I'm reasonably sure that one of the nightmare scenarios for algorithmic transcription is a field with lots of multilingual, domain-specific and obscure terminology, participants from a bunch of different countries, playing on teams from a bunch of different countries. To wit, pro cycling. Please join me in this adventure of watching Lanterne Rouge with the subtitles turned on: https://www.youtube.com/watch?v=3C6u5_0M3mU