In the late third or early second century BC the off-glide of the diphthong /ai/ was lowered to /ae̯/, leading to a change in spelling from <ai> to <ae> (see p. 40).Footnote ¹ The use of <ai> for <ae> in inscriptions of the first–fourth centuries AD, especially in genitive and dative singulars of the first declension, is actually not particularly difficult to find, even in quite large numbers (although given the thousands of examples of <ae>, the frequency is probably still very low).Footnote ² Some, but not all, of these will be due to Greek influence,Footnote ³ misreadings, or mistakes by the stonemason. Use of <ai> seems to have been one of the spellings favoured by Claudius (Reference BiddauBiddau 2008: 130–1), but examples can still be found long afterwards.

It is clear that Quintilian considers the <ai> spelling already highly old-fashioned:

The <ai> spelling is attributed to the antiqui by Velius Longus (5.4 = GL 7.57.20–58.3), Terentius Scaurus (5.2.2 = GL 7.16.7–10) and Festus (Paul. Fest. 24.1–2) but Marius Victorinus suggests that it may have been in vogue in the fourth century, which is not impossible given its presence in inscriptions, as already mentioned (although Marius recommends his charges always to use <ae>):

Use of <ai> for <ae> is extremely rare in the corpora. There are only 2 instances in the curse tablets from the first to fourth centuries AD.Footnote ⁴ The use of Maicius beside Maecius (Kropp 1.7.1/1, mid-first century AD, Altinum) could perhaps be attributed to Greek influence, since many of the names listed on the tablet are Greek. In 1.5.2/1 (around AD 50, Capua) quaistum is the only instance of /ai/ in this tablet, which otherwise shows a number of substandard spellings: ilius for illius, uita for uītam, ipsuq for ipsumque, mado for mandō, Sextiu for Sextius. Greek influence is of course possible, but there is no other internal or external evidence for it.

In the Vindonissa tablets, we have the dative Secundi{i}na<e> (T. Vindon. 41), where only one stroke of the final letter II <e> is observed. Given how rare the use of <ai> is in the corpora it seems unlikely that that it is intended here (although the editor notes the fashionability of <ai> under Claudius). The possibility that the second stroke of the <e> was simply not preserved on the wooden backing of the tablet must be strong.

There is also a single example of <ai> in the dative in the Isola Sacra inscriptions, in Marcianai (IS 308, undated). The editors suggest that this is a morphological borrowing from Greek, perhaps to distinguish the dative from the genitive. The (surviving) inscription reads d(is) m(anibus) Marcianai Donatus, so there is no evidence that the composer was a Greek-speaker (nor was the inscription found in situ, so there are no other known inscriptions from the same tomb). It is the case that many of the people commemorated in these inscriptions have Greek names, which sometimes have Greek morphology. There are also three instances of a Latin name with a Greek first declension nominative (Saluiane 89, Manteiane 196, Galitte 288; see Reference AdamsAdams 2003: 490), and there is at least one instance of the hybrid Greek/Latin first declension genitive -aes attached to a Roman name, in Aureliaes (74).Footnote ⁵ Greek influence, while by no means certain, seems at least as likely as an old-fashioned spelling.

Around the middle of the third century BC, the diphthong /ɛi/ underwent monophthongisation to close mid /eː/; about a century later, this /eː/ was raised further to /iː/, thus falling together with inherited /iː/ (see p. 40). We will see that neither <ei> nor <e> for /iː/ – and for /i/ which results from shortening of /iː/ – are very well attested in the corpora. Nonetheless, even after the first century BC/Augustan period a few plausible examples of each do pop up, in the case of <ei> in one of the Claudius Tiberianus letters, which in general often seem to preserve old-fashioned spelling, and, in the case of <e>, at Vindolanda.

<ei> for /iː/

The effect that this monophthongisation had on spelling was a subject of considerable discussion in the second and first century BC and beyond. The advantage of <ei> was that it provided a spelling that allowed /iː/ and /i/ to be distinguished, but Roman writers disagreed on the exact contexts in which <ei> should be used, with Lucilius, Accius and Varro all apparently taking differing positions (Reference SomervilleSomerville 2007; Reference NikitinaNikitina 2015: 53–8; Reference Chahoud, Pezzini and TaylorChahoud 2019: 50–3, 57–9, 67–9).

The use of <ei> for /iː/ was still extant in literary contexts towards the end of the first century BC and perhaps later. The Gallus papyrus (Reference Anderson, Parsons and NisbetAnderson et al. 1979), probably from c. 50–20 BC, with the reign of Augustus particularly likely, contains the spellings spolieis for spoliīs, deiuitiora for dīuitiōra, tueis for tuīs and deicere for dīcere, all with <ei> for *ei̯ beside mihi (whose final syllable scans heavy) and tibi (with light final syllable) < *-ei̯;Footnote ¹ <ei> for /iː/ is attested in manuscripts of authors as late as Aulus Gellius (13.4.1, writing in the second century AD), along with corruptions which suggest scribes dealing with the unfamiliar spelling (Reference Gibson and ClacksonGibson 2011: 53–4). According to Reference NikitinaNikitina (2015: 58–70), legal texts and ‘official’ inscriptions of the first century BC show a tendency to prefer <ei> for /iː/, especially from original /ɛi/, whereas from the Augustan period there is a clear move to using the <i> spelling, with very occasional instances of <ei>.

The Roman writers on language send mixed messages about the status of <ei>. In the late first century AD, Quintilian says:

It is not clear from this passage whether or not there are contemporaries of Quintilian who still use <ei>; although he only mentions Lucilius, the fact that Quintilian feels the need to argue against it may suggest that in fact there are. The same is true of Velius Longus’ discussion:

However, Charisius (Reference BarwickBarwick 1964: 164.21–29) attributes to Pliny (the Elder) a rule for explaining when third declension accusative plurals are in -eis, suggesting that at least some writers in the first century AD used <ei>, at least in this context, and in the late second or early third century AD, Terentianus Maurus also implies that <ei> is still in use, at least in particular lexemes and endings:

Marius Victorinus (Ars grammatica 4.4 = GL 6.8.13–14) attributes the <ei> spelling to the antiqui, and at 4.59 (GL 6.17.21–18.10) notes that the priores used it to represent the nominative plural of second declension nouns as opposed to the genitive singular. He follows this with the observation that the use of <ei> is a topic which has exercised all writers on orthography, although without making it clear whether any modern writers use it (he himself appears to be opposed).

Diomedes, however, in the late fourth century, is much more explicit that he considers this spelling out of use:

Any kind of conclusive, or even representative, survey of the use of <ei> in the inscriptional context is made extremely difficult by the problems involved in searching on the online databases. The string ‘ei’ in standard spelling represents several sequences of phonemes, while ‘i’ represents (at least) /j/, /i/ and /iː/, all highly frequent phonemes. It is thus extremely difficult to get results which are restricted to the use of <ei> which is desired, and completely impossible to compare it with instances of <i> for /iː/ (and even more impossible, so to speak, to isolate cases of /iː/ < /ɛi/). I carried out a search for the sequence ‘{e}i’ on a plaintext copy of all the inscriptions in the EDCS downloaded on 18/06/2019. After removing cases of <ei> which did not represent /iː/ or /i/, I identified a maximum of 15 dated to the first four centuries AD.Footnote ² However, this is highly likely to undercount the total instances, partly because of the usual problems with this database, partly because of my own decisions of what to include.Footnote ³ Nonetheless, this does not suggest that <ei> was in common usage in this period (as of 06/04/2021, the database finds 150,594 inscriptions dated from the first to fourth century AD).

In general, the corpora agree with this picture, since <ei> for /iː/ is entirely absent from the Vindonissa, Vindolanda and London tablets, the TPSulp. and TH² tablets, the Bu Njem ostraca, the Dura Europos papyri, and the graffiti from the Paedagogium.Footnote ⁴ There are a couple of examples where <ei> is used for the sequence /i.iː/: one in the Caecilius Jucundus tablets ([ma]ncipeis, CIL 4.3340.74, for mancipiīs), in the scribal portion of a tablet which is undated but presumably belongs to the 50s or 60s AD; and one in the Isola Sacra inscriptions (macereis, IS 1, for māceriīs).

There are two possibilities to explain these spellings. The first is that the sequence /i.iː/ has contracted to give /iː/ (Reference AdamsAdams 2013: 110). In this scenario, <ei> is being used to represent the remaining /iː/. The second possibility is that the speaker has undergone raising of /ɛ/ to /i/ before another vowel in words like aureus ‘golden’ > /aurius/ (Reference AdamsAdams 2013: 102–4). If this were the case, <ei> could be a hypercorrect spelling for /i.iː/. An example of such hypercorrection can be found in Terenteae for Terentiae in another Isola Sacra inscription (IS 27). There is no way to distinguish between these possibilities: neither inscription shows any other old-fashioned or substandard features (and nor do any of the other inscriptions from the same tomb, in the case of IS 1).

Otherwise, only the curse tablets and letters provide a certain amount of evidence for the continuing use of <ei> either to represent /iː/, [ĩː],Footnote ⁵ or etymological /iː/ which became /i/ by iambic shortening (on which, see p. 42), in words like ubi ‘when’ < ubī < ubei.Footnote ⁶

In the curse tablets <ei> is found, on the whole, in fairly early texts: in Kropp 10.1.1, from the later second century BC, and 1.4.4/3, 1.4.4/8, 1.4.4/9, 1.4.4/10, 1.4.4/11, 1.4.4/12, 2.1.1/1, 2.2.3/1, all from the first century BC, <ei> is used frequently (but not necessarily consistently), including for iambically shortened /i/ in tibei (1.4.4/3). There are a handful of other examples, although some are in undated tablets (see Table 1).Footnote ⁷ 1.4.4/1, dated to the first two centuries AD, has the spelling suom twice, and otherwise is entirely standard (including nisi, whose final vowel would have been /iː/ prior to iambic shortening). 1.5.3/2 has largely standard spelling, though it is possible that the writer had the /eː/ and /i/ merger, given the spellings Caled[um, Cale[dum] for Calidum and niq[uis] for nēquis and possible ni[ue for nēue if correctly restored. However, the author could also be using the variant nī for nē in the latter two. In 3.6/1, sanguinei reflects the (originally i-stem) ablative *-īd; <i> is used for the iambically shortened final vowel of tibi and for /iː/ in ni ‘if not’. Substandard spellings are found in domna for domina and hyper-correct palleum for pallium. In the case of the very late Deidio (4.3.2/1), the use of <ei> may have been preserved in the family name: <i> is used for /iː/ in oculique. The writer shows substand-ard spelling in bolauerunt for uolāuē̆runt and pedis for pedēs.

Likewise, in the letters most examples of <ei> are found in texts dated to the first century BC or the Augustan period. In the case of CEL 3, from the second half of the first century BC, the writer seems not to have learnt the rule of when to use <ei> well, deploying it for short /i/ in sateis for satis and defendateis for dēfendātis alongside /iː/ in s]ẹi for sī and conserueis (twice) for conseruīs. If the author Phileros is also the writer he may not have been a native speaker of Latin. CEL 12, dated to 18 BC, has <ei> in eidus for īdūs. CEL 10, from the Augustan period, includes <ei> for iambically shortened /i/ in tibei beside <i> in mihi, and, mistakenly, in the second person singular future perfect uocāreis for uocaris by confusion with the perfect subjunctive uocārīs. Alongside these, the genuine cases of /iː/ (originally < /ɛi/) in quī and sī are spelt with <i>. This letter is characterised by both conservative spelling features alongside substandard orthography (on which, see pp. 10–11). CEL 13, from the early first century AD, includes tibei (beside tibi); other instances of etymological /ɛi/ are spelt with <i>in ịḷḷịṣ and ṣc̣ṛịp̣si.

Probably the latest example of <ei> in the corpora is reṣc̣ṛeibae (P. Mich. VIII 469.11/CEL 144) for rescrībe in the Claudius Tiberianus letters, in which most other instances of /iː/ are spelt <i>, including original /ɛi/ in uidit, [a]ttuli, tibi (with iambic shortening). This letter also includes three instances of the dative singular of illa written illei, as well as ille[i]. This could represent Classical illī but scholars have instead suggested that it be understood as an innovative feminine dative /illɛːiː/ which lies behind Romance forms such as Italian lei ‘she’ (Cugusi, in the commentary in CEL; Reference AdamsAdams 1977: 45–7; Reference Adams2013: 459–64).Footnote ⁸ Adams makes the point that in the other letters of the Claudius Tiberianus archive the masculine dative is always spelt illi, and that illei in this letter is therefore more likely to be a specifically feminine form. This is not a strong argument, however, since none of the other Tiberianus letters is written by the same hand. None of the other scribes uses <ei> at all, while that of 469/144 also uses it in reṣc̣ṛeibae. So the fact that in the other letters the masculine dative is illi tells us nothing about the spelling illei in 469/144.

Similar forms appear in the letters of Rustius Barbarus from the first century AD: a feminine dative illei (CEL 75), and a genitive illeius (CEL 77), the gender of whose referent cannot be determined. Nowhere else in these ostraca written by the same hand is there any example of /iː/ being spelt with <ei> (and there are very many examples of /iː/); this includes one instance of illi (in the same letter 77, probably masculine). Now, it is conceivable that this has something to do with the unique genitive and dative endings of pronouns: perhaps <ei> was preserved as a spelling in the educational tradition to mark out these curious endings; this would be particularly relevant in the genitive where original illīus underwent shortening to illius. The spelling with <ei> could then preserve a memory of the original length. However, in this case, given the existence of other evidence for similar feminine genitive and dative forms in Latin put forward by Adams, and the absence of other instances of <ei> for /iː/ in Rustius Barbarus, there is a strong possibility that illei and illeius represent special feminine forms rather than illī and illīus. Since these forms are therefore present in a corpus of similarly early date, it cannot be ruled out that illei in the Claudius Tiberianus letter is also a form of this type, rather than having <ei> for /iː/.

<e> for /iː/

In addition to the continuing use of <ei> for /iː/, <e> too apparently remained an infrequent possibility to represent /iː/. Quintilian provides the relevant examples leber for līber and Dioue Victore for Iouī Victorī in his list of old spellings at Institutio oratoria 1.4.17, implying that they are no longer in use, and I have not found any other reference in the writers on language. There are significant difficulties in finding examples of <e> for /iː/ in the epigraphic record as a whole; I have found 114 instances on LLDB in the first four centuries AD.Footnote ⁹ This number is almost certainly too high since some cases will be things like a mistaken use of the ablative -e in place of dative -ī, and many examples are for original /iː/, not from /ɛi/: these may be hypercorrect of course, but may also suggest that a different explanation should be sought. Nonetheless, these results imply that the <e> spelling did survive to some extent, although the spellings with <e> must represent a tiny proportion of all instances of <i> for /iː/.

Unsurprisingly, then, in the corpora there are only a very small number of examples of <e> for /iː/, of which a couple are very plausible. One is deuom (CEL 10) for dīuom ‘of gods’ in Suneros’ letter from the Augustan period characterised by conservative as well as substandard spelling (see pp. 10–11). Another is amẹcos (Tab. Vindol. 650) for amīcōs ‘friends’ in a letter at Vindolanda authored by one Ascanius who was apparently a comes Augusti, and hence of relatively high rank. The text is all in a single hand, but we cannot tell if it was that of Ascanius himself or a scribe. The letter was sent to Vindolanda and therefore does not necessarily reflect the same scribal tradition. The spelling is all otherwise standard, as far as we can tell. Reference AdamsAdams suggests (2003: 535) that the maintenance of pronunciation of /eː/ < /ɛi/ ‘was seen as a regionalism and belonged down the social scale’, but this does not fit well with the social context of the Vindolanda letter (although of course it could be a feature of the scribe’s Latin rather than Ascanius’).Footnote ¹⁰ However, Festus (Paul. Fest. 14.13) notes amecus as an old spelling,Footnote ¹¹ so it may be better to see the spelling as old-fashioned.

In the curse tablets, a first century BC instance of <e> for /iː/ may be nesu (Kropp 1.4.2/2), if this stands for nīsum ‘pressure, act of straining’, although this text has a number of errors of writing (see p. 132 fn. 2).Footnote ¹² Otherwise we find only 4 instances of <e>, all from Britain: deuo for dīuō (Kropp 3.15/1, 3.19/3) ‘god’, demediam (3.15/1) for dīmidiam ‘half’, and requeratat (3.7/1) for requīrat ‘may he seek’. Reference AdamsAdams (2007: 602) suggests that deuo could reflect a British pronunciation of deo ‘god’, or be a code-switch into British Celtic, for which dēuos would have been the word for ‘god’;Footnote ¹³ at any rate, it is not a good example of <e> for old-fashioned /iː/. We could have a hypercorrect old-fashioned spelling with <e> for /iː/ < *ī in demediam, but perhaps instead one should think of confusion between the prepositions dī- and dē-. Nor is requeratat a plausible example, since the writer of the text has made a large number of mistakes in the writing of the text (as distinct from substandard spellings), such as memina for fēmina, capolare for capitulāre, pulla for puella, uulleris for uolueris, llu for illum, Neptus for Neptūnus etc.

