2.1 Special factors

Misspellings may have nothing to reveal about the state of the language at a particular time, for a variety of reasons. A few factors will be listed here, but no attempt is made to be comprehensive.

Old deviant spellings linger on, sometimes for centuries, and may be picked up by anyone who is literate, and used to confer respectability on writing without having any basis in speech. If the poet Manilius spelt nisi in one place with e in the second syllable (see iii.7.1.1), he would not have been attempting to represent the sound of speech but using an old form. If the writer of an informal letter on papyrus in Egypt adopted the same spelling, we need not assume that he, by contrast with Manilius, did so because that is how he heard the second vowel. He, like Manilius, might have seen the old spelling somewhere and assumed it was particularly correct. Substandard texts often contain deviations from a norm that are not literacy errors but quite the opposite: they may be determined by the recollection of a rule imbibed by the writer from instruction he had received. For example, in the letters of Terentianus there are signs of the grammarians’ rule that [k] should be represented by q before u, k before a and c before other vowels (see Adams Reference Adams1977a: 32–3).

Mechanical writing errors are common. Compiling a typology of these would be an undertaking in itself, and we can only cite a few instances here.

Geminated consonants may cause a problem. A writer may be aware that the written form of a word should contain a geminate but double the wrong consonant (e.g. tyrrany for tyranny; cf. Turranium at Terentianus 468.54).

Writers may be influenced by a form that they have just written or one that they are about to write. A sequence of accusatives in -m may cause an -m to be attached to a following term that should be in the ablative. In inscriptions the expected -s of a third declension name in the genitive is often omitted in the neighbourhood of a second declension genitive in -i.

There may be a physical reason for a misspelling, such as lack of space at the edge of a stone as the inscriber reaches the end of a line. This factor often effects the omission of a final consonant as the stonemason runs out of space (for details see viii). Since some final consonants were lost in the transition from Latin to the Romance languages, there may be a temptation to treat an omission as linguistically significant when it is due to this extraneous factor.

2.2 Varying frequencies of misspellings

Phonetic misspellings are variable in frequency from type to type, and difficult to interpret if they are not numerous. Also difficult to interpret is the absence, or virtual absence, of a misspelling that would be taken, if it had occurred, to represent a known development of the language. Should we conclude from the non-appearance of a misspelling in the inscriptions and documents of a particular period that the phonetic change that would have inspired it had not yet taken place, or alternatively that writers were aware of the change and careful to avoid representing it in writing? Some different patterns will be illustrated here, and the methodology of assessing such evidence discussed. A detailed account of the evidence will be found in the appropriate chapters later in this Part.

(i) Certain misspellings are so frequent that there can be no doubt that they reflect the state of the language. Cases in point are the omission of -m (viii) and the writing of ae as e (iv.2.2). But the state of what varieties of the language? Those spoken by a restricted educational/social class, or those spoken by the majority of the population? This is a question that cannot be answered merely from an examination of texts and their misspellings or absence thereof, because good spellers will stick to traditional spellings whether they are an accurate reflection of their own speech or not. If, roughly speaking, we are to place the pronunciation lying behind a misspelling in a particular social class, we need additional evidence, such as remarks by grammarians or other speakers. There is such evidence for -m that is clear-cut (viii.1.1). The information about e/ae is less straightforward, but there are indications that for a while grammarians attempted to preserve the diphthongal pronunciation (that is, among their clients, the highly educated) (iv.2.3). That would seem to imply the coexistence of two different pronunciations, with the monophthong presumably the norm down the social scale but the diphthong preserved by some purists. However, this duality may be idealised. Although grammarians might have been ‘guardians of the language’, their resistance to change in a particular case might have had limited aims. Were they always setting out to uphold the old ways in the speech, even informal, of their pupils, or sometimes simply trying to preserve a traditional articulation of the poetry that was the medium of their instruction? It will be suggested later that their efforts to maintain a quantitative vowel system related particularly to the recitation of poetry, and that even in their own speech the quantitative system was under threat (iii.6). When ae was monophthongised it soon became a form of short e, as early hypercorrect spellings show, and again the preservation of the diphthong might mainly have been for the sake of the reading of poetry.

(ii) Some misspellings are very infrequent, and raise the question whether they reflect a phonetic reality or are mere slips. Certain types of contact assimilation, for example (kt > t(t), ks > s(s), pt > t(t), ps > s(s)), are rarely attested (ix), and then not until the imperial period (if problematic cases are left out). If all misspellings of these types were combined and put alongside omissions of -m, their frequency would appear small in comparison, yet no one, it seems, has questioned the reality of the sound changes behind the misspellings. There are good reasons for this. Misspellings presented as mere statistics are not enough to provide reliable information about linguistic developments. There are, however, various telling facts about these assimilations and their attestation. First, structurally all of them are identical, as we will see, and of a type long avoided in the history of Latin. They form a system, and therefore attestations of the four categories may be treated as a unity. Their collective frequency may not come near to that of the omission of -m, but is rather more substantial than that of each of the categories considered in isolation. Second, the pattern of attestations is interesting. Republican examples have been adduced but are problematic. The first certain instances, and in some numbers, are in the graffiti of Pompeii, which provide the first real evidence that we have for non-standard varieties of Latin. The accumulation of examples in such documents almost certainly allows us to glimpse a feature of ordinary spoken Latin, albeit in an area where there was possible substrate influence. This impression is confirmed by continuing attestations in low-register documents during the Empire. Third, most of the assimilations have some sort of outcome in the Romance languages. The facts are consistent with a conclusion along these lines. The structural type represented by the four assimilations left no mark on the lexicon inherited from the prehistoric period (unlike many other types). It was long resisted but became influential at lower social levels by the early Empire, and probably earlier in submerged varieties of the language. It then influenced many of the Romance languages, although during the Empire it must have been kept at bay by many educated speakers, to judge from the limited attestations.

We may conclude that infrequent occurrences of a misspelling may hide the linguistic reality, and that the pattern of those occurrences must always be considered, along with extraneous evidence such as that of the Romance languages and metalinguistic comments (though of the latter there are none in this case).

(iii) Some misspellings may appear to be frequent but a lot of the evidence collapses when special factors are taken into account. This is particularly the case with the omission of final -s. Several scholars who have considered its omission in inscriptions with some care have observed that a majority of cases can be explained away as abbreviations or as due to lack of space at the edge of a stone or as determined by the presence of s at the start of the next word (viii.2.5.1). Not infrequently the omission of final -t is due to the same factors (viii.3.4, p. 155), but there are also omissions that seem to be determined by the phonetic context (sandhi phenomena), and these point to an early stage in the loss of the phoneme in final position.

(iv) The phenomena referred to in (iii) may be used to raise a methodological point. In interpreting the significance of a misspelling or its frequency a comparison of the treatment of related phenomena in a text may be revealing. As ‘interrelated phenomena’ we might cite e.g. different consonants in final position, or different diphthongs. If one final consonant (-m) is omitted repeatedly in a group of texts but another (-s) rarely or never (once special cases of the types alluded to above have been eliminated), it may be reasonable to conclude that there was a difference in the speech of the writers between the treatment of -m on the one hand and -s on the other, with the first dropped but the other stable. If however both were regularly written, we would not know whether either or both were pronounced by the writers in final position, since the regularity of spelling might reflect not their speech but their level of literacy. Variable treatment of the consonants is more revealing because the presence of errors of one type would suggest that the writers were not good spellers, and the absence of the other potential error might be taken to show that such a misspelling had no basis in speech.

This comparative method (for which see Adams Reference Adams2007: 635–6) works particularly well in the assessment of the history of the diphthongs ae and au. The spelling e for ae is far more common than o for au: there are many texts in which the former is well represented but the latter non-existent (see iv.3). That the different treatment of the two related phenomena reflects the state of the spoken language is confirmed by the universal monophthongisation of ae in the Romance languages alongside the partial survival of au.

Such comparisons may be particularly useful in assessing negative evidence, that is the absence of a particular misspelling. If the misspelling is never committed in a text that also lacks misspellings of comparable types, the writer might have been a good speller. If he was in other respects a bad speller, the absence of the misspelling may be significant.

7.1 Letters of Terentianus (Youtie and Winter 1951)

There are twelve cases of e for CL short i (see Adams Reference Adams1977a: 8) (two in nese):

These examples fall into a number of categories.

7.2 O. Faw. 1–7, CEL 73–80, Cugusi (Reference Cugusi1981)

There are ten cases of e for CL ī. Four of these are in tibe, where there was an original ei. For tibei see the letter of Suneros, P.Oxy. 44.3208 = CEL 10, line 5, a text which also has deuom (see above). Though tibe might be old-fashioned, these letters are very badly spelt, with no sign of hypercorrection or other old spellings, and there is an outside chance that tibe here is a phonetic spelling. Tibe alternates with tibi in the corpus. The spelling entro may show Greek influence (since Gk ἐν- corresponds to Lat. in-); there is one other definite case of Greek interference in the corpus (see Adams Reference Adams2006). The remaining five examples are all in verb endings (present), in second person forms. Apart from entro all instances are in final syllable.

7.3 Archive of the Sulpicii (Camodeca Reference Camodeca1999)

The only such spelling is ube at TPSulp. 45 (ad 37). This looks like old orthography (ubei is an attested spelling).

7.4 Vindolanda (Tab. Vindol.)

At 250 there is a spelling debetorem = debitorem. This may be a special case, influenced by debet (see Adams Reference Adams1995a: 91). The spelling is in an unstressed syllable. Ṿbe is at 642 (see 7.3). At 617 ịḅe may be of the same type (see the editors ad loc.). Two letters by a certain Florus (643) have an accumulation of misspellings in the final syllable of verb forms, none of them perfects (dabes three times, dabet and signabet once each: Adams Reference Adams2003b: 533–4). The same letter has the spelling benifeciario (on which see above, 7.1.3). In a new tablet (861; see Bowman, Thomas and Tomlin Reference Bowman, Thomas and Tomlin2010) a spelling uexellarius is found (possibly a vocalic assimilation, or influenced by the diminutive suffix -ellus). There are only ten possible examples in the Vindolanda tablets, five of them in verb forms and two possibly reflecting old orthography (ube, ibe). Three of the remaining examples (debetorem, benifeciario, uexellarius) may be special cases.

7.5 Vindonissa tablets (Speidel 1996)

Dabes occurs three times (15, 31, 53): cf. above on Vindolanda.

7.6 La Graufesenque (Marichal Reference Marichal1988)

Marichal (Reference Marichal1988: 59) cites tesanares (76.8) < ptisanarium (which is found in Horace: see Marichal Reference Marichal1988: 91). There are several examples of tisanares in the corpus. The only other fairly certain example is magedes (163.3). The word, apparently an archaic equivalent of lanx (see Marichal Reference Marichal1988: 88, citing Plin. Nat. 33.146 lances, quas antiqui magides uocauerant), is of Greek origin (< μαγίς) and ought to have an i in the second syllable. Varro (Ling. 5.120) presents a popular form magidam, which is derived from the Greek accusative.

The relevance of these forms to real popular Latin is open to question, because in Gaulish short i is attested as opening to e in final or unstressed syllables (see Lambert Reference Lambert1995: 41). Since both words are technical terms, they had perhaps taken on a Gallicised form first when borrowed into Gaulish and then in the Latin of the potters.

7.7 Four Latin letters from Mons Claudianus (O. Claud. 2, 131, 135, 367)

These letters, all dated to the second century ad (131 and 135 c. 107), are without misspellings of this type.

7.8 Second-century legal document transliterated into Greek letters (SB iii.1.6304, CPL 193)

In this text, which has numerous cases of ĭ written with an iota, there are two instances of epsilon instead. One of these is a grecism (εντερρογατιωνε, with the Greek prefix corresponding to Lat. in-), and the other is in βιγεντι for uiginti. Is this the same as those republican transliterations (see above, 5) showing epsilon for Lat. ĭ, motivated by the relative openness of the Latin i to Greek ears?

7.9 Bu Njem ostraca (Marichal 1992)

In this corpus of the mid-third century there are no examples, but many cases of (short) i correctly written (about 183). It is possible that the absence of misspellings of this type is a reflection of a feature of the African vowel system (see below, 13).

7.10 Spanish curse tablet (Corell Reference Corell1993)

A curse tablet from Carmona (Seville) has two front-vowel misspellings, one in the final syllable of a verb form (recipiates), the other old-fashioned (meretis: see above, 5 for mereto in early Latin).

7.11 Five early letters on papyrus (Cugusi Reference Cugusi1973)

In the five letters collected and discussed by Cugusi (Reference Cugusi1973) there are no such misspellings.

7.12 Two early letters from Myos Hormos (Cuvigny Reference Cuvigny2003)

In the letters M689 and M1107 (Cuvigny Reference Cuvigny2003: 409, 405), dated between the end of the first century and the time of Hadrian, there are no examples.

7.13 P. Rainer Cent. 164

In this text (late first century bc) there are no examples.

7.14 Graffiti del Palatino (Väänänen 1966–70)

These graffiti are of uncertain date, but may belong roughly to the second/third centuries.²² The only misspellings (see the linguistic appendices at i.253, ii.255) are i.304 dicet, ii.283 futuet, 289 omnes (nominative singular).

7.15 Some statistics

Forty-five (or forty-six) misspellings showing e for CL ĭ were considered from the fourteen corpora. About twenty-three were dismissed as special cases of one type or another. A further twenty (or twenty-one) were in the final syllable of verb forms.

The Pompeian graffiti must now be considered.

7.16 Pompeian graffiti

The relevant misspellings, which are numerous, have been analysed by Väänänen (Reference Väänänen1966: 21–2). He divides the material into three categories: cases (a) under the accent or pretonic, (b) after the accent but in non-final syllables, and (c) in final syllables. This last category is divided into spellings in nominal endings and those in verbal endings. Much of the material in the first two categories is problematic or open to special explanations, and there would be no point in counting or listing the examples again. But from (a) there remains prauessimus (CILiv.8259) and from (b) domene (iv. 1871), muli<e>rebu (iv.4137) and munerebu (iv.6900).²³ The last two examples are identical in type to tempestatebus in an early Latin inscription (above, 5). Väänänen (Reference Väänänen1966: 21) refers to ‘le timbre indécis que devait avoir toute voyelle dans cette syllabe, qui était aussi la plus sujette à la syncope’.

Under (c), first section (‘désinences nominales’), Väänänen lists eight examples of the type omnes for omnis (nominative), and one example of aled for alid. Here again there is an exact parallel in an early Latin misspelling (aidiles). Väänänen (Reference Väänänen1966: 84) categorises the e as indicating ‘une prononciation ouverte de i’. One of the spellings in the Palatine graffiti seen above is of the same type.

Finally Väänänen lists (c, β) numerous examples of e for ĭ in verb endings of the third conjugation, first and second person singular (-es, -et), in both the present and perfect tenses, and also the future. In this list there are about forty-three examples in the present tense (which include three instances of dicet, a spelling found in a letter of Terentianus), and ten in the perfect.

