My understanding is that English is the intermediary language, but not on purpose- they tried to create an "interlingua" but they trained with too many English-to-X and X-to-English data and ended up with a model of English as their intermediary language model.

And it's hard even for them to know because their models are opaque and hardly amenable to principled examination.

(I mean, you can't prove anything about them, so you can't really know what they're doing in there).