In the tablets of the Sulpicii, C. Novius Eunus writes dede (TPSulp. 51.2.13) for dedī‘I gave’. This could well be an old-fashioned spelling, since Eunus uses other old-fashioned spellings (see pp. 187, 202–4, and 262).Footnote ¹⁴ But we could also imagine repetition of the first syllable by accident (and Eunus is prone to mechanical errors in his writing, as shown by ets, 51.2.9, for est, Cessasare, 52.2.1, for Caesare, and stertertios, 68.2.5, for sestertiōs).

In the Isola Sacra inscriptions, there is one example of coniuge (IS 249, second–third century AD) in place of dative coniugī ‘for his wife’. There are no other substandard features in this short text, but perhaps this is a slip into the ablative (or even the accusative with omission of final <m>) rather than an old-fashioned spelling. The dative ending is spelt with <i> in merenti ‘deserving’.

A much more complicated situation arises where <e> represents short /i/ from long /iː/ by iambic shortening (see p. 42) or other types of shortening. This could be an old-fashioned spelling, reflecting the mid-point of the development /ɛi/ > /eː/ > /iː/. But from the first century AD onwards, at least some speakers in certain contexts use <e> for /i/, presumably due to a lowering of /i/ to [e] and the raising of (original) /ɛː/ to /eː/; with the loss of vowel length distinctions these phonemes would end up falling together as /e/ in the precursor of most Romance varieties (see p. 40). In cases where original /ɛi/ > /eː/ > /iː/ underwent iambic or other types of shortening, it is then difficult to tell whether <e> for <i> is old-fashioned or substandard, and each example needs careful investigation.

Reference AdamsAdams (2013: 51–5) entertains the possibility that several <e> spellings in the Rustius Barbarus and Claudius Tiberianus letters may be old-fashioned, and notes the following observation by Quintilian:

Quintilian clearly thought this spelling was old-fashioned (in addition to what he says in this extract, it forms part of a list of archaic spellings), and Reference AdamsAdams (2013: 54) follows him, saying that ‘[i]t is inconceivable that Livy and other literary figures used such spellings as a reflection of a proto-Romance vowel merger that was taking place in speech. They must have been using orthography with an old-fashioned flavour to it’.Footnote ¹⁵ According to Adams, use of <e> is retained from the time when sibi was still /sibeː/.

However, I have my doubts about this. Quintilian himself is aware, as shown by his comment ‘sed, an hoc voluerint auctores, nescio’, that it was possible for an author’s spelling to become corrupted by subsequent copyists (as noted by Reference De MartinoDe Martino 1994: 743). It is striking that the examples given by Quintilian are of <e> in absolute word-final position. Reference AdamsAdams (2013: 51–62, 67) has identified /i/ in closed word-final syllables as showing evidence of lowering to [e] in the first and second centuries AD. The apparent frequency of <e> in words like tibe and nese might instead be taken as showing that this lowering also affected /i/ in absolute word-final position. If lowering of /i/ to [e] had already happened, at least in words like sibi and quasi, by the first century AD, it is not impossible that it could have entered the manuscript tradition of earlier literary authors by the time of Quintilian.

In either case, it is probably not coincidental that the words in question all involve final /i/ resulting from iambic shortening. If Adams is right that this is an archaism, the old-fashioned spelling could have been retained in these words because iambic shortening applied to forms like /sibeː/ and produced a variant /sibe/; after /sibeː/ became /sibiː/ (and then /sibi/) the standard spelling sibi followed, but sibe remained as an alternative spelling.Footnote ¹⁶ This would explain, for example, why <e> is only found to write synchronic short /i/ in tibe in the Rustius Barbarus letters, despite a large number of instances of synchronic /iː/ < /ɛi/, which is what we might expect an old-fashioned use of <e> to represent. If, on the other hand, <e> in these words is due to lowering of /i/ in final syllables, it is also not surprising that the examples are in originally iambic words: iambic shortening of /iː/ is one of the very few sources of absolute word-final short /i/ in Latin.

The explanation by lowering seems particularly likely in the case of the Rustius Barbarus letters. As Reference AdamsAdams (2013: 55) notes, ‘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’. We find 4 examples of the spelling tibe (CEL 73, 74, 76, 77) for tibi beside 8 of tibi (these are the only examples of absolute word-final short /i/ in the letters). This compares with 4 (certain) examples of <e> for /i/ in final closed syllables of a polysyllabic word (scribes for scrībis ‘you write’, CEL 74, scribes, mittes for mittis ‘you send’ 75, scribes 76),Footnote ¹⁷ and 6 examples with <i> (dixit, enim 73, talis, leuis 74, possim 75, traduxit 77). The rate at which <e> is written for /i/ in these contexts is therefore almost identical,Footnote ¹⁸ so it makes sense that the same explanation, lowering of /i/ to [e] in final syllables, should apply to both. Consequently, it seems more probable that the tibe spellings are substandard rather than old-fashioned.

The same explanation could pertain in most of the other examples of <e> for /i/ by iambic shortening in the corpora, and cannot be ruled out in any of the examples I now discuss, from the tablets of the Sulpicii, the Isola Sacra inscriptions and the Vindolanda tablets.

In the tablets of the Sulpicii we find ube for ubi < ubei ‘when’ in the chirographum of Diognetus, slave of C. Novius Cypaerus (TPSulp. 45.3.3, AD 37). Although there are no other examples of <e> for /i/ in final syllables, Diognetus also spells leguminum ‘of pulses’ as legumenum, suggesting that /i/ may have been lowered to [e] more generally in his idiolect (although short /i/ is otherwise spelt correctly several times, including in a final syllable in two instances of accepit).Footnote ¹⁹ There are two examples of sibe in the Isola Sacra inscriptions, and these too are likely to be due to lowering. IS 27 contains several other substandard spellings, including Terenteae for Terentiae, filis for filiīs, qit for quid, aeo for eō. IS 337 has mea for meam and nominae for nōmine. These may reflect carelessness on the part of the engraver rather than lack of education, since mea comes at the end of a line (and in space created by erasure of a previous word or words), while nominae follows poenae and looks like the result of eyeskip. But the same carelessness could presumably have allowed <e> to be used, reflecting his pronunciation, instead of <i> in his copy of the text.

In the Vindolanda tablets ụbe (Tab. Vindol. 642) for ubi is the only instance of <e> for short word-final /i/.Footnote ²⁰ The spelling in this tablet seems otherwise standard, and note in particular an instance of ṭibi. Reference AdamsAdams (1995: 91; Reference Adams2003: 533–5) has emphasised the general lack of confusion between /i/ and /eː/ in the Vindolanda tablets. But given that this appears to be the only text written in this hand, it was probably not written by a Vindolanda scribe,Footnote ²¹ and that it has no other examples of /i/ in a word-final syllable, we cannot be absolutely sure that <e> is not due to lowering rather than being old-fashioned.

The final case of <e> for short final /i/ is nese (P. Mich. VIII 468/CEL 142, and CEL 143) for nisi in the Claudius Tiberianus archive. Could this be due to lowering? In 468/142, there are two other instances of <e> for <i>, both in a final syllable: uolueret for uoluerit ‘(s)he would have wanted’ and aiutaueret for adiūtāuerit ‘(s)he would have helped’ (beside 3 cases of <i>: nihil, [ni]hil, and misit). But apart from nese there are 10 examples of short /i/ spelt <i> in an open final syllable: tibi (twice), [ti]bi, ṭibi, mihi (4 times), [mih]i, sibi. In CEL 143, written by the same scribe, apart from nese there are no other instances of <e> for /i/, and 5 of short /i/ in an open final syllable: tibi (twice), mihi (twice), miḥi. On the one hand, therefore, the writer of these texts did seem to have lowering in word-final syllables followed by a consonant. On the other, nese is the only example of possible lowering of /i/ in absolute word-final position, compared to 15 examples spelt with <i>. I am not certain whether use of <e> is due to lowering or is old-fashioned.

The situation is further complicated by the fact that the letters include some further instances of <e> being used for /i/ which are not due to iambic shortening but could also reflect old-fashioned spellings. The first vowel of sene (468/142) for sine was originally /i/, but the writer may have thought of sine as being connected with sī ‘if’. A hypercorrect spelling seine (CIL 1².583), presumably resting on this false etymology, is attested in the second century BC. The forms nese, nesi in 468/142, and nese in 143 also have <e> for /i/ in their first syllable. This could be an old-fashioned spelling accurately reflecting original /ɛ/ here, since nisi probably came from *ne sei̯ (Reference FriesFries 2019: 94–7). The spelling nesei is attested in the two copies of the Lex luci Spoletina (CIL 1².366 and 1².2872, probably from the mid-second century BC), and nesi is mentioned by Festus (Fest. 164.1).Footnote ²² Alternatively, it could be a hypercorrection, with analysis of the ni- in nisi as being derived from the alternative negative nī < /neː/ < /nɛi/.

However, given that the writer (or author, dictating) of the text does have lowering of /i/ to [e], at least in final syllables followed by a consonant (i.e. in a position of minimal stress: Reference AdamsAdams 2013: 60), it is still possible that it is lowering that is to blame for the spelling with <e> in sine, nese, nesi, especially because these are function words, which are particularly likely not to receive phrasal stress (see p. 42), and hence might have undergone the same lowering seen in the final syllable despite not being in the final syllable of the word.

There is not enough evidence to draw completely certain conclusions. However, I do not think that we can be sure that the various types of <e> for /i/ used by the writer of P. Mich. VIII 468/CEL 142 and CEL 143 are to be attributed to old-fashioned spelling (which would be of several different types). I would be inclined to explain all instances as due to lowering of /i/ to [e] in relatively unstressed position (in function words and in final syllables).

There are a number of processes which led to the possibility of using <o> as an old-fashioned spelling to represent /u/. I will begin by discussing examples which may be attributed to the following sound changes:

1. (1) /ɔ/ in final syllables was raised to /u/ before most consonants and consonant clusters in the course of the third century BC, e.g. Old Latin filios > filius ‘son’;

2. (2) /ɔ/ was raised to /u/ in a closed non-initial syllable during the second century BC, e.g. *eontis > euntis ‘going (gen. sg.)’;

3. (3) /ɔ/ > /u/ before /l/ followed by any vowel other than /i/ and /eː/ in non-initial syllables (Reference SenSen 2015: 15–28), e.g. *famelos > *famolos > famulus ‘servant’ (cf. familia ‘household’).

For various reasons, including the relatively late confusion of /ɔː/ and /u/ discussed directly below, and the difficulty of removing false positives from searches in the database, trying to establish the rate at which old-fashioned <o> for /u/ appears in the epigraphic spellings of the first four centuries AD is not practical. In his list of changes wrought by time, all of which seem to be seen as deep archaisms, Quintilian includes some examples of type (1):

Apart from after /u/, /w/ and /k^w/, where the raising of /ɔ/ was retarded until the first century BC, which is discussed below (Chapter 8), there are few cases of <o> for <u> arising from these contexts in the corpora. Even where we do find <o>, a confounding factor in identifying old-fashioned spelling of /u/ of these types is the lowering of /u/ to [o] which eventually led in most Romance varieties to the merger of /u/ and /ɔː/. According to Reference AdamsAdams (2013: 63–70), this can be dated to between the third and fifth centuries AD, and did not take place at all in Africa.Footnote ¹ This requires him to identify a number of forms which show <o> for /u/ as containing old-fashioned spelling (or having other explanations) in the Claudius Tiberianus letters (Reference AdamsAdams 1977: 9–11, 52–3; Reference Adams2013: 63–4).

He sees posso (P. Mich. VIII 469/CEL 144) not as a rendering of possum, but a morphological regularisation with the normal first singular present ending which shows up elsewhere; this is quite plausible. Also plausible, given the spelling with final <n>, is the influence of the preverb con- on the preposition cum, which is spelt con in 468/142 (8 times) and 471/146 (twice), but old-fashioned spelling could also be a factor.

The origin of the adverb minus ‘less’, is extremely uncertain. It could go back to *min-u-s, since /u/ is also found in the stem of the related verb minuō ‘I lessen’; this may also be the origin of the neuter of the comparative minor ‘smaller, lesser’ (Reference LeumannLeumann 1977: 543; Reference SihlerSihler 1995: 360). If this is correct, the <o> in the final of quominos (470/145) would not be old-fashioned. But alternative explanations would see adverbial minus and the neuter of the comparative both coming from *minos > minus (although this would have to be somehow secondary, since the usual comparative suffix is *-i̯os-; Reference MeiserMeiser 1998: 154; Reference WeissWeiss 2020: 384). If minus does come from *minus rather than *minos, quominos could be a false archaism, by influence from comparative minus, since the neuter of comparative adjectives normally did come from *-i̯os-.

However, sopera (471/146) for supra certainly never had *o in the first syllable. Reference AdamsAdams (1977: 10–11) originally saw <o> here as due to the merger of /u/ and /ɔː/, but subsequently (Reference AdamsAdams 2013: 64) suggests that the scribe may have seen <o> as also being old-fashioned, bolstered by the possibility that the lack of syncope is also an old-fashioned feature (supera is found in Livius Andronicus, in Cicero’s Aratea and in Lucretius; Adams loc. cit. and OLD s.v. supra).Footnote ² It is of course not impossible that the scribe was hypercorrect here, but in the absence of other evidence for this false etymology this explanation is not particularly appealing.

In my view, it is more likely that sopera, and perhaps quominos, suggests that either the author, Claudius Terentianus, or the scribe, was an early adopter of lowered [o] for /u/ (and note that both cases of <o> are in paradigmatically isolated formations which may have made it difficult for the scribe to identify which vowel was involved).Footnote ³ It is also possible that one or both were not native speakers of Latin, which may also have led to problems in identifying the vowel for the scribe.

There are quite a number of cases of <o> for /u/ in the curse tablets (see Table 2), but I am doubtful of how many really reflect archaisms. con (Kropp 1.4.2/2), con (twice), co (twice, 11.1.1/37) for cum can be explained in the same way as in the Tiberianus letters, while second declension nominative singular masculine forms in -o(s) may be due to influence from other languages: Celtic in the case of Secundo, Secuno, Ssecundo (8.1/1, Pannonia, first half of the second century AD), if it really represents Secundus.Footnote ⁴ This tablet also has uolontas for uoluntās, but it uses <u> for /o/ in lucuiat (apparently for loquiat or loquiant, an active equivalent of loquātur or loquantur), as well as a number of spellings which cannot be explained as due to normal features of spoken Latin (<i> for /ɛ/ in ageri for agere and limbna for lingua), so we cannot be sure uolontas does not arise from the writer’s problems with spelling or abnormal phonology.

The fairly late date of cor]pos (1.4.4/13) for corpus and paretator (3.3/1) for parentātur allow them to be attributed to lowering of /u/; both also contain other substandard spellings. Undated capolare and capeolare (3.7/1) for capitulāre could also be due to lowering (in an inscription whose spelling is anyway highly deviant). And nascitor for nascitur (twice, 1.4.4/5), dated to the second or third centuries AD, could also be a precocious example of this; so could conscios (5.1.7/1) for conscius if it belongs towards the end of its date range. Both contain other substandard spellings. Having accepted that [o] for /u/ might be attested in the Claudius Tiberianus letters, I would suggest that this may also be a possibility for eorom (4.3.1/1) for eōrum, and nullom (5.1.2/1) for nūllum, also from the second century AD. The preponderance of cases in the final syllable might point to lowering occurring there first, in parallel to the situation of /i/ to [e].

Given their datings to the first century BC and first century AD, capilo (1.4.2.3) for capillum, questo (2.1.3/2) for quaestum, cos[i]lio, morbo (2.2.2/1) for cōnsilium, morbus, dioso (2.2.3/1) for deorsum, granom (4.4.1/1) for grānum could be old-fashioned spellings. Some of these texts also contain substandard spellings. None of them is in a context in which a mistake of ablative for nominative or accusative is very easy to envisage.

