I’m trying to determine the accuracy of alignments that I got from fast_align. I can see the cross_entropy and perplexity scores when I generate those alignments with fast_align.
After training translation model with that, is there any statistical way to determine how accurate my alignments are from translation model?