More on the collapse of contrastive vowel length in late Latin: a review of the metalinguistic evidence

1 Introduction: vowel length from Classical Latin to Proto-Romance

The exact way in which the demise of contrastive vowel length (henceforth CVL) occurred is an ever-topical vexed question in Latin and Romance linguistics.^[1] In (1a) and (1b) we schematize the two main contemporary opposing views (see, e.g., Bramanti 2022: 401; Mari 2021: 194; Probert 2021):

The first ([1a]), which dates back to at least Schuchardt (1866–1868; see also e.g. Loporcaro 1997: 55–70, 2015: 18–60; Marotta 2022: 186; Papini 2024: 49; Weinrich 1958: 181–182), suggests that the demise of quantity, which possibly spread out of African Latin (cf. Schuchardt 1866–1868: III, 43 and especially Herman 1982: 229, 1990; Mari 2021: 195–199), occurred through the establishment of an allophonic rule of open syllable lengthening (henceforth OSL) that was in force in Proto-Romance and is maintained in Italian today.

The alternative position ([1b]), upheld by Mancini (2014: 977–978, 2015b, 2019, followed by De Angelis 2022: 439, n. 5),^[2] maintains instead that the loss of CVL was accompanied by a generalized lengthening of all stressed vowels, in both open and closed syllables, which occurred during the imperial age:

The very first outcome of the collapse of vowel quantity in spoken Latin was a neutralization, which surfaced as a general lengthening of stressed vowels, both in heavy and light syllables, a structural representation still encoded in the conservative Western and Eastern varieties of Romania. (Mancini 2019: 47; see also Mancini 1994: 625 and 2001: 319)

The evidence that can be brought to bear is mainly of two kinds: on the one hand, the deviations from classical prosody that are observed in metrical epigraphs from the imperial age; on the other, the metalinguistic evidence offered by grammarians, particularly that contained in the so-called third part of the Artes Grammaticae, which was dedicated to vitia et virtutes orationis (on which see at least Baratin and Desbordes [1986] for an introduction). Further evidence, somehow intermediate between the two kinds just cited, can be gained from the metalinguistic accounts of the composition of metrical texts, the data of which must be interpreted in the light not only of the “prosodic revolution” we are concerned with, but also of the progressive fading of the knowledge of classical metrics (and, hence, would require separate treatment).

Here we will focus exclusively on the second type of evidence, reviewing only the non-metrical testimonies. We will scrutinize grammatical passages from the 3rd to the 5th century CE, i.e. the period that, according to Herman’s (1998) chronology, preceded the fragmentation of the Romance languages. A first observation is that the grammarians’ testimonies are consistent with both positions (1a) and (1b) in that they offer numerous examples of shortening of unstressed vowels, both in open and closed syllables.^[3] As for stressed syllables, a collection of passages from Latin artigraphers was provided by Mancini (2019), whose list is reproduced in the synopsis in Table 1.^[4] We have classified the testimonies according to the syllabic-prosodic structure of the words at issue (horizontally) and to whether they indicate lengthening or shortening with respect to the expected VL specification in classical Latin (vertically).

As is apparent from Table 1, neither of the views ([1a], [1b]) is confirmed straightforwardly by the data, as the cases of stressed vowel lengthening in closed syllables (Table 1, a.ii) contradict (1a), while, conversely, vowel shortening in open stressed syllables (Table 1, b.i) is at odds with both (1a) and (1b).

Based on a thorough analysis of the passages, Mancini concludes that they are compatible only with the hypothesis of a generalized vowel lengthening in stressed position independent of syllable structure ([1b]). Our discussion will examine anew the relevant testimonies, expanding, with few omissions,^[5] on Mancini’s collection and placing the passages within the artigraphic tradition to which they belong. It will be shown that Mancini’s conclusion cannot be upheld, and the opposite is the case, i.e. that the evidence rather supports scenario (1a).

The article is organized as follows. After briefly recalling (Section 2) a caveat that one needs to keep in mind while investigating the grammarians’ metalinguistic testimonies, in Sections 3–5 we move on to the close reading of such testimonies in chronological order (with a few motivated exceptions) to demonstrate the thesis that, if read correctly, the passages of the grammarians are only compatible with hypothesis (1a), not (1b); in Section 6 we depart from the chronological order and move on to consider the independent testimony provided by an author removed from the artigraphic tradition, Aulus Gellius (2nd c. CE), and that by a last grammarian, Virgilius Maro Grammaticus, who wrote at the very end of the history of Latin. Finally (Section 7), we offer our twofold conclusions, on the one hand reasserting the superiority of (1a) as an explanation for the demise of CVL, and on the other hand proposing two general principles of method that can guide the use of the artes grammaticae as sources for the reconstruction of pre-Romance Latin.