The instances in the perfect (which include e.g. fecet) are easy to account for. The e-spelling in the third person is early (both with a following -d and -t), and was very persistent into the imperial period, particularly in certain verbs, most notably fecet. Numerous examples from early Latin can be found in the index to CILi² (p. 820), and from later Latin at Adams (Reference Adams2003a: 51). The latter material, for example, includes fourteen instances of fecet from CILvi, and also cases of the transliterated form φηκετ. The regularity of this transliterated form would not favour the view that there had been a merger of CL ē and ĭ in the early Empire, since the two vowels are rendered by different letters in Greek.

It follows that in the perfect the spellings with e at Pompeii may simply be old-fashioned. It would not be convincing to put them down at Pompeii of this period to Oscan influence (Oscan retained the ending -ed, as in deded),²⁴ because the e-forms go back in Latin itself to a period much earlier than that of the Pompeian graffiti in the first century ad, and they clearly had a traditional character by the time when they were written at Pompeii.

In the present tense the e-forms are more problematic. In early inscriptions third person singular verbs tend to be in the perfect rather than the present, and there is not the textual evidence to determine whether an e-spelling in a verb form such as dicet (for CL dicit) was established in early Latin (though it can be reconstructed as original: < *deik-e-ti). In Oscan proper e is not attested in the third person present of the third conjugation (see Buck Reference Buck1904: 59). Väänänen (Reference Väänänen1966: 22, 130) is nevertheless inclined to see the influence of Oscan at Pompeii in the present tense e-forms. However, e-spellings are found at Pompeii in the final syllable of futures as well, in which Oscan could not have been influential (for the futures ibet and pugnabet see Väänänen Reference Väänänen1966: 22; an Oscan future is didest, i.e. dide-s-t, with syncope of the vowel in the ending).²⁵ In the material presented from elsewhere in the Empire in this chapter there are both futures (dabes, dabet) and future perfects with e (aiuataueret, uolueret; the Oscan future perfect is quite unlike the Latin).²⁶ There are at least four ways of explaining a form such as dicet at Pompeii: (i) it might be a traditional spelling which merely happens not to be attested in early texts; (ii) the e might have spread from the perfect (dedet) to other tenses; (iii) it might show Oscan influence; (iv) it might reflect weak articulation of the vowel in a final syllable. It is not obviously justifiable to latch onto (iii) amid all the possibilities. The simplest explanation of the e-forms in tenses other than the perfect (including the subjunctive recipiates in the Spanish defixio) is that they all reflect the same phonetic feature, namely an openness of ĭ in verb endings (in final syllable). Certainly our material above from non-Oscan areas cannot be explained from Oscan influence, and it is preferable to account for the evidence by a single hypothesis.

7.17 Further statistics

If the Pompeian forms considered in the previous section are added to the statistics given from the other corpora, the number of e-spellings for ĭ rises to about 111, and of these a very high number (seventy-three) are in verb endings.²⁷ A further twenty or so cases (some of them classified above as special in some way) are in final syllables in other types of words, nominal, adjectival or adverbial. More than ninety of the 111 spellings considered above are in final syllables. That statistic suggests the manner in which readjustment to the front-vowel system must have been getting under way in the early centuries of the Empire. Final syllables never bore a stress accent, and for that reason vowels in this position were historically subject to modifications, such as loss (*homce > hunc, face > fac: see e.g. Meiser Reference Meiser1998: 73–4), opening (facili > facile) and shortening. Some of the testimonia from grammarians collected in section 6 refer to shortening of a long vowel in final syllable, and we also saw other evidence for the phenomenon, some of it quite early. We therefore have at roughly the same period (the early centuries ad) evidence both for shortening in final syllables, and for a change in the quality of short i in such syllables. It is likely that in a verb form such as dabis the final front vowel was articulated with reduced tension, and that it tended to open slightly. As ĭ opened it will have moved close to ē without merging with it.

The proto-Romance vowel merger of ē and ĭ would only have occurred once phonemic distinctions of vowel length were lost across a wider spectrum, and evidence for the part played by the stress accent in undermining the quantitative system only appears in abundance from the third century. The transliterated form φηκετ shows that, though the final short vowel had an open quality, it had not merged with ē. Similarly the mid second-century transliterated text mentioned above (7.8), which, given that the Greek alphabet distinguishes better than the Latin between long and short vowels, is virtually a phonetic transcription, shows no sign of a breakdown in the quantitative system. Long e (η) and short e (ε), and long o (ω) and short o (ο) are consistently (and correctly) distinguished throughout, sometimes in the same word (οκτωβρης). In κωνσουλιβους ō and ŭ are correctly distinguished. Iota is used repeatedly for ĭ, and the one departure from classical norms in this respect is consistent with a tendency to opening in the short-vowel system, as distinct from a merger across the short- and long-vowel systems. Yet this text has numerous errors of other types, and there are signs that it was partly taken down from dictation.²⁸ It could only be concluded that the proto-Romance merger was fully in progress if misspellings were more widely distributed within the word than they are in our early imperial documents, and if there was evidence for a disturbance to the quantitative system.

10.1 Letters of Terentianus (Youtie and Winter 1951)

There are seventeen examples of o for ŭ, but almost every one is a special case.³⁵ There are four instances of uo (as in saluom, nouom, fugitiuom), eleven of con (which may either retain the old Latin vowel spelling, or reflect the influence of compound forms such as confido), one of the old-fashioned quominos, and finally sopera (471.21). Only the last is difficult to classify. It would not do to argue that here is a sign of the late back-vowel merger, because the non-syncopated form is suggestive of early Latin (supera for supra is in Livius Andronicus), and the o too may have been conceived of by the scribe as old-fashioned.

10.2 O. Faw. 1–7

For con (3.12) see above.

10.3 Archive of the Sulpicii (Camodeca Reference Camodeca1999)

There are no examples (though many cases of short u), apart from the uo-spelling fugitiuom (with uncertainties of reading but the ending is clear) at TPSulp. 43.ii.3.1.

10.4 Vindolanda (Tab. Vindol.)

No examples.³⁶

10.5 Vindonissa tablets (Speidel 1996)

No examples.

10.6 La Graufesenque (Marichal Reference Marichal1988)

No examples.³⁷

10.7 Four Latin letters from Mons Claudianus (O. Claud. 2, 131, 135, 367)

No examples.

10.8 Second-century legal document transliterated into Greek letters (SB iii.1.6304, CPL 193)

No examples.

10.9 Bu Njem ostraca (Marichal 1992)

No examples (see Adams Reference Adams1994b: 103–4). Fornus for furnus is a special case (see Adams Reference Adams1994b: 104).

10.10 Spanish curse tablet (Corell Reference Corell1993)

This contains morbo(s), which, if it has been correctly interpreted, is another old spelling.

10.11 Five early letters on papyrus (Cugusi Reference Cugusi1973)

No examples.

10.12 Two early letters from Myos Hormos (Cuvigny Reference Cuvigny2003)

No examples.

10.13 P. Rainer cent. 164

No examples.

10.14 Graffiti del Palatino (Väänänen Reference Väänänen1966–70)

No examples.

10.15 Pompeian graffiti

The evidence is treated comprehensively by Väänänen (Reference Väänänen1966: 26–9). The vast majority of spellings showing o where ŭ might have been expected are placed under the heading (28) ‘o pour u classique = o archaïque’, and are old spellings mainly of the types seen above in Terentianus. At 27 just four instances are cited of ‘o pour ŭ originaire’, but all of these are special cases, or of doubtful interpretation. At Pompeii there is no sign of a back-vowel merger.

10.16 Conclusion

There is not a single case in this material of an o-spelling that might be interpreted as showing opening of short u in anticipation of a vowel merger. In the same material in the front vowels there are cases of opening. The usual view of the relative chronology of these vowel changes must be correct.

2.1 The Republic

The diphthong ai developed to a monophthong represented by e in a number of the languages of Italy in the Republican period, and also in the Latin of areas outside Rome.¹ At Rome itself the diphthong was narrowed to ae,² but full monophthongisation seems to have been resisted for some time (at least in educated varieties of the language), though eventually monophthongisation became general and the outcome of ai > ae in most words in Romance languages was an open e identical to the outcome of CL short e (see however below, n. 10).³

To take first Italic languages rather than Latin dialects. In Oscan ai remained as a diphthong (Buck Reference Buck1904: 43–54), if the texts are anything to go by (though spelling is not a reliable guide). In Umbrian ai developed to a long open e (see Buck Reference Buck1904: 44 on the evidence for its open quality). In Faliscan texts (if Faliscan is to be regarded as a language rather than a dialect of Latin) there are three spellings, ai, ei and e (see Bakkum Reference Bakkum2009: i.110, and for the instances of the three forms, 111); for interpretation of the data see Bakkum (Reference Bakkum2009: i.112–14). There are also apparent cases of monophthong-isation in non-Latin inscriptions from the territory of the Marsi and Aequiculi, such as Vetter (Reference Vetter1953) no. 223 = Rix (Reference Rix2002: 66) VM 3 pa.ui.pacuies.medis | uesun.dunom.ded(e) | ca.cumnios.cetur (Antinum), 225 = Rix (Reference Rix2002: 66) VM 5 sos (Marruvium, S. Benedetto dei Marsi)⁴ and 226 = Rix (Reference Rix2002: 67) VM 8 stat (dat.) (Collemaggiore).⁵

As for Latin, in republican inscriptions from outside Rome e-spellings are attested at Praeneste and Tusculum in Latium, at Falerii Novi (the region of Faliscan), around the Lacus Fucinus in the territory of the Marsi, and in Umbria and the ager Gallicus (a collection of material may be found in Adams Reference Adams2007: 82–7). On the question whether there is any sign of monophthongisation in the inscriptions of Rome itself at this period see Adams (Reference Adams2007: 81–2). There is an absence of clear-cut evidence.

There is some literary evidence that throws light on the situation at Rome (see e.g. Lindsay Reference Lindsay1894: 42): Varro Ling. 5.97 hircus, quod Sabini fircus; quod illic fedus, in Latio rure hedus, qui in urbe ut in multis A addito haedus; Lucilius 1130 (Varro Ling. 7.96) Cecilius <pretor> ne rusticus fiat. In the first passage Varro makes a contrast between the city of Rome, where haedus was current, and rural Latium, where hedus was the corresponding form. To a city dweller in the late first century bc the monophthong was rustic. The other passage belongs to the second century bc. A certain Caecilius is ridiculed as rustic, and his rusticity is brought out by the use of the monophthongal spelling in reference to him. The two pieces of evidence suggest that the lack of definite inscriptional evidence for the monophthong at Rome in the last two centuries of the Republic does not give a false impression. The monophthongisation started outside the city but was retarded within, and was stigmatised by the urban upper classes. Dating the change at Rome itself is impossible, as evidence is not available. Since there was always movement of people from the countryside to Rome, the monophthongal pronunciation might have been heard at some social levels in the city, but Ferri and Probert (Reference Ferri and Probert2010: 21) are rightly cautious in stating that it ‘is not very clear…whether -e- for -ae-…already belongs to a low sociolect in the first century bc’. When it caught on among the educated is anyone's guess (see below).

The next evidence to turn up, a good deal of it new, belongs to scattered places, mainly outside Italy, dated to the first and the early second century. There are also some problematic statements in later grammarians, which raise the question whether the educated attempted to preserve the diphthong in speech over several centuries; they certainly preserved the digraph in writing.

In what follows these two categories of evidence are discussed in that order.

2.2 Non-literary texts of the early Empire

The earliest evidence in the first category raises again the question of usage in provincial Italy compared with that at Rome. In a coherent part of the archive of the Sulpicii from Pompeii, that is the documents written in the hand of the freedman C. Novius Eunus and dated to the first half of the first century ad (TPSulp. 51, 52, 67, 68), e is already the norm (see Adams Reference Adams1990a: 230). The e-spelling occurs seventeen times in what is only a small corpus, against just one case of ae, and that hypercorrect (51 petiaerit). The name Caesar is constantly spelt Cessar, and that is striking because a particular effort was made to get the spelling of personal names right; signs of this effort will be seen below (this section) in the ostraca from Bu Njem. It is possible that Eunus was taking the texts down from, or with the aid of, dictation. The correct versions of the same tablets (written on the exterior by scribes) always have ae. Here is a sign of the distinction between speech and writing. In the speech that Eunus heard around him the monophthongal pronunciation was well established, but well-trained scribes wrote the digraph.

It has been stated on the evidence of this archive (Adams Reference Adams1990a: 230) that ‘ae had been monophthongised in ordinary speech by the early first century ad’, but this is a misleading way of describing the situation revealed by the tablets. It is likely that the spelling reflects not a recent change but a long-standing feature of the speech of this Campanian region, distinguishing Campanian Latin still at this date from the speech of the city. The evidence of the archive is supported by a literary testimonium. According to Varro (Ling. 7.96) rustics (in Atellan farce) called the character Pappus/pappus Mesius not Maesius (see Adams Reference Adams2007: 154–5 with n. 100 on the passage and its difficulties), and these rustic characters can only have been Campanians. Since Oscan, unlike Umbrian, seems to have preserved the diphthong ai (see above, 2.1), such a monophthong in Campania cannot with any confidence be put down to the influence of the substrate (Adams Reference Adams2007: 154–5). The Pompeian graffiti, also of the first century, show up the same characteristic as the archive of Eunus. In these the e-spelling abounds (Väänänen Reference Väänänen1966: 23–4 for the evidence).

The non-literary evidence that next turns up is about a century later, and from places well beyond Italy.

In the Vindolanda tablets of the early second century ad the digraph is usually used correctly, but there are several documents with clusters of e-spellings, which hint at developments in speech that have been obscured by the literacy of the military scribes. In a letter by an entrepreneur Octavius (Tab. Vindol. 343) there are four cases of e and a hypercorrect case of ae for long e (see Adams Reference Adams1995a: 87). There is only one correct instance of ae in this text. In the account 186 there are three instances of e and none of ae. Fragmentary letters by a certain Florus in the third volume (643) contain one case of e and one of ae, and an abundance of errors of other types (Adams Reference Adams2003b: 534). These various texts might not have been written by the usual scribes. The material is not extensive, but extensive enough to suggest that in the Latin speech of the camp the monophthong was the norm. There were Batavians and Tungrians at Vindolanda, but nothing is known about the background to their acquisition and use of Latin.

In the letters of Terentianus from Egypt of much the same date e and ae occur with about the same frequency (see Adams Reference Adams1977a: 12), but there are variations from letter to letter that must reflect the practices of different scribes (on this phenomenon see Halla-aho Reference Halla-aho2003), or the influence of different lexical items. In letters 467 and 468 only ae is written, fourteen times. By contrast in 469 e occurs seven times, and the only instance of ae is hypercorrect, for short e (reṣṛeibae). In letter 471 ae occurs seven times and e eight. Here the influence of particular word forms is apparent. Aes is three times spelt correctly, presumably because the phonetic spelling es would have lacked distinctiveness,⁶ and twice the prefix prae- has the correct digraph; evidence will be seen below (2.5) from a later period suggesting that writers were taught to use ae after pr-. On the other hand the locatival–directional Alexandrie is six times spelt with e (never ae). Syntactically the locative expressing the goal of a verb of motion must have been slangy and idiomatic (see xv.4), and the scribe for that reason was happy with the non-standard spelling.⁷ The only example of the correct Alexandriae in the archive is in a letter of Tiberianus (472), the ‘father’ of Terentianus. In the circles in which Terentianus moved monophthongisation had taken place. The origin of Terentianus and of his scribes is obscure.

