This suggestion originates from a post on LinkedIn about fully digraphic languages and how to go about translation processes when part of the source text is in Serbian (Cyrillic) and part in Serbian (Latin).

I asked the author about how they approach:

1. handling this issue in a TM, and not being able to leverage all their TM content due to the digraphic issue (e.g. a shift between a preference between Cyrillic and Latin between iterations of a document).
2. handling this issue in a Termbase. In this regard it was simple enough to have Serbian (Cyrillic) and Serbian (Latin) as separate languages for an entry.

Could this not be a potential use for AI in languages where there is complete transliteration? E.g. Serbian or Kazakh.

Alternatively, could a three column mode (with two source columns) be an approach (one workaround I suggested for a list of strings was a multilingual Excel). One potential idea could be an AI-based Cyrillic to Latin transliteration.