A Neural Network for Factoid Question Answering over Paragraphs

I believe that training methods for recursive neural networks will be some of the most interesting future research in the field. We've seen some pretty revolutionary techniques come out for feedforward models in the last 30 years, but with recurrent/recursive networks it's often difficult to see what's going on inside the network, let alone properly train it. We've already seen what Deep RNNs can do when trained [1], but I think this is only the beginning.

[1] http://arxiv.org/pdf/1303.5778.pdf

"Factoid"?

Technically speaking, a factoid is something which seems like a fact, but which is actually false/unconfirmed.

I just took the last sentence from the paragraph, "Name this German author of The Magic Mountain and Death in Venice", put it in Google, and the answer popped right up.

I guess I'll have to read the paper and play with the data set, but the example wasn't very convincing. :)

I just started downloading the code and training data, I'm interested in what quiz bowl packets were used as the training data. Most college level quiz bowl packets are posted publicly on the web but are copyrighted by the original authors and are unlikely to be licensed for such a purpose.

can't find the code. Is only the paper being published?