Each request is split into different tasks. Each task is routed to a chain of translators until we are confident with the quality of the translation.