I am reading the BERT paper. In the paper, they say that:
Although this allows us to obtain a bidirec- tional pre-trained model, a downside is that we are creating a mismatch between pre-training and fine-tuning, since the [MASK] token does not ap- pear during fine-tuning. To mitigate this, we do not always replace “masked” words with the ac- tual [MASK] token. The training data generator chooses 15% of the token positions at random for prediction. If the i-th token is chosen, we replace the i-th token with (1) the [MASK] token 80% of the time (2) a random token 10% of the time (3) the unchanged i-th token 10% of the time.
I am wondering, what will happen when they keep the token unchanged? Is it necessary for better training? I cannot see the mismatch clearly. If they do not do that, what will happen? Also, when the model is given the unchanged token, does it predict the exact token correctly in your experience?
Thanks in advance.