Somewhat later the Bu Njem ostraca from Africa (dated to the middle of the third century) show a marked predominance of e over the digraph (see Adams Reference Adams1994b: 103; the figures given here are corrected slightly). E is written about fifty-five times, compared with sixteen examples of ae. One of the latter is hypercorrect (O. Bu Njem 133 aegregia), and twelve of the others are in proper names (Caecilius twice, Aemilius four times, Aemilianus five times, Aelius once). It is possible that even bad spellers made an effort to use the ‘correct’ form of personal names (see also x.1 for this point; contrast the remarks above about Eunus’ spelling of Caesar). The remaining three correct cases of ae are in praepositus (O. Bu Njem 13, 15, 34); on the significance of this form see the last paragraph and below, 2.5.

These various corpora show provincials in the first three centuries of the Empire writing e for ae with such regularity that monophthongisation must have been widespread across the Empire. Evidence of the same (non-literary) type is lacking for Rome for this period.⁸ Some remarks by grammarians must now be considered. It may be assumed that at Rome at the social and educational levels represented in the above corpora from outside the city monophthongisation was also the norm,⁹ but the question arises whether careful or educated speakers under the influence of grammarians tried to maintain not only the digraph but the diphthong.

2.3 Grammarians on ae

Bakkum (Reference Bakkum2009: i.110) includes Quint. 1.7.18 among imperial testimonia suggesting that ae was still (?) a diphthong in Roman or upper-class Roman Latin (see too Lindsay Reference Lindsay1894: 43): ae syllabam, cuius secundam nunc e litteram ponimus, uarie per a et i efferebant (‘[t]he syllable ae, which now has e as second letter, was formerly expressed by a and i’, Russell, Loeb). The passage is carelessly expressed, with ponimus suggestive of writing and efferebant more suggestive of speech, but overall the impression given is that Quintilian is commenting on varieties of spelling, and he certainly tells us nothing specific about speech at his time.

Next, there is Terentius Scaurus De orthographia GLvii.16.6–10 a igitur littera praeposita est u et e litteris, ae au…et apud antiquos i littera pro ea (e littera) scribebatur, ut testantur μεταπλασμοί, in quibus est eius modi syllabarum diductio, ut ‘pictaï uestis’ et ‘aulaï medio’ pro pictae et aulae. sed magis in illis e nouissima sonat, ‘well then, the letter a is prefixed to u and e, as in ae and au…and among the ancients i was written for e, as is shown by alterations of sound, among them the separation of syllables as pictaï uestis and aulaï medio for pictae and aulae. But in those words it is rather an e that sounds in final position’ (translation of Sturtevant Reference Sturtevant1940: 124 n. 55 with modifications). Terentius (attributed to the second century ad, if the De orthographia is genuinely his) seems to be saying that in pictae and aulae a diphthong was heard, with e as its second element. The passage is taken thus by Sturtevant (Reference Sturtevant1940: 124).

Another passage that might be read in the same way, though its interpretation is problematic, is Diom. GLi.452.17–19 (detractione) litterae, ut si detracta a littera pretor dicamus, ut Lucilius ‘pretor ne rusticus fiat’, cum debeat ae pronuntiari, praetor (‘(by the omission) of a letter, for example if we were to say pretor with the a omitted, as Lucilius (wrote) pretor ne rusticus fiat, when ae should be pronounced, i.e. praetor’). Diomedes may be dated loosely to the late fourth/early fifth century (Kaster Reference Kaster1988: 270–1). He does not make it clear whether he is distinguishing satisfactorily between writing and speech. He illustrates the detractio of the a in pretor from a written text, but then says that the word should be ‘pronounced’ praetor. If the latter remark could be taken at its face value it would constitute evidence that grammarians were trying to preserve the diphthong, but it is possible that he was using pronuntiari loosely.

Marius Victorinus GLvi.32.4–6 also on the face of it refers to the pronunciation of the sound represented by ae as a diphthong: rursus duae inter se uocales iugatae ac sub unius uocis enuntiatione prolatae syllabam faciunt natura longam, quam Graeci diphthongon uocant, ueluti geminae uocis unum sonum, ut ae oe au.

Servius GLiv.421.19–21 is not in line with the (possible) interpretation of the above three passages: e quando producitur, uicinum est ad sonum i litterae, ut meta; quando autem correptum, uicinum est ad sonum diphthongi, ut equus (see Kramer Reference Kramer1976: 22), ‘when e is long, it is near the sound of the “letter” i, as in the word meta; but when it is short, it is near the sound of the diphthong, as in equus’. This sentence means that long e was closer than short e. The openness of the short e in equus is likened to that of the ‘diphthong’, presumably in this context that of aequus. By ‘diphthong’ Servius must mean in modern terminology ‘digraph’. It would appear that aequus had an open vowel, not a diphthong, in the first syllable, represented in traditional orthography by the digraph ae. It is hard to believe that Servius would have compared the vowel sound in the first syllable of equus with a genuine diphthong (or just the second part of a diphthong), but since in the passage to the Romance languages the original ae diphthong merged with original short e, it is plausible that in the late Empire the sound of equus should have been likened to that of aequus. Servius might even have been suggesting that the pronunciation of the two words was identical, which would have been the case if the long open e that may be assumed to have been an intermediate stage in the monophthongisation of original ai/ae had by now been shortened (on this question see below, 2.4). An alternative possibility is that aequus still had a long open e, which could be likened to the e of equus in its degree of aperture if not its length.

This second possibility is perhaps supported by Pompeius GLv.285.8–9 si uelit dicere aequus pro eo quod est equus, in pronuntiatione hoc fit (‘if [anyone] were to say aequus instead of equus, this represents [a barbarism] of pronunciation’). The passage is about barbarisms of vowel length, consisting either of the lengthening of a short vowel or the shortening of a long. The example given represents the first type. Pompeius must be saying that, though the two words sounded much the same, the vowels in the first syllable should differ in length.

The evidence of these two passages suggests that in the fourth and fifth centuries grammarians felt that a monophthongal pronunciation of the ae was the norm and correct, though (in the eyes of Pompeius) the e of aequus was long.

‘Sergius’ Explan. in Don. (GLiv.520.28–9) is much the same as the passage of Servius above.

The evidence of the seven passages from grammatical works (including that of ‘Sergius’) is a mixed bag. Terentius Scaurus seems to be referring to a diphthongal pronunciation, as does Marius Victorinus. Diomedes possibly is, but there is another way of taking his remarks. Quintilian seems to talking about writing. The last three passages show grammarians accepting the monophthongal pronunciation. It may tentatively be concluded that grammarians for a while tried to keep the diphthong going in educated speech, but that by the fourth century they had given up the attempt. There is, finally, a slight disagreement between Servius and Pompeius, as the passages have been interpreted here. Pompeius, who possibly was writing anachronistically, found a difference of length between the vowels of equus and aequus, whereas there is no hint of a distinction in Servius.

For some further testimonia from grammarians, which do not affect the conclusions above, see Lindsay (Reference Lindsay1894: 42–3).

Another piece of evidence has been brought to bear on the question whether the diphthong lasted into the Empire in educated speech. Flobert (Reference Flobert1990: 105) notes cases of the digraph ae marked with an apex in the Laudatio Turiae (praéferam, patriáe), Res Gestae, and particularly in inscriptions from Vienne and Lyons. There are no errors in the last two corpora. Flobert concludes: ‘Cette insistance sans faute montre que ae, comme au, est encore une diphtongue sous le Haut-Empire.’ The evidence is not decisive. The apex might have been an artificial appendage of the digraph, acknowledging the original length of the diphthong, whatever the pronunciation that lay behind it. The absence of errors is interesting but might reflect not pronunciation but the practice of good spellers.

2.4 Long versus short open e

When ae was monophthongised it is assumed that there was a stage during which the resultant vowel was long, given that diphthongs regularly developed to long vowels in the history of Latin. It is also assumed that this long e was open, rather than a closer e equivalent to the inherited long e of Classical Latin. The reasoning behind the second assumption is that the Romance reflex (in most words)¹⁰ of original ae was the same as that of CL short (i.e. open) e rather than of CL long (i.e. close) e. The hypothetical long open e (< ae) must have disturbed the vowel system, in that there would for a time have been a long close e, a long open e and a short open e, contrasting with the usual long/short pairs. Symmetry was restored by the shortening of the new long open e, such that it merged with the CL short e.

The (temporary) existence of the new long (open) e seems to be confirmed by the hypercorrect use of the digraph to represent original long e (see R. Coleman Reference Coleman1971b: 186–90; also e.g. Pirson Reference Pirson1901: 19, Gaeng Reference Gaeng1968: 251–2). The hypercorrect use of ae for short e for its part is a sign that the long open e had undergone shortening. This second hypercorrection turns up quite early. Petiaerit for petierit is attested as early as June ad 37, in the part of the archive of the Sulpicii bearing the name of C. Novius Eunus (TPSulp. 51) (see above, 2.2, p. 73). There is a similar example elsewhere in the archive, in a document of a certain Diognetus, slave of C. Novius Cypaerus (TPSulp. 45 aeodem = eodem). The same, short monophthongal, outcome of ae is also attested to by Greek transliterations of Latin employing epsilon to represent the sound of the original ae. There is a case in the second century in a document of Aeschines Flavianus of Miletus (SBiii.1.6304 = CPL 193), who writes βετρανε for the genitive of ueteranae, a spelling that is phonetic in three different ways (see the discussion of the text at Adams Reference Adams2003a: 53–62). The witness to this same document, Domitius Theophilus, writing Latin in Latin script, has aeadem for eadem; cf. e.g. aeorum for eorum at O. Faw. 1.12. The earliest inverse spelling (ae for short e) in the Spanish inscriptions according to Carnoy (Reference Carnoy1906: 77) is Naerua in an inscription from the end of the first century ad. He notes (75, 76) that inverse spellings in Spain both in inscriptions before the fifth century and in Christian inscriptions always show ae for short rather than long e, and seeks to set up a regional distinction between Spain and other provinces in this respect, arguing that material in Schuchardt (Reference Schuchardt1866–8: i.223–460) from elsewhere in the Empire shows a large number of cases of ae for long e. The argument is unsafe. R. Coleman (Reference Coleman1971b: 186) offers examples of ae for short e from various regions, and we saw early inverse spellings from Italy above; see also above, 2.2, p. 74 on the letters of Terentianus. The Spanish attestations are not many, and it will not do to argue for regional variations from limited evidence.

2.5 Spelling

For centuries most writers aspiring to correctness attempted to go on using the digraph ae, long after it had ceased to be representative of speech. In the manuscripts of Gregory of Tours the hypercorrect substitution of ae for e is more common than the phonetic spelling e for ae (Bonnet Reference Bonnet1890: 97), and that betrays the importance still attached to correct spelling at a time by which the ae-spelling had no basis in pronunciation. In the late period in Italy (in the Edictus Rothari, Ravenna papyri and Jordanes) there are signs that writers resorted to mechanical rules to get the digraph right at least some of the time. In these texts ae predominates over e in case endings (feminine) and in the prefix prae-, whereas in other positions in the word e outnumbers ae (Adams Reference Adams1976a: 43–4). A writer could be taught to write ae in case endings and after pr-, but no rule could instruct him when to use the digraph in other positions. A mechanical rule offered some help but might sometimes lead a writer astray. Bonnet (Reference Bonnet1890: 97–9) notes that in the manuscripts of Gregory of Tours after pr- ae is often written instead of the correct e (e.g. in praece, praecor, praemo, praetium and praesbiter). See too Adams (Reference Adams1976a: 44) on praetium. Vielliard (Reference Vielliard1927: 38) observed that in the Merovingian documents which she examined ae was most often preserved in the prefix prae-. Jeanneret (Reference Jeanneret1918: 26) notes in curse tablets several inverse examples (for short e) after pr- (repraeensionem, praecatio). The latter is at Audollent (Reference Audollent1904) no. 222.b.9, from Carthage. On the spelling prae- for pre- see also B. Löfstedt (Reference Löfstedt1961: 103). The confusion that had set in is revealed by a fifth-century grammatical passage cited by Löfstedt: Agroecius GLvii.114.21–115.1 praemium cum diphthongo scribendum; pretium, precor sine diphthongo.

There are by contrast other late texts in which all attempts to preserve the digraph had been abandoned, such as the Tablettes Albertini (Väänänen Reference Väänänen1965: 27–8, Adams Reference Adams1976a: 43).

2.6 Conclusions

The writing of e for ae was probably considered a vulgarism of spelling for many centuries. ‘Correct’ texts, such as the exterior versions of legal documents written by professional scribes (so the documents in the archive of Eunus), regularly have the digraph. In speech in the Republic the monophthong (probably representing an open e, long or short) was a feature of some regional varieties of Latin outside Rome, in rural Latium and further afield. It was also found in some Italic languages, such as Umbrian, Faliscan and Marsian. To urban observers in the last two centuries bc it was rustic if heard in Latin, and stigmatised, as is clear from the line of Lucilius directed at a Caecilius (above, 2.1). The Romance evidence shows that eventually the monophthong spread across the whole Latin-speaking world, including Rome. Its spread seems to be a case of a ‘wave effect’, with the wave originating in rural parts of Italy rather than in the city. What remains unclear is whether the monophthong had reached Rome from fairly early on in lower social dialects. It had rural associations, but was that all? We do not know. The spelling pedagogus (learned Latin paedagogus) in a specimen of the simple style at Rhet. Her. 4.14 is unrevealing, because in Koine Greek the ai- diphthong had long since turned into a monophthong represented by epsilon (in anticipation of Modern Greek), and the interchange of the two is common in the papyri (Gignac Reference Gignac1976: 191–3, Biville Reference Biville1995: 40–2). Popular Latin, in contrast to the learned language, might have borrowed the koine form.

What little evidence there is in grammarians suggests that in the early centuries of the Empire there was an attempt to maintain the diphthong, but that by about the fourth century the monophthong was so established that it was acceptable even to grammarians. Attitudes had changed since the Republic, when it was stigmatised in the eyes of educated Romans. e for ae was more a mark of poor literacy than of substandard speech by the later imperial period. If there were genuinely efforts to preserve the diphthong earlier in the Empire in educated speech, the consequence might have been similar to that seen above (iii.14) in the vowel system: inconsistent use of ae in careful speech.