2 A(n obvious) caveat about the Roman grammarians and spoken Latin

Before proceeding any further, let us stress an obvious caution that must be adopted: of course, the materials drawn from ancient grammars do not always reflect the coeval linguistic situation. While Barwick’s (1922) idea that the Artes Grammaticae were highly stereotypical and thus incapable of providing any insight at all into the linguistic repertoire of the time is exaggerated and now outmoded, a wealth of studies have shown that much of this corpus rests on “an accumulated body of grammatical lore” that “was handed down from one teacher to another” (Law 1982: 13), so that

each extracted from it what he found useful. Such innovation as there was consisted only of the choice of sources to be followed and of the degree of integration to which material from various sources was subjected. The presentation and ordering of material afforded some room for individual variation, but even that was subject to the traditional schematisation which was derived ultimately from the Stoics and Dionysius Thrax. (Law 1982: 13)

Thus, by far not any form stigmatized by the grammatical tradition can be attributed to pre-Romance Latin. To ascertain whether such forms should be attributed to the repertoire of contemporary Latin speakers, one must bear in mind that

[i]n alcuni passi gli artigrafi si limitavano a citare forme riprese da una lunghissima tradizione precedente, forme che molto probabilmente non avevano ormai corrispondenza alcuna con quanto si stava verificando nel latino parlato. In altri, pur alludendo a forme riscontrabili nel parlato contemporaneo, le riprendevano da testi anteriori epitomati o commentati. In altri casi ancora – certo meno frequenti – si trattava di citazioni episodiche tratte dall’esperienza diretta del grammatico/parlante [in some passages, the artigraphers merely quoted forms drawn from a very long earlier tradition, forms that most probably had no correspondence by then with what was occurring in spoken Latin. In others, while hinting at forms found in contemporary speech, they took them from earlier epitomised or commented texts. In still other cases – certainly less frequent – they were episodic quotations taken from the direct experience of the grammarian/speaker]. (Mancini 2005: 139–140; cf. also De Paolis 2010: 57)

Against this background, in order to sift the grammarians’ testimonies and come up with the tiny subset of the forms they stigmatize that have a chance of being attributed to the verbal repertoire of contemporary Latin speakers, one must first exclude examples that the grammarians draw from texts of classical auctores rather than from everyday Latin. Moreover, sometimes the forms stigmatized do come from everyday Latin, yet not from the language of the writer’s own time, but rather from earlier stages in the history of Latin, and grammarians repeated them by drawing on the exemplification of their colleagues from previous periods. Finally, careful attention must be paid to the fact that ancient grammatical theorization also differs significantly from today’s linguistic science and, therefore, the data that the artigraphers hand down must always be examined in the light of the context in which they are placed.

In short, at the cost of stating the obvious, one must stress that the evidence delivered by grammarians must always be weighed up and sifted, in order to gain information on contemporary spoken Latin. This is what we are going to do in the following sections.

3 At the beginning of the artigraphic tradition: Plotius Sacerdos and Marius Victorinus

Our review begins with Plotius Sacerdos, probably at the end of the 3rd c. (Bramanti 2022: CV–CVIII; Kaster 1988: 352; Zetzel 2018: 318). In the chapter of his grammar entitled De barbarismo, he lists eight types of faults which can occur in the spoken (proper barbarismus) as well as in the written language (barbarolexis). Only three of them are possibly relevant to the issue of VL: they are productio ([2]), correptio ([3]), and immutatio accentuum ([4]).

If all three passages could be relied upon as proofs of contemporary sound changes, we would have one example ([4]) compatible with both reconstructions (1a) and (1b), one ([2]) compatible only with (1b), and one ([3]) compatible with neither (1a) nor (1b). However, none of them holds up to closer examination.

