Nick Nicholas proposes that for Graeco-Latin nouns, Lojban follow the lead of Latino sine Flexione (and to a lesser extent Interlingua), and borrow words from Latin in the ablative singular case.
Advantages:
- Forces retention of stem consonants often swallowed up in the nominative. This makes the loanword more recognisable both given the Romance languages, and related words.
- For example, leone is more recognisable than leo, given Italian leone (and English lion), and the adjective leonine.
- Guarantees the word will end in a vowel.
Disadvantages:
- The ablative is not the citation form.
- People shouldn't have to be Latinists to borrow words into Lojban.
- They do if they're going to write a dictionary of them, though.
- Any Latin dictionary that is not utterly worthless will list the nominative and genitive of all nouns (Lewis and Short, for example). The following map gives the ablative case of every noun.
- A few words will end up less recognisable, not more.
- The best example is that of elements: radium would become xukmrnradio, not xukmrnradiumu.
- jinmnradio or jinmnradiume. xukmi is for drugs.
- Disagree, but it looks like we'll need clarification at some point according to the supplicatory model anyway. But that's not really relevant here...
- In any case, they still have one r too many, canonical would be xukmnradio or xukmnradiumu
- I use sodna for the alkali metals, jinme for other metals, navni for noble gases, kliru for halogens, and jicmu for other nonmetals. For radium, I use a lujvo: dircyjinme (not to be confused with dircynavni which is radon). Since so many element names in Latin end in -ium, and so many in Lojban begin with jinm-, I drop -um and let the rafsi take its place: jinmrstronti. — phma
- Disagree, but it looks like we'll need clarification at some point according to the supplicatory model anyway. But that's not really relevant here...
- jinmnradio or jinmnradiume. xukmi is for drugs.
- The best example is that of elements: radium would become xukmrnradio, not xukmrnradiumu.
Rules of thumb (Latin has exceptions and whatnot, but this will do in the general case.)
Nominative Ablative
-a -a
-us -o
-us -u (Very rare; virus is the only good example)
-us -ore, -ere (Not rare enough)
-um -o
-on -o
-es -e
-o -one
-er -re
-er -ero (rarer)
-is -e
-x -ce, -ge, or -cte 
-Cs -Ce (C for consonant)
-C -Ce (C for consonant)
-u -u
If you know the genitive of the noun (and Latin dictionaries always list the genitive next to the nominative, precisely because Latin gets so wacky), the procedure is much simpler:
-ae -> a, -i -> -o, -us -> -u, -ei -> -e, -is -> -e
Examples:
- kriterio (criterion, criterio)
- introduktione (introductio, introductione)
- specie (species, specie)
- indice (index, indice)
- genere (genus, genere)
- opere (opus, opere)
- krise (crisis, crise)
- antena (antenna, antenna)
- foko (focus, foco)
- medio (medium, medio)
- arbitro (arbiter, arbitro)
- amoniace (ammoniac, ammoniace)
For verbs, the infinitive is as good a form to borrow as any.
See also Transliterating Graeco-Latin
On second thought, don't: See The Complete Lojban Language, Chapter 4.8.
Bear in mind, however, that Church Latin is to be preferred to Classical Latin, because it is the Church Latin forms that have achieved scientific internationalisation. In other words, verbum, not uerbum, acer, not aker, and trajectorium, not traiectorium.
- Why not vulgar latin, to preserve recognizability for non-latinists who speak a romance language?
As a supplement thereto, oe tends to be treated in Modern European languages as a front, rather than a back vowel. Thus, co'elakanto for C(o)elacanth, not ko'elokanto.
- Why is <oe> /o'e/? If early classical it shd be /oi/. If postclassical/Church it should be /e/, shouldn't it? If K > C before front vowels, churchily, why isn't <oe> too treated churchily? --And
- Erk. You're right; I'm hesitant to formalise it for orthographic recognisability, but will probably be convinced otherwise... — nitcion
I'd have thought that most words from classical Latin could be translated into Lojban by lujvo rather than borrowed as the basis for fuhivla. Or is the idea that Lojban will be like, say, English, with many near-synonym pairs, one member of which is native and the other borrowed from Greek/Latin? (mi'e And)
I have in mind explicitly things like Linnaean terms, and some scientific terminology - nominal-like entities. Because the place structures of fu'ivla are vague, I anticipate Lojbanists will much prefer lujvo in any case, and only use fu'ivla where the lujvo is unworkable. Thus, for example, I don't expect introduktione would really be used as a fu'ivla mi'e nitcion