3.1 Special Cases

3.2 Dissimilation

Forms such as Agustus for Augustus (e.g. CILiv.2124 Agusto) display a dissimilatory loss of part of the diphthong before u in the next syllable (see Schuchardt Reference Schuchardt1866–8: ii.308–13, 316 for numerous examples, Lindsay Reference Lindsay1894: 41–2, Väänänen Reference Väänänen1966: 32, Reference Väänänen1981a: 39–40, Leumann Reference Leumann1977: 73), which continued in most Romance reflexes (It., Sp. agosto, OFr. aost; the diphthong was retained in Romanian). The dissimilation is seen too in ausculto > asculto, a process resisted by at least one grammarian: Caper GLvii.108.6 ausculta non asculta. The dissimilated form lies behind e.g. It. ascoltare. The start of the word was reinterpreted as a prefix, and a change of prefix occurred (*excultare), which lies behind Fr. écouter.

3.3 Conclusions

The anecdote from Suetonius shows that there were speakers who attempted to preserve the diphthongal pronunciation au even in the lexical items in which the monophthong was widely established. Note too Servius on Virg. Georg. 2.30: ‘quin et caudicibus sectis’ pro codicibus, sicut caulem pro colem, sauricem pro soricem dicimus;¹⁸ clearly the commentator was struck by the abnormality of au for o in this term, but dicimus suggests that some used the au-form even in this special set of words. Such evidence, along with that of the Romance languages, shows that au was no mere grapheme concealing a monophthongal pronunciation. Even in non-literary texts such as writing tablets, ostraca and papyri, texts in which every type of phonetic spelling recurs, au is well preserved, except in significant lexical items. There could not have been a submerged variety of Latin in which long o < au was a general feature. The o-variant (whether in speech or in writing) was associated with certain words, but attitudes to the o-forms were complex, as the anecdote in Suetonius and generic variations in written texts show. There were purists who thought that they should be avoided even in the terms with which they were associated, and others clearly allowed them only in appropriate contexts. These might include letters, proverbs and practical as distinct from high literary genres. The o-forms must have had a conversational flavour suited to casual style. They had a limited acceptability even among the educated classes, but only in the appropriate style or context.

2.1 Some perfect verb forms

Between /w/ and /t/ a post-tonic short i tended to be lost. Thus the past participle of caueo is cautus < *cauitus. The types of words most frequently producing this phonetic environment (though not following word-initial stress) are perfect verb forms such as amauit, exiuit (see also vi.6 (iv)). Since -it was a morpheme and the u here is a marker of the perfect, syncope was resisted, at least in the educated language or by careful speakers. But in badly spelt texts of the Empire syncope turns up even in this position. For exiut see O. Faw. 2.9. In first-conjugation verbs note CILiii.12700 curaut, iv.2047 pedicaud, vi.24481 donaut (Väänänen Reference Väänänen1966: 45, Reference Väänänen1981a: 44, Cugusi Reference Cugusi1981: 744). The question arises whether the persistence of the correct spelling in literary Latin masks a development that had occurred in educated speech. This, however, is not a type of syncope that shows up in classical versification (for cases of syncope determined by metre see e.g. some of the material collected by Kramer Reference Kramer1976: 38–40), and it may tentatively be interpreted as substandard in speech, at least for a period. It is reasonable to suggest that the -it perfect ending was long preserved as correct by many speakers, supported by the analogy of the -i- in other parts of the paradigm. But there were clearly varieties of the language, such as those represented in the graffiti at Pompeii, in which the ending was not kept intact, and here we see some evidence for social variation.

The syncopated form did eventually catch on in an extensive part of the Romance world. See Väänänen (Reference Väänänen1981a: 143) on cantau(i)t > OIt. cantao (mod. cantò), Sp. cantó, Pg. cantou.

One of the factors inhibiting syncope must have been the morphological system: there was pressure to preserve a functional morpheme.

2.2 Cal(i)dus

Quintilian has a note on cal(i)dus: 1.6.19sed Augustus quoque in epistulis ad C. Caesarem scriptis emendat quod is ‘calidam’ dicere quam ‘caldam’ malit, non quia id non sit Latinum, sed quia sit odiosum (‘Augustus also, in his letter to Gaius Caesar [his grandson], corrects him for saying calidam rather than caldam, not on the ground that it is not Latin, but as being repulsive’, Russell, Loeb).³ On this passage see now Ax (Reference Ax2011: 259–61).

Caldam and calidam are emendations (Keil) for the transmitted caldum and calidum, and are not accepted by all editors. The point is that syncope is particularly common in the feminine calda, which means ‘hot water’ (sc. aqua) (see e.g. Baehrens Reference Baehrens1922: 14–15). For a list of examples of calda see TLLiii.151.82ff. In this sense calida is not invariably syncopated, common as the syncopated form is; there are as many examples of calida quoted by TLLiii.151.75ff. as of calda. In other applications calidus is rarely syncopated in manuscripts (see the material collected at TLLiii.151.44ff., with Baehrens Reference Baehrens1922: 14–15). As Colson (Reference Colson1924: 81) puts it, the emendation to cal(i)dam is therefore tempting, but not necessarily right, ‘in view of the possibility that A.'s view on these words may not have been the normal view’.

Whichever text is adopted, some general points emerge from the case of calda. Syncope might have become acceptable, even to the educated, in an idiomatic or specialised use of a term without being acceptable in other uses of that same term. The idiomatic/specialised usage might have been so familiar that the original form was considered pedantic or pretentious. The adjective ‘hot’ is perhaps applied more often to water than to anything else, and frequent usage favours phonetic reduction. The alternation between calda and calidus also suggests that speakers must have been conscious of the coexistence of the syncopated and unsyncopated forms, and as long as there was such consciousness different speakers would have had different views about the acceptability of the syncopated versus the unsyncopated. Debate was possible, and a ‘sound change’ could not be said to have fully occurred while it was being monitored by speakers.

Two further observations may be based on the data provided by the TLL about calidus versus caldus (see Baehrens Reference Baehrens1922: 14 for these points). First, the syncopated form is rare in poetry (twelve examples, according to TLL lines 32–4), but there is a significance to the genres in which it does turn up (mainly satire and epigram: Lucilius, Horace, and particularly Martial, who prefers caldus by 7:3). Second, in the manuscripts of prose writers it is mainly in mundane or technical genres that caldus occurs. There is a hint here (particularly in the evidence from metrical texts) that in the Republic and early Empire caldus was still considered non-standard or at least casual in tone. But caldus eventually became the norm, because it is the form surviving in the Romance languages: any purist movement, such as it was, eventually failed.

2.3 Audac(i)ter

A similar case is the syncopated adverb audacter alongside audaciter. Quintilian tells us (1.6.17) that there were certain pedants (he makes his views clear on this point) who insisted on saying the full form, whereas the orators all use the syncopated form: inhaerent tamen ei quidam molestissima diligentiae peruersitate, ut ‘audaciter’ potius dicant quam ‘audacter’, licet omnes oratores aliud sequantur (‘some however cling to it [analogy] with such perverse and irritating pedantry that they say audaciter rather than audacter, even though all the orators follow another course’). A few cases of audaciter appear in the manuscripts of Cicero (see Landgraf Reference Landgraf1914: 203 on S. Rosc. 104). Niedermann (Reference Niedermann1931: 48) points out that in an adverb of parallel structure (tenaciter < tenax) the unsyncopated form is the norm, and there are many other words of the same type without syncope (see Gradenwitz Reference Gradenwitz1904: 418, citing e.g. contumaciter, efficaciter, fallaciter, mendicaciter, minaciter, mordaciter, procaciter).⁴ Clearly syncope was up to a point lexically determined: it caught on in some words but not others. We would not hear of anyone accused of molestissima diligentiae peruersitas for using tenaciter. Quintilian's testimony shows that syncope in some lexical items might be something of a battle ground among the educated, with some insisting on preservation of the old form but others regarding it as pretentious. A distinction between educated and Vulgar Latin is simply not an issue in this case.

2.4 Dom(i)nus

The influence of idiom on syncope (see above), and the ability of speakers to monitor the phenomenon and either resist or accept its working, are also shown by the case of dominus/domina, as discussed by Väänänen (Reference Väänänen1981a: 42; cf. Reference Väänänen1966: 43), following the treatment of the word in TLL and some other works. In inscriptions the syncopated form domnus/domna is found mainly as a title (TLLv.1.1907.34ff.; cf. Carnoy Reference Carnoy1906: 114, Väänänen Reference Väänänen1966: 43, the last citing a number of examples in the Pompeian inscriptions). The referents include emperors (TLLv.1.1907.49ff.) and also clerics and saints (40ff.). By contrast in reference to the Christian God the syncopated form is rarely used (TLLv.1.1907.76ff.). The full form must have been felt to have a particular dignity; Christians had a habit of reviving archaic usages in creating a technical terminology.⁵ The fact that as a title domnus/domna was respectful did not inhibit its syncopation.⁶ We cannot conclude that the syncopated form had a substandard flavour, even if the full form was felt to be on a higher plane. On this evidence dominus and domnus coexisted for a while, though it was the syncopated form only that survived into Romance languages (Väänänen Reference Väänänen1981a: 42). The material provided by literary manuscripts cannot be trusted (TLLv.1.1907.68f.), and is not dealt with systematically by the TLL.

2.5 Ualde

Also revealing is the case of ualde (< ualide), alongside ualidus, which remains unsyncopated. Validus in Classical Latin was a high-register word, a synonym often of the more mundane firmus (see Adams Reference Adams1974: 59–60). It possibly had little place in speech. Valde on the other hand was a familiar term, used particularly in Cicero's letters (about 250 times) but much more sparingly in the speeches (about twenty-five times: see Pinkster Reference Pinkster2010: 192 for statistics; cf. Wölfflin Reference Wölfflin and Meyer1933: 134–5, Hofmann and Ricottilli Reference Hofmann and Ricottilli2003: 202–3). In Plautus the unsyncopated form ualide is still preferred (by 10:4) but thereafter ualde is the norm. On the other hand ualidius and ualidissime had some currency as the comparative and superlative forms of ualde, to judge from the material assembled by Pinkster (Reference Pinkster2010: 192–3). This then is a complex case. Syncopation was avoided in the high literary term ualidus but was the norm in the intensifier ualde, which is ten times more numerous in Cicero's letters than his speeches; but there seems to have been some resistance to the forms ualdius and ualdissime. The latter pair do occur (Neue and Wagener Reference Neue and Wagener1892–1905: ii.759), and it would be interesting to have full statistics for the reduced versus unsyncopated comparative and superlative forms of the adverb.

2.6 New evidence

The frequency of syncope in speech is confirmed by new subliterary texts, which were in at least some cases taken down from dictation. For examples in the letters of Terentianus see Adams (Reference Adams1977a: 21), adding now 468.11 sṭṛ[a]glum and the same form straglum in the fragment of the same letter (Rodgers Reference Rodgers1970 = CEL 143). For the numerous examples in the legal documents in the name of C. Novius Eunus (TPSulp. 51, 52, 67, 68; see Camodeca Reference Camodeca1999) of the first half of the first century ad, texts that have a version in the hand of Eunus himself, probably written from dictation, see Adams (Reference Adams1990a: 231–2). In the exterior, scribal, versions of the same documents these syncopes do not occur. There is a distinction hinted at between writing and speech. The scribes knew how to spell, but they might well have dictated the texts to Eunus, and if so he was presumably recording their pronunciation as he heard it. For syncope in the ostraca of Wâdi Fawâkhir (including the unusual and surely substandard type exiut) see Cugusi (Reference Cugusi1981: 744). Syncope is now well attested in Vindolanda tablets (Adams Reference Adams2003b: 536–7, 539–40; also Reference Adams1995a: 92), most notably in the form pestlus for pessulus (597b), which not only shows omission of u but also the insertion of t to render acceptable the non-Latin consonant cluster -sl- (Adams Reference Adams2003b: 539–40). Pestlus is transitional between the form of CL (pessulus) and that reflected in Romance (pesclus, with a further change tl > kl, for which phenomenon see e.g. Adams Reference Adams1977a: 33–4, Reference Adams2003b: 540, Cuzzolin Reference Cuzzolin2010, and below, 2.7). A high proportion of the syncopations in the Vindolanda tablets (nine of the fifteen) show loss of u between c/g and l (such as euericlum for euerriculum at Tab. Vindol. 593), which was clearly a favoured environment for syncope (Väänänen Reference Väänänen1981a: 42–3); in some cases the full form was itself secondary, an original cluster -tl- having been modified by an epenthetic vowel as well as by change of the first consonant to k. In the Bu Njem ostraca there are ten occurrences of speclis (Adams Reference Adams1994b: 106). This type of syncope is so banal in written texts that it must be doubtful whether there were many speakers who resisted it.

The same (sometimes unusual) syncopated form turning up in different corpora may suggest that the syncopation had become standard in ordinary speech. A pewter tablet from London (Hassall and Tomlin Reference Hassall and Tomlin1999: 375) with (accusative) list of valuables has the form Alxadri (pastellos auri viii sabanum Alạḍṛị | mappam…). Whether this is an abbreviation remains unclear (see the editors ad loc.; the text has many uncertainties). But the form of the second syllable can be paralleled, in the Pompeian tablets associated with Eunus. Twice the form Alxadrini occurs (TPSulp. 51, 52). The loss of the nasal before a stop (-ad- for -and-) is commonplace, and can be paralleled in this name (see Adams Reference Adams1990a: 241). One might be tempted to treat the syncope in the second syllable as a slip, but the pair of examples suggests that the spelling does reflect a common pronunciation.

A new Vindolanda text (Tab. Vindol. 596) has uirdem for uiridem, a form (= ‘green’) that was to pass into Romance (see Baehrens Reference Baehrens1922: 14, 16, discussing the term in the Appendix Probi). Three times the same syncopated form has turned up in military contexts from the eastern desert of Egypt, as a neuter plural noun uirdia ‘greens’ = CL holera: O. Max. inv. 80, O. Faw. 2, text published by Bülow-Jacobsen et al. (1994: 35; with discussion of the meaning). The second vowel of uiridia is under the accent and would not usually be subject to syncope, but syncope must have occurred in the base form uíridis and then spread to other forms.

Vetranus for ueteranus occurs in a letter of Terentianus (468.6; see Adams Reference Adams1978: 21). Similarly a transliterated text (in Greek letters) in the hand of a slave trader Aeschines Flavianus of Miletus (SBiii.1.6304 = CPL 193) has the phonetic spelling βετρανε (feminine genitive singular). There are signs that an element of dictation went into the composing of this text (see Adams Reference Adams2003a: 60).

Two other spellings in Terentianus are postae (467.23) and singlare (468.14–15). Both can be paralleled in classical verse (Lucr. 1.1059, 6.1067, Sil. 13.553),⁷ and must have been widespread.

