BERTHA: Video Captioning Evaluation Via Transfer-Learned Human Assessment