Numeri and cunei command structure?

[WuZhuiQiu]

A modelling question has prompted these historical questions:

If numeri and cunei would have been commanded by centurions, might there have been some additional Roman elements to their command structure?

Also, might there have been some parallel native or barbarian command structure, much as in the pre-independence Indian Army?

Moroever, if the Roman centurions spoke Latin, and the native or barbarian warriors spoke their own language, how would they have bridged the linguistic gap? Might the assigned centurions have been chosen from among descendents of Romanized people of the same cultural group? Might a native or barbarian sub-commander who spoke both languages have transmitted the centurion's orders?

Thanks in advance for any insight!

[Mad Doc Morris]

Simple answer: We don't know.

Not so simple answer: There's obviously a lack of formalisation. Whatever the reason, Roman administration had no need for that. So we might assume that 'irregular' units, which were mustered just for a certain (set of) campaign(s), were left to their own devices in terms of organisation. Most likely there had to be some kind of spokesperson (or a number thereof), acting as a contact for Roman administration. But this could well be limited to overall command; on average the Romans seemed to have a lot of trust in their confidants, once these had been exposed and accustomed to Roman culture (Arminius is a well-known if somewhat bad example).
The terms "numerus" and "cuneus" don't help either, since these were apparently used to describe, well, 'informal' formations of regular soldiers as well, i.e. a number (numerus!) of troops from different units mixed together.
There's one piece of evidence – which I remember – mentioning a Roman legionary centurion in charge of – apparently – foreign troopers. But the composition of these troops is unclear at best, and the purpose of their assembly was construction work not actual warfare.
In essence, there's no evidence for an 'in-depth' Romanisation of irregulars. In terms of modelling, that's liberating I guess.