2.7 Latin loan-words in Greek

Some idea of the extent of syncopated forms in imperial Latin may be obtained from Binder's account (Reference Binder2000: 153–214) of syncope in Latin loan-words into Greek, in which the phenomena are classified according to the phonetic environment (e.g. on cul-/-cl- see 156–172; a notable section (176–84) concerns -tul-/-tl-/-cl, for which see above, 2.6). In many cases it is only the syncopated form of a word that turns up in Greek, as Binder's classifications make clear: there are separate sections for terms attested only in a syncopated form, and for those that are attested in both a syncopated and unsyncopated form. Of particular interest here is Binder's observation (210) that sometimes an official title is found in unsyncopated form whereas the base term has a syncopated form, as in the pairs τάβλα but ταβουλάριος, κουβούκλιον but κουβικουλάριος, κορνίκλιον but κορνικουλάριος. Official titles occur mainly in writing and retain their original form whereas the base terms are part of everyday speech and the syncopated form becomes established even in the written language. There is manifested in the distinction a sense that there was a greater formality to the unsyncopated forms, and that sense may sometimes have had consequences in the spoken language, with speakers aspiring to use the full form when they thought that formality was required. The contrasting case of dominus above (2.4), however, which even as a respectful title was often syncopated (except in its Christian use), shows the impossibility of generalising about the status of syncopated versus unsyncopated forms. An individual lexical item may have its own distinctive history (on change tied to particular lexical items, referred to here as ‘lexical restriction’, see xi.4).

5.1 U before u in final syllable

Already in Pompeian graffiti forms such as fatus for fatuus, ingenus for ingenuus and mortus for mortuus are common (Väänänen Reference Väänänen1966: 41; cf. TPSulp. 68 (dated 15 Sept. ad 39) mutos = mutuos; for later examples see e.g. Niedermann [Reference Niedermann1944] Reference Niedermann1954b: 100, Stotz Reference Stotz1996: 71), and these anticipate Romance forms (It. morto, Sp. muerto etc.). Caesar in his De analogia (frg. 27 Funaioli) argued that the form mortus not mortuus was correct for the past participle, because forms with -uu- were nomina (adjectives) not participles: mortuus ex qua parte orationis declinetur incertum est. nam sicut ait Caesar, ab eo quod est morior, in participio praeteriti temporis in tus exire debuit, per unum scilicet u, non per duo. nam ubi geminata est littera, nomen est, non participium, ut fatuus arduus. Here is possibly a sign that the reduced form was already current in educated speech at the time of Caesar (cf. Willi Reference Willi2010: 232, 242 with n. 26), and that an attempt was made to draw a distinction between the functions of two competing forms, -uu- and -u-. However, the tense of debuit stands against such an interpretation. If the citation is correct, Caesar must have meant that the participle should have had (in theory) the form -tus (but did not).

Note too battunt for battuunt, attributed by Fronto to Marcus, p. 50.6–7 van den Hout, on which see Holford-Strevens (Reference Holford-Strevens2010: 337–8). Väänänen refers to the loss (chute) of u before u or o (i.e. another back vowel: cf. quattor at e.g. TPSulp. 51, 52, CILvi.13302), and adds in brackets with a question mark ‘ou contraction des deux voyelles’. His hesitancy in speaking about contraction is understandable, as the contraction of two identical vowels would normally produce a long vowel (see e.g. Leumann Reference Leumann1977: 119–20, Väänänen Reference Väänänen1981a: 44). But if a contraction parallel to that of ii did occur the vowel in a morphologically significant ending would have been shortened to bring the ending into line with the nominative -us. The mechanism of the ‘loss’ in this case is thus problematic,¹⁹ but the loss itself was well established early.

Superficially different are spellings such as tus and sus for tuus and suus, because here the loss is under the accent, and Romance scholars have tended to give the forms special significance. Such reduced forms are reflected in the modern period, above all in Gallo-Romance (mon, ma, son, ton etc.), and there is sometimes found in the Romance literature a view that they derive from an atonic use of the possessives, as if some sort of ‘clitic reduction’²⁰ had taken place in the Latin period in unstressed uses of the possessives (see e.g. Elcock Reference Elcock1960: 84, C. Lyons Reference Lyons1986). Certainly, in the semantic sense, in Classical Latin there was both an emphatic and an unemphatic use of possessive adjectives, marked to some extent by their placement in relation to that of the nouns qualified: when the possessive was emphatic, contrastive or the like it tended to be placed before the noun, but its unmarked position was after.²¹ If, therefore, in the phonetic sense as well, there was an unstressed (clitic) use, obviously the unstressed forms would be expected to be located after the nouns qualified. In an unpublished ostracon from Carthage²² containing a number of possessive adjectives the one reduced form (tus (benefic[iu])) is placed in the emphatic position, and it is emphatic in the context, in that there is a contrast between the favour conferred by the addressee on the writer, and the failure of a ‘father’ to be of any assistance. By contrast three of the five full forms (tua, meu, tuo, meus, tuis) of the possessives in the letter are found in the unemphatic, postponed position. On the evidence of this letter clitic reduction in unstressed possessives should not be invoked as the determinant of reduction in the early (Latin) period.

Similarly in a letter of Terentianus (471), while there is one case of tus in the supposed ‘clitic’ position (17 quo pater tus mi mandauit), there is also an example of sus which is both preposed and emphatic in the context (30 non magis qurauit me pro xylesphongium sed sum negotium et circa res suas). Notable in this second passage is the fact that in the next phrase suas, apparently in the unemphatic position, is given the full form.

A different determinant from a vague clitic reduction must be sought, and it is obvious what that determinant was. In the letters referred to the reduced forms are found when u is followed by another u, whether or not the possessive is emphatic and regardless of its position in relation to the noun. ‘Reduction’ thus seems to have had its starting point (at least in the imperial period: early Latin is possibly another matter)²³ as a form of loss or contraction (see above for these categories) when the u in hiatus was followed by another u. In the African ostracon when u is followed by a vowel of different quality the original form of the word is retained. Once sus and tus were established in this way analogy might have caused the emergence of sa, ta etc.; there is already an example of ma in Terentianus (471.34) and another, possibly, at Vindolanda (Adams Reference Adams1995b: 120). It is not clear whether the reduced forms of meus (see e.g. CILxi.746 and Svennung Reference Svennung1936: 16) had a different origin from those of tuus and suus (on ‘synizesis’ in forms of meus in Plautus see Lindsay Reference Lindsay1922: 61, Questa Reference Questa2007: 175).

The above conclusion is confirmed by practice in another African corpus, the Tablettes Albertini of the late fifth century (493–6). Suus retains its full form when the second vowel is other than u, but suum and suus are regularly contracted. Sum occurs seventeen times²⁴ and sus once (x.4); suum is found just once (vii.20) and suus not at all. By contrast suis (forty-six times), sui (four times), suoru(m) (six times), suo (three times), suam (eleven times), sue (once), sua (twice) and suos (four times) regularly have the full form and are never reduced. Compare, for example, xxxi.5 emtorem sum with xxxii.4 emtore suo, and xxxi.7 sum esse dixerit with xxxii.5 suam esse dixerit. The variable spelling of the forms of suus is exactly paralleled by that of the forms of perpetuus in the same corpus. uu is regularly contracted (in the expression in perpetum),²⁵ whereas the u is retained in the form perpetuo (see xii.12, xv.24).²⁶

The reduction of the possessives tuus, suus noted above as showing up particularly in Gallo-Romance was not universal in the Romance languages. Note for example OSp. súe, túe (feminine), which fall into the class described by Lloyd (Reference Lloyd1987: 191; see above, n. 1) as showing a high vowel under the accent which remained in hiatus.

7.1 Yod in initial position before a vowel or in intervocalic position

The yod of Classical Latin found at the start of words before vowels or intervocalically within words seems to have developed, at least in some varieties of the language, to an affricate that might be tentatively represented dz. The posited development was j > dj > dz.³⁶ The suggested final stage is represented by spellings of the type Zanuario (CILx.2466), Zouiano (xiv.1033), huzus (vi.37200) and Zerax = Ἱέραξ (x.3699.16), the intermediate stage by Diauolenus = Iauolenus (cited by Leumann Reference Leumann1977: 130 without a reference).

7.2 Yod after certain stops

The first signs of palatalisations are to be seen in misspellings showing s(i) or z(i) for di or ti, e.g.:

Further examples may be found in Svennung (Reference Svennung1936: 9), Sturtevant (Reference Sturtevant1940: 172),³⁷ Mras (Reference Mras1948: 92–5), Battisti (Reference Battisti1949: 147), Bonioli (Reference Bonioli1962: 116, 119), Kiss (Reference Kiss1972: 55–6), Väänänen (Reference Väänänen1981a: 53–4), Biville (Reference Biville1990: 277–8). These developments were so entrenched by the time of the African grammarian Pompeius (fifth century) that he presents the palatalised pronunciations of Titius, Auentius, Amantius, meridies and dies as correct (GLv.286.6–33; see the text as it is printed by Kramer Reference Kramer1976: 70); the spelling pronunciations were by then considered a uitium.³⁸ Nor was Pompeius the only grammarian who alluded to the changes and was tolerant of them (see too Servius GLiv.445.8–12, Papirianus ap. Cassiod. GLvii.216.8–9, Servius on Virg. Georg. 2.126, and Mras Reference Mras1948: 87–9).

The combination [dj] seems to show varying treatments in the same words in texts and inscriptions. For example, the frequentative adiuto appears in Terentianus in the form aiuto (468.41, 471.28) (cf. CILvi.37748 aiutori), but at (e.g.) CILviii.18224 there is a spelling azutoribus for adiutoribus. A parallel for aiuto is the form ios(s)um for deorsum (found e.g. in the Mulomedicina Chironis: TLLv.1.559.27ff.). Likewise on the one hand there are spellings such as zebus = diebus (CILxiv.1137) and on the other ies = dies (iii.2225).³⁹

There are various ways of explaining the variation. First, dj might have had two phonetic outcomes current in different social or regional dialects, which we might represent loosely as j and dz. As a variant on this view Grandgent (Reference Grandgent1907: 114–15) favoured a chronological explanation. In the ‘latter part of the Empire’ dy was reduced to y, but ‘towards the end of the Empire’ it ‘had another – doubtless more elegant – pronunciation, which was probably dz’. This is mere guesswork. A second possibility is to interpret the different symbols as representing the same sound, perhaps some sort of affricate (dz or the like). It was seen above (7.1) that the original intervocalic (and initial) yod developed (sometimes) to such an affricate, represented in texts as z. The letter i (originally representing j) when correctly written intervocalically in the traditional manner might have been seen by some to stand for (d)z in speech, and thus might sometimes have been used to represent the dz that developed from original dj: thus aiuto might simply be a spelling for a(d)zuto.

This second explanation does not ring true for the form aiuto in Terentianus, in the early second century ad. It was noted above (see n. 37) that assibilated misspellings do not seem to occur before the third century, and it is hard to believe in the absence of explicit evidence that the affricate had emerged by the time of Terentianus. It seems likely that an early outcome of dj was j. Its relationship to dz is uncertain.⁴⁰

According to Bonfante (Reference Bonfante and Bonfante1999: 29) the ‘assibilation of the cluster -dy- in intervocalic position…is observed only in the inscriptions of the imperial period of Italy and Africa’. Certainly most of the evidence comes from those places, but it is problematic on this basis to set up a regional variation in the Latin period. Even Bonfante's presentation of the matter raises doubts. The inverse spelling di for z is cited by Bonfante (Reference Bonfante and Bonfante1999: 30) as further evidence of the development, but (among other sources) from the Peregrinatio Aetheriae (baptidio for baptizo, five times), and that is a north-western text (from Spain or southern Gaul: Adams Reference Adams2007: 342–53). Correct spellings do not tell us what was happening in speech, and it may merely be the lower cultural level of a good deal of the writing that survives from Africa and parts of Italy that betrays the developments in speech there. The Romance outcomes of the cluster in areas other than Italy (Gallo-Romance, Ibero-Romance) can be derived from later evolutions of the same type of change as that attested by Italian and African misspellings (see Herman Reference Herman and Wright2000: 44).

Spellings of the type terciae for tertiae (CILxv.4376, a fragmentary text; see the material in Mras Reference Mras1948: 91–2, Bonioli Reference Bonioli1962: 113–14, Stotz Reference Stotz1996: 220–1) suggest that the combinations kj and tj were approaching each other in pronunciation, without necessarily becoming identical,⁴¹ though the interpretation of the evidence is problematic (see Carnoy Reference Carnoy1916: 146–7 and B. Löfstedt Reference Löfstedt1961: 172 for attempts to explain the significance of the misspellings). Indeed in parts of Romance original kj and tj were kept apart (B. Löfstedt Reference Löfstedt1961: 170, Väänänen Reference Väänänen1981a: 54, Lloyd Reference Lloyd1987: 259). The spelling ti for ci is rarer and later than the reverse spelling (B. Löfstedt Reference Löfstedt1961: 171, Stotz Reference Stotz1996: 188), and some cases (and indeed some cases of ci for ti) reflect not phonetic developments but popular etymologies or changes of suffix (B. Löfstedt Reference Löfstedt1961: 171).

1.1 ‘Omission’ of final -m and Vulgar Latin

According to Quintilian final -m before a following vowel was ‘obscured’ rather than dropped without trace, such that it became a sound that would need a ‘new letter’ to represent it. It is a reasonable guess that the vowel was nasalised, and possibly lengthened as well (see Allen Reference Allen1978: 30–1):

Quintilian was speaking of standard educated pronunciation, not that of the uneducated. So in classical verse elision occurs when a final syllable ending with -m precedes a vowel at the start of the next word.

See too Velius Longus GLvii.54.2–6, and particularly 54.13–15 (on the interpretation of which see Allen Reference Allen1978: 30 with 96):

The peregrina littera corresponds to Quintilian's noua littera. Velius likewise was not talking of lower-class speech. There is a new element in this passage. The examples show that Velius thought that final -m had been reduced not only before vowels but also before consonants. A feature of pronunciation that first emerged before vowels must have spread to pre-consonantal position.

Quintilian's ‘muffled’ articulation can be traced well back into the Republic, because the letter is sometimes left out in early inscriptions, as in two of the Scipionic elogia (CILi².7, 9) (see Leumann Reference Leumann1977: 223–6). In the first of these there are four omissions, none of them before a following vowel. In the second there are eight omissions, only two of them before a vowel.

The omission of -m in writing is not to be classified as a feature of lower sociolects, because it is a reflection of educated (indeed all) speech. Nevertheless the phenomenon is given a prominent place in handbooks of Vulgar Latin.¹ Omissions reflect rather poor literacy, at least in the imperial period; in earlier Latin there was a more relaxed attitude to spelling, and the phonetic spelling might not yet have been stigmatised. There is a certain interest to the mechanisms of omission in written texts, though this is a subject to do with literacy practices not language use.

