@georgejrjrjr Perhaps generate a set of responses and match approximate lengths up against one another for pairwise comparisons, but this does mean you're likely comparing unrepresentative samples…

Posters: xlr8harder. xlr8harder

2 likes recorded by the source corpus.