How to incorporate the copy mechanism in seq2seq generation?

[zhhongzhi]

In some seq2seq problems, copy mechanism is involved to enable the model generate words by copying from the input sequence, such as

• question/answering generation Neural Generative Question Answering Generating Natural Answers by Incorporating Copying and Retrieving Mechanisms in Sequence-to-Sequence Learning
• summary To The Point: Summarization with Pointer-Generator Networks it's tensorflow code is here. Incorporating Copying Mechanism in Sequence-to-Sequence Learning the released code is here.

I have no idea how to merge the weights from traditional seq output and the weights from the copy mechanism. Any idea or suggestion on how to implement the model in keras is welcomed.

[NidhiRustagi]

Hi! Did you get any pointers on how to resolve this?