Software Spirit of Service
March 2nd, 2006 4:39:14 pm pst by 
Sterling Camden
TDavid was just talking about voice recognition yesterday and the state of the “art” of software’s understanding of human verbiage. Later, I had to call Qwest because I recently switched back to them for land-line (I never really liked Qwest, but they swear they have improved service, and for a $100 savings per month I’ll let them back up that claim), but my toll-free number was still being billed by my previous carrier. So I call the business customer help line, and of course a speech-recognition engine answers the phone, using the voice of a twenty-something male with slightly over-regulated testosterone.
“Welcome to Qwest.”
At which point a Spanish-speaking lady jumps in and informs me (in Spanish) that I should press 9 to hear everything in Spanish.
The young male takes the helm again, “What’s the number you’re calling about? Or..” at which point I interrupt him by keying in the number. He repeats it (very naturally speaking, compared to earlier implementations I’ve encountered), and asks me if that’s correct. I guess I have to say “Yes”, because there’s no option given for a button to push. But, he understands “Yes”.
He continues, “OK, in a few words, tell me why you’re calling today. It can be something like ‘I’d like to pay my bill.’ So, how can I help you?”
Let’s test his limits, thinks I. “My toll-free number needs to get moved over.”
“It sounds like you’re calling to report unwanted phone calls. Is that right?”
“NO.”
“Sorry, my mistake. Let’s try that again.”
“HELP.”
“It sounds like you’d like technical support. Is that right?”
“NO.”
He then apologizes again, and runs down a list of possible options, the last one being “other”.
“OTHER.”
This finally puts me in a queue to talk to a human. Even though my automated triage representative spoke English pretty well, we finally understood each other when I started speaking Computer. I’d say this technology has a ways to go yet.
Posted in Get Outta Here, Get Real | 
2 Comments » RSS 2.0 | Sphere it!
Subscribe by Email
It’s a bit like the opposite of how it tends to work with humans: things tend to work best with humans when each speaks his or her own language and the other understands it. With human-computer interactions, it seems each does best speaking the other’s language. I wonder if there’s a lesson here for artificial intelligence programmers.
Good thought, apotheon, and thanks for stopping by.
Constructing a thought in language presents more difficulty than understanding one. It’s natural, then, that the human speaker does better in his/her native tongue, while the listener can not only passively parse the result but can also probably pick up some other verbal and non-verbal cues from the reader to help in interpreting the message.
Computers like the one I was talking to have no clue about cues that go beyond the vocabulary and semantics of the language. They’re still struggling with generalizing different pronunciations and alternate syntax. When it comes to framing speech, so far they are only saying what they’re programmed to say. When a computer can frame a truly novel utterance in English or any other language, then we can say that they truly have consciousness.