interestingasfuck

It would be better to have a layer in between. That layer would consist of concepts.
E. g. "I" <-> self-reference & object <-> "watashi"
or "I" <-> self-reference & subject <-> "watashi wa".

That's how translation software often works.

So when do I get a UT?

Maybe for an ideal translation, but I think even real-time transliteration would be awesome... and might even start blending grammar patterns across borders.