Spelling correction model deletes word instead of fixing

I’m trying to fine-tune a spelling correction model using a MBart pre-trained model of my language. However, when testing the model I notice that in some cases the model just deletes the incorrect word and does not even return the original one.


dong ho caso

Expected result:

đồng hồ casio


đồng hồ

I also checked my training dataset to make sure that there is no case of adding an extra word to incorrect sentence. Can someone help me with this case? Thanks