1.2 Final -m in monosyllables and ‘grammatical’ words

A distinction must be made between final -m in general and that in monosyllables or ‘grammatical’ words. In the latter it was assimilated to a following consonant.² Cicero twice (Orat. 154, Fam. 9.22.2) says that the combination cum nobis should be avoided because it is open to an obscene interpretation (cunno…). The m of cum was assimilated to the following n (and in educated speech). At Fam. 9.22.2 he also suggests that the combination illam dicam once uttered in the senate (probably by Cicero himself) might have been taken as obscene (= landicam ‘clitoris’). Misspellings in inscriptions and writing tablets imply assimilations, such as tan durum (CILiv.1895), where the nasal was presumably alveolar under the influence of the alveolar stop. Cf. O. Faw. 2.3, 6 tan leuem, tan leuis. In tan cito at O. Faw. 2.4 the final consonant of tan must represent a velar nasal before the following velar consonant. There are traces of the nasal in a few monosyllables of the Romance languages (e.g. Fr. rien and mon before vowels, but with change to n: see Väänänen Reference Väänänen1981a: 67).

Like final -m, final -n was also assimilated thus in proclitic monosyllables. Terentianus writes im mensem (468.26), im perpetuo (468.65) and im bia (470.26). At Pompeii im balneum (CILiv.2410) showing n > m before b contrasts with tan durum (see Väänänen Reference Väänänen1966: 66). Assimilation affected not only nasals at the end of monosyllables, but stops such as t and d (see below, 3.5).

Spelling was not consistent. Once assimilation had undermined spelling regularity a scribe might sometimes mechanically write tan or con even when the -n was not determined by a following consonant.

The Ciceronian evidence shows that this phenomenon too was a feature of the language in general, not of a lower sociolect.

1.3 Orthographic tendencies

B. Löfstedt (Reference Löfstedt1961: 226–8) showed that in the Sangallensis of the Edictus Rothari (seventh century) -a was far more often written for -am than were -u/-o for -um and -e for -em. This orthographic tendency shows up in other late texts as well, such as the Anonymus Valesianusii (Adams Reference Adams1976: 51–2). It can now be confirmed for a much earlier period, from the Bu Njem ostraca of the mid third century ad (see Adams Reference Adams1994b: 106–7). There the -a ending predominates over -am by 18:3, whereas the -em ending is preferred to -e by 19:2. These latter figures may be misleading, because eighteen of the examples of -em are in the greeting salutem, but the abnormal frequency of -a for -am is nevertheless upheld by the figures for -u versus -um: these two forms are equally common (22 against 19). In the letters of Terentianus in accusative singular endings -m is omitted about once in every five instances (of 149 accusatives, including those in participles, twenty-nine are written without -m: see Adams Reference Adams1977a: 22), but there are variations of frequency according to the declension. -a is outnumbered by -am by about 2:1, -e by -em by 3:1 but -u by -um by more than 7:1 (see Adams Reference Adams1977a: 23). Scribes were more likely to drop the -m in first declension feminines than in the second and third declension accusative singular, and from quite an early period.

It is unlikely that there is a phonetic significance to this tendency. Löfstedt (Reference Löfstedt1961: 228) relates the frequency of -a to the passage of neuter plurals into the feminine singular. Those conscious of this development were perhaps more tolerant of -a in the accusative because of a sense that it might be an original neuter. The relative rarity of -u may have something to do with the fact that it did not look like a second declension form.

There is a full account of omissions in inscriptions by Diehl (1899). He deals with non-linguistic determinants (190–242), such as abbreviation (211–20), lack of space at the edge of a stone (220–31), a following -m (231–2) and slips by stonemasons (232–5), but once these are allowed for there remain large numbers of omissions that cannot be explained away (243–86). Apart from inscriptions, in those documentary texts on materials other than stone in which omissions are common there seems to be little or no sign of such mechanical determinants. In later sections (2.5.1–4) it will be seen, for example, that the omission of -s in inscriptions is predominantly due to lack of space at the end of a line, and other factors come into it too (such as abbreviation), but -m by contrast is often left out in sentence-internal or phrase-internal positions where extraneous factors cannot be invoked as the cause. Nor are there restrictions on the phonetic environments in which omission is found or on the lexical classes. Letter 468 of Terentianus is revealing in these respects. The degree of spelling error in this corpus varies from letter to letter in reflection of the practices of the different scribes (see Halla-aho Reference Halla-aho2003). Letter 468, in contrast to 467, is full of mistakes (Halla-aho Reference Halla-aho2003: 248), and here omissions occur not only in accusatives but also in particles (aute) and verbs (iacuisse, speraba). Phrase-internal omissions in accusatives occur freely before consonants (e.g. in 22, 51, 52, 59, 60–1) as well as vowels, and it could not be convincingly argued that the weakness (whatever its phonetic details) of the phoneme in final position was conditioned by the phoneme that followed.

1.4 Conclusions

The evidence concerning the treatment of m in final position is clear-cut, the one uncertainty being the phonetic character of the ‘new’ or ‘foreign letter’ that replaced it. There is one point on which grammarians are unequivocal. Developments affecting -m were a feature of the speech of the class to which the grammarians themselves belonged and not restricted to uneducated varieties. There was no advocating of spelling pronunciations. The occasional omission of -m goes well back in the Republic and is found in high-style inscriptions, and that hints at the state of elite speech at an early period. The omission of final m has found its way into handbooks of Vulgar Latin because of a tendency to confuse writing with speech. It was by the classical period merely a vulgarism of writing.

2.1 The Romance languages

In the Romance languages there is a split between the West and the East in the treatment of final -s (see e.g. Rohlfs Reference Rohlfs1966: 431; cf. Bourciez Reference Bourciez1946: 51, B. Löfstedt Reference Löfstedt1961: 129, Väänänen Reference Väänänen1981a: 68, Weiss Reference Weiss2009: 514). In the West (France, Spain, Sardinia and part of Raetia) -s was retained up to a point (Fr. le fils, Sp. tres libros), though with a good deal of regional and diachronic variation. In the East (Romanian, Italian and dialects) it was lost (e.g. It. meno, peggio, meglio, credi, lodiamo, foste, fuori < minus, peius, melius, credis, laudamus, fuistis, foris; on the other hand after an original long vowel in the monosyllables nos and uos it was vocalised as [j], noi, voi). There are signs that in parts of Italy it was retained into the medieval period, and in some northern dialects it was retained under certain circumstances, but the general tendency was towards loss (details in Rohlfs Reference Rohlfs1966: 431–4).

2.2 Early Latin

In early Latin inscriptions final -s after a short vowel is often omitted, and in early verse it often does not make position after a short vowel and before a consonant, a feature that points to its omission in speech in a restricted phonetic environment (for details with bibliography see Adams Reference Adams2007: 51–2, 74–5, 104 n. 319, 140–1 with n. 74; add Niedermann 1931: 131–2, E. P. Hamp 1959).³ In twenty plays of Plautus 1,058 cases of omission have been identified, compared with 770 cases where the s is present, a rate of deletion of 57.9 per cent (see Wallace 1984a: 217; also 1982), though the -s was retained in orthography (see Questa Reference Questa2007: 32–5). By the end of the Republic the evidence of verse suggests that its restoration was widespread, though a passage of Quintilian (9.4.38), which cites various republican authorities, shows that there was some controversy in the late Republic about whether and under what circumstances -s should be articulated in final position (for this passage see below, 2.5.5). On the relatively low number of omissions in Lucretius (forty-four cases in 7,400 verses, or one every 169 lines, compared with one every five lines in Ennius) and the limited environments in which they occur (mostly in the fifth foot, with exceptions found mainly before forms of esse beginning with s) see Butterfield (2008: e.g. 194, 198 with n. 36). The sole instance in Catullus (116.8dabi’ supplicium) precedes an s at the start of the next word (a phenomenon that will be seen repeatedly in this chapter: see also above, previous sentence, and below, 2.4), and may contain an allusion to a passage of Ennius (see Timpanaro 1978: 179 n. 45). In late republican and Augustan poetry there are a number of verses in which every word ends with an s (see Housman Reference Housman1937b: 106 on Manil. 4.780, Courtney 1980: 305 on Juv. 6.O3), and that is an effect that may depend on the articulation of the sound in final position. Further investigation is needed, as there may be many more lines of this type, and if so cases would have to be classified according to the length of the preceding vowel and the phoneme following.

It is in theory possible that by the late Republic deletion of -s before a consonant was still taking place in speech, but that poets had come to eliminate from verse any sign of this tendency. There is, however, further evidence suggesting that -s was now articulated, both testimonia in literary sources (see the next paragraph) and the constant writing of the letter in otherwise badly spelt non-literary documents (see 2.4). Late republican poetic practice is likely to reflect the restoration of -s in all environments in speech.

This restoration of final -s between the time of early republican literature and the late Republic looks like the product of a standardisation movement.⁴ Indeed Messala wrote a book on the letter s (Quint. 1.7.23), and the other passage of Quintilian (9.4.38: see above) shows that the ‘letter’ in this position was actively discussed. According to Cicero (Orat. 161) the omission of -s was by his time ‘rather rustic’ (subrusticum). Did he mean that he had noticed the phoneme dropped in rural dialects, or was he simply disparaging those educated late republican scholars who went on omitting -s? It would be going too far to attach a regional significance to his remark. He probably meant that the omission was now out of keeping with educated practice (see Adams Reference Adams2007: 140–1, 146).

In Plautus some evidence has been found that deletion is less common in cantica than in spoken metres and recitative (see Wallace 1982: 123, Reference Wallace1984a: 221–3). Since the sung parts of Plautine comedy are considered to be more formal in style than the other parts it is possible that in careful speech there was some effort made to maintain -s, and the later restoration of the consonant in final position would represent a movement to impose the prestige variant across all stylistic levels (as distinct from a movement to reintroduce a phoneme that had been completely lost in a particular environment). Statistically, however, the distinction between cantica and the other parts is not striking (Wallace 1982: 123 notes a deletion rate of 60.2 per cent in spoken verses and recitative, and 43.9 per cent in cantica), and it would not do to place too much reliance on it in making generalisations about the nature of the later change. The scansion of cantica is notoriously problematic, and any figures to do with deletion of -s must be subject to considerable scepticism. Ideally the evidence should have been set out in detail and not in the form of raw figures.

Drexler (1973: 132) observed in early inscriptions a similar stylistic distinction between presence and absence of -s to that argued for Plautus by Wallace. He undertook a rough survey of the sacral and funerary inscriptions in Diehl's (Reference Diehl1930) selection of early inscriptions and found that 119 sacral inscriptions had complete preservation of -s and only twenty-three showed some omissions. By contrast of the funerary inscriptions considered only seven had complete preservation but thirty-two had some omissions. Funerary inscriptions often aspire to an intimate tone. Drexler also noted that in the laws and decrees assembled by Diehl he had found just one case of an omission. These figures are suggestive of a different treatment of -s in formal versus informal discourse, though one would like to see an exhaustive survey of all the inscriptions down to about 150 bc. There is also the question whether the variations of spelling are relevant to the spoken language. Drexler's evidence would be consistent with a conclusion that there was widespread loss of -s in speech, which was obscured more carefully in formal writing than in less formal.

A finding of Wallace's second paper on Plautus is that deletion was to a considerable extent lexically determined, i.e. that it is far more common in some lexical items than others (Wallace 1984a: 218, 223–5). This is the phenomenon again of lexical restriction (see above, iv.4): a sound change will not suddenly occur in all terms in which it is possible but will begin in some words and spread gradually (see Wallace 1984a: 223–5 for discussion). The change may stop and fail to affect all lexical items. In our case it seems to have been reversed.

2.3 Questions

Was there any continuity between usage in the earlier Republic, when -s was often left out in inscriptions from Italy, and Romance, where it was lost in Italy? If -s was restored in educated speech by the late Republic (except perhaps in that of eccentrics who preferred to follow the auctoritas of the ancients), was the restoration also widespread in non-educated usage, as revealed perhaps by imperial documents that are otherwise badly spelt? Or was Cicero right in the strict sense to say that the omission was ‘rather rustic’? Was there a distinction between educated and non-educated practice?

The case can be made that there is only one conclusion fully justified by the evidence. Final -s was restored right across the social spectrum and in all areas by the early Empire, and it was maintained for centuries. The loss that shows up in Italian and some other Romance languages must have occurred very late, and was not a direct continuation of the situation obtaining in the archaic period.

In the following sections we will consider, first (2.4), non-literary documents (some of them of relatively recent publication) excluding inscriptions on stone, which raise special problems of interpretation; then inscriptions themselves (2.5), drawing on discussions of -s that go back more than a hundred years; and finally (2.5.5, 2.7.2) the view that there was indeed a direct link between the early period and late Latin/Romance.

2.4 The evidence of non-literary documents other than epigraphic

The stability of -s during the Empire is suggested by the rarity of its omission in non-literary texts that by contrast omit -m constantly (for this methodological point see Väänänen Reference Väänänen1966: 81: it is the contrast that is important), and display many other phonetically inspired spelling errors.⁵ These texts are spread over about four centuries, and all present the same picture. They have tended to be disregarded by those discussing -s even in recent times.

In the Bu Njem ostraca there are about 363 cases of -s written against no cases of its omission. By contrast -m is in effect more often omitted than written (Adams Reference Adams1994b: 106–7, and above, 1.3). It is left out forty-four times and written fifty-four, but eighteen of the latter cases are accounted for by the formulaic greeting salutem, which scribes knew how to spell. It may be left aside.

In a much later African corpus, the Tablettes Albertini (of the 490s), the contrast is similar. The omission of -m is ‘très fréquente’ (Väänänen Reference Väänänen1965: 29; cf. Courtois et al. 1952: 68), but Väänänen (Reference Väänänen1965: 30) lists only four cases of the omission of -s in the nominative masculine ending -us, all of them in names, three in a long list of names (Tabl. Alb.xxx.9) of which most have the correct spelling. The omissions must be slips. A few other apparent omissions, such as eiusdem gemioni (xv.17) and ponderi plenum (vii.16), reflect confusions of case or declension rather than a phonetic development (ponderi is obviously an ablative: see further Väänänen Reference Väänänen1965: 30, 35).

In the letters of Terentianus, in which -m is constantly omitted (see above, 1.3), there are forty cases of -s written after a short vowel (it is written correctly even more often after long vowels: see further below) against just one omission. That is a special case (Terentianus 471.21 pater meu sopera), with an s following; for this type of omission see above, 2.2 and Carnoy (Reference Carnoy1906: 183), Proskauer (1910: 108), Väänänen (Reference Väänänen1966: 77, 79), Adams (Reference Adams1977a: 30 with n. 112), Butterfield (2008: 194), and particularly Quint. 9.4.37 (below, 2.5.5). There is a second omission at 468.25 subtalare ed udones. Here the preceding vowel is long, and in that position final -s is preserved to an overwhelming extent throughout Latin (see e.g. Väänänen Reference Väänänen1966: 77, Adams Reference Adams2003b: 538). It is not clear what lies behind occasional omissions after a long vowel: a regional variety of Latin, a social dialect or a mere slip? The last possibility is not unlikely, given the infrequency of the phenomenon. Also worth noting is the use of si qui for the expected si quis at 468.41, before sibi. This is not a mere omission, because qui, with a long vowel, is a different word from quis, with a short, but the switch from the one word to the other was from the early period (Plautus) often motivated by a feeling against s + s at a word boundary (see E. Löfstedt Reference Löfstedt1956: ii.82–3 on si quis versus si qui).

