Automatic generated intralingual and interlingual subtitling and the need for human intervention

Automatic speech recognition technologies (ASR) are language-specific computer programmes that convert spoken input into written text in the language of the original speech. Introduced in 2009, YouTube’s auto-captioning technology, built on Google’s speech recognition technology, allows users to automatically provide intralingual subtitles for the videos they upload on the platform. Since their early beginnings, speech recognition technologies have improved remarkably. Nevertheless, they still face several challenges, mainly related to linguistic issues, such as the disambiguation of homophones, the lack of recognition of named entities (people, institutions, brands), and the specificities of spoken language (among which different accents or pronunciations). This obviously has an impact on another form of subtitling, namely automatic interlingual subtitling. The integration of ASR and Machine Translation into platforms like YouTube has the purpose to supply auto-generated subtitles in instances where official ones are unavailable. However, despite improvements in technology and the vast resources of Google and YouTube, automatic captioning can fail to convey the message accurately.