetic, hopefully something that can be strung together without becoming incomprehensible. I also think that tonal audio output may be preferable to spoken output as the amount of data increases. For example, imagine a service monitor that hums along as requests are processed, becoming discordant when it starts seeing error messages. This lets us internalise the status of the system, noticing immediately when something is out of the ordinary. An equivalent speech system could only alert us when certain conditions are met, eg. "Warning, 10% error rate". Cheers, Chris _______________________________________________ fonc mailing list [email protected] http://vpri.org/mailman/listinfo/fonc
