Working on a Hip Hop Ghost Writer NLP

Show and Tell
1

So I am working on training a NLP text generation model on a curated Hip Hop lyric dataset (Bbrown44/hiphop-ds-v4 · Hugging Face) and was wondering if anyone can provide some insight on the following questions.

  1. What techniques could I leverage to increase the text generation of the model?
  2. Is there a way that I can improve the model by rating the text generation on a Likert scale and using that results to improve text generation performance? Is there any research being done on that?
  3. Is there any research on NLP training for cultural cadence?

Thanks for your input in advance!
~Bryon