Do we know how 'ng' was pronounced in classical Latin?

Question

How was 'ng' pronounced in classical Latin and how do we know? I believe metric considerations strongly indicate that it was not a short consonant (/ŋ/ or other), but I can still think of two reasonable (and many unreasonable) options: /ŋg/ and /ŋ:/. The second option is widely used in today's Finland — in all languages, not just Latin.

In Greek the corresponding — at least roughly corresponding — sound was spelled 'γγ' instead of 'νγ'. In Latin, however, 'gg' denotes /g:/ unless I am badly mistaken. This makes me wonder: Was 'ng' pronounced any differently in Greek loanwords than in originally Latin words? If the pronunciations in the two languages were identical, I assume the answer is negative.

Please do not confuse this question with the one about pronouncing 'gn'.

Answer 1

The first thing worth noting here is the pronunciation of n before certain consonants, including g:

Before a velar or a labio-velar, however (as in uncus, ingens, relinquo, lingua), [n] stands for a velar nasal [ŋ] (as in English uncle or anger). [Allen, Vox Latina, 27]

One way we know this is from the descriptions of the sounds provided by Nigidius Figulus, who was quoted by Marius Victorinus ("intermediate between n and g"; see Allen, 28 and Tafel, Latin Pronunciation, 46).

Furthermore, Accius attempted to associate the sound with the g character instead of n, following the Greek practice, by writing aggulus for angulus, for example (Allen, VL, 28; Allen, Vox Graeca 35ff., Haldeman, Elements, §242–43). And apparently the Greeks did the same thing in reverse: Haldeman cites Plutarch using gamma to represent the n in principia (§246), indicating the similarity of the sounds between the two languages.

From this, then, we can have confidence that the n of the ng pairing is best represented by /ŋ/ and that therefore the entire pairing as /ŋg/, as pronunciation guides indicate:

Latin ng corresponds to the double sound ('ng' + g) in English finger, and not the single sound heard in singer [...] Greek γγ = Latin ng = English or Welsh ng + g. [Arnold and Conway, 11, 13]

Here ng = ng + g; congero = cong-gero. [...] Lat. ng is never like ng in Eng. singer or hanger. [Westaway, §37]

Answer 2

For Classical Latin, I have never seen any pronunciation given for <NG> aside from a cluster consisting of a nasal consonant followed by a voiced plosive. I would assume the reason a long nasal consonant /ŋ:/ is used in Finland is simply because that's how the digram "ng" is pronounced in Finnish.

For example, the word pango has the reconstructed Classical Latin pronunciation [ˈpaŋɡoː].

Classical Latin did not have a phonemic distinction between velar and palatal plosives. Words like gingiva are analyzed as containing the phoneme /g/, but it's likely that in this kind of context, the voiced plosive transcribed as /g/ was phonetically fronter (or in other words, more palatal) than the /g/ in a word like pango. So it's possible that the phonetic form of a word like gingiva might have been something like [g̟iŋ̟ˈg̟iːwa] or [ɟiɲˈɟiːwa]. However, allophonic fronting of velar plosive phonemes next to front vowels is often not transcribed even in phonetic transcriptions, because to speakers of languages like English and French the front quality of a plosive in this context is not perceptually salient.

Classical Latin is also usually analyzed as lacking a velar nasal phoneme, so strictly speaking there is no /ŋ/ here or in any Latin word. In this context, sounds like [ŋ], [ŋ̟] and [ɲ] did not contrast with [n], so the usual phonemic analysis of the cluster is /ng/. Some phonological analyses make use of entities known as an "archiphonemes" to account for lack of contrast in certain specific contexts; if we use the symbol |N| to represent a nasal "archiphoneme" in Latin, we might instead say that the phonological form of the cluster is |Ng|.

the case of "-ngu-"

The situation is basically the same for non-syllabic <NGU> in words like lingua, although there are some arguments about the phonemic analysis of non-syllabic <GU> and the allophony may have worked a bit differently.

In theory, we could analyze such words as containing a unitary phoneme /gʷ/. Allen (Vox Latina) somewhat hesitantly endorses such an analysis, by analogy with voiceless /kʷ/, for which it is easier to make a case for a single-phoneme analysis (p. 25). However, I believe most linguists prefer to analyze Latin non-syllabic <GU> as a cluster /gw/, on the grounds that a phoneme /gʷ/ would have an unusually highly restricted distribution. In Latin words, non-syllabic <GU> only occurs after N (in around 11 words and their derivatives) and after R (in the single word “urgueo”, a variant form of urgeo) ("Aspects of the Phonology and Morphology of Classical Latin", András Cser (2016), p. 17).

For words with <NGU>, allophonic palatalization from a following front vowel may have affected the offglide. Allen cites a passage in Priscian in support of the proposition that the labial element represented by <U> in <QU> and <NGU> may have had a front articulation before a front vowel (p. 17).