If Adams is right that the lowering of /u/ to /o/ did not take place in Africa (but see footnotes 1 and 10), old-fashioned spelling becomes a plausible explanation for a few instances there of <o> for /u/ from the second to third centuries AD. Incleto for Inclitum (11.2.1/31) I attribute to influence from the Greek second declension ending -ον, since the text has other examples of interference from Greek spelling.Footnote ⁵ 11.1.1/32 has eo(?)]rom for eōrum, but in addition to substandard spellings has a number of mechanical errors. So does 11.2.1/6, and there is also the possibility that p]er deo meo reflects an error in what case goes with per rather than an old-fashioned spelling of deum meum. The other example of a noun following per in this text is Bonosa for Bonōsam, so the author may have thought that per took the ablative. But we do find illoro, iloro for illōrum (11.1.1/5), Cusconio for Cuscōnium (11.1.1/6), cursoro[m] for cursōrum (11.1.1/20). There is no reason to imagine that Cusconio is an ablative since it forms part of a list of four names in the accusative. Many of these texts also contain substandard spellings.Footnote ⁶

Overall, it is striking how many cases of <o> for /u/ there are in the curse tablets. I am reluctant to see them all as old-fashioned features, especially since few of the texts show any other such spellings; in practically all of the texts there are several other endings in -us or -um, and there seems no reason why a single word should be marked out in this way, especially in cases like the three names in sequence Cosconio Ianuarium et Rufum (11.1.1/6). I have more sympathy than Adams does with the idea that we may be seeing early signs of the lowering of /u/ to [o] that is better evidenced in much later texts; one could even suppose that in African Latin /u/ did lower to [o] in final syllables as elsewhere, but since /ɔː/ did not merge with /u/, [o] simply remained an allophone of /u/. One should also note that the intrinsic difficulties in the writing and reading of (often very damaged) curse tablets do make them more unreliable than other types of epigraphic evidence (Reference KroppKropp 2008a: 8).Footnote ⁷ On the other hand, all the examples of <o> for /u/ from the curse tablets do come from original /ɔ/, unlike with sopera and perhaps quominus in the Claudius Tiberianus letters, and a few of them are very early compared to the emergence of good evidence for the merger of /ɔː/ and /u/. So I do not rule out the possibility that some of these <o> spellings are old-fashioned.

There is one Augustan example of <o> for /u/ < /ɔ/ in the letters, in the form of Dìdom (CEL 8, 24–21 BC), which appears to represent the name Didium. It seems strange that an old-fashioned spelling should be used here but not in the name [I]ucundum with which it is conjoined (or indeed in the other accusative singular second declension form in this letter, decrìminatum), but no other explanation arises.Footnote ⁸ Another damaged letter (CEL 166, around AD 150) has epistolám, which is reasonably likely to be an old-fashioned spelling (assuming this is not an early example of lowering of /u/).

There are two possible examples in the Bu Njem ostraca, but kamellarios (O. BuNjem 76) is very uncertain: it could be a nominative singular for accusative, which is common in the ostraca (Reference AdamsAdams 1994: 96–102), but could also be an accusative plural (this is how Adams understands it). The other is ]isṭolạ[ (114), which may reflect epistola for epistula; but apart from the fact that the reading is not certain, there is a certain amount of evidence for confusion between /ɔː/ and /u/ at Bu Njem (see fn. 10).

Another case of <o> for <u> occurs in cui, the dative singular of the relative pronoun quī and the indefinite pronoun quis. An older form is quoiei (CIL 1². 11, 583, 585) and it had come to be spelt quoi around the start of the first century AD, according to Quintilian:

The passages of Velius Longus (13.7 = GL 8.4.1–3) and Marius Victorinus (4.31–32 = GL 6.13.11–12) quoted on pp. 166–7 suggest that the spelling quoi, and its genitive equivalent quoius, was also old-fashioned for these writers. We find no examples of quoi in the corpora, but the strange spelling cuoì, [c]ụ[o]i (TPSulp. 48) in the parts of a tablet written by the scribe presumably reflects a sort of compromise between quoi and cui (the non-scribal writer spells the word cuì). I know of no other examples of this spelling in Latin epigraphy.

Another conceivable instance of an old-fashioned spelling involving <o> are fornus (O. BuNjem 7), for[num (49), fornarius (8, 25), foṛ[narius] (10) for furnus, furnārius. The sequence /ur/ < *or or *r̥ in the standard forms of these words is unexpected; a dialectal sound change or borrowing from other languages is often supposed (I have argued for the latter: Reference ZairZair 2017). So these words could represent the original, older form, which is attested in manuscripts of Varro and by writers on language (Reference ZairZair 2017: 259). However, influence from fornāx ‘furnace, oven’ is also possible (thus Reference AdamsAdams 1994: 104);Footnote ⁹ another possibility is that /u/ was lowered by the following /r/ in syllable coda, which by this time was ‘dark’ in Latin (Reference Sen and ZairSen and Zair 2022). This might be particularly likely if there was some confusion of /ɔː/ and /u/ at Bu Njem.Footnote ¹⁰

The diphthong /ɔu/ became /oː/ and then /uː/ by the third century BC (see pp. 39–40). On the use of <o> in place of <u> in poplicos > pūblicus and words derived from it, see Chapter 17. It is possible that iodicauerunt (Kropp 11.1.1/26) in a curse tablet from Carthage, in the second century AD, for iūdicāuē̆runt < *i̯oudik- is an old-fashioned spelling representing the mid-point of the change (at any rate, no other explanation springs to mind). Since we have long /uː/ in the first syllable, this spelling cannot be explained by confusion of /ɔː/ and /u/. At least in legal texts, derivatives of iūs were particularly favoured for this (Reference DecorteDecorte 2015: 160–2), although use of <o> for /uː/ rather than <ou> was never common, even in the archaic period.

There are two environments in which spelling with <u> and <i> alternated in Latin orthography, with, on the whole, a movement from <u> to <i>, although in certain phonetic, morphological or lexical contexts the change in spelling either did not take place at all or took place at different rates. These environments are (1) original /u/ in initial syllables between /l/ and a labial; (2) vowels subject to weakening in non-initial open syllables before a labial. Both the question of the history and development of the spelling with <u> and <i>, and what sound exactly was represented by these letters is lengthy and tangled (especially with regard to the medial context; for recent discussion and further bibliography, see Suárez-Martínez 2006 and Reference WeissWeiss 2020: 584).

/u/ and /i/ in Initial Syllables after /l/ and before a Labial

In initial syllables we know on etymological grounds that the words in question had inherited /u/. In practice, there are very few Latin words which fulfil this context, and only two in which the variation is actually attested: basically just clupeus ~ clipeus ‘shield’ and lubet ~ libet ‘it is pleasing’ and its derivatives such as lubēns ~ libēns ‘willing’ (which is part of a dedicatory formula and makes up the majority of attestations of this verb), *lubitīna ~ libitīna ‘means for burial; funeral couch’, Lubitīna, Lubentīna ~ Libitīna ‘goddess of funerals’. No forms with <u> are found in liber ‘the inner bark of a tree; book’ < *lub^h-ro-, whose earliest attestation is libreis in CIL 1².593 (45 BC, EDR165681), as well as being attested in literary texts from Plautus onwards. Strangely, lupus ‘wolf’ does not become ^×lipus, as Reference LeumannLeumann (1977: 89) points out, although as it is attested in Plautus it was surely borrowed from a Sabellic language (as demonstrated by /p/ < *k^w) early enough to have been affected.

As we shall see, both spellings are attested from the third century BC onwards in lub- and clupeus, with <u> predominating initially and slowly being replaced by <i>. Some scholars view this as a sound change from /u/ to /i/ (e.g. Reference WeissWeiss 2020: 153), others as the development of an allophone of /u/ to some sound such as [y], leading to variation in spelling with <u> and <i>, but with <i> eventually becoming standard (e.g. Meiser 1988: 80; making it more or less parallel with the development in non-initial syllables, which we shall discuss later).

In epigraphy other than my corpora, the <i> spelling is attested early in the lub- words (see Table 3): libes (CIL 1².2867) for libēns is about the same time as the first instances of lubēns, but <u> outnumbers <i> by 13 (or 14, if CIL 1².1763 is to be dated early) to 3 in the third and second centuries BC. In the first century BC, however, there are only 4 (or 5 if CIL 1².1763 is to be dated later) instances of <u> to 4 of <i>, and subsequently <u>, with 2 instances in the first century AD (or 1 in the first, 1 in the second if CIL 3.2686 is to be dated late), is completely swamped: there are 16 (or 17 if CIL 5.5128 is to be dated early) instances of <i> in the first century AD, and in subsequent centuries the numbers are too massive to be included in the table.Footnote ¹ These have not been thoroughly checked, and some are mere restorations, but the vast majority do indeed belong to the lexeme libēns.Footnote ² Overall, then, it seems clear that the spelling with <i> was becoming more common in the course of the first century BC, becoming the usual spelling in the first century AD, and subsequently overwhelming the <u> spelling, although the latter is still occasionally found in the first, and perhaps second, century AD.

The spelling of clupeus ~ clipeus (Table 4) has a rather different profile: the lexeme is not found before the first century BC, when only the <u> spelling appears (2 or possibly 3 examples); in the first century AD there are 6–8 inscriptions which use <u>, but only 1–3 with <i> (and possibly quite late in the century), and still 4–5 <u> in the second century AD to 7 of <i>, with 1 <i> in the fourth.Footnote ³ It is perhaps surprising, given the common formulaic usage of lubens in dedicatory contexts, that clupeus appears to have retained the <u> spelling longer. Perhaps this is connected to the influence of the Res Gestae of Augustus.

Unsurprisingly, given the restricted number of lexemes containing the requisite phonological environment, there are very few instances of this type of <u> spelling in the corpora. However, lubēns ~ libēns is used occasionally in letters at Vindolanda, where <i> outnumbers <u> 5 to 1. The sole use of <u>, in lụḅẹṇṭịṣsime (Tab. Vindol. 260), occurs in a letter whose author Justinus is probably a fellow prefect of Cerialis, and which the editors suggest may be written in his own hand, as it does not change for the final greeting. Towards the end of the first century AD, it seems fair to call this an old-fashioned spelling.

The examples of <i> are libenter (291, scribal portion of a letter from Severa), libenti (320, a scribe who also writes omịṣẹṛas, without old-fashioned <ss>), libente[r (340), libentissime (629; probably written by a scribe)Footnote ⁴ and ḷibent (640, whose author and recipient are probably civilians, and which also uses the possibly old-fashioned spelling ụbe).

The <u> spelling also occurs in a single instance in the Isola Sacra inscriptions (lubens, IS 223, towards the end of the reign of Hadrian or later). There is a good chance that this is the latest attested instance of the <u> spelling.The inscription is partly in hexameters, the spelling is entirely standard, and <k> is used not only in the place name Karthago but also in karina ‘ship’. Again, it is reasonable to assume that the <u> spelling in this word might be considered old-fashioned.

/u/ and /i/ in Medial Syllables before a Labial

The second context for <u> ~ <i> interchange is short vowels which were originally subject to vowel weakening before a labial. Hence we are not dealing only with original /u/ as is the case in initial syllables, and hence the subsequent development is not necessarily the same as in initial syllables.

In order to utilise the evidence of the corpora it is necessary to first examine the highly complex evidence both of inscriptions and of the grammatical tradition, which descriptions in the literature such as Reference MeiserMeiser (1998: 68), Suárez-Martínez (2006) and Reference WeissWeiss (2020: 72, 128) tend to oversimplify.Footnote ⁵ Reference LeumannLeumann (1977: 87–90) provides a more comprehensive discussion. I will begin with the evidence of inscriptions down to the first century AD. In the first place, it is important to make a distinction which most of those writing about the <u> and <i> spellings do not make clearly enough. There are certain words in which the vowel before the labial was always written with <i> or <u> (as far as we can tell); presumably in these words the vowel had become identified with the phonemes /i/ or /u/ early on.Footnote ⁶ By comparison, there are some words in which the vowel before the labial shows variation in its spelling. The first instance of <i> before a labial is often attributed to infimo (CIL 1².584) in 117 BC (thus Reference NikitinaNikitina 2015: 19; Reference WeissWeiss 2020: 72), or testimo[niumque (CIL 1².583) in 123–122 BC (thus Reference Suárez-MartínezSuárez-Martínez 2016: 232). However, these are in fact the earliest examples of <i> in a word in which <i> and <u> variation is found. Probably earlier examples of the <i> spelling actually occur in opiparum ‘rich, sumptuous’ in CIL 1².364 (200–171 BC, EDR157321) and recipit ‘receives’ in CIL 1².10 (170–145 BC, EDR109039), for which a <u> spelling is never found.

In Table 5 I provide all examples of the use of <u> and <i> in this environment in some long official/legal texts of the late second century BC.Footnote ⁷ As can be seen, both spellings are found in these texts, but the distribution is not random. Most of the words with an <i> spelling never appear with a <u> spelling in all of Latin epigraphy: compound verbs in -cipiō,Footnote ⁸ -hibeō, -imō,Footnote ⁹ and forms of aedificium and aedificō,Footnote ¹⁰ uadimonium and municipium. Outside these particular texts, the same is true of pauimentum (CIL 1².694, 150–101 BC, EDR156830), animo (CIL 1².632, 125–100 BC, EDR104303). It looks as though by the (late) second century certain lexical items had already generalised a spelling with <i>.Footnote ¹¹

By comparison, <u> spellings are found only in words which either show variation with <i> in the later period or which are subsequently always spelt with <i>,Footnote ¹² such as testimonium, which across all of Roman epigraphy is found with the <u> spelling only in CIL 1².582.Footnote ¹³ The <u> spellings predominate in these words in these inscriptions: with <i> we have only infimo beside the far more common superlatives in <u>, the ordinal undecimam beside uicensumam, testimo[niumque beside testumonium, and eidib(us), which, as a u-stem, is also found spelt elsewhere with <u>.Footnote ¹⁴

The same pattern is found in other inscriptions from the third and second centuries: in Table 6 I have collected all instances that I could find of <u> spellings in inscriptions given a date in EDCS, along with examples of <i> spellings of those words (other than those from CIL 1².582, 583, 584 and 585 and 2924). It seems clear that at this period the <u> spellings are dominant, although we do find a few <i> spellings (perhaps more towards the end of the second century). Nonetheless, all of these words do subsequently show <i> spellings (although the extent to which the <i> spelling is standard varies, as we shall see).

Overall, the picture seems to be a much more complex one than simply a move from early <u> spellings to later <i> spellings.Footnote ¹⁵ Although <u> spellings outnumber <i> spellings in some words and morphological categories in the second century BC, certain words have already developed a fixed <i> spelling by this period, with no evidence to suggest that they were ever spelt with <u>. Most other words will go on to see <i> supplant <u> as the standard spelling, although at varying rates as we shall see, but some, like monumentum, postumus and contubernalis, will strongly maintain the <u> spelling.

For the later period, Reference NikitinaNikitina (2015: 10–48) examines the use of <u> and <i> in words which show variation in a corpus of legal texts and ‘official’ inscriptions from the first centuries BC and AD. In the legal texts, she finds only <u> down to about the mid-first century BC, after which <i> appears: in a few texts only <i> is attested, but many show both <u> and <i>. The lexeme proximus seems to be particularly likely to be spelt with <u>, perhaps due to its membership of the formulaic phrase (in) diebus proxumis. Even in AD 20, the two partial copies of the SC de Cn. Pisone patri (Reference Eck, Caballos and Fernández GómezEck et al. 1996) contain between them 24 separate <u> spellings and 2 <i> spellings, while CIL 2.1963, from AD 82–84, has 7 instances of <u> (5 in the lexeme proxumus), and none of <i>. There are only two ‘official’ inscriptions of the first century BC which contain words with <u> or <i> spellings, but in the other ‘official’ texts of the first century AD, <i> spellings are heavily favoured (73 examples in 25 inscriptions) over <u> spellings (8 examples across 4 inscriptions).

An interesting observation is that in the first century BC, superlatives in -issimus are often spelt with <u>. By comparison, in law texts of the first century AD, except in the SC de Cn Pisone patre, all 10 attested superlatives in -issimus have the <i> spelling, whereas the irregular forms like maximus, proximus, optimus etc. show variation. Although the switch between <u> and <i> in the -issimus superlatives is probably less abrupt than Nikitina perhaps implies,Footnote ¹⁶ it does seem likely that the <i> spelling became particularly common in this type of superlative around the Augustan period: as we shall see below, in imperial inscriptions <u> is used vanishingly seldom.

Nikitina’s study makes it clear that there was a movement from <u> spellings to <i> spellings in some words in high-register inscriptions over the course of the first century BC and first century AD. This movement probably took place more slowly in the more conservative legal texts,Footnote ¹⁷ and more quickly in certain lexical items (notably superlatives in -issimus) than in others.

