Appendix.A ProX training data construction.

[dsj96]

Thanks for your great work! I have some questions regarding the construction of the ProX training data (Appendix A):

As shown in Table 1, ProX includes:

Document Level: drop_doc(), keep_doc()

Chunk Level: remove_lines(line_start, line_end), normalize(source_str, target_str), keep_chunk()

Were the implementations of these functions generated by the model or manually written?
After training, would ProX generate functions other than these five mentioned above?