I find that Opennmt description page shows that Opennmt-py contains the n-best rescoring feature, but I can not find it~
If verbose is set, will output the n_best decoded sentences
Here is the information
Hi thanks for your kind reply~ but I mean re-score with multiple features(e.g., r2l model) rather than just output the top n hypos.
Which Opennmt description page do you mean?
What do you exactly want to do?
There are plenty ways of rescoring translations from a universal vector space model, a Language-Agnostic SEntence Representations for embeddings comparison, or a scoring with a model like BERT as Language Model.
Thanks for you kind reply~ I refer to the n-hypothesis re-ranking mechanism conditioned on the given multiple scores (e.g., LM score, r2l model score or other self-defined feature scores).
Not sure what description page you are refering to.
The only way so far is to do it manually.
You output the n-best and with a R2L model you rescore the n-best output of the L2R model.