TL;DR: Latin words ending in -que should be accented on the syllable before -que only if that syllable is (or has become) heavy; otherwise, the word should retain its original accent. If this sounds new to you, that’s probably because you’ve been following Allen & Greenough and other nineteenth century scholarship.
The rules of accent in Classical Latin are usually very simple. Almost all words follow the formula of the ‘Penultimate Law’, which states that the accent in multiple syllable words falls on the second-last (the penultimate) syllable if this is of heavy quantity, and otherwise on the third-last (the antepenultimate).
But the enclitic –que (and the other enclitics, –ve, –ne, –ce) complicates these rules.
I was first taught, following Allen & Greenough’s nineteenth-century grammar, that any word ending in the enclitic conjunction –que should be accented on the syllable before –que, whatever its quantity. This advice seems to come mainly from a generalised rule stated in late antique grammarians.
However, at least as early as 1908 this has been a matter of vigorous contention. It is not simply that it sounds “off” (you try routinely putting a stress, not a pitch, accent on the short ă in bellaque while preserving the brevity of the vowel for metrical purposes). In poetry, it often clashes with moments where an accent is expected.
Charles Newcomer, in an article The Effect of Enclitics on the Accent of Words in Latin (1908), noted that there was “much confusion and inconsistency” in the rules of grammars on this question, where “some follow the ancient grammarians blindly; others in certain respects only; yet others in still different regards.” He also pointed out that the always-accent-before-que rule disrupts the expected accent on the fifth foot of Vergil’s hexameters. Vergil fails to put the accent at the start of the fifth foot in “less than one in two hundred lines” – i.e. half a percent of all cases. And yet Vergil quite liberally places words such as armaque, bellaque in the fifth foot; indeed, he uses them “much oftener in the fifth foot than in any other”.
Newcomer was quite ruthless in his condemnation of the late antique grammarians on pronunciation. “The grammarians, who follow Greek models, are all post-classical. Their accumulated heap of grammatical rubbish needs sifting. Such statements as do not represent at least a silver Latinity should be stricken from our school grammars.”
Nevertheless, Newcomer acknowledged that natural pronunciation changes probably occurred in the centuries between Plautus and the writings of the grammarians. He claimed (despite his rejection of the late antique grammarians’ statements for earlier periods) that there was universal late Latin testimony for sceleráque, and suggests on this evidence that a secondary accent was gradually added to the syllable before –que especially in forms of words that ended in heavy syllables (e.g. scélerìque), and that this secondary accent was gradually generalised to other forms and later became the main accent: scéleràque > sceleráque. His recommendation for Golden Age Latin, however, was: “pléraque, ítaque, béllaque; but probably bellúmque, scelerúmque.”
In the mid-twentieth century, R. Whitney Tucker continued the debate in Accentuation before Enclitics in Latin (1965), this time focusing on expected accentuation in Plautus (2nd century BC) and Seneca (1st century AD). Tucker came to a different conclusion, stating that the normal accent in Plautus is unaffected by the addition of –que, but that in Seneca the composite word with –que should be accented just as if the total were a normal word (!). For Seneca, he recommends natósque (and also magísque, utróque, etc.), where the accent is attracted to a long penultimate, and vérbaque, mémbraque, preserving the accent of disyllabic words ending in a light syllable.
Incidentally, Tucker gives no recommendation or evidence from Seneca’s period for what to do with the case of a word like límina + –que, which according to the Penultimate Law would become *limínaque. I am hesitant to apply the Penultimate Law to such a combination without first seeing some type of evidence from the poets.
The most highly regarded reference for Latin pronunciation, Vox Latina, by W. Sidney Allen (1965), provides yet another suggestion. I quote him at the bottom of this post in full, but his conclusion after considering the various scholarly positions is as follows: vírum + –que should become virúmque; béllă + –que should probably become béllaque; and límina + –que could possibly become líminaquè, with a main accent on the first syllable and a secondary accent on the enclitic –que. His reasoning for the last note is that Aen. III.91 begins with líminaquè laurúsque, the first –que being made to stand in for a metrically heavy syllable; this would seem more likely if it had some kind of accent on it.
In any case, my main area of interest on this question is in the intended pronunciation of Vergil’s Aeneid. So out of curiosity, I decided to do a survey of the accent of –que in book I of Vergil’s Aeneid. In the future I might extend the analysis to the other books of the Aeneid, but I wanted to see first if there were any signs of meaningful data.
