Open baragona opened 7 years ago
I am still looking into ways to generate better output or to aggregate what we already have. From what I've seen the second hypothesis is the best. h1 is very conservative, making small changes, sometimes containing the same content as the input, while h3 and h4 are sometimes ungrammatical.
@ojotoxy What is test.en
and test.sen
used?
Did you use the pre-trained model, in that case, what was the torch version used?
Which of these is the best general purpose model to use for text simplification? Is there any way to improve the result grammatical quality? Thank you!
Inputs:
Outputs: