There's likely a good bit of analysis that could be done on TLAs. Consider TLA itself is {Adjective : Count} {Noun} {Noun}. Meanwhile, DUI is {Gerund} {Preposition} {Noun} with the stop word 'the' removed.

It might be interesting to take a sample of TLAs used and look what words can be used in those spots. If the third position is 90% likely to be a noun, that could change the distribution... guessing not in a significant way itself but it could be interesting to see.

This is the best work I know on the topic (admittedly having done no literature review): https://gwern.net/tla