Is there any small and interesting research directions of NLP recommended?

Popular NLP model(bert, GPT) is getting more and more bigger, the cost can't affordable for single person not rely big company, it's bad for diversity in research.

I admit bigger model have better performence, but I think explicable and modifiable technology is more important, bigger model seem that do nothing more in explanatory of model.

More and more people can put into NLP, help NLP to Artificial General Intelligence faster, if the cost is lower.

Thanks for any advice.