When
(and not if) robots are going to be a regular feature helping us out, as we
need some looking after, we need them to communicate efficiently. Now that
doesn’t mean pre-programmed responses, but means that they have good voice recognition
and can come up with useful replies rather like a super-Siri in Apple speak.
The
golden chalice would be for your robot to actually learn the language so it
could use the correct grammar, understand and pronounce words correctly. Thus
teaching your robot to talk then becomes like teaching a young child to speak.
For a robot to master that skill with language would mean that it could
interpret nuances and be of much more use as you personal friend and helper.
A
step in this direction has been reported in last week’s issue of PLoS ONE by Lyon et al from U o Hertfordshire (1).
They chose to work on iCub, an open source robot project run out of Italy that combines the efforts of 18 EU academic establishments with the U o Illinois
(2).
The
experiment involved a group of 34 participants who had to teach the iCub going
under the nom de guerre “DeeChee” about colors and patterns. DeeChee babbled
away like a young child while the teachers taught, or did their best. They
congratulated him (apparently they all decided he was a boy with no physical
evidence whatsoever) when he managed an appropriate word sound. He did like to
be told “well done.”
There
were some problems. Some teachers weren’t as good as other, some spoke over the
top of him, but others did quite well. In their 8 minute slots DeeChee managed
to learn some one syllable words. I hasten to say that none of these were the
four letter words that you might teach to your parrot to shock the neighbors.
Clearly
you wouldn’t want to teach your robot to talk individually. One would hope that
you would only have to do it with one or two who could then telepathically pass
it on via their Bluetooth.
- http://www.plosone.org/article/info%3Adoi%2F10.1371%2Fjournal.pone.0038236#pone.0038236
- http://www.icub.org/
