What should I know when building pointer Bi-LSTM?

I have a project that user pointer Bi-LSTM for text segmentation. The theory is from Neural Models for Sequence Chunking.

Anyway… I’m still studying TensorFlow. When I looked at an open-source pointer network LSTM for ideas, there are multiple functions I still don’t know. So I want to know more about the functions.

What should I know?

What functions I should build?

What’s the purpose of each function?