If we turn to the evidence of the writers on language, the question of the spelling of these words was clearly one of great interest for some time.Footnote ¹⁸ Quintilian briefly mentions sounds for which no letter is available in the Latin alphabet, including the following comment:

This appears to imply that the vowel in this context was not the same as either of the sounds usually represented by <i> or <u>.Footnote ²⁰ The spelling with <u> was however apparently ‘old-fashioned’ for Quintilian (at least in the words optimus and maximus):

Cornutus (as preserved by Cassiodorus) appears also to think that the <u> is old-fashioned, and suggests that the spelling with <i> also more accurately reflects the sound. He gives as examples lacrima and maximus, as well as ‘other words like these’:

Velius Longus discusses the vowel in this context in several places. What he says about it provides an important caution against us assuming that the ancient writers on language thought, like us, that the words with <u> and <i> variation formed a single category for which a single rule was necessarily applicable. Instead, it seems likely that they looked at each word, or category of word, individually (an approach which accurately reflects usage, on the basis of the epigraphic evidence). Note that he also includes among his examples lubidō and clupeus (discussed above, pp. 75–82). The first passage which touches on this issue is a long and complex one:

My understanding of this passage is that Velius Longus is saying that the pronunciation of the words he discusses involves a sound which is not the same as the sound represented by <i> in other contexts, and is apparently ‘fuller’, but not as ‘full’ as it used to be, when <u> was a common spelling. Nowadays, the usual spelling is with <i>, but people who prefer to use the old-fashioned spelling <u> may do so. However, they should not extend this to actually pronouncing the sound as [u], because if they did, they would also by the same logic have to say [u] in words like manibiae and libidō. This final point is rather surprising. Does it suggest that by the early second century AD, the vowel of the first syllable of libidō had already developed to /i/, and hence a spelling pronunciation of [u] would sound wrong? Perhaps the same could be true of manibiae, if the development of the medial vowel before a labial was very sensitive to phonetic conditioning, such that here the pronunciation had again fallen together with /i/, unlike in the superlatives.

The following passage suggests that variation in both spelling and pronunciation still existed, with mancupium, aucupium and manubiae (again!) containing a sound which some produced in an old-fashioned ‘fuller’ manner and spelt with <u>, while others used a more modern and elegant ‘slender’ pronunciation, and wrote with <i>. Unlike in the previous passage, it is not explicitly stated here that the pronunciation of the relevant sound is different from /u/ and /i/.Footnote ²²

The next four passages give examples of which letter to use in particular words which show variation: clipeus, aurifex, contimax and alimenta are better than the spellings with <u>, but aucupare, aucupium and aucupis are better than spellings with <i> (contradicting the previous passage with regard to aucupium). It is implied at 13.1.1 that the <u> spelling actually corresponds with a different pronunciation, but it may simply be as /u/.

Terentius Scaurus has little to add, except for some other examples of <u> and <i> interchange (in two of which, the dative/ablative plurals of the u-stems artus and manus, analogy with the rest of the paradigm is the cause of the continuing oscillation, as Scaurus goes on to note):

The fourth-century grammarians Diomedes and Donatus use almost exactly the same wording, no doubt due to reliance on the same source. They both imply that only <u> is used in optimus, but that it does not have the same sound as in other words:

Marius Victorinus, although in the fourth century, suggests that optimus maximus is presently written with <u>, but that a number of other words, including maximus again, should be written with <i>, not <u>. This may be carelessness, or be due to differences in the sources that Marius Victorinus used. If we want to exculpate him of inconsistency, we might note that the sequence optimus maximus is a traditional epithet of Jupiter, in which the <u> spelling may have been maintained for longer than in maximus in other contexts.

The final passage contains various words in which Victorinus says that others have thought that they contain a sound between u and i, of which only proximus is relevant here. He suggests that in fact this sound is no longer used, and recommends a spelling either with <u> or <i>:

In dealing with these extracts from the writers on language of course the usual problem arises of to what extent the authors are reporting the situation in their own time, and to what extent they are reacting to spellings long out of use but still found in manuscripts and inscriptions, and passed down in grammatical writings. Nonetheless, it seems reasonable to me to deduce that even in the fourth century AD there were some people who used <u> in at least some words. However, Quintilian, Cornutus, Velius Longus, and to some extent Marius Victorinus, all imply that at least in some words this spelling was old-fashioned. On the basis of what Quintilian and Velius Longus say, there may have remained a sound not easily identifiable as /i/ or /u/ in some words into the second century AD (for more on the possible phonetic developments, see pp. 276–9).Footnote ²⁴

If it is true that this sound continued in at least some words, it makes identifying old-fashioned spelling somewhat difficult. Since we do not know precisely at what point in a given word the sound became identified as /i/, continuations of the <u> spelling in words which are generally written with <i> may reflect an attempt to represent the sound as spoken, particularly if the writer has other substandard spellings, rather than knowledge of an older orthography, and this must be borne in mind when analysing the data.

We can now turn to the inscriptional evidence of the first to fourth centuries AD, and the use of the <u> spellings in the corpora. The lexicalised nature of the spellings with <u> or <i> makes it important that we do not assume that the spelling of all words containing the variation developed in the same way. This is also convenient, since it is difficult to carry out searches in the EDCS for sequences like ‘um’, ‘im’ etc. without including far too many false positives. I have therefore restricted searches to the words and categories which show variation in the corpora: these are largely the lexemes monumentum, contubernium and contubernalis, superlatives, and the ordinals septimus and decimus, and derivations thereof (including names).

I shall start with these last two categories. Very few superlatives in -issimus are found with <u> spellings in the epigraphy. I have found 39 inscriptions containing a <u> spelling in the first four centuries AD,Footnote ²⁵ of which 1 is dated to the fourth century, 8 might be as late as the third, 19 as late as the second, and 11 are dated to the first century. This might suggest a general decline over time, although it would be necessary to know the frequency of superlatives in -issimus in these centuries to be sure of this (since in principle use of superlatives in general in inscriptions might have decreased over this period). In any century, however, the <u> spelling is clearly rare when compared with use of -issimus, which is found in thousands of inscriptions.Footnote ²⁶ Combined with the evidence of a change in official inscriptions in the first century AD discussed above, it is reasonable to suppose that the standard spelling was <i>, and that the sound before the labial had become identified with /i/ in this morphological category.

In other superlatives in -imus and words derived from them, the <u> spelling, while uncommon in all lexemes except postumus, is far more frequent than in -issimus superlatives (as we can see in Table 7).Footnote ²⁷ Compared to the dominance of <u> spellings in the second century BC, it is clear that for most lexemes the <i> spelling becomes the standard in the imperial period, although to varying degrees. This may partly be because the sound before the labial remained different enough from /i/ to inspire <u> spellings for longer than in the -issimus superlatives. In postumus, conversely, it may at some point have been identified as /u/, but the (few) spellings with <i> suggest that this analysis was not inevitable: some people still heard a sound closer to /i/. A confounding factor is that several of these superlatives and their derivatives are very frequent as personal names; the same is true for ordinals: Septimus, Decimus, Postumus etc. It might be assumed that old-fashioned spellings are more likely to be preserved longer in names, but there is no easy way to search only for examples as names.

We also see a difference in the use of <u> and <i> in the ordinals in -imus: septimus is found in 1525 inscriptions, while septumus in only 83, given a rate of <u> of 5%; by comparison, decimus appears in 221 inscriptions and decumus 53, so that <u> is found in 19%.Footnote ²⁸

On this basis, it is unclear at exactly what point use of <u> spellings in the corpora in these words, other than postumus, will have become old-fashioned rather than being a possible spelling for a living sound. Certainly not in the first century BC (e.g. Maxuma Kropp 1.7.2/1, optumos CEL 7.1.21, maxsuma CEL 10). However, the corpora tend to match quite well the distribution we see in the epigraphic record more generally. At Vindolanda, there are no examples of <u> spellings in 54 examples of superlatives in -issimus or 12 of superlatives and ordinals in -imus, and at Vindonissa 1 in -issimus and 1 in -imus (a name). At Dura Europos, 50 superlatives and ordinals in -imus (and -imius) are found, almost all the name Maximus. In the Isola Sacra inscriptions, 98 instances of this type are found with <i>, some names but the majority not; there is a single instance of <u> in the name Postumulene (IS 364). This consistent use of <i> rather than <u> is clearly not particularly remarkable.

In the curse tablets, there are a few examples of <u> spellings in names, where the spelling was probably maintained for longer: Septumius (Kropp 1.11.1/11, second century AD, Sicily), Postum[ianus] (3.2/77, third or fourth century AD, Britain), Maxsumus (5.1.4/10, first half of the second century AD, Germania Superior); likewise in the tablets of Caecilius Jucundus the names Postumi (CIL 4.3340.56, 74, 96) and Septumi (92) have <u> (there are no other examples of these names).

In the letters there is more variety, with three non-name instances of <u>. We find amicissumum (CEL 2, second half of the first century AD) in a very broken text apparently using a model letter of recommendation as a writing exercise. This is striking, since <u> is found so seldom in -issumus superlatives, and particularly so, since the same text apparently also includes an <i> spelling in plurị[mam]. The writer is not yet expert, going by the spelling Caesarre for Caesare. It seems likely that use of <u> was old-fashioned at this point, and the vowel had probably already merged with /i/ in -issimus by this period. Another damaged letter (CEL 166, around AD 150) has plúruma[m alongside the old-fashioned spelling epistolám (assuming this is not an early example of lowering of /u/; see pp. 66–71). The combination of the relative lateness, another old-fashioned spelling, and the fact that plurimus is written with <u> so rarely, suggest that this too is an old-fashioned spelling. On the other hand, this is far less clear for proxumo in a letter from one soldier to another dated to AD 27 (CEL 13). While this letter does have an old-fashioned spelling in tibei for tibi, proximus does seem to have maintained the <u> spelling for longer than the other superlatives, given the relatively high frequency of the <u> spelling for proximus seen at large, and Reference NikitinaNikitina’s (2015: 26–7) observation that this lexeme was particularly likely to maintain <u> in official inscriptions. So its use here may not be very old-fashioned. The writer’s spelling is otherwise standard.Footnote ²⁹

In the tablets of the Sulpicii, we find <u> spellings in the epithets of Jupiter Optum<u>m, Maxumu, Optumum (TPSulp. 68), the first 2 by Eunus, the last by the scribe, where the <u> spelling is probably supported by tradition. And, again, we also have 2 other examples of <u> in pr[o]xum[e] (15) and prox]ume (19), both written by scribes. The same lexeme has the <i> spelling in proximas (87, 89), the former by a scribe, the latter not. Again, proximus show signs of having maintained its <u> spelling longer than some other words. Apart from these, the <i> spelling appears in duodecimum (45), uicẹṇsimum (46), the lexeme mancipium (85, twice, and 87, 3 times), and the perfect infinitive mancipasse (91, 92, 93), all written by scribes.Footnote ³⁰ There are also 5 instances of the name Maximus (25, 50, twice, and 66, twice), of which 4 are written by a scribe.

Moving on to other lexemes, a curse tablet has alumen[tum] (Kropp 3.23/1, AD 150–200, Britain), which is found in two other inscriptions of the second century AD (alumentorum, AE 1977.179; alument[a]r(iae) CIL 9.3923= EDR175389).Footnote ³¹ This is therefore probably an old-fashioned spelling, since it compares with 98 inscriptions from the first to fourth centuries AD containing the <i> spelling (not to mention Velius Longus’ advice to use <i>),Footnote ³² although it is just possible that this word maintained a vowel for which <u> could be a plausible representation. Another has an<n>uuersariu (Kropp 1.4.1/1, c. AD 50, Minturno) for anniuersārium. No other examples of the <u> spelling are found, versus anniuersali (AE 1992.1771, AD 193–195, anniuersarium (CIL 6.31182, AD 101–200, EDR166509), anniuersaria (CIL 11, 05265, AD 333–337, EDR136860) and [ann]iuer[sarium (Res Gestae Diui Augusti; Reference ScheidScheid 2007; CIL 3, pp. 769–99, AD 14).

Lastly, a letter of the third century AD (CEL 220) has estumat for aestumat. I find no other instances of a <u> spelling, and no instances of the <i> spelling either, dated to the first four centuries AD in the EDCS, other than in the Lex Irnitana, which has both, but with <u> predominating (see fn. 17). In my corpora, aestimatum is found Dura Europos in a list of men and mounts (P. Dura 97.15, AD 251), and aestimaṭurụm in a copy of a letter sent by a procurator (P. Dura 66B/CEL 199.2, AD 221). If the vowel in the second syllable had not yet merged with /i/, estumat could be an attempt to represent the sound rather than an old-fashioned spelling, especially since the author has substandard <e> for <ae>.

Apart from these, fairly infrequent, examples, most words which are found with <u> in the corpora are those for which <u> seems to have been maintained as the standard spelling, i.e. monumentum, and contubernalis and contubernium. The evidence of the corpora provides an interesting hint that by the late first century or second century AD, use of <u> in these words was associated with writers whose orthographic education hewed closer to the standard and/or included old-fashioned features.

The Isola Sacra inscriptions, being funerary in nature, are the only corpus to include monumentum (see Table 8). There are 4 instances spelt with <u> and 10 with <i>, a reversal of the pattern in all of first–fourth century AD epigraphy, in which <u> spellings make up nearly two thirds of the examples, with 420 dated inscriptions, while the <i> spelling is found in 240.Footnote ³³ There is no evidence that <u> is used in earlier inscriptions than <i>.

It is possible that there is a correlation between use of monimentum and substandard spelling. The inscriptions with the <u> spelling use an orthography which is otherwise standard, with the exception of filis for filiīs in 228; the stonemason has also made several mistakes in the lettering, so an accidental omission of an <i> is also possible. All but 228 also feature Greek names containing either <y> or aspirates which are spelt correctly. However, the text of IS 30 is very damaged. By comparison, of the inscriptions containing monimentum, 206 has Procla for Procula, preter for praeter and que for quae; 284 also has filis for filiīs; 320 has que for quae, 337 has mea for meam and nominae for nōmine (both may be stonemason’s mistakes, however; see pp. 62–3), and sibe for sibi (on which, see pp. 59–64). 106 and 107 have Ennuchis for Ennychis, 337 Afrodisius for Aphrodisius (Agathangelus and Tyche are spelt correctly in 240; Polytimus, Polytimo and Thallus in 284; and Zmyrnae in 320). None of these inscriptions, even those which use the <u> spelling, features any other old-fashioned spellings (e.g. <c> rather than <k> before<a> in cari[ssimae, IS 30, huius with single <i>, IS 125). While not being conclusive evidence, all this would be consistent with the possibility that monumentum was the standard spelling at this period, and that monimentum was substandard.Footnote ³⁴

A search on the EDCS finds 433 inscriptions containing contubernium and contubernālis in the first four centuries AD, and only 17 with contibernālis (there were no examples of contibernium).Footnote ³⁵ The earliest dated example found for contibernālis is contibernali (AE 1975.226), from between 31 BC and AD 30 (EDR076061), although the earliest examples of contubernālis are not necessarily much earlier: contubernal(i) (CIL 6.39697, 50–1 BC, EDR072515) and contubernali (CIL 5.1801, Augustan period). It seems, therefore, that the spelling of these words with <u> is not old-fashioned in terms of usage: it remained current throughout the imperial period and was apparently never replaced by the <i> spelling in standard orthography. The epigraphic evidence does not even allow us to be certain that the <u> spelling is the older spelling.

For the spelling of this word in the Vindolanda tablets see Table 9. The <u> spelling in this word is used by the writer of 181, who also wrote 180 and 344; the author was a civilian, and the writer of these texts also uses <ss> and <xs> (see p. 263) as well as some substandard spellings. 310 is the letter of Chrauttius, whose scribe also uses <ss>. 311 is written by a scribe who also uses apices. 343 is the letter whose author is Octavius, possibly a civilian, and which combines use of <xs>, <ss> and <k> with a number of substandard spellings. 349 is a fragmentary letter, presumably written by a scribe. It includes an instance of <x>. Note that two of these texts also include superlatives spelt with <i>: felicissimus (310), plurimam, inpientissime (311).

The <i> spelling is used by the writer of the letters 346, 656, 657, and perhaps also the fragmentary 708, presumably a scribe. The spelling is entirely standard (n.b. solearum twice at 346, not soliarum). In 655 the same writer has misi rather than missi, and in 657 <x> rather than <xs>. The writer of 641, a letter, is presumably also written a scribe. It contains misi and an example of <x>. 698 is too fragmentary to say anything about. The number of instances of <i> is surprising, but less striking when we observe that 4 out of 6 (probably) belong to the same writer. All instances of <i> are likely to belong to scribes, whereas <u> is used both by scribes, and, possibly, civilian writers. The use of <u> may correlate with the other old-fashioned spellings <ss>, <xs>, and <k> – but also substandard spellings.