As for (2), i.e. the lengthening of the first vowel of pernix in closed stressed syllable, scholars usually compare (and so do we, in violation of the chronological order) a passage from the grammarian Pompey (5th c.; see [14], Section 4) attesting to the lengthening of the stressed /a/ of arma in a closed syllable ([5]):^[6]

Pompey’s testimony leaves room for at least some uncertainty as to whether the example was actually drawn from a spoken variety of Latin;^[7] however, even setting aside for the moment the problem of “flag-words” (on which see Section 4), if we frame (2) and (5) within the whole diachrony of Latin, we find that they do not indicate a redistribution of VL in the imperial age, pace Mancini (2019: 37), but rather are a belated testimony to a late-Republican sound change, i.e. the lengthening of vowels before /rC/ and particularly /rN/. As shown in (6), this lengthening is attested in the epigraphic record through the use of apex or gemination, but did not find its way into late Latin and, hence, into the Romance languages (with the possible exception of Italian forma [ˈforma] < lat. forma [ˈfoːrma] rather than [ˈfɔrma]):

Neither Sacerdos nor Pompey thus provide a description of coeval Latin, but rather at best record an innovation that was older and, more importantly, recessive at their time (though it had perhaps established itself in a few lexemes, such as forma). We must then conclude that neither (2) nor (5) can be counted among the evidence regarding the redistribution of VL in the imperial age.^[8]

As for (3), i.e. the apparent shortening of the second vowel of steterunt in stressed open syllable, both the provenance of the example and its linguistic interpretation suggest that it should not be given credit as a reliable testimony to late-Latin VL. On the one hand, Sacerdos found his example in the text of Vergil, where this scanning of steterunt occurs twice (Aen. 2.774, 3.48). As a matter of fact, he mentions this very example also as a case of systole ‘shortening’ (VI 452.16 K), a type of metaplasmus related to VL, and his interpretation was followed by Pompey (III 64) and the anonymous Auctor ad Caelestinum (259.10–11). On the other hand, neither (1a) nor (1b) could account for the change Sacerdos illustrates, if it were such. In fact, it does not refer to an alleged **[steˈtɛrunt], but rather, obviously, to the archaic [ˈsteːterunt] (Bramanti 2022: ad loc.), with antepenultimate stress because of the retention of the original ending -ĕrunt, a form documented in Archaic Latin (see e.g. Plaut. Bacch. 928, Most. 281), which coexisted with classical -ērunt (and -ēre), resurfaced in Classical Latin poetry for metrical requirements, and lives on in Romance (e.g. It. dìssero, Fr. dirent < dīxĕrunt, It. fecero, Fr. firent < fēcĕrunt; see Leumann 1977: 608).

Of all the examples quoted by Sacerdos, therefore, all that remains is the immutatio accentuum of Ceres ([4]), reporting lengthening in stressed open syllable and shortening in closed syllable. As already stated above, this is compatible with both (1a) and (1b). Actually, even this piece of evidence could be doubted, since Ceres is a flag-word. The issue of flag-words is not critical here, but it is crucial in assessing the weight of the evidence provided by the scholastic tradition that sprang from the teaching of Aelius Donatus, active in Rome during the fourth century. Hence, it may be useful to introduce it here before moving on to the next author. Flag-words have been aptly defined by Mancini (in a different context):

nel metalinguaggio degli artigrafi venivano impiegate alcune parole-bandiera che troviamo costantemente ripetute nei diversi trattati, un po’ come ancor oggi in una qualunque lezione o manuale di linguistica comparativa si lavora magari su pochi esempi particolarmente significativi memorizzati e memorizzabili in modo agevole [in the artigraphers’ metalanguage some flag-words were employed that we find constantly repeated in the various treatises, much in the same way that even today, in any lecture or textbook of comparative linguistics, we may work with a few particularly significant examples that get memorized and easy to remember]. (Mancini 2001: 316; original emphasis)

They are specific recurrent words that were automatically paired with the phoneme(s) they were supposed to illustrate and thus constituted an effective mnemonic tool for school teaching. It can be assumed, therefore, that once certain words were biuniquely associated, for teacher and learner, with certain phonemes, they could just as easily be reused to illustrate any mistakes in their pronunciation.

As a matter of fact, the occurrence of Ceres in (4) is not isolated in the grammatical tradition, but Sacerdos himself had used it earlier in his work to exemplify the prosodic structure of iambic disyllables and Diomedes would use it for the same purpose ([7a] and [7b]):

Even for this example, therefore, there is a legitimate suspicion that it was not included because it aptly reflected the coeval language, but due to the requirements of symmetry and correspondence on which the Artes grammaticae were typically built.

Similar problems are involved in three passages from Marius Victorinus’ Fragmentum de barbarismis et soloecismis (end of the 4th c.; cf. Kaster 1988: 437; Zetzel 2018: 328) ([8]–[10]):^[10]

Example (8), the apparent lengthening of an unstressed vowel, is clearly drawn from the poetic tradition, once again from Vergil’s Aeneid (1. 498: exercet Diana choros, quam mille secutae), and is indeed typically mentioned in the grammatical tradition as an example of metaplasmus, more specifically, through ectasis ‘lengthening’ (though indeed ī was long etymologically; see Mari 2021: 152; EDL 168); the shortening of the second syllable of emerunt ([9]), instead, is linguistically identical with Sacerdos’ steterunt ([3]): both examples can be safely dismissed as uninformative for the reconstruction of spoken late Latin.

As for (10), it attests to the erroneous pronunciation of the verb rĕpōno, with the first syllable long and the second short. Mancini (2019: 40; cf. also Mari 2021: 193, 195) interprets this passage as witnessing the shortening of a stressed vowel, but this explanation would require admitting the simultaneous lengthening of a vowel in unstressed (open) syllable, an unparalleled phenomenon at odds with the phonetic tendencies of Latin at all stages.

A better interpretation is suggested by the fact that in imperial Latin the stress on prefixed verbs that had a short root vowel in the penultimate open syllable was shifting from Classical Latin cómmŏvet, ḗmŏvet to late Latin/Proto-Romance commŏ́vet, emŏ́vet, without retraction on the prefix, as Italian commuove or French émeut ‘s/he moves’ etc. still show today.^[11] In view of this, we will read rḗpŏno [ˈreːpɔno(ː)] and interpret it as a case of hypercorrection reacting to the ongoing change, not involving a shortening of the vowel in a stressed penult, but rather a lengthening of the vowel in an open syllable (the first) that had become stressed, and a shortening of the vowel in an unstressed syllable (the second), which was originally stressed. This is compatible with both theses in (1a) and (1b).

4 The Donatian tradition

Many of the further testimonies to which we now turn, following chronological order, belong to the Donatian tradition. Donatus wrote an Ars minor, an introductory school handbook about the partes orationis, and an Ars maior, which was more extensive and advanced. Both works soon became best-sellers and played a crucial role in the subsequent development of grammatical tradition (see e.g. Holtz 1981; Zago 2022). The third book of the Ars maior begins with a chapter De barbarismo providing a sketchy illustration of the matter: the grammarian first of all describes the two levels at which a barbarism may occur (pronuntiatione et scripto), its types (species: adiectio, detractio, immutatio, transmutatio), and the five categories of items (res) that are subject to it (litterae, syllabae, temporis, toni, adspirationis), to then move on to exemplify their various potential combinations, offering examples of most of them. As far as VL is concerned, Donatus provides four examples ([11a]–[11d]):

They show the lengthening of an unstressed vowel ([11a]), a stress shift aligned with the shortening of a formerly stressed vowel ([11b], [11c]), and the lengthening of a stressed vowel with the simultaneous shortening of an unstressed vowel ([11d]); no examples of barbarismus toni are included (“Pour un désir de faire bref” [for a desire to keep it short] according to Holtz [1981: 144]).

The first three examples have no bearing on the reconstruction of the development of VL in imperial Latin, as they are drawn from poetic language (unsurprisingly, the Aeneid: see [12a]–[12c]) and had already been employed in grammatical treatises (e.g. Quint. I 5.18):

The sole case of deos ([11d]) does not seem attributable to a specific poetic context. However, despite the trustworthiness autoschediastically accorded to it by Sedulius Scottus (Comment. in Don. artem mai., 330.91–92 Löfstedt 1977: Et notandum, quod hoc exemplum non poetica auctoritate, sed e vulgari consuetudine sumpsit), it is unlikely that Donatus resorted to coeval language for this example alone. The testimony, then, must be taken cum grano salis.^[12]

The later grammatical tradition tried to remedy the paucity of examples in Donatus’ treatment. First Servius (fl. 4th c. ex.), to whom we owe the canonization of Donatus’ text, took up the chapter De barbarismo (Zago 2016, from which we quote) and repeated some of his predecessors’ examples ([13c], [13d]). What is more, Servius also added two examples unattested in previous grammatical treatises ([13a], [13b]):

The last two examples ([13c] and [13d]) concern metrical licenses rather than coeval sound changes and coincide with Donatus’ ([11a]) and Sacerdos’ ([3]) examples, respectively. The other two ([13a] and [13b]) are the shortening of the first vowel of Roma in a stressed open syllable and the mirror case of the lengthening of the first syllable of rosa, which deserve further discussion.

