Streams

Computers and Language

Wednesday, August 25, 2010

Computers can speak and understand spoken language. They can also convert text to speech, speech to text, and so much more. Julia Hirschberg, Professor of Computer Science, Columbia University, and Steve Lohr, New York Times technology reporter, explain how computers can converse with humans, identify voices and dialects, detect deception, recognize human emotions, and generate images to illustrate text.

Guests:

Julia Hirschberg and Steve Lohr

Comments [16]

Amy from Manhattan

On the Google translations of recipes, the main thing is that in Spanish, instructions are often given in the infinitive, not the imperative, so if the program isn't set up to recognize the context as a set of instructions, it will give translations like "To add 1/2 teaspoon of cinnamon."

Aug. 26 2010 11:34 AM
peter from vancouver

open the pod bay doors please hal

Aug. 25 2010 06:19 PM
Lslo from ny

who is the promo voice on wnyc is that a computer

Aug. 25 2010 01:57 PM
kk from nyn@

You only mentioned IBM how about apple? didn't they have this tech first? remember talking moose?

Aug. 25 2010 01:53 PM
Paul from Weehawken

Good show. Paying Kraftwerk was a nice touch.

Aug. 25 2010 01:49 PM
Paul

Good show. Paying Kraftwerk was a nice touch.

Aug. 25 2010 01:48 PM
Amy from Manhattan

It's not so much having a computer try to sound human that bothers me. But when an ATM screen signs off by saying "It's a pleasure to serve you," I find that very irritating. It's a machine. It doesn't get any pleasure out of serving me. I guess its programmers think that comes off as polite, but it doesn't to me.

Aug. 25 2010 01:47 PM
Jon from NYC

Those automated computer phone answering systems that you get when you call a company are not actually intended to benefit you in any way. Their real purpose is to get you to hang up. Even if they can dissuade only 10% of callers from continuing with the call, the company has saved money.

Aug. 25 2010 01:46 PM
Bill M from New Rochelle

Please mention internet browsers and computer text readers for the blind...

Aug. 25 2010 01:45 PM
Bill M from New Rochelle

I am reminded of a job I worked years ago with a fellow named Tom. We were working 78 hours a week, and ate 3 meals on the job.

Tom would often order Chinese food, and always used a made-up name; "Larry Lolly." I asked him why he gave a phony name, and Tom told me he never got tired of hearing them reading it back.

Aug. 25 2010 01:39 PM
Rob from Brooklyn

I am dysphonic and these voice recognition systems never work for me. If the system does not have a back up that allows for push button tones to answer yes or no questions, then I have to hang up and I hate that.

Aug. 25 2010 01:39 PM
Betty Anne from UES

What about computer speech being too close to that of a human, does the uncanny valley phenomenon happen?

Aug. 25 2010 01:37 PM
Amy from Manhattan

At first I thought Leonard & Dr. Hirschberg were saying "prosity" & wondering what it was...took a little while to realize you were saying "prosody," which I've actually heard of! And that was with human beings saying it....

Aug. 25 2010 01:36 PM
Ash in Chelsea

One of my most annoying pet peeves in this computer age is when I encounter one of those voice-based computer systems on the phone that asks you something that requires a 'yes' or 'no' answer. Then, when I give my most articulate 'yes' or 'no' have it chime, "I'm sorry, but I did not understand that answer. Please try again." :-)

Aug. 25 2010 01:35 PM
Daphna from The Upper West Side

What about speech to text programs? Do they use the same technology, and how accurate are they? I would like to find a good speech to text program for my dysgraphic child to use for school work. Any recommendations for what type of program to look for?

Aug. 25 2010 01:35 PM
Don from Long Island

Thank you for bringing on some computer people to discuss technology! As a software person, it's nice to hear an informed perspective on my trade. I wonder when we'll commonly be able to have a conversation with a system about a subject domain and get useful information out of a lengthy dialog rather than saying over and over, "I'd like to speak with an operator".

Aug. 25 2010 01:29 PM

Leave a Comment

Email addresses are required but never displayed.

Get the WNYC Morning Brief in your inbox.
We'll send you our top 5 stories every day, plus breaking news and weather.