I hope this gets through to the group; apparently Google decided to break something and has a pretty embarassingly bad excuse for "help center" for such a major corporation; I wouldn't believe that any software devlopers work at Google if I didn't know some.
Anyway, thanks for bringing this to our attention. That particular query was chosen to go in to the documentation becuase it's such
common knowledge we thought NELL would always at least have an answer, even if it might not be the right answer. It just so happens that
at present NELL has no answer. The service itself is still up and running properly.
But on closer inspection you have found a bug of some sort. Consider posing the query in the other direction: "Apple has what CEO?" rather
than "who is the CEO of Apple?"http://rtw.ml.cmu.edu/rtw/api/json0?lit1=Apple&predicate=companyceo&lit2=*&agent=KB
Now you get a nonempty answer. It is possible to get different answers depending on which way you ask the question because NELL is using a
variety of learning / predicting methods, and it can for instance be fundamentally easier to predict a continent given a country than to
predict a country given a continent. But in this particular case the answer shouldn't be this different because the "KB" agent is a direct
lookup from basically a database.
At the moment we're getting close to finishing an overdue overhaul of NELL's low-level KB routines that might remediate this behavior as a
side-effect, so I'm just going to add this to the list of test cases to use afterward. If in the meanwhile you find other queries that
appear to be malfunctioning, please let us know and we'll use those as well to try to smoke out any lingering bugs.bki...@cs.cmu.edu