Masked language modelling with specific entities or POS

For a MLM task say with this input “There are 42 [MASK] in [MASK]?”. Is it possible to constrain the predicted [MASK] token to be a specific entity or POS? For example the second [MASK] token i would like the prediction to be a company name only so it may output something like:

  1. “There are 42 employees in Microsoft?”
  2. “There are 42 developers in Microsoft?”
  3. “There are 42 employees in Apple?”
    … etc