Mismatch in -valid_src and -valid_tgt in the provided training script for forward and backward tasks

[EronHou]

Hello,

Thank you for sharing this great repository. I have been following the instructions in the training script provided, but I noticed a potential mismatch in the validation dataset configuration for both the forward and backward tasks.

Forward Task:
In the onmt_preprocess command for the forward task:
-valid_src "${DATASET}/src-valid.txt" -valid_tgt "${DATASET_TRANSFER}/tgt-valid.txt"
Here, the -valid_src comes from ${DATASET}, while the -valid_tgt comes from ${DATASET_TRANSFER}. Shouldn't both src and tgt for validation come from the same dataset?

Backward Task:
Similarly, in the backward task:
-valid_src "${DATASET}/tgt-valid.txt" -valid_tgt "${DATASET_TRANSFER}/src-valid.txt"
Here, the -valid_src comes from ${DATASET}/tgt-valid.txt, and -valid_tgt comes from ${DATASET_TRANSFER}/src-valid.txt. This seems inconsistent, as the validation source and target do not match in a straightforward way.

Questions:

1. Is this mismatch intentional? If yes, could you explain why the validation datasets for src and tgt are taken from different sources (${DATASET} and ${DATASET_TRANSFER})?
2. If this is an error, how should we correctly set the -valid_src and -valid_tgt for both forward and backward tasks?