Abstract
We propose Constrained Deep Neural Network (CDNN) a deep neural model for answer sentence selection in the context of Question Answering (QA) systems. To produce the best predictions, CDNN combines neural reasoning with a kind of symbolic constraint.
It integrates pattern matching technique into sentence vector learning. When trained using enough samples, CDNN outperforms the other best models for sentence selection.
We show how the use of other sources of training can enhance the performance of CDNN. In a well-studied dataset for answer sentence selection, our model improves the state-of-the-art significantly