The <u> spelling of contubernalis is found also in the letters of Tiberianus (P. Mich. VIII 467.35/CEL 141), which features some old-fashioned spelling (<uo> for /wu/, <k> before <a>), and some substandard features, although the spelling is overall closer to the standard than some of the letters in this archive.Footnote ³⁶ This combination leads Reference Halla-aho, Solin, Leiwo and Halla-ahoHalla-aho (2003: 248) to suggest that the writer was ‘a military scribe, trained to write documents for the military bureaucracy’. This letter also provides evidence for the independence of <u>/<i> spellings across lexemes: it includes plurimam, optime, optimas, libenter.

The <u> spelling is also found in an early private letter (contubernálés, CEL 8, 24–21 BC), which has completely standard spelling (apart possibly from Nìreo for Nēreō, see p. 209 fn. 6), and also includes ualdissime. A much later letter has c]ọn[t]ubernio (CEL 220, third century AD), and has estumat for aestimat (see above). The use of <e> for <ae> in the latter is substandard; we do not have enough evidence to be sure that <u> is old-fashioned.

In the course of the second century BC, /ɔ/ became /ɛ/ after /w/ and before a coronal, other than a single /r/ (Reference WeissWeiss 2020: 152), for example uoster > uester ‘your’, uoto > ueto ‘I forbid’, aduorsom > aduersum ‘against’. The earliest inscriptional example comes in the Lex repetundarum of 123–122 BC (CIL 1².583), where we find a single example of auersum beside five cases of the spelling <uo>. I have found 6 instances of the <uo> spelling dated to the first century BC, beside 52 examples of <ue>.Footnote ¹

This suggests a fairly rapid replacement of the <uo> spelling by the <ue> spelling (although diuortia apparently remained the standard spelling for this word), which is supported by the fact that only 13 instances of <uo> are found datable to the first four centuries AD.Footnote ² Of these, 5 are instances of the divine name Vortumnus, in which archaic spelling might be expected to be retained longer than in other items. Two late cases of uostras (ICUR 5.14057), uostrum (CIL 8.9081) may well reflect the analogical effect of uōs and noster which led to the *o of the Romance languages in this word (e.g. Spanish vuestro, Italian vostro, French vôtre). We find uortice (AE 2015.1186), uorsum (twice, CIL 6.20674) in verse inscriptions of the second century AD, where the effect is probably intended to be archaising; the latter inscription also features the spellings paussa for pausa ‘pause’, gnatam for nātam ‘daughter’, ollim for ōlim ‘once’, and ollis for illīs ‘them’.

The writers on language make it clear that the <o> spelling is outmoded. Quintilian makes the following comment:

While this passage is expressed somewhat cryptically, I take it to mean that he considers the <uo> spelling to be absurdly old-fashioned, and it comes as part of a list of such spellings. Cornutus and Marius Victorinus also address the topic:

Accordingly, none of the corpora preserves the <uo> apart from the curse tablets. This phonological context appears only in a very few lexical items so there are not that many tokens of it in most corpora (I count 7 instances in the tablets of the Sulpicii, for instance, and 6 in P. Dura, although 5 of these are in the same text), but the curses as a genre happen to contain many instances of uerto ‘I turn’ and lexical items derived from it, so there are particularly large numbers of examples. Only 3 of these (across 2 tablets) appear to show the spelling with <uo>, as opposed to 93 with <ue>. Two belong to the same tablet (Kropp 1.4.2/1, Latium), aruosarius for aduersārius, and aruosaria for aruersāria. Since the tablet is dated to the first half of the first century BC, this usage is not relevant for spelling under the empire. The spelling is probably already old-fashioned, but not overly so, as both /wo/ > /we/ and ar- for ad- before a labial fricative are characteristic of the second century BC. The remaining instance is uostrum for uestrum (1.1.1/1, Arretium, second century AD). Although this text does also include another old-fashioned spelling in the form of uoltis for uultis,Footnote ⁴ uostrum could again instead be an instance of analogy from uōs and noster.

A number of sound changes raised original /ɔ/ to /u/ in Latin in the course of the third and second centuries BC (in addition to those given on p. 65, this raising also took place before the sequence -lC-, for example *solkos > sulcus ‘furrow’, in the second century). However, the raising was delayed in all cases after /u/, the labial glide /w/ and the labiovelar stop /k^w/ until the first century BC.Footnote ¹ I have found no certain examples of a spelling <uu> in these sequences prior to the first century. It is often stated or implied in modern scholarship both that original /wɔ/, /k^wɔ/ and /uɔ/ became /wu/, /k^wu/ and /uu/ at the same time, and that the use of <uo> fell out of use extremely quickly.Footnote ² However, neither of these statements appears to hold true.

As to the former, the inscriptional evidence, including that after the first century BC and some of the corpora (as we shall see), suggests that /uɔ/ became /uu/ earlier than /wɔ/ and /k^wɔ/ became /wu/ and /k^wu/.Footnote ³

As can be seen in Table 10, 5 inscriptions, all likely to be from the first half of the first century BC,Footnote ⁴ show 6 examples of the spelling <uu> for original /uɔ/, alongside 2 inscriptions showing three examples of /uɔ/ being spelt <uo>. Conversely, there are 4 instances of original /wɔ/ and /k^wɔ/ being spelt <uo>, and only 1 instance of <uu> which might be dated to between about 100 and 50 BC.Footnote ⁵ This suggests that /uɔ/ became /uu/ towards the start of the first century BC, whereas /wɔ/ and /k^wɔ/ became /wu/ and /k^wu/ towards the middle of that century. Of course, the numbers are small, but this does fit in with the later spelling conventions, as will be seen.Footnote ⁶

In the rest of the first century BC and till the end of the Augustan period, <uo> remains the majority way of spelling both /uu/ and /wu/ and /k^wu/. Leaving aside the aqueduct inscriptions from Venafrum, the Fasti Consulares and Triumphales, and the Res Gestae of Augustus, which would distort the figures and will be discussed below, in inscriptions dated between 49 BC–AD 14 I have found the following figures:

• 5 (16%) instances (from 5 inscriptions) of /wu/ and /k^wu/ are spelt <uu>

• 27 (84%) instances (from 27 inscriptions) of /wu/ and /k^wu/ are spelt <uo>

• 8 (32%) instances (from 8 inscriptions) of /uu/ are spelt <uu>

• 17 (68%) instances (from 13 inscriptions) of /uu/ are spelt <uo>.

There are very few inscriptions which contain both /wu/ or /k^wu/ and /uu/, but just as we will see in the corpora there is none which contains both /wu/ or /k^wu/ spelt <uu> and /uu/ spelt <uo>. All three other possibilities are attested: the Laudatio Turiae (CIL 6.41062), from the last decade or so BC, uses <uo> for both /wu/ (uolneribus) and /uu/ (tuom). An inscription on a marble tablet from Herculaneum (CIL 10.1453), shows <uo> for /wu/ (seruom) but <uu> for /uu/ (perpetuum), and it is the same almost consistently in an Augustan edict from Venafrum regarding an aqueduct; across the three copies of the inscription plus a number of cippi marking the route (CIL 10.4842 and 4843; Reference CapiniCapini 1999 no. 1a, 1b, 1c, 2a, b, c, d, f, g, i, l, 17–11 BC), there is 1 instance of riuos and 10 of riuom, alongside 1 instance of [u]acuo[m, 1 of uacuum, and 6 of uacuus. In the Res Gestae of Augustus (Reference ScheidScheid 2007; CIL 3, pp. 769–99), <uu> is used in both contexts: sụụm, annuum, ṃ[agistratu]um, riuum, uiuus.

That there was some confusion about when to use <uo> and when to use <uu> in this period is suggested by the Fasti Consulares (CIL 1² pp. 16–29, FC) and Triumphales (Reference DegrassiDegrassi 1947 no. 1h, FT), erected by Augustus. These in general show a mixture of more old-fashioned and more up-to-date spellings, presumably partly due to their composer working from a range of earlier sources, and partly due to the tendency for names to retain older spellings anyway.

For /wu/ and /uu/ we consequently find an interesting mixture of spellings. In both Fasti we have <uu> used for /uu/ in mortuus (twice, FC) and triduum (FT), and <uu> used to represent /wu/ in the personal names Vulso (4 times, FC, once FT), Ca]luus (FC) and Coruus (3 times, FT), and in the name of the non-Roman people Vulcientib(us) (FT). There is also <uo> for /wu/ in the names of non-Roman peoples: Volsceis (twice, FT), Volsonibus (FT), where the <uo> spelling would remain standard, and the abbreviation uol(nere) (FC). But in addition to these we also find the personal name usually written Scaeuola as Scaeuula (FC), [Sc]aeuula (FT), and the names of the peoples generally known as the Volsinienses as Vulsiniensibus, V]ulsiniensibus (FT). Although these spellings do indeed reflect the expected development of the sequence /wɔ/ to /wu/ before dark /l/ before a back vowel or a consonant, the older spelling Scaeuola appears to have been generally retained, with no other instances of Scaeuula attested, while there are 23 epigraphic instances of the spelling Volsinii and Volsinienses as late as the third century AD. The only other instance with <uu> in this word is Vulsinios (4 times) in a copy of rescript of Constantine (CIL 11.5265, AD 333–337), with many non-standard features.

I would attribute these spellings to an (inconsistent) tendency to modernise the spelling of the sequence of /wu/ to <uu> in the Fasti, even in those lexemes where the old-fashioned spelling would in the end be continued as the standard spelling. Whether this was an idiosyncrasy of the writer of the inscriptions or whether it reflects a more wide-ranging movement towards the use of <uu> for /wu/ amongst whatever body was responsible for the composition of the Fasti cannot be known, although it does fit in with the preference for <uu> also demonstrated by the Res Gestae.

On the basis of this epigraphic evidence, therefore, there is already significant support for the conclusion that /uɔ/ had become /uu/ around the start of the first century BC, while /wɔ/ and /k^wɔ/ only became /wu/ and /k^wu/ around the mid-point of the century. For both contexts, the spellings with <uo> remained more common to the end of the Augustan period, although /uu/ was more frequently written <uu> than /wu/ and /k^wu/ were. In the Augustan period, there are signs of <uu> becoming the standard spelling in official inscriptions for both contexts.

The idea that /uɔ/ became /uu/, and adopted the spelling <uu> earlier, and more thoroughly, than /wɔ/ and /k^wɔ/ is supported by the evidence of both the writers on language and of my corpora. Starting with the former, a well-known passage of Quintilian states that <uo> for /wu/ was still used by his teachers towards the middle of the first century AD, who presumably also passed on this spelling at that time, although he subsequently prefers to use <uu>. The examples he gives for the <uo> spelling are of /wu/ (seruos and uolgus). This is not a coincidence: as we have seen, the epigraphic evidence suggests that his teachers might well have already been using <uu> to represent /uu/, and this in fact provides the key to understanding the passages in which he talks about use of <uo>. The two relevant passages are extremely complex:

And:

The exact meaning of these passages is somewhat complicated, and is discussed by Reference ColsonColson (1924) and Reference AxAx (2011) in their commentaries, as well as, for the first passage, Reference ColemanColeman (1963: 1–10). The first passage states that when a vowel is added to another vowel within a syllable this either represents a long vowel (in old writers), or a diphthong (but not a triphthong: three vowels can only go together in the same syllable if one is consonantal <i> or <u>). In addition, <i> and <u> can occupy both vocalic and consonantal positions, as shown by the interchange between /j/ and /i/ in iacit and conicit respectively,Footnote ¹⁰ and by /wu/ in uulgus and seruus. If these letters are considered always to be vowels, one then has to explain why they can (nowadays) appear consecutively in the same syllable, when two identical consonants cannot do this (or indeed any two consonants, except in muta cum liquida sequences).

In the second passage, Ax explains unus sonus as the onset and nucleus of a syllable (‘eine neue eigene silbische Toneinheit’). He concludes that, since it was acceptable to use <u> to write /w/ plus a vowel other than /u/, Quintilian’s teachers, not being prepared to countenance <uu> for /wu/, fell back on <uo>, which was acceptable. However, in the absence of other information, this leaves us in the dark as to why <uu> for /wu/ was not to their liking.

Colson takes unus sonus to refer to a diphthong, and says of the second passage:

But if it is true that the rule is that two vowel letters in a syllable must form a (rising) diphthong, the sequence <uo> for /wu/ and /wɔ/ ought to have been just as forbidden as <uu>, since these also did not form a diphthong (and the same would be true of <ua>, <ue>, <ui>).

The missing piece to the puzzle is the fact that, once gemination had ceased to be used to represent vowel length, doubled vowel letters generally could only represent two vowels in two consecutive syllables, as in words like cooperatio and anteeo;Footnote ¹¹ <uu> of course also represents /uu/. These sequences of vowels in separate syllables unquestionably represent two sounds. I take it, therefore, that unus sonus refers to a sequence of sounds within the same syllable, as in <uu> for /wu/. So, Quintilian’s teachers accepted the use of <uu> to represent /uu/, since this was a sequence of two sounds across two syllables, as in all other sequences of two vowels, but not <uu> to represent /wu/, since this would be considered unus sonus.Footnote ¹² And in fact, this analysis will be supported when we turn shortly to other writers from shortly before and after Quintilian, who make it explicit that the problem with <uu> is that it ought to represent two vowels in two syllables.

Combining and expanding on Quintilian’s two passages, his argument is as follows: it is necessary to consider whether i and u are to count as vowels or consonants. At least some of the time, i and u should be considered consonants, as in iam and uōs, where they occupy the syllable onset. It is true that when vowel letters are combined in a single syllable they represent either a long vowel (in the olden days) or a (rising) diphthong (e.g. ae, au etc.), but an ostensible combination of three vowels (e.g. seruae) in fact can only be analysed as containing a consonantal i or u. The analysis as vowels is also problematic if we assume that two of them can be combined in a single syllable, when two identical consonants have to be split across a syllable boundary (and indeed two non-identical consonants, except in muta cum liquida sequences); in addition (and more relevantly), as Quintilian’s teachers maintained, two identical vowels have to be split across two syllables too (as in words like cooptō, praeeō, ingenuus). However, now it is recognised that i and u can sometimes function as consonants, allowing the spelling seruus (although consonantal u does somehow sound different from u as a vowel, so it would be sensible to use the digamma for consonantal u).

This analysis is supported when we turn to other writers who talk about the <uo> spelling: Cornutus, Velius Longus and Terentius Scaurus all refer to the old belief that two consecutive identical vowel letters could only represent vowels in separate syllables. This point is made very clearly by Cornutus:

We can see that Cornutus, a decade and a half older than Quintilian, does indeed follow the rule that Quintilian ascribes to his teachers that <uu> must reflect two vowels in different syllables. Direct evidence that Cornutus used <uo> for /wu/ may come from the following passage; however, the manuscripts are all corrupt here, so that the reading is not certain, and due to its brevity the passage is also difficult to understand:Footnote ¹³

Velius Longus also explains the rule concerning <uu> more clearly than Quintilian:

Like Quintilian, Velius Longus, writing probably slightly later, sees the use of <uo> for /wu/ as old-fashioned. In addition to the reference to superiores in the passage above, he subsequently makes the comment

Terentius Scaurus mentions the rule more briefly, and again makes it clear that the <uo> spelling for /wu/ is old-fashioned:

Interestingly, Pseudo-Probus, probably largely repeating Sacerdos’ late third century AD Artes grammaticae, treats <uo> simply as an alternative spelling, with no suggestion that is old-fashioned or unusual:

Marius Victorinus also does not make an explicit statement about whether the <uo> spelling is old-fashioned, although he does go on, after the following passage, to recommend the use of <uu>, to his pupils, with spelling matching pronunciation:

Donatus does not mention the <uo> spellings, but gives seruus and uulgus as examples of consonantal plus vocalic <u> (Donatus, Ars maior 2, p. 604.5–6 = GL 4.367.18–19).

Looking at the inscriptional evidence after the Augustan period, it seems likely that Quintilian, Velius Longus and Terentius Scaurus’ objections to <uo> may actually be a response to its survival relatively late, even in official and elite inscriptions, throughout the first century AD and into the second. As we have seen, the Res Gestae already uses <uu> for both /uu/ and /wu/ and /k^wu/, but <uo> could still be used for both on the gravestone of a high-status woman towards the end of the first century BC, and <uo> for /wu/ in particular is found in a number of inscriptions which could be considered to represent the elite standard.Footnote ¹⁵

In legal texts:

• aequom (CILA 2.3.927) in a Senatus consultum from Spain, AD 19–20

• aequom (CIL 2.5.900) in a Senatus consultum from Spain in several copies from AD 20

• clauom (twice, CIL 2.5181; second half of the first century AD) in a lex from Lusitania

• uacuom, diuom (CIL 2.1964), diuom (6 times), seruom, suom (CIL 2.1963), diuom (12 times) beside seruum duumuir, duuuiri, suum (CILA 2.4.1201) in several versions of a Lex Flavia municipalis from Spain, with parts dating back to legislation of Augustus

• riuom (3 times), alongside riuum, riuus (twice) in the Lex riui hiberiensis from Hispania Citerior, during the reign of Hadrian (Reference Beltrán LlorisBeltrán Lloris 2006)

• diuos and –]ụom (CIL 6.40542) on a legal text on a marble tablet, Rome, during the reign of Antoninus Pius.

