Welcome to MSDN Blogs Sign in | Join | Help

Say *@$#! for the operator

No, surely not. A reader at Lifehacker claims:

Many IVR (interactive voice response) systems are programmed to recognize key words. Among those keywords are frequently a list of swear words, like the FCC's dirty 7. When asked to respond, use on of those epithets and you will likely be transferred directly to a live human being. 

The supporting anecdotal data in the comments is shaky, to say the least. Can we bury this as an urban myth? Anyone out there ever deployed a speech application that explicitly models swear-words in its grammars?

(via Keith Dawson)

Published Friday, June 09, 2006 3:22 PM by Stephen Potter
Filed under:

Comments

Saturday, June 10, 2006 11:27 AM by Mark

# re: Say *@$#! for the operator

Not sure how you would test such a grammar without having your co-workers think you have some emotional problems. ;-)

Seriously, I've never created a voice application that used this approach.  There are better ways to determine when callers are getting angry – see http://www.voiceingov.org/blog/?p=93  
Monday, June 12, 2006 4:37 PM by Stephen Potter

# re: Say *@$#! for the operator

Thanks Mark, yes, we'd certainly never be able to ship a sample with that stuff in it, there are 'release police' to catch that sort of things...

How robust is your bookmark method to detecting frustration? It seems to me that there could be many possible causes of early barge-in: line echo, background noise, power users and so on. That said, for most applications the fix for a frustrated user is usually also a good fix for line echo, background noise, etc., so maybe the general catch-all is good.
Tuesday, June 13, 2006 11:34 AM by Terry Gold

# re: Say *@$#! for the operator

We've joked about building the grammar for it so that others wouldn't have to think of all the words that people might say when they are angry.  I really think that if a system got deployed this way, it would get enough press that it would be tied up with people just calling to see if they could think of new words to try.

Years ago our speech recognition auto-attendant would recognize "Pain in the ass" as "Terry Gold".  I found it humorous and figured it was in the grammar.  That doesn't work anymore, but then we've deployed on a new recognition engine so maybe the new recognizer is just doing a better job.

As the myth busters say, I would mark it "plausible."
Tuesday, June 13, 2006 5:08 PM by Stephen Potter

# re: Say *@$#! for the operator

Heh heh. Or maybe there's another pseudonym for you in the current grammar.  :-)
Friday, June 16, 2006 12:41 PM by Mark

# re: Say *@$#! for the operator

Your right about the approach I examine in my blog posting -- it's far from perfect.  I'm not aware of any really good ways of *accurately* and *consistently* detecting emotions, including anger, in speech recognition systems.

What's an even more interesting question is, if we are able to accurately detect frustration (by matching a naughty word grammar), what do you do with the caller.

I've talked with people who have told me that if the caller has descended to the level where they are using foul language, it might actually be preferable to keep them in the IVR instead of inflicting them on an agent.  Directing them to a complaint dialog where they can leave a message, or a call back number when they have cooled down are options.
New Comments to this post are disabled
 
Page view tracker