Take a RNN network fed with Shakespeare and generating Shakespeare-like text.

Once a model seems mathematically fine, as can be assessed by observing its loss and accuracy over training epochs, how can one assess and refine the goodness of the result ?

Only human eyes can judge of the readable character of a text, its creativity, its grammatical correctness etc.

QUESTION : Which systematic approach can be used to refine a generative model (text) ?

  • $\begingroup$ How would you measure the quality of the output text? $\endgroup$ Commented Sep 24, 2021 at 13:28
  • $\begingroup$ That's a philosophical question, at first, it should be readable and correct, which is already a challenge when tuning generative models $\endgroup$
    – kiriloff
    Commented Sep 24, 2021 at 20:18
  • $\begingroup$ So you could, as a very basic metric, use readability scores (though they have loads of faults). $\endgroup$ Commented Sep 25, 2021 at 18:25


You must log in to answer this question.