Other inscriptions of an official or public character:

• equom, ]uom beside suum, magistratuum (AE 1949.215) in a tablet recording the honours paid to Germanicus Caesar, from Etruria, AD 20

• au]onc[ulus], diuom (CIL 13.1668; Reference MallochMalloch 2020) beside diuus, patruus, arduum. A tablet recording a speech of Claudius, Lugdunum, AD 48 or shortly afterwards

• riuom (CIL 6.1246) in an inscription commemorating Titus’ rebuilding of the Aqua Marcia, Rome, AD 79

• aequom (AE 1962.288). A bronze tablet recording a rescript of Titus, from Spain, AD 79 or shortly afterwards

• diuos (AE 1988.564) in a marble fragment of an imperial Fasti from Etruria, in the reign of Trajan or later

• aequom (CIL 3.355). A bronze tablet recording a rescript of Trajan, AD 125–128, from Asia.

In general, <uo> was by no means uncommon. It was perhaps particularly frequent in uiuos for uiuus ‘alive’ on tombstones, where it appears in the formula uiuus fecit ‘(s)he made it while still alive’. A search on the EDCS finds 209 inscriptions dated between AD 1 and 400 which contain uiuos, and 372 containing uiuus. So <uo> represents /wu/ in 36% of these inscriptions.Footnote ¹⁶ Of course, some instances of uiuos may be accusative plurals rather than nominative singulars, but given the vast frequency of the uiuus fecit formula, this will make up a very small part of the total. A search with much smaller numbers allows for checking the inscriptions, and confirms this proportion of <uo> to <uu>: there are 55 inscriptions containing (con)seruos in the nominative singular and 118 of (con)seruus = 31% dated between AD 1 and 400.Footnote ¹⁷

It also survived for a long time, although its use for /uu/ is rare in later inscriptions. Not including instances of quom for cum (on which see pp. 165–8), I find 68 inscriptions dated between AD 150 and 400 which contain <uo>, of which only 3 have <uo> for /uu/ (CIL 5.4016, AE 1989.388, CIL 3.158); the rest are all /wu/ or /k^wu/.Footnote ¹⁸ These are found in inscriptions from a range of genres (funerary, honorary, dedication, building, a contract on a wax tablet, a statue base etc.) from all over the empire. To give an extremely rough idea of the frequency of <uo> at this period I searched for <uu> on the EDCS, which found 979 inscriptions containing this sequence from between these dates, giving a frequency of 7%.Footnote ¹⁹ A couple of inscriptions from this period suggest that the convention of using <uo> for /wu/ and <uu> for /uu/ may have been maintained: CIL 10.1880 has [P]rimitiuos and [p]erpetuus, CIL 3.5295 (= 3.11709) has uolnus and suum.

The corpora confirm the tendency among some writers to use <uo> to represent /wu/, and <uu> to represent /uu/ (and never the other way round) and hence the implication that /uɔ/ to /uu/ took place earlier.Footnote ²⁰ This is most clear in the Vindolanda tablets, where the distinction is consistent: /wu/ is always spelt <uo> and /uu/ always spelt <uu> (see Table 11).Footnote ²¹ Most of the instances of <uo> appear in letters to and from the prefect Cerialis; it correlates with instances of etymological <ss> for /s/ in 225, a draft letter from Cerialis, probably written in his own hand, and in 256, a letter to Cerialis. Most of the latter was written by a scribe, whose spelling is otherwise standard, and it comes from a certain Flavius Genialis of unknown status, but who is probably not the prefect Flavius Genialis. In 261, another letter to Cerialis, presumably from someone of similar rank, it appears in the formula annum nouọm f̣ạuṣṭum felicem ‘a fortunate and happy New Year’. In 720 too little remains to say anything about the contents.

Although this distribution might imply that use of <uo> is associated with high-status individuals, it also appears in 154, which is an interim strength report, unlikely to have entered the official archives of the unit. Although it does not contain a large amount of text, its spelling is standard except for the contraction of original /iiː/ sequences in is (six times beside eis twice) and Coris ‘at Coria’. Of the tablets using <uu> for /uu/, 187 is an account, whose spelling is, as far as one can tell, standard. 270 is a letter to Cerialis, likewise. 291 and 292 are letters from Severa to Lepidina; the main hands of each are described by the editors as ‘elegant’ and ‘rather elegant’ respectively, and use standard spelling. 631 is a letter to Cerialis from an Ingenuus, who addresses Cerialis as domine ‘my lord’; very little remains, although an apex is used in the greeting formula, which may imply that it was written by a scribe (see pp. 226–32). 735 is fragmentary, but also includes the word dịxsịt. It looks as though use of <uo> is associated with use of etymological <ss> and of <xs>, and both <uo> and <uu> with standard spelling; most of our examples come from texts associated with high-status individuals, but their appearance in a strength report and an account suggest that this is a coincidence, and the absence of <uu> for /wu/ or <uo> for /uu/ suggests that <uo> is the normal way of spelling /wu/ and <uu> /uu/ at Vindolanda.

The same distinction is found in one of the Claudius Tiberianus letters (P. Mich. VIII 467/CEL 141), where <uo> is used for /wu/ in saluom, no]uom, fugitiuom and <uu> is used for /uu/ in tuum (twice). A confused version of the rule also seems to appear in bolt (469/144) for uult ‘wants’ (<o> after /w/ according to the rule, but /w/ spelt with <b>), while <uu> is also used by the same writer in tuum (468/142), but there is no example of /wu/ or /k^wu/. All of these letters feature substandard spelling to varying degrees (Reference Halla-aho, Solin, Leiwo and Halla-ahoHalla-aho 2003: 247–50), but they also feature (other) old-fashioned spellings (<k> before /a/ in 467/141; <q> before /u/ inconsistently in 468/142; <q> before /u/ inconsistently, <ei> for /iː/ once in 469/144). Also from a military context, but significantly later, the Bu Njem ostraca show one example of seruu (O. BuNjem 71.5) for seruus or seruum and one example of ṭụụṃ (114.5)

In the tablets of Caecilius Jucundus all 5 instances of /wu/ are spelt with <uo> (see Table 12);Footnote ²² all in the word seruos and by five different writers, one a scribe. All 8 instances except 1 of /uu/ are spelt with <uu>; 2 instances are written by a scribe, and the remaining 6 are by Privatus, slave of the colony of Pompeii, who uses <uo> once in duomuiris, which he otherwise 4 times spells duumuiris.

In the tablets of the Sulpicii, /wu/ is commonly spelt <uu>: there are 5 instances (TPSulp. 26, 46, 51, 56, twice) in the word seruus, and 1 in seruum (TPSulp. 51), all written by scribes, between AD 37 and 52. There is 1 example with <uo>, also written by a scribe, at the early end of the date range of the tablets: fụg̣ịṭ[i]ụom (TPSulp. 43, AD 38). There are 4 examples of /uu/ in the lexeme duumuir (TPSulp. 23, scribe; 25, twice, scribe; 110, non-scribe). The tablets from Herculaneum have a single example of [ser]ụụṣ (TH² A10).

In the curse tablets, there are only two instances of <uo> for /wu/, but Primitiuos (11.1.1/18, second–third century AD, Carthage) is not a certain example, because the writer of the curse also writes Romanous for Rōmānus, suggesting some confusion as to the vowel in the final syllables of these names (and perhaps Greek influence). The remaining form uoltis (1.1.1/1, second century AD, Arretium) also features another old-fashioned spelling, uostrum for uestrum (unless this is analogical on uōs and noster, see p. 106), and shows substandard spelling in interemates for interimātis and interficiates for interficiātis, as well as nimfas for nymphās ‘nymphs’. There are 34 instances of <uu> (from the first century BC to third or fourth century AD, from Rome, Hispania, Britannia and Africa).

There are 6 instances of <uo> for /uu/ beside 18 of <uu> (from the first or second century AD to the fourth century AD; two in Germania, one in Italy and the rest in Britannia), but 4 of the examples of <uo> for /uu/ are dated to the first century BC, when the change was only just taking place (1.4.4/3, 1.7.2/1). The remaining text has 2 examples of suom (1.4.4/1, first–second centuries AD). It also contains old-fashioned <ei> for [ĩː] in eimferis for inferīs. All we can really deduce from the evidence of the curse tablets is that use of <uo> was uncommon in texts of this type, but could be found as late as the second century AD in texts which showed other old-fashioned spellings and, in one case, substandard orthography.

In the corpus of letters, <uo> for /wu/ and <uu> for /uu/ is found in CEL 10 (the letter of Suneros, Augustan period), which contains uolt and deuom beside tuum. This distinction can hardly be considered old-fashioned at this time; the spelling as a whole might be considered conservative, as well as including substandard features (see pp. 10–11). Another letter from the last quarter of the first century BC contains sa]ḷuom (CEL 9). By comparison, CEL 167, a papyrus of c. AD 150 from Egypt, contains dịụus and annuum (as well as ]uum, which could represent /wu/, /k^wu/, or /uu/), and CEL 242, an official letter on papyrus from Egypt of AD 505, has octauum and Iduuṃ. In addition, uult (CEL 75) is found in a letter of Rustius Barbarus, tuum (CEL 1.1.18) perhaps third or fourth century AD, suum (CEL 226) in a papyrus from Egypt, AD 341, and ambiguum (CEL 240), a papyrus from Ravenna of AD 445–446. This confirms that at least in the Augustan period there were writers who used <uo> for /wu/, and, in one case, distinguished it from <uu> for /uu/. However, it seems that from the first century AD, <uu> was used also for /wu/.

In the London tablets, there is only a single example of /k^wu/, spelt ]equus (WT 41). In the Isola Sacra inscriptions, there is only a single example of /uu/, spelt suum (Isola Sacra 285). There is also only 1 instance form Dura Europos (e]quum, P. Dura 66PP/CEL 191.42, AD 216).

In the second half of the second century BC, the Romans adopted the practice of writing long vowels with double letters from the Oscan alphabet (Reference OliverOliver 1966: 151–5; Reference VineVine 1993: 267–86; Reference Wallace and ClacksonWallace 2011: 18; Reference WeissWeiss 2020: 32). However, it did not remain a standard part of Latin orthography past the end of the Republic. According to Oliver, Wallace and Weiss, the double spelling of long vowels can be found as late as the early fourth century AD. Oliver points out uii (CIL 3.4121) = uī ‘by force’ from AD 312–323,Footnote ¹ exercituus (CIL 6.230) for exercitūs ‘of the army’, from AD 222–235, and aara (Reference LemerleLemerle 1937 no. 12) for ārās,Footnote ² fourth century according to Oliver, third century according to Lemerle.

It is difficult to find really plausible examples for the first to fourth centuries AD, partly because the possibility of false positives when searching the EDCS is very high, and partly because it is hard to be sure that a particular instance is not a mistake in the writing. Searches for <aa>, <ee>, <oo> and <uu> provide a small number of at all plausible examples:Footnote ³ Spees (CIL 4.5127, prior to AD 79), [I]uunia (CIL 4.8029, prior to AD 79), lacuus (CIL 12.2606, 2607, first century AD), domuus (CIL 9.4794; first century AD, Reference DessauDessau 1892–1916, 7332), Iuulius (AE 1976.700, AD 201), porticuus (ILA 531, AD 298), Ianuariaas (CIL 11.4033, AD 345, but note a mistake in palcis for pacis), Erclaanius (ICUR 10.26742 = EDB18026, AD 301–349), uoot(a) (AE 1977.540), feceerunt (AE 1972.709), dieebus (CIL 14.1212), duouiratuus (CIL 3.9768).

The writers on language who mention this feature at all consider it old-fashioned. Quintilian mentions it in passing:

Unsurprisingly, therefore, use of double letters to write long vowels in the corpora is extremely rare, if not non-existent. The curse tablets provide two possible examples: uoos for uōs ‘you’ (Kropp 11.1.1/26) from Carthage, dated to the second century AD, and ceernis (Kropp 6.1/1) for cernis, from Noricum, mid-second century AD. In neither case can a mere dittography be ruled out. In the case of 6.1/1, additional letters are also written in siuem for sīue (due to anticipation of the following word Iouem?), oporno/tet for oportet and quom/modi for quōmodo (dittography across a line divide). It seems unlikely that ceernis is an intentional use of double letters. In 11.1.1/26 uos is written thus several more times, and no other long vowel is written with double vowels. We also find in this tablet the old-fashioned spelling iodicauerunt for iūdicāuē̆runt (see p. 74). The spelling on this tablet is substandard, but mostly reflects the spoken language. However, there is an unmotivated geminate spelling in coggens for cōgēns ‘forcing’, and a scrambled spelling in Atsurio for Asturiō. I do not think we can be sure that uoos reflects an old-fashioned spelling rather than an accidental dittography.

In addition, we find an instance of q̣uụr (Tab. Vindol. 652) for cūr at Vindolanda between AD 104 and 120. However, while a possible analysis here is that <q> represents /k/ before /u/ and that <uu> represents /uː/, it is more likely that this is a quasi-etymological spelling whereby /qu/ represents *k^w (cf. the spellings quom and quum for cum; pp. 165–8). In the Vindonissa tablets, we have the dative Secundi{i}na<e> (T. Vindon. 41). Again, dittography seems more likely than intentional double writing of the vowel.

The writers on language were aware that <c> had previously been used for /g/ (e.g. Terentianus Maurus 210–211, 894–901 = GL 6.331.210–211, .351.894–.352.891). Instances of <c> for <g> are occasionally found in my corpora, but it is hard to take them seriously as examples of old-fashioned spelling.Footnote ¹ In the curse tablets there are scores, if not hundreds, of instances of <g> in the corpus as a whole, and in most cases the few apparent cases of <c> are probably to be put down to the difficulty of distinguishing <c> from <g>, either in the writing or reading of small letters on a thin piece of soft metal which is generally then subject to folding and unfolding, abrasion, water and other types of damage etc. – cf. Reference VäänänenVäänänen’s (1966: 53) comment that instances of <c> for <g> at Pompeii are ‘simple writing errors’ (‘simples erreurs d’écriture’). For similar reasons, the few instances of <c> for <g> in the graffiti from the Palatine are not to be taken seriously.

Most of the curse tablets have only one or two apparent examples of <c> for <g>. Kropp 1.4.3/2 has colico for colligō beside two other cases of <g>, 1.10.2/1 has [r]oco for rogō beside rogo twice. 3.2/25 has no other instances of <g> beside sacellum for sagellum, but is fairly short. 3.14/1 has defico for defigō and also a number of mechanical errors: intermxixi/ta for intermixta, fata for facta (if this is not due to assimilation), sci for sīc, possitt for possit, amere for amārae. 11.1.1/3, along with two examples of Callicraphae for Calligraphae, has seven other examples of <g>.

In the first to fourth century AD, only 1.4.1/1 (Minturnae, c. AD 50) gives the impression that <c> for <g> could be intentional:Footnote ² it is used for all instances of /g/, in acat for agat, ficura for figūram, dicitos and ticidos for digitōs, and cenua for genua. However, it also has hbetes for habētis, tadro for tradō, uitu for uultum, fulmones for pulmōnes, dabescete for tabēscentem, and ticidos for digitōs. In particular, the use of <f> for <p>, <d> for <t>, and <t> for <d> suggests that the writer, as well as being prone to errors such as omitting or transposing letters, had particular problems in identifying stops correctly. Something weird is going on, but use of <c> for <g> cannot be attributed certainly to the type of education the writer received, rather than linguistic problems (or, conceivably, a form of ‘magical’ writing).

Visible in a small number of the Isola Sacra inscriptions is a curious tendency for /g/ to be represented by <c> when there is another <c> in the word (see Table 13). Whatever the explanation for this, it seems unlikely that it has anything to do with old-fashioned spelling.

Apart from in these corpora, the only other example of <c> for <g> is q]uadrincẹnto (CEL 157, AD 167, Egypt) for quadringenti, where the editor is probably correct to suspect influence from centum ‘one hundred’ (the letter contains another 5 examples of <g>).

