I kinda wonder if you can encode knowledge in Lojban and have a more efficient language model using it then only translate to English at the output (if needed).

I kinda wonder if you can encode knowledge in Lojban and have a more efficient language model using it then only translate to English at the output (if needed). Is there any study on the informational density of Lojban/loglan?