Experimental research in automatic subtitling

Recent developments in neural machine translation, and especially speech translation, are gradually but firmly entering the field of audiovisual translation (AVT). Automation in subtitling is extending from a machine translation (MT) component to fully automatic subtitling, which comprises MT, auto-spotting and automatic segmentation. The rise of this new paradigm renders MT-oriented experimental designs inadequate for the evaluation and investigation of automatic subtitling, since they fail to encompass the multimodal nature and technical requirements of subtitling. This paper highlights the methodological gaps to be addressed by multidisciplinary efforts in order to overcome these inadequacies and obtain metrics and methods that lead to rigorous experimental research in automatic subtitling. It presents a review of previous experimental designs in MT for subtitling, identifies their limitations for conducting research under the new paradigm and proposes a set of recommendations towards achieving replicability and reproducibility in experimental research at the crossroads between AVT and MT.