Single *i̯ between vowels was lost very early in Latin (possibly at the Proto-Italic stage). Consequently, the sound represented by consonantal <i> between vowels was actually geminate /jj/ from various sources (Reference WeissWeiss 2020: 67–8). I have not been able to find any epigraphical examples of <ii> prior to the first century BC, and Reference WeissWeiss’ (2020: 68 fn. 64) statement that ‘[g]eminate spelling … is frequently encountered on inscriptions’ seems exaggerated. A search for ‘cuiius’, one of his two examples, on the whole of the EDCS, finds 13 examples, as opposed to 793 for ‘cuius’. The other is maiiorem (CIL 2.1964.3.10): a search for ‘maiior’ finds 12 examples, including derived names, beside 948 for ‘maior’.Footnote ¹

The same infrequency applies specifically to the dated inscriptions from the first to fourth centuries AD. I searched for selected forms either mentioned by the grammarians or which appear in the corpora: there are 7 instances of Maiia to 336 of Maia (encompassing the month, divine name, and personal names), 2 of huiius to 192 of huius, 6 of Pompeiius and Pompeiianus to 914 of Pompeius and Pompeianus.Footnote ²

The writers on language often discuss use of <ii>; from Velius Longus, and especially Terentianus Maurus, there are some hints that it might still be in use in the second century AD, but implying that the single spelling is standard. Others do not give the impression that it is much in use in their own time:

Use of <ii> to represent intervocalic /jj/ is very uncommon in the corpora; for example in the Caecilius Jucundus tablets, in which the genre and location of the texts mean that eius, Pompeius, Pompeianus, and the month Maia appear frequently, there are 31 instances of <i> and none of <ii>, and in the Isola Sacra inscriptions, 29 instances of <i> in the words cuius, eius, huius, and in the names Cocceius, Manteiane, Maiorice, Septeius and Tonneius, but none of <ii>. Generally, therefore, I did not count instances of <i> . Where <ii> is used, this may be one of the times when an old-fashioned spelling corresponds to a spelling produced by a writer with lower education, since words like eius really did contain a double /jj/, which might be spelt <ii> simply by a writer who closely produced what they pronounced.

The Vindolanda tablets have a single example of <ii> in Coceịió (Tab. Vindol. 645). Either old-fashioned or substandard spelling is possible: the writer uses old-fashioned <ss> in fussá, but also has substandard features in Vindolande for Vindolandae ‘at Vindolanda’ and resscribere for rescrībere ‘to write back’. He writes the name Maior without <ii>. There are 4 instances in the curse tablets, across 3 different texts. Pompeiius appears in the undated Kropp 1.3.1/1 from Maruvium, eiius in Kropp 3.11/1, fourth century AD, from Britain, and huiius and eiius in Kropp 11.2.1/36 from Africa, perhaps the third century AD.

The brief text of Kropp 1.3.1/1 shows no other substandard or old-fashioned features. In the case of Kropp 3.11/1 a substandard spelling seems most likely: there are a number of others in the text, most notably, since they suggest particular attention to representing glides, puuer for puer [puwɛr] and puuella for puella [puwɛlla]: this is not part of the old-fashioned spelling tradition. There are also straightforward mistakes such as omitted and transposed letters. So I do not think that eiius here should be taken as an old-fashioned spelling. The same could be true for 11.2.1/36, although the writer here produces several spellings which presumably do not reflect his or her speech (initial <h> in hac, hora, hoc despite hypercorrect haera; final <m> in omn]ium, omnium twice, caelum, terram, Veram, de]tinentem, sempiternum, amorem, neminem, alium, quem, consummatum despite immobile for immōbilem; double <ll> in nulli despite coliga for colligā), so the likelihood of its being old-fashioned is higher; no other old-fashioned spellings are found, however.

The Latin alphabet inherited from its Etruscan model a superfluity of signs to represent the phoneme /k/: <c>, <k> and <q>. It also inherited, to some extent, the convention in early Etruscan inscriptions whereby <k> was used in front of <a>, <q> before <u>, and <c> before <e> and <i>, although consistent usage of this pattern is found rarely even in the oldest Latin inscriptions (Reference HartmannHartmann 2005: 424–5; Reference Wallace and ClacksonWallace 2011: 11; Reference Sarullo, Moncunill Martí and Ramírez-SánchezSarullo 2021). Over time, <c> was preferred for /k/ in all positions, while the digraph <qu> was used to represent the phoneme /k^w/. Nonetheless, both <k> before <a> and <q> before <u> (with the value /k/) lived on as optional spellings into the imperial period.

Since the writers on language often mentioned <k> and <q> together, I compile here their comments for both usages (organised in rough chronological order). Especially as regards the use of <k> there is considerable variation between the authors, and the discussion continued past the fourth century AD. Consequently, I have included some writers from later than the fourth century, without carrying out a complete survey. I will refer back to these in the separate discussions of <k> and <q> below.

<k> before /a(ː)/

Modern scholars usually say that in the imperial period, K. was the standard abbreviation for the praenomen Caeso (C. being used for Gaius), and the usual spelling of kalendae was with a <k>.Footnote ³ It is often noted that Carthago, Carthaginiensis was frequently spelt with a <k>,Footnote ⁴ which could also act as an abbreviation for this and for other words. Other words are occasionally mentioned in which <k> is used before /a/.Footnote ⁵

The writers on language have an unusually broad and complex range of views about use of <k> (for the relevant passages, see pp. 138–43). Until the fourth century it seems likely that its use, at least in some contexts, was not uniformly seen as old-fashioned. Quintilian, the only one who says that he believes <k> should not be used (except in words for which it can stand alone as an abbreviation), accepts that others maintain that it should be used whenever <a> follows, as does Velius Longus (who, however, suggests its use is pedantic). Most of the writers say it should be used, either whenever followed by <a> (Donatus, Charisius, Maximus Victorinus), short /a/ (Diomedes),Footnote ⁶ short /a/ at the start of a word (Marius Victorinus) or short <a> at the start of a word when the following syllable begins with a single consonant (Ps-Probus). Terentianus Maurus only gives the examples of kalendae and kaput, along with Kaesōnēs in older writers. Even in those authors who do not restrict its use to the beginning of a word, this may be implicit, since all the examples are in fact at the start of a word.

Terentius Scaurus is the only writer prior to the fifth century not to mention use of <k> in full words: he states that some deny it altogether, and that there are others who approve it only as an abbreviation. Servius and Cledonius say that, while earlier writers used it whenever an <a> followed, now (in the fifth century) it is used only in kalendae. Use of <k> as an abbreviation is identified as representing Caesō, caput and calumnia (Charisius, Diomedes, Dositheus, Velius Longus, Terentius Scaurus), kalendae (Charisius, Dositheus, Velius Longus, Terentius Scaurus), Carthāgō (Charisius and Dositheus), and clādēs (Diomedes).

How does the information in the writers on language fit with the epigraphic record? First it is worth mentioning that <k> is also used to represent /k/ in proper nouns in two contexts which are not relevant to the present discussion; in both cases use of <k> is not restricted to position before /a(ː)/. The first group is Greek names, or Roman names which, however, appear in a highly ‘Greek’ context, such as bilingual inscriptions or inscriptions which contain other Greek names, for example:

• Britanniae / Sanctae / p(osuit) Nikomedes / Augg(ustorum) nn(ostrorum) / libertus (CIL 7.232, York)

• Quintio Ḳrassi / Frugi sumptuarius / Κοιντιων Κρασσου / Φρουγι σουμπτου/αριος (CIL 3.12285, Athens)

• D(is) M(anibus) / L(ucio) Plautio Heli/o filio qui ui/xit annis duob/us mensibus / X diebus XII Iso/krates et Markel/la filio pientissi/mo fecerunt (CIL 6.24272, Rome).

The second group is non-Roman and non-Greek names, in which use of <k> may represent some sound different from the Latin /k/ written with <c>, or is felt to be appropriate to mark out ‘foreign’ names, for example:

• Bodukus f(ecit) (AE 2002.885, Edgbaston)

• D(is) M(anibus) / Galulircli / et omnes / an<t>ecessi / Duetil Tiblik / Eppimus Soris / omn(i)bus co(m)p/otoribus / bene (CIL 13.645, Bordeaux).

Outside these proper nouns, use of <k> overwhelmingly occurs before /a(ː)/,Footnote ⁷ and as a single letter stands as an abbreviation of a number of words (sometimes before other vowels). A large number of personal names that contain /ka(ː)/ are found spelt with <k>,Footnote ⁸ as well as the place name Carthāgō. Otherwise, the use of <k> appears to be lexically determined, with frequency varying significantly across lexemes, as can be seen in Table 14. The table contains all examples of words in which the <k> spellings appear at a rate greater than 1% in inscriptions dated from the first to fourth centuries AD which I found in the EDCS (except for kalendae, where <k> is standard, and words only found abbreviated as k.). I have not removed proper names where these seem originally to have been the same lexeme.Footnote ⁹

The most frequent word with <k> is kalator, which makes up 60% of all instances of this lexeme in inscriptions dated to the first–fourth centuries AD in the EDCS. By far the greatest number of tokens of a word spelt with <k> are found for carissimus ‘dearest’, and it also has a high frequency relative to spellings with <c>: a search for karissimus on the EDCS finds 1258 inscriptions containing this spelling beside 1916 with carissimus in the first to fourth century AD, a frequency of 40%. The adjective from which it is derived, carus ‘dear’, is also common (though far less so than carissimus), especially as a name, and shows a somewhat lower frequency of <k> spellings, at 27%.Footnote ¹⁰ However, the derived noun karitas is found only in 2 inscriptions, with caritas and Caritas in 22 in the same period (8%), while the extremely numerous causa is, as far as I can tell, never spelt kausa after the start of the first century AD;Footnote ¹¹ for many words, there are no spellings with <k> at all.

The relatively high numbers and frequency of <k> in these lexemes compare with the more haphazard occasional uses of <k> in other words. The forms dedicauit, dedicauerunt, dedicatus appear in 1246 inscriptions compared to 6 for dedikauit, dedikauerunt, dedikatus (0.5%); there are single examples of kalumnia,Footnote ¹² kapsarius,Footnote ¹³ kasus,Footnote ¹⁴ kanalicularius ‘a clerk in the Roman army’Footnote ¹⁵ and katolika (ILCV 1259, probably under the influence of Greek orthography).

As an abbreviation, <k> can stand for cardo ‘baseline (in surveying)’, which never appears written in full and is found 11 times in inscriptions dated to the first to fourth century AD, and in fact 49 times across all inscriptions regardless of period.Footnote ¹⁶ It also often stands for castra and castrensis, for caput, and for casa in 3 inscriptions (2 of which are undated in the EDCS). Twice in the same inscription we find b. k. for bona caduca (Reference IhmIhm 1899: 141). The abbreviation k. c. for cognita causa appears in two inscriptions from Ostia (CIL 14.4499, 5000), and ḳ. k. once (AE 2003.703c = 2016.468) for citra cardinem.Footnote ¹⁷ These cases where it stands as an abbreviation of a word where /k/ is followed by a vowel other than /a(ː)/ are surprising, but perhaps connected to the fact that in each case they form a syntagm of two consecutive words beginning with /k/. There is a single instance of kos for consulibus (CIL 6.2120). In CIL 5.1025, the abbreviation k(oniugi) is directly followed by karissime.

The use of <k> in inscriptions to some extent matches what the writers on language say, but shows somewhat different applications. As many state, it is uniformly found before /a(ː)/ (except in abbreviations), but Diomedes, Marius Victorinus and Ps-Probus’ further restriction to short /a/ is not followed, since we find <k> before /aː/ in arcānus, arcārius, cārissimus, cāritās, cārus, euocātus, uicārius. The range of words written with <k> is greater than those given as examples by the writers, but they do not say that their examples are exhaustive (although the repetition of the same examples through the tradition might imply this); there are also examples of <k> in non-initial position, again unlike the examples, and against the explicit advice of Marius Victorinus and Ps-Probus. Likewise, <k> can stand as an abbreviation for caput, but also a wider range of words than implied by the writers, who do not mention <k> for castra and cardo despite their appearance in relatively large numbers (perhaps because the language of the army and of surveying fell outside the literary focus of the writers).

Since I have carried out searches for individual lexemes it is not possible to see to what extent spellings with <k> in the inscriptions reflect an orthographic practice of writing every word containing the sequence /ka(ː)/ with <k>, rather than only writing particular words with <k>. However, the very fact that frequency of spelling with <k> is so variable across lexemes implies that many of those using <k> some of the time did not do so in every context, as some of the grammarians imply is (or should be) the case. In particular, the high frequency of <k> in the word cārissimus (very heavily concentrated in funerary inscriptions) suggests that it was not solely the ‘religiosi’ who were using it in this lexeme, as Velius Longus states (nor was it restricted to letters).

We can now move on to the evidence of the sub-elite corpora, beginning with the instances of <k> at Vindolanda (Table 15).Footnote ¹⁸ The lexeme karissimus, karissime clearly predominates, mostly appearing in the closing greeting of letters, occasionally in the opening greeting (Tab. Vindol. 670, 893 add.), and once in the main text (331). The closing formula is often written in a second hand, presumably that of the author: these are 242 (probably the prefect Cerialis), 247, 285, 623 (Aelius Brocchus, an equestrian officer), 291, 292, 293 (Severa, wife of Aelius Brocchus), 611 (Haterius Nepos, an equestrian, probably a prefect), 613 (unknown), 869 (a Secundus), 875 (unknown). So, many of these instances of <k> are in parts of texts that are not written by scribes. There is no evidence as to whether the instances of karissimus, -e in closing greetings reflect scribal orthography or not in some letters. 341 and 531 preserve only the closing greeting; on 355 the editors comment ‘it is not clear whether the hand changes for the last two lines’. In 632 and 661 the same hand writes the final greeting and the rest of the letter. There are only two other instances of /ka(ː)/ written by the same hands in these letters; 632 also includes the word caballi, while 670 has Cataracṭoni. In 613 the other hand writes capitis, and in 875 Candịḍị, C̣ạ[, Carạṇṭ[.

In the main, the hands which use <k> in this lexeme otherwise use standard spelling (as far as we can tell, since often there is little other text), as we might expect since several are of equestrian rank. In 292 Severa writes ma for mea (which could be a reduced form of the possessive pronoun or dittography after anima; Reference AdamsAdams 1995: 120). In 341 and 531 the geminate seems to be simplified in karisime for karissime, but there is not enough remaining text in either case to tell if this is a feature of these writers. In 661, we perhaps find a substandard spelling if muḍetur is for mundētur, although the editors are uncertain of the reading.Footnote ¹⁹

The remaining four letters have <k> in other contexts: in 250 and 251, Karus is the cognomen of the author, probably also a prefect; the letters are in different hands, and there is no reason to think either is his. There is no other instance of /ka(ː)/. The spelling of 250 is standard except for ḍebetorem for debitōrem ‘debtor’; given the general infrequency of i and e confusion at Vindolanda, Reference AdamsAdams (1995: 91) suggests the influence of debet, which seems possible.Footnote ²⁰ 343 is the letter sent to Vindolanda from Octavius, who may have been a civilian, and whose spelling has both substandard and old-fashioned features. Apart from karrum and ḳarro (and K(alendas)), /ka(ː)/ is otherwise spelt <ca> in Candido, explicabo, spịcas, circa, erubescam, and Cataractonio. 879 has uikario in a highly broken context: the spelling is otherwise standard, but there is little text, and no other instances of /ka(ː)/. In the accounts 583 and 597 there are no other instances of /ka(ː)/ (except for K(alendis) in the former). 597 shows a substandard spelling in the form of laṃṇis for laminīs ‘sheets’ and pestlus and p̣eṣṭḷ[us] for pessulus ‘bolt’ (Reference AdamsAdams 2003: 539–41).

The pattern of <k> use at Vindolanda appears pretty much as we might expect on the basis of the inscriptional evidence. Although strictly speaking we usually do not have evidence that writers who spell cārus, cārissimus with a <k> do not use <k> consistently before /a(ː)/, the match with the inscriptional evidence (and that of Velius Longus) suggests that this spelling is probably specific to this lexeme (there are only 2 examples spelt with <c>, at 255 and 306). We also find uicārius attested with <k> in the inscriptions (although not very frequently); there are no other examples at Vindolanda. The 3 instances across 2 tablets of karrum ‘wagon’ compare to 4 across 3 tablets of carrum (488, 642, 649), 4 across 3 tablets of carrulum (315, 316, 643), 1 of carrārius (309) and 1 of c̣arr[(721). There are no other instances of canis. Apart from in cārus, cārissimus, then, <k> is rare in the tablets: I count 139 instances of <ca>, of which 90 are word-initial, across a great range of lexemes, including castra (3 instances, tablets 300, 668) and castrēnsis (337), castellus (178), caput (613), casula (643), and personal names such as Candidus, Cassius and Caecilius.

