OpenNMT Forum

Feature Requests

Supporting Tokenize/Detokenize automatically by Translation Server (18)

SDL Trados Studio integration ( 2 ) (22)

BPE issues with unknow target translations (1)

Predictive translation server (5)

Any plan for an objf based on other than ppl? (1)

Multi task learning after the encoded layer (1)

Scheduled Sampling for Sequence Tagger Model (1)

Use language model in translation (1)

Teacher forcing (12)

Extending the translation server to add features in source text (14)

Creating custom cost metric (5)

Local Attention (4)

Improving word feature labels by block text labeling for context (3)

Adding ROUGE to Validation metric (5)

Adding perplexity or negative log-likelihood to CTranslate (4)

Layers config settings (1)

Weight table:set weight of some vocabulary (6)

Support other forms of dropout/regularization (9)

Minimum Risk Training for NMT (1)

Ensemble translation (3)

A w2v-coupled training procedure (1)

Model compression via weight tying (2)

Pre-trained/fixed embeddings for features (1)

Support n_best option for CTranslate (5)

It will be better if there is a Coverage Attention Module (3)

Same source/target embeddings for intralingual task (3)

Creating phrase tables out of alignment output for translating UNKs (3)

Support of Translation History (1)

External reward signal for human-in-the-loop training (3)

Use CuDNN RNN for faster training (2)