Tucker had rather summarily dismissed evidence from Golden Age poets, including Vergil, as inconclusive, because “a contradictory accentuation may in rare instances be found here too, as in Aen. 1.65 …hóminum rex, and 1.105 praeruptus áquae mons.” I find this argument unconvincing, not because these exceptions don’t exist – everyone knows that they do – but because of the extreme rarity of such exceptions, occurring in less than half a percent of Vergil’s hexameters. If a phenomenon appears more often than the expected rate of anomalous fifth foot accents, then the phenomenon should be worth investigating.
Therefore, I set out to measure two things in Aeneid I: The rate of fifth foot ictus not coinciding with an accent, and the rate that words ending in –que seem to break Allen and Greenough’s accentuation rule.
The accent before -que in Vergil’s Aeneid I
My results, even at this early stage, were rather telling.
Aeneid book I contains 756 lines. Of these lines, there were three incomplete lines which I excluded from this study. I also found one line which contained a fifth foot spondee, which I also excluded from the study as being far outside the normal pattern. This brings us down to 752 lines.
Of the remaining lines, 102 contained a fifth foot –que. For establishing a baseline for the rarity of unaccented fifth foot ictus, I made my control sample the 650 complete lines which do not have a fifth foot –que or a fifth foot spondee.
Only two out of these 650 lines exhibited no accent on the fifth foot ictus (1.65, 1.105). This suggests a 0.3% rate of occurrence (even lower than Newcomer’s suggested rate of 0.5%).
Out of the 102 fifth foot -que words, 94 contained heavy syllables before the –que (after the pattern virúmque and viróque), and the expected accent coincided with the syllable before -que. However, eight words expected an accent elsewhere, contrary to the rule of “always before -que”:
8 out of 102 represents a 7.8% rate of occurrence, which is roughly 25 times more often than the expected rate of contradiction of ictus and accent. This appears to be a significant phenomenon.
In these eight cases, the accent fell where the word would have been accented if it didn’t have a –que attached. All of them contain a light syllable before the que (as opposed to, say, inconcessósque of line 651). When the expected accent fell on the fourth-last syllable, –que was elided with the next syllable (but then again, the rules of hexameter would make that necessary).
Interestingly, while preparing this data, I found that the sixth foot also normally has an accent on the ictus, with only one line out of the 650 control lines being exceptional (141). However, out of the nine lines which contained a sixth foot –que, two had the ictus fall in an unexpected place. These were:
With such a small sample size, I am hesitant to draw conclusions from these sixth foot examples. However, they are interesting in that, even though they contain a heavy syllable before the –que, the accent appears to fall where it would have been on the word without a –que. I would be interested to find out if there are any instances in the rest of Vergil’s Aeneid of a sixth foot ending in the pattern of ármaque (with a short syllable before the -que).
My general conclusions at this point are that the most sensible advice for accenting words with –que is mostly what is suggested by Allen in Vox Latina. In short: the accent moves to the syllable before –que only if that syllable is (or has become) heavy (bellóque); otherwise, the normal accent of the word is retained (ármaque). It is just possible (on very slight evidence) that a composite with -que ending in four light syllables might make take on a secondary accent on the enclitic itself (líminaquè). At any rate, in Vergil, the apparent accent seems to fall most often on either a heavy syllable before –que or on the originally accented syllable of the word.
Bonus: A cringe at recent scholarship
While digging through scholarship for this post, I found an article from as recently as 1997 which seems to have somehow completely missed a century of debate about accent of words with the enclitic –que. Haike Jacobs, in Latin Enclitic Stress Revisited (1997), begins his abstract with: “The well-known fact that Latin enclitics induce stress on the immediately preceding syllable…” [emphasis added].  The main introduction then begins with:
In Latin, enclitics such as –que, ‘and’ always induce stress on the immediately preceding syllable. This fact, well documented both by Latin grammarians (most notable Priscian) and by detailed 19th-century philological studies such as those of Corssen (1870) and Lindsay (1894), has been analyzed in metrical theory in a number of different ways… [emphasis added].
The whole body of the article then goes on to examine the implications of this “fact”, using it to test the reliability of metrical theories by how well they can explain it. It is extraordinary that a linguist who wrote on this subject in 1997 (and his reviewers!) seem not to have read any articles from the preceding century on the controversy surrounding Latin enclitic stress, relying on only a late antique grammarian and nineteenth century scholarship. With an article titled Latin Enclitic Stress Revisited one would have expected some awareness of developments on the topic in the last hundred years.
