Is 42 the Answer to Everything in Subtitling-oriented Speech Translation?

Proceedings of the 17th International Conference on Spoken Language Translation (IWSLT), pages 209–219 July 9-10, 2020. c⃝2020 Association for Computational Linguistics https://doi.org/10.18653/v1/P17 209 Is 42 the Answer to Everything in Subtitling-oriented Speech Translation? Alina Karakanta Fondazione Bruno Kessler University of Trento Trento - Italy akarakanta@fbk.eu Matteo Negri Fondazione Bruno Kessler Trento - Italy negri@fbk.eu Marco Turchi Fondazione Bruno Kessler Trento - Italy turchi@fbk.eu Abstract Subtitling is becoming increasingly important for disseminating information, given the enor- mous amounts of audiovisual content becom- ing available daily. Although Neural Machine Translation (NMT) can speed up the process of translating audiovisual content, large man- ual effort is still required for transcribing the source language, and for spotting and seg- menting the text into proper subtitles. Cre- ating proper subtitles in terms of timing and segmentation highly depends on information present in the audio (utterance duration, natu- ral pauses). In this work, we explore two meth- ods for applying Speech Translation (ST) to subtitling: a) a direct end-to-end and b) a clas- sical cascade approach. We discuss the benefit of having access to the source language speech for improving the conformity of the generated subtitles to the spatial and temporal subtitling constraints and show that length1 is not the answer to everything in the case of subtitling- oriented ST. 1 Introduction Vast amounts of audiovisual content are becom- ing available every minute. From films and TV series, informative and marketing video material, to home-made videos, audiovisual content is reach- ing viewers with various needs and expectations, speaking different languages, all across the globe. This unprecedented access to information through audiovisual content is made possible mainly thanks to subtitling. Subtitles, despite being the fastest and most wide-spread way of translating audiovisual 1Speaking of subtitles and their optimum length of 42 characters per line, we could not help but alluding to the book and series The Hitchhiker’s Guide to the Galaxy by Douglas Adams, where the number 42 is the “Answer to the Ultimate Question of Life, the Universe, and Everything”, calculated by a massive supercomputer named Deep Thought for over 7.5 million years. content, still rely heavily on human effort. In a typi- cal multilingual subtitling workflow, a subtitler first creates a subtitle template (Georgakopoulou, 2019) by transcribing the source language audio, timing and adapting the text to create proper subtitles in the source language. These source language subti- tles (also called captions) are already compressed and segmented to respect the subtitling constraints of length, reading speed and proper segmentation (Cintas and Remael, 2007; Karakanta et al., 2019). In this way, the work of an NMT system is already simplified, since it only needs to translate match- ing the length of the source text (Matusov et al., 2019; Lakew et al., 2019). However, the essence of a good subtitle goes beyond matching a prede- termined length (as, for instance, 42 characters per line in the case of TED talks). Apart from this spa- tial dimension, subtitling relies heavily on the tem- poral dimension, which is incorporated …

Content truncated to 500 words