Later Pompey (probably African, 5th c.), himself belonging to the Donatian tradition and very close to Servius (Holtz 1971: 48 n. 5; Zago 2017: I XCV–XCVI), retrieved the by-now traditional examples Italia ([14b]) and steterunt ([14d]), but also included words very close to Servius’ innovations, introduced as barbarisms in pronuntiatu ([14a]), and the case of arma discussed above ([14c] = [5], Section 3):

While (14b) and steterunt in (14d) can be easily dismissed as traditional examples from poetry (see once again discussion on [3] and [11]), Roma in (14a) and (14d) is one of the most controversial passages in our dossier due to serious problems with the transmission and constitution of the text. Yet, the latest editor of Pompey’s third book, Zago (2017: ad loc.; but see already Zago 2013 with further bibliography), proved that Pompey’s two examples concern the mispronunciation of the nouns Roma and equus, the former with a short stressed vowel, the latter with a long stressed vowel.

While neither the lengthening in rosa nor in equus require particular explanations inasmuch as they can be intuitively reconciliated with both (1a) and (1b), the case of Roma has attracted much scholarly attention. If Roma with the shortening of the first vowel were to be taken at face value as indicating a phonetic fact, it would be compatible with neither (1a) nor (1b) and should hence appear problematic to Mancini as well. Nonetheless, Mancini (2019) attaches great importance to this example and interprets it as a hypercorrection, allegedly supporting his thesis of vowel isochrony ([1b]):

in fact, if the R[omance]Q[uantity] rule applied, within a very predictable pattern of syntagmatic distribution of syllabic quantity and its consequent shortening of long vowels in closed syllable and lengthening of short vowels in open syllable, the shortening of old, long vowels in open syllables would not make sense. (Mancini 2019: 41)

Actually, if one were to appeal to hypercorrection, one would have to admit that this would be also compatible with the opposite thesis ([1a]): it is easy to imagine that a speaker/writer, noticing the widespread tendency to lengthen vowels in open stressed syllables, reacted to this tendency by shortening them indiscriminately, including those that were etymologically long.

However, the problem is cut off at the root when one observes that the examples of barbarism chosen by Servius and Pompey are not random, but also occur in another section of the Ars in Donatus’ tradition, namely in the first part, which was devoted to the description of the sounds of Latin. In this section, each phoneme is associated with a word illustrating it. Thus, Servius distinguishes the pair ŏ ≠ ō through the nouns rosa and Roma/orator ([15a]), and similar examples are also provided by the so-called Sergius (to whom the Explanationes in Donatum I and II have been attributed; cf. De Paolis 2000; Zetzel 2018: 321–322) and Pompey ([15b] and [15c]):

The examples of barbarisms cited by Servius and Pompey (and particularly the unusual Roma with a short first vowel) contain what, following Mancini (2001: 316), we have defined as flag-words (see the discussion of [4] in Section 3). Their choice can be motivated by the tendency not to introduce new words and to resort to stock examples already provided in the first part of their work.

Thus, the thorough examination of the relevant passages, not taken individually but framed within the whole of the work to which they belonged, allows for the conclusion that all the examples discussed so far should not be interpreted as reflecting a phonetic change taking place at the time of the grammarian who cites them, but rather as depending on an arrangement of grammatical theory based on the demands of internal symmetry and correspondence. Of course, nobody would deny that the abundance of examples in the Donatian tradition derives from a progressive fading of the ability to correctly assign VL, as amply documented in various testimonies by late antique grammarians (see e.g. Ult. syll. gramm. IV 222.28–30 K; Serg. IV 522.24–27 K; De primis, mediis et ultimis syllabis 174–6, Corazza 2011) who at times propose tricks to determine it (see e.g. Diom. gramm. I 347.19–22 K; Ad Basilium amicum Sergii VI 242.20–31 K). Certainly, however, one cannot take all these examples as direct documents of the ways in which this fading was realized at the phonetic level. Thus, the probative value of both rosa, equus and, especially, the aberrant Rŏma is considerably reduced.^[13]

