BLEURT or BERTScore added to NLG datasets.

[jordiclive]

As in #50, maybe having a machine-learned metric would be a good dimension for this work to have. As it is a recent trend in evaluation. The downside is that learned metrics take a while to run.

I'm not sure if COMET is just for MT, BERTScore or BLEURT supposedly can be used for all NLG tasks. Versions are in HF metrics (not sure about the implementation). I can see BLEURT is used for one EH dataset already.

[StellaAthena]

We are choosing to not do this as they don't tend to generalize beyond the specific context in which they were developed.