td,dr: Just use English rules — unless nobody else does. Then do whatever they do.
Perhaps the best rule is to ignore the language of origin that we pinched the newly borrowed word from and just use the same basic English rules for all inflections (usually noun plurals), as though it were fully naturalized into English. It’s certainly the easiest rule to remember, since you don’t have to learn the inflectional morphology of dozens of donor languages that way.
The truth is that it depends in part on how well assimilated the word is into English, and what the tradition is in the particular domain where that word is being used. Sometimes more than one version of a plural inflexion exists, such as both the English indexes and the Latin indices for the noun index, where mathematics tends to use the more conservative form.
Greek and Latin loanwords used in science and other technical disciplines often conserve their classical plurals. Some are now relatively common words like species and series, which due to their Latin origin (and perhaps also other phonologic factors serve as both the singular and the plural of each. So too does French chassis, although at least that one has a pronunciation difference between one chassis and two chassis.
Other terms with non-English plurals whose special Latin or Greek plural forms may not always be observed outside of technical writing include such singular/plural pairs as genus/genera, antenna/antennae, vertebra/vertebrae, mare/maria, thesis/theses, crisis/crises, emphasis/emphases, taxon/taxa, mythos/mythoi, dictum/dicta, paramecium/paramecia, phenomenon/phenomena, penis/penes, clitoris/clitorides, crysalis/crysalides, octopus/octopodes, fungus/fungi, nucleus/nuclei, radius/radii, criterion/criteria, matrix/matrices, cicatrix/cicatrices, larynx/larynges, stigma/stigmata — and many, many more.
Some exotic inflections seem more resistant to naturalization than others, such as with crisis > crises and emphasis > emphases. I don’t know why that should be. Loanwords that start out with classical inflections often lose them as we become more used to them. For example, the OED observes that specimina was fairly common as the plural of specimen during the latter half of the 17th century. However, the “special” plural form specimina instead of specimens is now so vanishingly small as to be virtually non-existent; you now only find it, if ever, in highly technical treatises on biology or typography.
You also sometimes see special plurals from French like plateaux, châteaux, and tableaux, or from Italian like mafiosi, tempi, or paparazzi. For whatever reason, non-English plurals for words coming from languages other than Latin or Greek don’t tend to last as long as the truly classical ones sometimes do. My own theory is that this is because compared to your average joe, writers in the sciences are not only more “precisionist” in their writing, they also are more likely to have studied Latin (and sometimes Greek) in their own studies.
It just depends on the field, though. For example, in computational linguistics the word corpus comes up all the time, meaning a body or collection of texts. In that domain, the expected form of the plural is the Latin one, so you have several corpora. However, we do not (to my knowledge) use the Latin genitives unless we are forming the species name of a particular taxon. So the possessive of the singular corpus is simply corpus’s in English instead of the Latin genitive singular corporis, and the plural possessive is corpora’s in English instead of the Latin genitive plural corporum. (Again, taxa are excepted, but those aren’t really English away; they’re “modern” Latin.)
Now and then you find some imported German nouns retain their German plurals for a little while, depending on the field. John Maynard Keynes actually used Spielräume for a particular philosophical term, being the German plural of Speilraum. Or when talking about the semi-autonomous local governmental units of Germany or Austria, each of which is a Land, you may see English writers using Länder for several of these. But these are markedly foreign; they are unassimilated imports. You can tell a German noun is on the road to being more fully Englished as soon as it loses its initial capital, as we see people writing zeitgeist in English instead of Zeitgeist. If that needed to be made plural, I would not be surprised to encounter the uncapitalized version to be the Englished zeitgeists, as it stands out rather less than writing Zeitgeister would.
The OED attests hundreds of different singular-to-plural irregulars for imports, but most of these don’t last long in English. It might have been bruschetta/bruschette and pannino/pannini right at the start, but those didn’t lost long, such now people look at you funny if you speak of “a graffito” in the singular, and it is not uncommon to get “double” plurals, like “two panninis”.
Just use English rules on loanwords’ plurals — unless you’re working with one that is expected to retain its irregular form. Those you should be able to find in a dictionary.
And the plural of virus is viruses. Always. :)