At this point, a first conclusion can be drawn: the entire grammatical tradition examined so far, whether it is Sacerdos or the Donatian strand of Schulgrammatik, does not provide reliable data about the concrete ways in which the redistribution of VL occurred in Late Latin, which will come as no surprise, since as early as Quintilian ([16]) the grammarians themselves point out the fictitious character of many of the examples they offer (apart from those that they drew from poetic texts), and Pompey himself ([17]) admits that many examples could disprove the rules laid out in the grammar:

As far as reconstruction is concerned, then, the text of professional grammarians is itself a vivid testimony to the uncertainty in the assignment of VL, but provides no real data about its allophonic (re)distribution.

5 Two “heterodox” grammarians: Consentius and Augustine

In light of the foregoing discussion in Sections 3–4, it is easy to understand why the method of selection and illustration of the exempla employed by the grammarians was the object of criticism on the part of some ancient authors who departed from the artigraphic tradition. Among these Consentius and Augustine also provide valuable evidence on coeval changes in the vowel system, given their autonomy with respect to the practice of their “colleagues”. Such, admittedly rarer, testimonies turn out to be all the more significant against the background of the general unreliability – for our present purposes – of the grammatical tradition reviewed in Sections 3–4.

6 At the two ends of the “second dialectalization”: Aulus Gellius and Virgilius Maro Grammaticus

7 Conclusion: the artigraphic evidence and the demise of Latin CVL

We are now in a position to take stock of the foregoing discussion. In Table 1 we reproduced Mancini’s (2019) catalogue of the relevant testimonies. As we tried to show, though, testimonies by ancient grammarians that deal with deviations from the classical norm possibly related to VL are far more abundant. Our analysis has taken into account 31 passages against Mancini’s 15. Yet we have also argued that the grammarians’ remarks cannot simply be collected and compared with one another, but require careful sifting to assess their place in a very conservative tradition, the provenance of their examples (that is, whether they are extracted from metrical texts, selected for reasons of consistency throughout the entire work, or taken from contemporary language), and the linguistic information they provide. Close reading of the passages has made it clear that in most instances Latin grammarians, especially when sticking to the orthodox methods and practices of the artigraphic tradition, hand down no reliable testimonies about the demise of CVL. Modern appraisal of the data they provide must be wary of the difficulties involved in handling these texts. Our review has revealed that nine testimonies refer to metrical texts ([3], [8], [11a]–[11c], [13c], [13d], [14b], [14d]), five (or perhaps six) are concerned with other linguistic phenomena ([2], [5], [9], [14d], [24], [3]?), and seven (or eight) contain flag-words ([4], [13a], [13b], [14a], [14d], [25], [26], [5]?). Table 2 collects all and only the passages that we feel have some chance of describing the development of VL in pre-Romance spoken Latin (the testimonies also included in Table 1 are in bold, even when our interpretation requires putting them in a different category).

Here is Mancini’s (2019: 40) analysis of his collected data:

After our discussion, this assessment must be modified as follows:

Mancini’s three categories ([34a]–[34c]) are all heavily downgraded from our investigation, since most of the artigraphic testimonies, except Consentius’s and Augustine’s remarks, do not withstand closer examination; on the other hand, Gellius’ and Virgilius’ accounts are to be added to the list, thus introducing a new category ([35d]). Globally, the relevant evidence thus sifted is only compatible with scenario (1a), that is, with the hypothesis that Latin CVL was ousted by the rise of an OSL allophonic rule: shortenings of long vowels in closed stressed syllables ([35d]) do not reconcile with (1b), whereas the dissolving of the evidence which would allegedly attest to “lengthening also in closed syllables” and “shortening of long vowels in open syllables” ([35b] and [35c]) shows that scenario (1b) makes an unnecessary assumption that is not needed to account for any of the relevant data.^[24]

This conclusion is in keeping with what we know independently about long-term trends in Latin phonology (see Section 6.1), where superheavy syllables tended to be eliminated and where – as demonstrated by all counts, based on both texts (see Kiss 1971; Marotta and De Felice 2023) and dictionaries (see e.g. Marotta and De Felice 2019: 448) – only open stressed syllables hosted a higher percentage of long vowels, while short vowels prevailed, to varying degrees, in all other contexts, thus foreshadowing the further Romance development and, in particular, the allophonic distribution determined by the OSL ([1a]). By contrast, scenario (1b) assumes (without valid evidence, neither textual – as we have shown, see Table 2 – nor structural)^[25] that at a certain point this long-term tendency was reversed and that new superheavy syllables were systematically created by lengthening short vowels even in stressed closed syllables, not only in open ones, as foreseen in (1a).