In the ostraca of Wâdi Fawâkhir (letters bearing the name of Rustius Barbarus) of about the second century -s is written fifteen times after a short vowel and thirty-one times after a long vowel, and never omitted (see however Cugusi Reference Cugusi1981: 743 on the interpretation of fasco coliclos at 2.11). On the omission of -m in this small corpus see Cugusi (Reference Cugusi1981: 743).

In the Vindonissa tablets there are fifty-six cases of -s correctly written after a short vowel (often in personal names in -us), with no omissions.

In the Bath curse tablets there are about 123 instances of -s correctly written after a short vowel. These are often in personal names in -us. Tomlin (Reference Tomlin and Cunliffe1988: 76) notes two omissions, Suli for Sulis alongside deae at 94.5 (but here there is probably confusion with the dative), and sua alongside mentes at 5.5 (but here the omission is after a long vowel).

In the Vindolanda tablets there is just one omission found to date (Tab. Vindol. 643 (ii) habea), and that too is after a long vowel (Adams Reference Adams2003b: 538).

In the Carlisle writing tablets (see Tomlin 1998) there is an omission in a nominative name in -us (Tab. Luguval. 18.1 Primu et Anoncletus). In this small corpus there are about thirty-six instances of -s correctly written, and not infrequently in the nominative of personal names (as in Anoncletus above alongside Primu; cf. e.g. Tab. Luguval. 16, a long text with a number of such names; also 17, 38).

In the documents of C. Novius Eunus (TPSulp.) -s is written eighty-three times and omitted twice (see Adams Reference Adams1990a: 237). Both omissions are after a long vowel. In one case (TPSulp. 68.3.3 de singulos for dies singulos) an s begins the next word. In the other (68.3.4 nummo obligatum) there might have been a confusion of case in a syntactically difficult legal formula (cf. 68.5.14 HS xx nummos obligatum iri). On the omission of -m see Adams (Reference Adams1990a: 236–7).

It is also worth observing the practice of the curse tablets that had been published by the second decade of the twentieth century (notably by Audollent 1904). Jeanneret (Reference Jeanneret1918: 58–60) collects and discusses omissions in these tablets. Eighteen omissions are noted, compared with 650 omissions of -m (59), and the omissions of -s can mostly be explained away (60), because the letter would be at the end of a line (on the significance of this factor, see below, 2.5.1), or because s follows, or because the interpretation of the tablet is not certain.

The evidence of this substantial group of documents suggests that after short vowels there had been a restoration of -s. The stray examples after long vowels are so infrequent that no linguistic theory can be based on them.

It is, however, worthwhile to raise at this point another methodological question. It is in theory possible that a single misspelling or small cluster of misspellings in a corpus that otherwise spells the form in question correctly may reveal the true state of the spoken language, which had otherwise been concealed by scribal correctness (on this point see Adams Reference Adams1995a: 87, Reference Adams2007: 44, and below, 3.4, p. 154). One might therefore suggest that (e.g.) the spelling Primu in the Carlisle tablet above betrays the deletion of -s that had taken place in speech. There is an argument that can be advanced against such a conclusion in the case of -s. Much depends on whether there are other texts in which the misspelling is common. A tiny handful of cases of e for ae in one or two Vindolanda tablets can with confidence be taken to represent a feature of speech intruding for once into the written language, because we know from a host of other misspelt texts that ae had been monophthongised. The instance of -s omitted at Carlisle is rather different. We do not know from a host of other texts of the period that -s had been deleted at all. Omissions are consistently very few or non-existent in corpus after corpus, or subject to mechanical explanations, and that despite the fact that in the same corpora phonetic spellings of other types are commonplace. To undermine the view that is being put forward here about the generalised restoration of -s in the imperial period one would need to find at least one text in which instances of deletion were frequent and not readily explained away.

The most remarkable evidence to do with final -s comes from the graffiti of La Graufesenque in southern Gaul, of the first century ad, as described by Marichal (Reference Marichal1988: 68–70); see also Flobert (1992: 108–9). In this archive there are texts in Gaulish and texts in Latin. Personal (masculine) names either have the Celtic ending -os or the Latin -us. The incidence of reduction by omission of the final consonant varies strikingly between the two types. Marichal (Reference Marichal1988: 69) summarises as follows. If Latin personal names or names that have been Latinised with the ending -us, and common nouns in Latin texts with an s-ending, are considered together, there are 126 occurrences of -s, with only a single omission. On the other hand if Gaulish personal names or names that have been Gallicised with the ending -os, and common nouns in Gaulish texts that ought to have an s-ending, are considered together, there are 349 occurrences of -s, with fifty-eight omissions, a proportion of 17 per cent. Marichal convincingly rules out the possibility that the omissions may reflect abbreviation. There seems to be a difference between the state of the two languages, with Gaulish showing a marked tendency to drop the final consonant but Latin showing no such tendency. However the language situation is to be interpreted in the pottery, the contrast between the two languages must be significant. There are various ways of explaining it. Final -s might have been so entrenched in Latin speech (ordinary speech, it should be stressed, given the humble social level of the potters) that it was resilient even when under pressure from the tendency to omission in the other language spoken by the same potters. Alternatively Gaulish potters learning Latin and Latin literacy might have had it drummed into them that they had to write s in final position in that language. But it is hard to accept that a mere spelling rule could have had the effect of preserving -s in Latin written by such imperfect literates if they had not had Latin pronunciation to guide them.

We may mention finally the Visigothic slate tablets edited by Velázquez Soriano (2004). These are so late (sixth to eighth centuries) that arguably they might be treated as evidence for early Ibero-Romance rather than for practices of the Roman period, but they are worthy of comment because of conclusions that have been based on their apparent treatment of -s. Velázquez Soriano (2004: 503) notes that in this corpus -s is for the most part preserved, which is in line both with the survival of the consonant in this position in Ibero-Romance (see above, 2.1), and with the practice of the other non-standard Latin documents above. Herman (1995: 71) finds six ‘sure cases’ of omission, five of them in masculine nominative singulars supposedly written with -u. The number is described as remarkable, given the small size of the corpus, and Herman goes on (72) to offer a morphological explanation, arguing that there had been in Spain a neutralisation of nominative and accusative singular in the second declension, causing orthographic uncertainty. An examination of the five cases reveals that none will stand up to scrutiny. At 65.6 the editor prints Paulus, not Herman's Paulu, and at 40.iii.4 rogitus not rogitu. At 48.4 Valeriu is taken by Velázquez Soriano (Reference Velázquez Soriano2004: 274) to show omission of -m not -s. Vitellu at 46.15 precedes a word beginning with s. Finally, at 6.7 the incomplete -egiusnu is certainly a personal name with -s omitted, but the document contains a list of twenty masculine names, nineteen of which are correctly given the -us ending, and it follows that the one exception can only be a careless slip. Omission of -s is virtually non-existent in this corpus (which is far from small), and the data do not support deductions about the case system.

It would not do to argue on the evidence of the above corpora that there was a continuing loss of -s throughout the Empire in rustic (the word of Cicero) or lower-class speech, as all texts are well down the educational scale and from outlying places and -s is regularly written. The evidence for the stability of -s in African Latin provided by two of the above corpora may be supported from Latin loan-words into Berber dialects (see Bonioli 1962: 109). In Tashelhiyt (the Berber dialect of southern Morocco) the word for ‘donkey’, asnus, retains the -us ending, which is lost in Romance reflexes (see Adams Reference Adams2007: 572 with bibliography). There is no sign of continuity between the early Republic, where the omission was after short vowels, and the above imperial corpora, in which the occasional examples are as often as not after long vowels.

It is unfortunate that all the texts extant of this type are from provincial regions, and that we have nothing to go on but inscriptions for Rome. Inscriptions raise particular problems of interpretation.

2.5 Inscriptions

2.6 Manuscripts

Neither non-literary documents nor inscriptions of the Empire support the idea of continuity between the early and late periods, but it would not do to imply that there was never much evidence for the weakening of -s in Latin writing. Evidence is to be found in some early medieval manuscripts, particularly from Italy (see B. Löfstedt Reference Löfstedt1961: 130, 132), and if this has been correctly reported we see here the beginning (albeit very late) of the Romance development. In the Compositiones Lucenses, for example, a work which survives in a single manuscript copied in about 800 at Lucca, omissions are frequent, and there are also hypercorrect additions of -s (see Svennung 1941: 116). The argument is sometimes advanced that frequent omissions in a manuscript imply an Italian origin (see B. Löfstedt Reference Löfstedt1961: 132; cf. Josephson 1940: 82–3).

2.7 Conclusions

3.1 Introduction

The discussion of t in final position may be split into two topics. The first concerns the omission of final -t (or -d in the early period: see below, 3.2) mainly in third person verb forms, singular (uixi for uixit, dede for deded) or plural (curauerun). The second concerns the replacement of -t by -d or of -d by -t in certain other types of words, particularly monosyllables and grammatical words (see 3.5). Various questions are raised by the first phenomenon. Is an omission in writing necessarily a reflection of a speech habit? Did those who wrote -t correctly necessarily pronounce it? Can any continuity in the dropping of -t be established between either the republican or the imperial period, and the Romance languages? Are there regional variations to be found either in inscriptions or manuscripts? Were omissions in speech ever socially stigmatised?

3.2 Omission of final -t/-d in early Latin and at Pompeii

In early Latin the original secondary ending of the third person singular, -t, became -d (see e.g. Sihler Reference Sihler1995: 461), and in inscriptions is sometimes omitted (though it may be unclear in a given case whether it is -d or -t that has been omitted: see Bakkum Reference Bakkum2009: i.93). Omissions are also found in other Italic languages. The same voiced form -d was retained in Oscan but lost in Umbrian (Oscan deded but Umbrian dede: Buck Reference Buck1904: 80). In Early and Middle Faliscan -d is usual, but -t also occurs in both Middle and Late Faliscan (Bakkum Reference Bakkum2009: i.93). -t from earlier -ti, the original primary ending, is retained in Latin, Oscan and Umbrian, though sometimes left out in Umbrian, as in habe = habet (Buck Reference Buck1904: 80).

In Faliscan there is sometimes omission of -t (the primary ending) in the third person singular of the present tense, as cupa = Lat. cubat ‘lies’ at Bakkum (Reference Bakkum2009) no. 305 (see Joseph and Wallace Reference Joseph and Wallace1991: 166, Bakkum Reference Bakkum2009: i.93, 310 (on the formula in which the verb occurs)).

The early Latin examples of omissions in the secondary ending of the third person singular, with provenance, are as follows (references are to CILi² unless otherwise indicated):

In the Forum inscription CILi².1 the form kapia may be equivalent to CL capiat (i.e. = kapiad), but the next word starts with d and there may have been a false word division.

-t was also sometimes omitted after n in the third person plural of the perfect tense:

In the perfect third person plural the complete consonant cluster -nt was also sometimes dropped:

These last two phenomena are also a feature of Faliscan (Bakkum Reference Bakkum2009: i.93).

The inscriptional material from early Italy is not extensive, and the above examples represent a suggestive corpus. In the perfect tense there seems to have been some tendency in different parts of Italy for final -t/d not to be heard.

In the first century ad there are further Italian cases of omission, most notably in the Pompeian inscriptions, but these are not restricted to the perfect tense: the examples cited by Väänänen (Reference Väänänen1966: 70) are mainly in the present tense, singular and plural, of both the indicative and subjunctive. They include the following: CILiv.1173 add. p. 204quisquis ama ualia, peria qui nosci ama(re) (= quisquis amat, ualeat, pereat qui no(n)scit amare), 4434 fela, 8314 habitan = habitant.⁷ See also below, 3.4, for further possible evidence from the same place and period (the archive of the Sulpicii), and 3.5.2 for -d for -t in present tense forms at Pompeii. It must be asked whether these omissions (and others not yet mentioned) may be related to developments in the Romance languages.⁸

3.3 Romance

In most Romance languages final -t was lost, for example in the present tense; the loss seems at least superficially to be related to the nature of the omissions just seen at Pompeii (found also in the present tense), which, as was noted, differ from those in early Latin (found in the perfect). It was mainly in the Romance of France (also Sardinia) that -t was preserved, into the medieval period (see e.g. Rohlfs Reference Rohlfs1966: 434–5 (on Italy), Vincent Reference Vincent1988a: 37, Harris 1988: 213 (on Old French), Jones Reference Jones1988: 326 (on Sardinian), Väänänen Reference Väänänen1981a: 69, Weiss Reference Weiss2009: 515). Thus cantat > OFr. chantet, but It., Sp., Pg., Prov. canta; cantant > OFr. chantent, but Prov., Cat., Sp. cantan, Pg. cantam (see Väänänen Reference Väänänen1981a: 69). According to Elcock (1960: 342–3) final -t in Old French was ‘presumably still pronounced during the ninth and tenth centuries, but the evidence of scansion confirms that it had ceased to be articulated by the end of the eleventh’ (see too B. Löfstedt Reference Löfstedt1961: 136). Opinions differ about the date of its loss in France (contrast Väänänen Reference Väänänen1981a: 69), but there seems to be agreement that it was maintained until quite late.

Given the signs that in the Latin of Italy from quite early onwards -t was prone to be dropped, and given that it was lost for the most part in Italo-Romance, there has been a tendency to argue for a continuity between usage in the Vulgar Latin of the Roman period in Italy, and that in the later Romance dialects of the peninsula, and also to imply that there might have been regional differences in Latin, between Italy and France.

3.4 The imperial Latin evidence and possible continuity

B. Löfstedt (Reference Löfstedt1961: 135) notes that the omission of -t in verb endings is attested very early in Italy, at least since the Pompeian inscriptions. The phenomenon goes back rather further, as was seen above (3.2), to republican inscriptions. Löfstedt says that omission is common in Italian inscriptions and in Italian manuscripts, and he relates this to later Romance developments, in that the loss of -t was standard in Italian. Rohlfs (Reference Rohlfs1966: 434) is cited with approval as tracing the Italian feature back to Vulgar Latin.⁹ Löfstedt did not go so far as to suggest that there might have been regional variation in the Roman period. On the contrary, he stated that the weakening of -t must have been a general feature of Vulgar Latin, and went on to observe that omissions are found in the inscriptions of Africa and Spain,¹⁰ and also Gaul (1961: 136). Similarly Pirson (Reference Pirson1901) lists omissions from Gallic inscriptions (101), and makes the general point (102) that in this respect (and in the omission of final -s, which is treated alongside that of -t) the epigraphic language of Gaul is not distinguished from that of other provinces. Löfstedt also cites some evidence of omission in manuscripts from Gaul, though noting that it is rare in these (1961: 136), in contrast to those from Italy (135).¹¹ Battisti too (Reference Battisti1949: 139) mentions examples in early medieval Gallic texts (the Formulae Andecauenses and Fredegar). Önnerfors (Reference Önnerfors1975: 9), writing of the Physica Plinii Bambergensis, was incautious in using omissions in a manuscript as evidence for its Italian origin (see Adams Reference Adams2007: 504–5). The evidence from Italian versus Gallic texts has not been presented in statistical form, and in any case medieval manuscripts are not a good guide to the state of the language, because scribes tried to get things right, and variations in educational level, rather than the regional origins of writers and their patterns of speech, might be the determinant of the varying frequency of an error. Indeed a medical text might just as well have been written in Africa (a significant source of Latin medical works) as in Italy, and Löfstedt cites evidence from Africa for the omission of -t (see above). The manuscript of the Physica may have been written in Italy, but the case is hardly to be proved on this evidence.

