During my Latin education (using classical pronunciation), I was taught that 'r' should be 'rolled', making a sort of growling sound.
For example, the r's (more the second than the first set) in errāre would be rolled.
What is the historical significance of this? What evidence is there that Romans pronounced the letter this way?
This paper talks about several primary sources (i.e. Roman texts) that describe rolling Rs:
Terentianus Maurus writes in De litteris that
vibrat tremulis ictibus aridum sonorem
the R vibrates with a dry sound from trembling blows
Martianus Capella writes
R spiritum lingua crispante corraditur
[R] is pronounced with difficulty (?), with the tongue vibrating the air
Capella's writing contradicts itself on this count in another place, leaving this testimony in doubt.
Lucilius writes, more colorfully,1
r: irritata canis quam homo quam planius dictat
which (i.e. the sound r) an irritated bitch pronounces more clearly than a man
This is supposed to imply a growling similar to the modern instruction or R-rolling.
Taken as a group, these sources would seem to indicate that the Romans did roll some of their Rs. However, other authors, including Varro, Charisius, Marius Victorinus, and Martianus Capella (referring to his contradictory account), describe more fluid pronunciations. What this implies above all else is that pronunciations may have differed among different groups and in different time periods, hence the different accounts.
Indirect evidence for R-rolling comes from the Romance languages, where this is quite common. In Spanish, for example, rr indicates this rolling (rather than simply r). The presence of rolling in these languages suggests a common origin, likely in Latin.
1 The above translations are given by the author of the paper, and are not necessarily word-for-word. For example, brianpck suggested translating Lucilius differently, as "An irritated bitch which a man pronounces as clearly as possible."
I would suggest that you could challenge the notion that rolling r in Latin is based on the fact that it is pronounced that way in other romance languages. Romance languages pronounce c and g as s a soft g in some instances, which is clearly not the case in Latin. If the pronunciation of those letters has been corrupted by local influences and uses in those languages, there's a pretty good case for the corruption of the pronunciation of r as well. The fact that it isn't rolled in English, which also has a strong Latin influence, might support this theory. Bringing ecclesiastical Latin into the argument doesn't hold either. That was spoken for many centuries and also exposed to local variances and influences. Priests were not born speaking Latin. Their first language influences any other language they learned thereafter. If it were rolled, then it would be more likely to have remained rolled in all communities in England as well, since they were fairly well isolated after the Romans left England. Each was influenced to a lesser or greater extent by those tribes that inhabited that region and the influence of any external raiding tribes. We know that English is a mishmash of many old languages. If the Romans had rolled the r, surely some parts of England would still do it? It seems improbable that they all lost it as English developed. Yes, it exists in Scotland but they were largely unaffected by Roman rule. As for the snarling of a dog, the way a dog snarls depends on what language you speak. In English, human versions of dog snarling don't sound like a rolling r, so we can't assume that a rolling r is what a Roman is hearing in a dog snarl either.