Our conclusion in favour of scenario (1a) is further corroborated by a crucial chronological argument, namely that that “certain point” when the long-term trend was reversed cannot actually be identified at any given moment in time. As seen in Section 6.1, Gellius describes in the 2nd century CE the harbingers of OSL: the latter is clearly operating in the imperial Latin of Africa (see Adams 1999: 114; Herman 1982, 1990), is well established in Rome at the end of the Western Empire (see Herman 1982, 1990), and is later described for Gaul in the 7th century by Virgilius Maro grammaticus. The same situation can be reconstructed for the pre-literary Old French of one to two centuries after Virgilius (Morin 2003: 121):

Diphthongization in (36c)–(36d) applied to a long vowel, which had been in turn lengthened through the same OSL described for the Latin of Gaul by Virgilius. Finally, OSL left unmistakable traces in the Romance varieties from the North Sea to Sicily and is still at work in Standard Italian, Central-Southern Italo-Romance dialects and Sardinian (Loporcaro 2015: 25–30).

Since all we have now said regarding the period from the 7th century to present rests on established facts (i.e. it does not rely on textual interpretation as strongly as the evidence – as valuable as it is – provided by Latin grammarians), proponents of (1b) must meet a non-negligible burden of proof. They assume that CVL did not collapse due to the emergence of the OSL, which is independently known to have arisen, but was lost when a completely different change (of which we have no independent evidence) occurred, which allegedly lengthened all stressed vowels regardless of syllable structure. To support their view effectively, they should answer the following questions: when exactly would such a generalized lengthening (also in closed syllables) have been established? And when did it yield to OSL?

We submit that such answers cannot be supplied. Given that (a) Gellius, in the 2nd century, shows the harbingers of OSL ([1a]), not of generalized lengthening ([1b]); that (b) this is the same situation that Virgilius Grammaticus depicts as having become the rule five centuries later; that (c) the Romance varieties of Italy and Gaul/France are known to have OSL, or to have had it in the past; and, finally, that (d) OSL is in itself logically sufficient to explain the demise of Latin CVL, the inescapable conclusion, in our view, is that there is no room in the chronology of Latin and the Latin-Romance transition for (1b) to ever have come into being. This change, in fact – unless one wants to postulate a back-and-forth of changes (without any evidence) – should have preceded Gellius’ time, but on the other hand it is reconstructed based on texts of artigraphers much later than Gellius, which is contradictory.

By contrast, our conclusion in favour of (1a) is free from contradiction and compatible with the chronology of sound change in Romance as well as with the data furnished by the artigraphers’ pertinent testimonies.

In addition to this conclusion on the specific issue of the demise of Latin CVL, we finally propose a conclusion on method, in the form of the two principles in (37) and (38), that are relevant to the study of the metalinguistic testimonies by Latin grammarians:

The first principle ([37]) states that the mere repetition of examples from the traditional stock is of no value to the diachronic phonologist when it comes to establishing and dating pre-Romance changes in the phonology of Latin, unless – of course – the information provided by the relevant testimony is backed up by independent evidence.

The second principle ([38]) gives all the more weight to the testimony of ancient grammarians that are demonstrably independent of previous tradition. In this way, while obviously recognizing the importance of evaluating individual testimonies locally, a hierarchy emerges whereby the testimony of authors outside the artigraphical tradition, such as Gellius, weighs more heavily than that of artigraphers of acknowledged originality, such as Consentius, and that of the latter in turn weighs more heavily than that of traditionalist artigraphers, particularly within the Donatian tradition, such as Servius or Pompeius. There is probably no need to stress that such hierarchies are local rather than global, and that authors that for some matters turn out to be unreliable may provide valuable material in other sections of their work.