Closed hanyaqian closed 7 years ago
What exactly do you need @hanyaqian? Some example of val.coco file? val.coco.txt
If you refer to translation performance in general, it should be comparable to other NMT systems, such as nematus or OpenNMT. A well-tuned NMT system should have a similar performance (or better) than Moses, especially using ensembing decoding.
In my experiments, results are slightly below Moses when using a single system (in general). Nevertheless, using an ensemble, I usually beat Moses. Note that these results depend on the task and the language.
Unfortunately, I (still) don't have any systematic comparison against other NMT/SMT toolkits. But it would be nice to have it :)
@Sasanita @lvapeab, sorry, did`t reply in time.
I recently prepared an interview, unfortunately, i did not pass it.
Thanks!
Hi @lvapeab
Can you show me some experimental results about the translation performances measured in BLEU score by using nmt-keras?
Thank you very much!