Vox Latina, W. Sidney Allen, Chapter 5 (pp.87-88):
When an enclitic (-que, -ve, -ne, -ce) was added to a main word, the resulting combination formed a new word-like group, and a shift of accent was therefore to be expected in some cases: thus, for example, vírum but virúmque (like relínquo). Such a shift is discussed by many of the grammarians, but is then generalized into a rule that when an enclitic is added the stress always shifts to the last syllable of the main word (e.g. Varro, cited by Martianus Capella, iii, 272: ‘… particulas coniunctas, quarum hoc proprium est acuere partes extremas vocum quibus adiunguntur’): thus, for example, Musáque, limináque, where the accented syllable is light and would not normally receive the accent if the combination were treated like a single word. The application of this rule to light syllables is expressly discussed by Pompeius (K. v, 131); but in fact nearly all the examples quoted by the grammarians are of the type virúmque – almost the only examples of the accented light syllable are the two cited above, which appear in more than one source.
It has been suggested that the general rule is in fact a grammarians’ rationalization (perhaps with some ‘squinting’ at Greek Μοῦσά τε, etc.), and that the accentuation of e.g. Musaque was Músaque. This is supported by the fact that such combinations are commonly found in the fifth foot of hexameters, where we expect agreement between ictus and accent (and similarly in the fourth line of sapphics). In the case of liminaque, etc., the expected accentuation would be limínaque; but it is possible that in combinations of this pattern the accent of the main word was maintained, perhaps with a secondary accent on the enclitic; one may note the common Vergilian pattern líminaquè laurúsque…, etc. (note: on these questions see especially C. Wagener in Neue Philologische Rundschau (1904), pp. 505 ff.)
One cannot, however, exclude the possibility of an analogical accentuation of the type bonắque after the pattern of bonúsque, etc. Priscian specifically mentions such an analogy in the case of the fused compounds utrắque, plerắque, after utérque, plerúsque (K. ii, 181: ‘communis trium vult esse generum’). But it is doubtful whether these analogies apply to the classical period.
Allen, W. Sidney. Vox Latina. 2nd ed. Cambridge [England]: Cambridge University Press, 1965.
Jacobs, Haike. “Latin Enclitic Stress Revisited”. Linguistic Inquiry 28, no. 4 (1997): 648-661.
Newcomer, Charles. “The Effect Of Enclitics On The Accent Of Words In Latin”. The Classical Journal 3, no. 4 (1908): 150-153.
Tucker, R. Whitney. “Accentuation Before Enclitics In Latin”. Transactions And Proceedings Of The American Philological Association 96 (1965): 449-461.
 W. Sidney Allen, Vox Latina, 2nd ed. (Cambridge [England]: Cambridge University Press, 1965): Ch.5, p. 83.
 Charles Newcomer, “The Effect Of Enclitics On The Accent Of Words In Latin”, The Classical Journal 3, no. 4 (1908): 150.
 Ibid. 151.
 Ibid., 153.
 R. Whitney Tucker, “Accentuation Before Enclitics In Latin”, Transactions And Proceedings Of The American Philological Association 96 (1965): 449-461.
 Sidney Allen, Vox Latina.
 Ibid., 456, footnote 27.
 In addition I found nine lines in which the fifth foot consisted of a monosyllable followed by disyllable. I did not count these lines as lacking an accent on the ictus, as I regarded the monosyllable to have provided an accent on the ictus of the fifth foot. These mono + disyllable combinations are: I.199 his quoque, 290 hic quoque, 328 o dea, 380 ab Jove, 407 tu quoque, 445 nam fore, 498 per iuga, 592 aut ubi, 734 et bona.
 Pronounced Lāvīnjăquĕ. The poets occasionally treat what is normally an i-vowel as consonantal; see Sidney Allen, Vox Latina, Ch.1 vii p. 38: “The close connexion between the vowel and consonant sounds of i in Latin is also seen in the occasional poetic interchange of their functions – e.g. on the one hand quadrisyllabic Ĭulius and on the other trisyllabic abiete (with i-consonant ‘making position’).”
 Haike Jacobs, “Latin Enclitic Stress Revisited”, Linguistic Inquiry 28, no. 4 (1997): 648.