Testing CX2 in production by translating "Atazanavir" from English to Spanish (trying to reproduce T195665), I found a paragraph where MT fails, and as a result gets eliminated from the translation document, causing the alignment of sections to break.
This is illustrated in this video with more detail, and in a short gif below.
As part of this ticket, it would be good to:
- Make MT failures less severe by not removing the paragraph. We can show an empty paragraph as if the user had selected "Don't use machine translation", allowing the user to switch to a different option using the MT card.
- Determine why the paragraph consistently fails to be translated automatically.