There is no clear-cut inscriptional evidence from an earlier imperial period for any regional variations across Spain, Gaul, Italy and Africa in the treatment of final -t. In the inscriptions of Britain too, which are generally well spelt and also less numerous than those from some other parts of the Empire, omissions are found (C. Smith Reference Smith1983: 926). Regional variation, if it existed, cannot be established. Something further must now be said about the possibility that there might have been continuity between republican/imperial Latin and Romance, whether in Italy or elsewhere.

What should be stressed about the omission of -t in inscriptions is its relative infrequency, compared again with the omission of -m. A list of instances cited in isolation may be misleading, given that the corpora from some areas (notably Rome) are vast. Comparative statistics are needed, or at least an indication of the size of the corpus. In inscriptions there are also extra-linguistic or special factors that could cause the letter to be left out (see below), and lists may take no account of these. The problem of interpreting omissions of -t is much the same as that of interpreting omissions of -s, though it is probably true to say that more attention has been given to -s.

Another question that must be addressed, as we will see, is whether omissions, if not due to practical causes, may sometimes be a sandhi phenomenon, in the sense of reflecting assimilation within a consonant cluster across a word boundary. There is a fundamental difference between this type of omission, and the generalised, unconditioned, loss of a phoneme in a particular position.

Documentary evidence, largely non-epigraphic, is assembled here from Rome, Africa,¹² Egypt, Gaul, Britain, Spain and Pompeii. Africa, as we saw, was mentioned by Löfstedt as a region in which omission occurs, and there is good non-literary evidence from there.

Cases of the omission of -t in verb forms (singular and plural) in the Roman inscriptions published in CILvi are collected by Gordon et al. (Reference Gordon, Gordon, Jansen and Krummrey2006: 288). A quick count turns up about eighty-eight examples. This may seem a high figure, but there are, as we have noted, about 40,000 Roman inscriptions, many of which have third person verb forms, and the eighty-eight represent a minute proportion. Many of the examples are in two verb forms, fecerun and uixi, and extra-linguistic factors (such as convention, limitations of space, syntactic conflations, stonemasons’ errors, the last alluded to by Gordon et al.) may have played a part. By contrast the omission of final -m is so constant that Gordon et al. (Reference Gordon, Gordon, Jansen and Krummrey2006: 279) are unable to list occurrences: they state that it occurs passim in the nominative singular neuter, the accusative singular and the genitive plural. A desideratum would be a contextual analysis of the Roman omissions of -t.

In the Bu Njem ostraca from Africa there are twenty-nine cases of the plural ending -nt and twenty of the singular ending -t, and no certain omissions of the final -t. At 94 the letters ]bun occur in a line otherwise containing nothing. At 110 accepi is certainly first person. There are thus forty-nine cases of -t written and no certain case of its omission. The figures contrast with those for the omission of -m, which is in effect more often left out than written (see above, 2.4).

Several centuries later in the African corpus the Tablettes Albertini the situation is much the same. The omission of -m is described by Väänänen (Reference Väänänen1965: 29) as very frequent (cf. 2.4), but he cites (30) just five cases of the omission of -t in what is a large corpus (cf. Courtois et al. 1952: 70, citing three examples). Four of Väänänen's instances are in monosyllables (au, es three times; for the latter form see below), and the other in a compound with es as its second part (abes). Final -t in monosyllables has its own tendencies (see below, 3.5), but what is most striking about this small group of examples is that in three of the five cases the -t would have been in a three- or fourfold consonant cluster across a word boundary (vii.5 es primo, vii.21 au questionem, xxviii.6 es Felix). There seems to have been simplification of a consonant cluster of much the same type as that seen in Cicero's posmeridianas for postmeridianas: Orat. 157et posmeridianas quadrigas quam postmeridianas quadriiugas lubentius dixerim (Cicero would himself rather say posmeridianas than postmeridianas). Simplification (contact assimilation) is not the same as a generalised loss of a final consonant (see further below, 3.4.1). There is hardly any evidence in this large, badly spelt corpus for the loss of final -t. It is obvious that in a comprehensive discussion of -t all examples should be quoted in context so that the possible influence of a following consonant may be assessed. Jeanneret (Reference Jeanneret1918: 63), dealing with another corpus, aptly remarks that the treatment of final -t ‘était sans doute subordonné à l'initiale du mot suivant, mais l'orthographe, dans la majorité des cas, maintenait -t’. See further below on his corpus.

In the letters of Terentianus there are fifty-one verb forms with final -t (forty-five singulars in -t and six plurals in -nt). The only example of an omission is at 469.3 si potes<t> fieri, but this may be a conflation of (si) potest fieri with si potes (which occurs later in the letter) (see however below on Audollent Reference Audollent1904 no. 272.a.7). Again the consistency with which -t is written in final position contrasts with the frequency with which -m is omitted (see above, 1.3). At 471.19 the -t of post is dropped before paucos, another simplification of a threefold consonant cluster. Pos, the etymon of Romance forms (see TLLx.2.158.25ff.), is well attested in low-register Latin texts (see Adams Reference Adams1977a: 29 with n. 109), including Pompeian graffiti (see Väänänen Reference Väänänen1966: 71, citing e.g. CILiv.6820 pos fata, 4966 posquam). See now the full treatment of the form at TLLx.2.156.60ff., from which it emerges that the vast majority of omissions are before a word starting with a consonant (particularly t), though the loss must eventually have spread to the pre-vocalic position, as there are some instances of pos obitum and one or two other such combinations (TLLx.2.157.38ff.).

The above three corpora cover the period c. ad 100–496. All are badly spelt, not least in the frequency with which another final consonant is left out. Together they display no trace of a general loss of final -t. The few examples seen are special cases, with sandhi phenomena notable.

There are no omissions in the documents of Rustius Barbarus in O. Faw., but that means nothing because t is rarely found in final position in the corpus.

In the curse tablets considered by Jeanneret (Reference Jeanneret1918: 63–4) omission is noticeable in third person verb forms, especially in the plural. Jeanneret cites instances of potes for potest, one of which (Audollent Reference Audollent1904 no. 221.7) is before a vowel (potes ilos), and cannot be explained as reflecting consonantal simplification. In the plural he lists eleven instances, including three in a single sentence (Audollent Reference Audollent1904, no. 272.a.11–12 cadan frangan disiungantur male guren palma). The same document also has possin (13, at the end of a sentence), but on the other hand cadant, Blandus (9). Almost all omissions are before a consonant (indeed in the first word of cadan frangan before a double consonant), but at 111.3 a compound of esse shows an omission before a vowel (adsin ad Plutonem).

Marichal (Reference Marichal1988: 67–70) deals with final consonants in the graffiti of La Graufesenque. The special case of -s came up earlier (2.4). For the dropping of -m see Marichal (Reference Marichal1988: 67). No omission of -t is noted.

In Vindolanda tablets no omissions have been noted.

Tomlin (Reference Tomlin and Cunliffe1988: 76) lists five omissions in the Bath curse tablets. One (38.11 pertuleri) is put down tentatively to lack of room at the end of the tablet, and another (97.6 habe apparently for habeat) is said to be due possibly to an oversight. It is not a straightforward omission, and may reflect a conflation of two verb forms, imperative for jussive subjunctive. The remaining three instances are all in the same tablet (5.2, 5.4, 5.9), a text which has no sure case of a verb form that keeps -t: at 5.6 perd[at] has a reconstructed ending. Two of the omissions are at the end of a line, but that is not so with [p]ẹrdidi mani(cilia) at 5.2. There is, however, a doubt hanging over this example, because a first person verb would suit the context: indeed perdidi in the next tablet is taken by the editor to be in the first person. A cluster of omissions in a single text might in theory hint at a feature of the spoken language that is otherwise obscured by the spelling system (see above, 2.4, p. 138), but in this tablet every instance is open to a special explanation.

In the Visigothic slate tablets from Spain Velázquez Soriano (2004: 501) notes that omissions are few (and one of them, aiute at 103, precedes a word beginning with t). There is one text (104: see also Herman 1995: 70) in which they are more common than usual, but it is extremely late (ninth or tenth century).

In the final line of the scriptura interior of TPSulp. 51, in the hand of C. Novius Eunus, there is an odd error that might just be explicable if Eunus (perhaps taking the text down from dictation) was not accustomed to hearing a final -t in verb forms: que ab omini ui periculo meo est, [[dico]] fateor. The correct formula has esse not est. Est is wrong also because the subject que (= quae) is plural. If the person assumed to be dictating tended to drop the final short vowel of esse or to articulate it weakly Eunus might have interpreted ess as = est, but only if the -t of est was regularly lost in pronunciation. Twice (52, 67) Eunus writes ets for est and once et, alongside several correct spellings of the term. He also wrote non{t} at 68 (see Adams Reference Adams1990a: 238). Could these various misspellings indicate that t was not heard in est or after n (in verb forms)? If Eunus heard est as es(s) but was aware that there was a t in the written form he might sometimes have put t in the wrong place. Similarly if he habitually wrote -t after n in third person plural verb forms without hearing the -t, he might mechanically have added -t to non. But it would not do to base a theory on evidence as problematic as this.

Omissions in the verb ‘to be’ are, however, not infrequent. Es = est was seen above three times in the Albertini tablets, and it also occurs in inscriptions from Dalmatia (see Mihăescu 1978: 211 §182); on its Romance outcomes see Väänänen (Reference Väänänen1981a: 69) and Wright (2011: 67). We also saw above potes for potest in Terentianus and in curse tablets, abes for abest in the Albertini tablets, and adsin and possin in curse tablets. Kiss (Reference Kiss1972: 46–7) cites two examples of es for est from inscriptions, another from the Ravenna papyri along with an instance of abes for abest from the same corpus, and es for abest from a bilingual Virgilian glossary (see CPL 8.945 effata es). His list also contains sun for sunt from an inscription. Gordon et al. (Reference Gordon, Gordon, Jansen and Krummrey2006: 288) list examples both of es and sun from Roman inscriptions.

A feature of lists of omissions from corpora of inscriptions on stone (see e.g. B. Löfstedt Reference Löfstedt1961: 127 for references to the indexes of volumes of CIL; also Diehl 1910: 163) is just how short they usually are in relation to the size of the corpora. Pirson (Reference Pirson1901: 101) lists from the inscriptions of Gaul instances that occupy just nine lines. Carnoy's Spanish examples (Reference Carnoy1906: 175–6) number scarcely more than a dozen. Kiss (Reference Kiss1972: 46–7) cites about a page of examples, but they are from all over the Empire. The instances cited by Hoffmann (1907: 23–4) from African inscriptions are not extensive.

There are also special factors that have to be allowed as possible determinants of omission. Carnoy (Reference Carnoy1906: 176) notes that omissions are sometimes made on the edge of an inscription. The same goes for omissions on instrumenta domestica. Other ‘omissions’ (or additions of -t where it does not belong) may be due to the conflation of two interchangeable constructions. In funerary inscriptions and some legal documents the agent of the action described (the one setting up the funerary monument or, say, acknowledging a debt) may use either the third or the first person of himself, and if the two constructions are conflated a verb form that is first person may look like a third person without final t (see Adams Reference Adams2003a: 59, and above on perdidi in a Bath curse tablet). A possible case of syntactic conflation in a letter of Terentianus was noted above.

Another factor came up above in relation to the Albertini tablets. A systematic account of the omission of final t in inscriptions and non-literary documents would record the following word. In environments in which the t would follow a consonant (as in est or curauerunt) and precede another its omission might represent not an unconditioned loss but the reduction of a consonant cluster across word boundaries (cf. Battisti 1949: 140, and above). That such reductions occurred even in educated speech is confirmed by the remark of Cicero at Orat. 157 quoted above. For comparable omissions across a word boundary see Terentianus 471.19 quoted above, and CILxiii.1791 pos templ(um). Much of the material cited by Jeanneret (Reference Jeanneret1918: 63–4) from curse tablets in Audollent (Reference Audollent1904) might be explained in the same way: e.g. (in addition to the sequence of omissions in a single sentence quoted above) 112.7 ommutuerun nec, 112.8 possun nec, 273.a13 cadan, precor. Note too CILiv.4966 tabificanque (Väänänen Reference Väänänen1966: 70). In this last inscription there is also a case of posquam (see above), and if we take the latter as a simplification of the cluster we should take the other in the same way. Hoffmann (1907: 23) found a tendency in African inscriptions for -t to be omitted before a consonant, as distinct from a vowel. The reduction of consonant clusters is admittedly only a partial explanation of the omission of -t; not all examples can be explained thus (see the full collection offered by Kiss Reference Kiss1972).

3.5 Monosyllables and grammatical words

A common misspelling in inscriptions and non-literary documents consists in the writing of -d for -t in final position, or conversely -t for -d. The d-spelling is sometimes interpreted as a form of weakening prior to loss, a view that may owe something to developments in Oscan and Umbrian. The secondary ending -t in the third person singular was voiced to -d in Oscan (deded) but lost in Umbrian (dede) (see above, 3.2). So Carnoy (Reference Carnoy1906: 175), after listing some misspellings from Spanish inscriptions showing interchange of -d and -t, cites Lindsay (Reference Lindsay1894: 123) as seeing in the use of -d for -t in final position ‘un indice d'un affaiblissement de cette consonne, précurseur de sa disparition’. However, the misspelling -d for -t does not occur haphazardly but is restricted in distribution (early Latin verb forms are not relevant here). There is hardly any such alternation, for example, in verb forms (but see below, 3.5.2, and n. 14). The misspellings can be put into one or both of two overlapping categories. First, many are in monosyllables or grammatical words, or alternatively in what might be seen, rightly or wrongly, as compounds containing such monosyllables or grammatical words (e.g. adque for atque or quoat for quoad). The final consonant of such grammatical words is regularly assimilated in some way to the following phoneme, even when that final consonant is not -t or -d (see e.g. 1.2 on final -m and -n). Second, misspelling was generated particularly when there existed a pair of words differing only in their final stop; the two were conflated orthographically by careless spellers (e.g. ad/at, quod/quot, id/it, reliquit/reliquid;¹⁴ into this category also fall the spellings idem for item and vice versa, though the confusion is not word-final). Thus for example in speech the stop of ad would have been assimilated in voice to t before, say, te (at te: see below, 3.5.2), but misspelling was further prompted by the existence of a form at.

There are testimonia in grammarians and others that throw light on these phenomena. Other evidence lies in misspellings in texts. The two sources are considered in turn.

3.1 Kt > t(t)