[ale] OT: speech recognition software

Joseph A Knapka jknapka at earthlink.net
Thu Aug 15 17:00:45 EDT 2002


Jim wrote:
> 
> David Pogue of the NYT has a wonderful article today explaining why speech
> recognition sotfware will never be 100% effective. He says this as an avid
> user of the software. But he listed a series of phrases that he spoke
> alongside what the software "heard".
> 
> What I said -> What was transcribed:
> 
> * bookmark it -> book market
> * Motorola -> motor roll a
> * modem port -> mode import
> * a procedure -> upper seizure
> * and then stick it in the mail -> and dense thicket in the mail
> * movie clips -> move eclipse
> * I might add -> I my dad
> * inscrutable -> in screw double
> * hyphenate -> -8
> * suffocate -> Suffolk 8
> * a case we summarily dismissed -> a case we so merrily dismissed
> * or take a shower -> Ortega shower
> * the right or left -> the writer left
> * oxymoron -> ax a moron
> * ArialPhone guy -> aerial fungi
> 
> So, how do you explain to the software "That ain't what I meant!"? (or "That
> ain't what I mint!")

This is a well-known issue in natural-language analysis. The
answer (well, *an* answer, and the one humans apparently use)
is to perform better top-down context analysis. That is,
as you parse the sentence, the semantic environment in which
the sentence occurs, and the partial syntactic parse of the
part of the sentence already processed, must be allowed to
inform the process of analyzing the speech stream. If the
conversation is taking place in Radio Shack, for example,
the phrase "modem port" is much more likely to occur than
the phrase "mode import". Humans do the lexing, the syntax
analysis, and the semantic analysis more-or-less simultaneously,
and rarely make mistakes (though it does happen - remember
that old AC/DC song, "Dirty Deeds, Thunder Chief"?). Computers
(the kind we consumers have access to, anyway) just aren't fast
enough yet to handle it all at once. Plus we don't really
understand the semantic level very well, in terms of converting
theory into code. But there's no doubt in my mind that
it will happen eventually, and (appeasing the topic
zealots :-) Linux will be critically involved in that
research.

Cheers,

-- Joe
 "I'd rather chew my leg off than maintain Java code, which
 sucks, 'cause I have a lot of Java code to maintain and the
 leg surgery is starting to get expensive." - J. Knapka

---
This message has been sent through the ALE general discussion list.
See http://www.ale.org/mailing-lists.shtml for more info. Problems should be 
sent to listmaster at ale dot org.






More information about the Ale mailing list