The big elephant in the AI room
by Volker Weber
I am watching all these demos from Google, Microsoft and IBM - yes, IBM - and everything looks oh so impressive. And then it fails as soon as you are not using a single language. Microsoft Cortana for instance fails as soon as you are using an English version Windows with a German locale. Game over. Cortana and Siri can only understand one language. How easy would it be to discover when I switch from German to English and back? It's just that nobody thought of that. Cognitive? Bah, humbug.
Comments
Representing a household in which German, Spanish and English is regularly written and spoken I can only agree wholeheartedly.
for a long time using Siri to play music did not work because I have Siri configured in German and it had a hard time with the English band names and song titles.
With iOS 9.3 this has gotten lot better, just tried about 20 different bands and 18 of them worked.
Eddie Cue is talking about these improvements in the Talk Show Episode 146:
https://daringfireball.net/thetalkshow/2016/02/12/ep-146
on the Apple TV it has gotten a lot better to with the last update
...and Microsoft's Tay just went amok again ;).
Slightly off topic:
This reminds me of a task back in the year of 2000 during the "Bundeswettbewerb Informatik". You had to write a program which analyzes and recognizes the language of any given text file. I wonder why this hasn't been implemented yet in voice recognition as even my solution was pretty simple and stupid but had even an option to add own language files, a self-learning module and a hit rate above 95% based on over 10.000 txt files. Good old times... ;-)
Here's the original task (page 6, task 3): http://www.bundeswettbewerb-informatik.de/uploads/media/download/191aufgaben.pdf
And for all interested: coded with Turbo Pascal 6.0 on a Pentium II 400 MHz as a pure DOS program ;-)
Thanks Volker for bringing back those memories.
Mit Siri geht es mittlerweile recht gut.
Wie bereits von Thomas geschrieben: In einem Interview spricht Apple direkt dieses Problem an und dessen Herausforderung - aber auch deren beabsichtigte Lösung.
Isn't voice recognition only one facet of AI?
