The locative is used to express a location in a city or a small island, e.g. Romae instead of in Roma. But it's not entirely clear which islands are small. I am currently on Malta and I'm curious to know which category this island belongs to. This island was known to (and occupied by) the Romans, so there must be classical examples. Do they all agree on the way to say "on Malta"?
Both Melitae and in Melita seem to appear:
Cicero, Epistulae ad Atticum 3.4.1:
Statim iter Brunddisium versus contuli ante diem rogationis, ne et Sicca, apud quem eram, periret et quod Melitae esse non licebat.
Cicero, Epistulae ad Atticum 10.7.1:
…Melitae aut alio in loco simili…
Cicero, Epistulae ad Atticum 5.19.3:
De Patrone et tuis condiscipulis, quae de parietinis in Melita laboravi ea tibi grata esse gaudeo.
A quick corpus search suggests that Melitae is more common, but there are not enough attestations of either one to draw very strong conclusions.
It is a fun exercise to apply some math to the issue (statistics/Bayesian inference specifically). I'll try to avoid technical language, for the sake of readability. Of course, math is not magic, so we need some assumptions and simplifications.
In the following, I'll assume,
- The only evidence we have is what brianpck says in the comments, that Cicero uses the locative Melitae 5 times and in Melita only once.
- Every time Cicero wanted to say what we would translate as in Malta, he chose at random between Melitae and in Melita, but the chances are not even, let's say he had p% probabilities of choosing Melitae and (100-p)% probabilities of choosing in Malta (e.g. 60% and 40%)
With these assumptions, math can help us tell whether a specific value of p is plausible or not (thanks to the binomial distribution), but for that we need two more definitions:
- A small value of p is implausible if there is less than 5% chances that the outcome of Cicero choosing at random gave the evidence we have or something better (i.e., more instances of Melitae)
- A big value of p is implausible if there is less than 5% chances that the outcome of Cicero choosing at random gave the evidence we have or something worse (i.e., more in Melita's)
Note that these assumptions could have been wildly different, leading to wildly different results.
The question on whether the choice of in Melita was forced by disambiguation remains open, so I'll solve the two extreme problems I can identify:
- Assuming Cicero didn't really care about disambiguation, and his choice of in Melita was just one more random instance, and
- Assuming disambiguation was the only explanation for choosing in Melita in that occasion, i.e., he was forced to do so, and only had a choice in the other five. In this second problem, all the evidence we have is 5 times Melitae, but we may still be interested to know to which extent we can rule out that he could have chosen otherwise.
1. Assuming no disambiguation wanted
- a value of p less than 41.8% is implausible. P(Melitae>=5 out of 6 given p=41.8%)=5%
- a value of p greater than 99.1% is implausible. P(Melitae<=5 out of 6 given p=99.1%)=5%
This leaves us with plausible values of p as defined above between 41.8% and 99.1%. Of course, statistics cannot tell with 100% certainty that p was not outside this range, it is just saying what is more reasonable to expect given our assumptions and the evidence we have.
This is undoubtedly disappointing since with as little evidence as we have, we cannot even tell whether one was actually preferred over the other. Had we more examples, we could narrow down that range.
2. Assuming disambiguation unquestioned
It sounds like a safe assumption to say that Cicero chose to use in Melita to avoid an ambiguous Melitae that could have been read as a genitive. Given that, we only have 5 choices out of 5 of Melitae. I'd blame no one if they say that's enough evidence for them to think Cicero would have chosen Melitae every time there was no ambiguity.
But a skeptic could ask whether there was any chance that Cicero would have chosen differently in a lost text or in speech, and we have no answer. The best we can do (in the absence of further evidence or any other reasonable assumption) is to ask math for help. Not that you cannot blindly rely on it, but it gives meaning answers if we keep our assumptions in mind. In this case:
- a value of p less than 60.7% is implausible, P(Melitae=5 out of 5 given p=60.7%)=5%
- a value of p of 100% is still plausible! Did I mention I'd blame no one? P(Melitae<5 out of 5 given p=100%)=0%
This leaves us with plausible values of p between 60.7% and 100%. In this case, we could say (with a lot of caveats, starting form the very assumptions), that it is reasonable to think that Cicero preferred Melitae, but still can't rule out he wouldn't have used in Melita.
Edit: regarding the title question, I think at least the optional use of the locative is out of question. Concerning its normative use, it seems more reasonable if we accept the disambiguation hypothesis (but p=99% sounds pretty much like normative, so we can't rule it out if we don't accept it).