fairseq
fairseq copied to clipboard
Question about Discriminative Reranker for NMT
❓ Questions and Help
Before asking:
- search the issues.
- search the docs.
What is your question?
How to use monolingual data ? Do I understand this correctly ?
- given parallel data
P
and monolingual data in target sideM
. - train a translation model
T1
inP
from target to source - translate
M
to source side withT1
- train a translation model
T2
inP
from source to target - using
T2
to generaten-best
list from source side ofP
andM
. - using generated
n-best list
to train a reranker
@an918tw Hi, Could you please help me out ? Many thanks
@Hannibal046 yes this looks correct to me.
@an918tw Thanks for your help !
@an918tw , Hi, would you mind share the trained checkpoint for DE-EN and EN-DE direction ? Thanks so much !