Is all that Glitters in Machine Translation Quality Estimation really Gold?