This is a first attempt at speeding up the the adapted composition by adding a first Levenstein edit-distance computation. This consumes RAM proportional to m*n (number of words in both ref and hyp) so it could come problematic.
The speedup is noticeable, I've seen 10x in some cases.
This is a first attempt at speeding up the the adapted composition by adding a first Levenstein edit-distance computation. This consumes RAM proportional to m*n (number of words in both ref and hyp) so it could come problematic.
The speedup is noticeable, I've seen 10x in some cases.