songmzhang / cbmi Goto Github PK

I noticed that your method introduced LM Loss, which has been verified by previous papers can improve the quality of translation. But does your baseline and other methods were also introduced LM loss? The improvement in this paper seems to be more due to the introduction of LM Loss, as in the paper mentioned above, the introduction of LM Loss resulted in a 0.75 improvement in the WMT EN-DE dataset(see Table 1), and your reported improvement was 0.91. Can you provide a baseline with LM Loss?