Question about OOV words.

Thanks for your great job, firstly.
I have some questions about how the Copy Mechanism solved OOV problem. In your code, you built a _**content_label_vocab_dict**_ size of 2w, and mapped the words out of the vocabulary to **UNK**. So, it means the OOV words are regarded as the **SAME** token, and can not map back to the real **DIFFERENT** OOV words precisely. Then how can we generate an OOV word when decoding?

Thanks again
Sam