What do bert word embeddings learn about the french language

HIGHLIGHTS

• What: To this the authors propose framework for comparing the quality of encoding in different models. Shifting the research focus from constantly increasing the training datasets towards understanding more about existing models, their parameters (and how they can be improved) might become necessary in the nearest feature. The authors propose an intrinsic metric evaluating the quality of information encoding (InfEnc) and suggest a framework that allows the identification of the best dimension candidates of word embeddings that potentially encode target information (in the experiments : grammatical number and gender for French nouns and adjectives, part of speech (POS . . .
  

   

  

  If you want to have access to all the content you need to log in!

  Thanks :)

  Username or Email

  Password

  Remember me

  Lost your password?

  If you don't have an account, you can create one here.