Translation results on Zh-En pre-trained model

[riddlehk]

Dear author,

The provided pre-train model produce strange translation results on single/short word translation, for example:

hi --> 1 。 Hi ，你好
hey --> 嘿，嘿，嘿，嘿
dog --> 1 。一条经过训练的狗
apple --> 如 ： (1) 苹果。

In that case, the model is not usable at all if it fails in translating common words accurately. Is there any solution on eliminating such problem via training/fine-tuning?