search

ictnlp / OR-NMT

Source Code for ACL2019 paper <Bridging the Gap between Training and Inference for Neural Machine Translation>
41 stars 10 forks source link

vocab size #3

Closed haorannlp closed 4 years ago

haorannlp commented 4 years ago

Hi, Zhang Wen! When I apply BPE using command subword-nmt learn-joint-bpe-and-vocab --input train.en train.de -s 37000 -o bpe_code.en.de --write-vocabulary en.vocab de.vocab, the vocab sizes of ENG, DE are 39320, 40364, respectively (In your paper, they are 39418, 40274). Did I miss something here? Besides, does 'tgru' mean 2 GRU cells? Any help would be greatly appreciated.

zhang-wen commented 4 years ago

@haorannlp Hi, I think this is reasonable, because we may not filter the training set in the same way, so the number of sentence pairs in the training set may not be the same. Anyway, differences in vocabulary size will not have a big effect. No, tgru means the model proposed in the paper "DTMT: A Novel Deep Transition Architecture for Neural Machine Translation", please refer to this paper for more details.

haorannlp commented 4 years ago

Thank you