As we will see in the case of <xs>, use of <k> does not provide as much evidence for a specific old-fashioned orthographic tradition among the scribes of Vindolanda as might be thought at first sight. In cārus, cārissimus, use of <k> is clearly standard in the greeting formulas of letters among writers of equestrian rank, including Severa, whose education was presumably not carried out in the army, as it is among scribes, if the letters all in one hand were written by them. Octavius, whose letter is apparently in his own hand, who is particularly fond of old-fashioned features, and who may or may not have received an army education, uses 2 out of 5 of the remaining instances of <k>. It is also conceivable that the use of <k> particularly in carrum may reflect this word’s Celtic origins, since <k> was in general associated with foreign words: the writers at Vindolanda were clearly in contact with Celtic speakers, so may have recognised it as a borrowing.Footnote ²¹

The other, much smaller, military corpora, show a different picture. There are 20 instances of <ca> at Vindonissa, including 2 examples of carus (both T. Vindon. 45) and castra (4, 40), and none of <ka>. At Bu Njem, there are 24 of <ca>, including once caṣtṛị[s (O. BuNjem 29), of which 5 are the lexemes camellus ‘camel’, camellārius ‘camel driver’ (3, 5, 10, 42, 78); all 6 examples of <k> are kamellus (8, 9) and kamellarius (7, 8, 76, 77). The preponderance of <k> in these words is probably because of their Greek origin. At Dura Europos, again, <k> is almost entirely lacking. Against 69 instances of <ca>, we find once Kastello (P. Dura 94, c. AD 240) as part of a place name (otherwise spelt with <c> several times) in a summary of dispositions of soldiers and once ]kas(tra?) (P. Dura 66SS/CEL 191.45, AD 216) in a letter. Both of these are lexemes which we saw occasionally receive the <k> spelling in the epigraphy more generally. Several letters in a military or official context from Egypt also use <k> in spelling kastra (CEL 207, AD 200–250), kastresia, kastrense (CEL 205, AD 220) or as an abbreviation for castrī (CEL 231, AD 395; 232, AD 396; 233, AD 401).Footnote ²² This may be a military convention which developed after the time of the Vindolanda tablets. In addition there is 1 instance of kapitum (CEL 234, 399).

The letters also match those of Vindolanda in showing use of <k> in cārissimus, which is used by writers of low and high educational level. 3 of the 4 instances of karissimus come from initial or final greetings formulas as at Vindolanda. 2 occur in the letters of Rustius Barbarus (k[a]ṛịṣṣịṃẹ CEL 74, karisimo CEL 77), which is interesting since he otherwise appears not to know any old-fashioned spellings and his orthography is highly substandard (pp. 35–6). The remaining 2 examples, both also from Egypt, are karissiṃe (CEL 140, AD 103, copy of an official letter from the praefectus Aegypti), and karissi[mum (CEL 177, AD 150–200, a military letter of commendation).

In the Claudius Tiberianus archive, only the scribe of one letter (P. Mich. VIII 467/CEL 141) uses <ka>, and does so inconsistently, with no particular rationale emerging: all 4 examples are word-initial, but the same is true of 1 out of 3 of the instances of <ca>; 2 instances consist of karus (in the main text) and karissimus (in the initial greeting formula), as we might expect, and <k> in Kalaḅ[el] might be particularly likely in a foreign place name, but kasus is not commonly spelt with <k> in the inscriptions. This writer is characterised by another old-fashioned spelling in the form of <uo> for /wu/, and substandard spellings (see p. 263).

Table 16 shows the cases of <k> in the Isola Sacra inscriptions, with <k> once again predominantly used with the lexeme cārissimus, although less frequently than at Vindolanda (where we find 22 cases with <k> and only 2 with <c>) or in the epigraphy in general (where 40% of tokens of cārissimus have <k>). There are 5 instances of <k> in this word against 15 of <c>. Unlike at Vindolanda, the cognomen Cārus appears with <c>: Carae (IS 296). Otherwise, <k> appears in personal names (Kallotyceni, Kania),Footnote ²³ the former of these Greek; but these are very much the minority: there are several dozen other names beginning with <ca>. Note in particular Callistianus (IS 85), Callityche (IS 133), Callisto twice, Callistion (IS 241), Callist[ (IS 282) and Canniae (IS 237). The single instance of <k> within the word is arka (IS 319), which has a <k> not infrequently in the epigraphy more generally. IS 223 has the only instance of a <k> spelling of carīna in all of Latin epigraphy (albeit compared to only three examples of carina), alongside the place name Karthago (but <c> is used in Caelestino). The use of <k> may be connected to the fact that the inscription includes a hexametric composition (and also includes the old-fashioned spelling lubens); its spelling is otherwise completely standard.

There is an interesting distinction between the use of <k> in cārissimus and in other words. As already mentioned, the context of 223 is poetic, while 27, 34 and 319 all show signs of substandard orthography: 27 has Terenteae for Terentiae, filis for filiīs, qit for quid, aeo for eō and sibe for sibi, as well as sarcofago rather than sarcophagō; 34 has mesibus for mēnsibus; and 319 has hypercorrect Lucipher for Lūcifer. Meanwhile, all the inscriptions with <k> in cārissimus have standard spelling. This pattern is not dissimilar from the situation at Vindolanda, where <k> is used in cārissimus by standard spellers, while in other words it is used by Octavius and the writer of Tab. Vindol. 597, whose spelling is substandard. This may imply that use of <k> before <a> is subject to some rather fine distinctions: in the word karissimus it is an acceptable and commonly used variant in standard orthography, but in other words it may have a poetic ring or be characteristic of substandard spellers.

The tablets from Pompeii and Herculaneum show complete avoidance of <k> before /a(ː)/, except for in kalendae, and its derivative kalendarịọ (TH² A13) ‘estate’. In TH² there are some 55 instances of <ca>, including 3 instances of Carum (2 in TH² 85, 1 in TH² 85 bis) as a personal name. In the Jucundus tablets there are approaching 200 instances of <ca>, almost all in personal names, and no instances of <k> (there are no instances of cārus or Cārus). In the tablets of the Sulpicii there are also large numbers of <ca>, including a handful of Cārus, while the only case of <k> is in Ḳ[innamo] (TPSulp. 62), where the use of <k> is presumably triggered by the Greek name.

In the curse tablets, use of <k> before /a(ː)/ is practically non-existent, compared to several hundred instances of <ca>. The only instance is Karkidoni (Kropp 11.1.1/37), from Carthage, which is presumably a foreign name; Greek letters are also used at the start of the tablet. There are a handful of cases of <k> before another letter: Greek influence must also be responsible for the spellings koue for quem and kommendo (11.2.1/32), also from Africa, and perhaps also for Klaudia (1.5.4/2, Pompeii), given that this woman’s cognomen is Elena; the divine name Niske (3.11/1) from fourth century AD Britain is presumably a case of <k> being used for a foreign name, although no similar explanation arises for Markellinum (3.2/45, Aquae Sulis, second or third century AD).

The graffiti from the Paedagogium contain 4 instances of <k>, 2 in the Greek word Nikainsis (297), Nikaẹnsis (332), and 2 in Kartha(giniensis) (322), Kart(haginiensis) (323); all perhaps from the reign of Septimius Severus. There are 10 other examples of <ca>. In the London tablets there are 16 instances of <ca>, including Caro as a personal name (WT 36) and castello (WT 39), and none of <ka>.

Use of <k> in the corpora depends very much on lexeme and genre. The word carissimus, which occurs frequently in fairly formulaic contexts in letters, is overwhelmingly spelt with <k>, in the letters at Vindolanda and elsewhere, in texts which show standard and substandard spelling, and which are written both by scribes and non-scribes, including those of relatively high and low social rank. The same lexeme is also spelt frequently with <k> in the Isola Sacra funeral inscriptions.

At Vindolanda, it is possible that carrum ‘wagon’ may also have become associated with a spelling <k>, although this may have been a peculiarity of individual writers rather than part of the scribal tradition in the army; as also perhaps in the very occasional other instances of use of <k>. The use of <k> in the word castra, and in particular as an abbreviation, seems to have been a feature of official/military spelling in Egypt and Dura Europos in the third–fifth centuries AD.

There is very little evidence for a general rule that <k> should be used before all instances of /a(ː)/, although perhaps the writer of one of the letters of Claudius Tiberianus had learnt such a rule (4 out of 5 instances of word-initial /ka(ː)/ are spelt with <k>). Interestingly, although there is a small number of personal names being spelt with <k>, it is not clear that it is their status as personal names that triggers the <k>: in the two instances of Karus at Vindolanda, the cognomen is also a lexeme which anyway tends to be spelt with <k>, and in the examples in the Isola Sacra inscriptions there are no other instances of /ka(ː)/. Outside the lexeme carus, use of <k> may have a remarkable twofold correlation: on the one hand with substandard spelling, and on the other hand with other old-fashioned spellings.

<q> before /u(ː)/

Reference Wallace and ClacksonWallace (2011: 27 fn. 29) notes that <q> before /u(ː)/ ‘is found with some frequency in late Republican Latin, particularly in the word for “money”’ [i.e. pecūnia], and that it is ‘also found sporadically in Imperial Latin texts’. On the whole, a more comprehensive investigation supports Wallace’s statements. Searches on the EDCS show that pequnia is found in 39 dated inscriptions up to the end of the first century BC, pecunia in 45, giving a frequency of 46%,Footnote ²⁴ which is higher than most other lexical items containing this sequence:Footnote ²⁵ Merqurius, Mirqurius is found in 4 inscriptions, Mercurius, Mircurius, Mercurialis in 18, giving a frequency of 18%; sequndum, Sequndus is found in 4 inscriptions, secundus, secundum, Secundius in 52, giving a frequency of 7%; cura and forms of curare are found in 71 inscriptions, where qura only appears once (1%), although the low number of instances of <q> is probably due to the existence of the alternative spelling coera-.

The numbers of inscriptions with <q> decline significantly in the period of the first to fourth centuries AD, but the relative preponderance of pequnia continues:Footnote ²⁶ it is found in 20 inscriptions dated from the first to the fourth centuries AD, while pecunia is in 650, to give a frequency of 3%. A search for curauit, curauerunt between the first and fourth centuries AD gave 750 inscriptions, while there were 6 with quravit, quraverunt, a frequency of 0.8%. Only 2 examples of sequritas are found (both from fourth century AD inscriptions), while securus, securitas, Securius are found in 243 inscriptions, also a frequency of 0.8%. 2 inscriptions contain Merqurius, and 773 have Mercurius, a frequency of 0.3%.Footnote ²⁷ There were 3 inscriptions containing Sequndus, Sequndinus (all names), and 3241 containing secundus (including as a name), Secundius, Secundulus, Secundinus, secundum, giving a rate of 0.09%. Not all words containing /ku/ have variants with <qu>. For example, there were 61 instances of secum between the first and fourth centuries, and none of sequm.

This data suggests that pecūnia is one of the most frequent words which appears spelt with <q>.Footnote ²⁸ This is true both in the earlier period up to the end of the first century BC and in the period of the first to fourth century AD, even though the rate at which <q> was used for /k/ declined significantly in the first to fourth centuries AD compared to the earlier period.Footnote ²⁹ In both periods, the rate at which <q> was used varies between lexemes.

The decline in the use of <q> before /u(ː)/ that we see in inscriptions is reflected in the relative lack of attention to this spelling provided by the writers on language (see pp. 138–43 for the relevant passages). Although it is often noted that <q> is only used before <u>, the examples given generally make it clear that the writer is thinking of the use of <qu> for /k^w/, rather than /ku(ː)/ (e.g. Maximus Victorinus, Ars grammatica, GL 6.195.19-23–196.1; the same may be true of Donatus, Ars grammatica maior 1.2, p. 604.16–605.2). Already in the mid-first century AD, Cornutus’ rule (at Cassiodorus, De orthographia 1.23–24) makes it clear that <q> is only to be used when it is followed by <u> and one or more vowels, i.e. when it represents /k^w/, and this is implied also by another passage (De orthographia 1.45–48); the same rule is stated by Curtius Valerianus (in Cassiodorus, De orthographia 3.1). Velius Longus (De orthographia 13.10), Marius Victorinus (Ars grammatica 4.36), and perhaps Terentianus Maurus (De litteris 204–209) are aware of <q> before /u/, and all deprecate it,Footnote ³⁰ suggesting that it may have had some continued currency, but was presumably a minority usage. Servius (Commentarius in Artem Donati, GL 4 422.35–423.4) describes it as an old custom, no longer in use.

The use of <q> for /k/ before /u/ is found occasionally in the corpora, but is neither frequent nor widespread. There are no examples in the TPSulp. (dozens of cases, including 18 of pecunia) or TH² tablets (31 instances, including 4 of pecunia), nor in the graffiti from the Paedagogium (20 instances of <cu>), the Bu Njem ostraca (35 instances of <cu>), the tablets from Vindonissa (14 instances of <cu>), or the Isola Sacra inscriptions (71 instances of <cu>). Among dozens of instances of <cu> from Dura Europos, the only exception is in the name Iaqubus (P. Dura 100.xxvii.12, 101.i.f.3, 101.xxix.14, Dura 101.xxxv.15), where it presumably reflects an attempt to represent non-Latin phonology.

In the Jucundus tablets, the scribe of the earliest tablet twice writes pequnia (CIL 4.3340.1, AD 15); this text does not contain any other cases of /ku/. All subsequent scribes use <cu> in this lexeme (38 instances); in total there are more than 200 instances of <cu> used by both scribes and other writers. There is only one other certain instance of <qu>: Iuqundo (45, undated but presumably in the 50s or 60s AD); there are no other cases of /ku/ in this part of the text, which was written by P. Alfenus Varus, the first centurion of the praetorian cohorts, and subsequently praetorian prefect under Vitellius. His spelling is substandard (problems with geminates: Augussti for Augusti, acepisse for accepisse, Pollionnis for Pollionis, acctum for actum; missing final /m/: noue for nouem; misplaced aspirate: Nucherina for Nucerina).Footnote ³¹ Both Privatus, the slave of the colony of Pompeii, and the scribe on his tablets use <q> several times in pasqua for pascua ‘pastureland’ (145, 146, 147), beside 1 instance of pascua[m] (146). However, this spelling could reflect the reduction of /ku/ to /kw/ before a vowel, and this seems particularly likely since there are another 15 instances of <cu> in tablets 145–147, used by both Privatus and the scribe.

There are two doubtful examples in the Vindolanda tablets beside more than a hundred instances of <cu>: q̣uụr (Tab. Vindol. 652) for cūr ‘why’ could be analysed as having <q> for /k/ and <uu> for /uː/, but since the use of double letters for long vowels is not otherwise found at Vindolanda, and seems to be very uncommon by the end of the first century AD (see pp. 129–31), it is more likely that this is an etymological spelling with <qu> for /k/ < *k^w before a back vowel (see pp. 165–8). In quequmque (Tab. Vindol. 643) for quaecumque ‘whatever’ it seems not improbable that <q> before <u> is triggered by the <qu> in the preceding and following syllables (<c> is used elsewhere in this letter in arculam, securem).

Compared to the absence of <qu> for /ku/ at Vindolanda, it is striking that it was clearly part of the orthography of several of the (presumably) military scribes writing the Claudius Tiberianus letters (Table 17), although it appears to be used consistently only in P. Mich. VIII 471/CEL 146 (and perhaps 470/145); all of these letters except 472/147 show substandard spelling.

In the other texts from CEL, there are two instances of <q> before <u>: ]q̣ụṣ, presumably for -cus (CEL 7),Footnote ³² dated to a little before 25 BC, and qum for cum (CEL 10, the letter of Suneros, Augustan period); both also contain 2 instances of <cu>. CEL 7 contains some substandard features (nuc for nunc ‘now’, c̣ic̣quam for quicquam), as well as <u> for <i> in optumos for optimōs ‘best’ (although this is probably not old-fashioned at this time), while CEL 10 has spelling which is conservative to old-fashioned, as well as substandard features (see pp. 10–11). Otherwise only <cu> is found, in large numbers, in CEL.

The curse tablets also occasionally use <q>, across the first to third (or fourth or fifth) centuries AD (Table 18). Again, the middle <q> in quiqumque (Kropp 11.2.1/3) may have been instigated by the <q> in the preceding and following syllables. In Kropp 2.1.3/3, pequnia and pequniam for pecūnia with <q> are found beside qicumqui for quīcumque. The spelling is substandard: Cr[y]se for Chrysē, uius for huius, onori for honōri, senus for sinus, o[c]elus for ocellus; substandard spelling occurs also in Kropp 4.1.3/9 (uinculares for uinculāris) and in Kropp 11.1.1/15 (demo[n– for daemōn, oc for hōc, [a]c for hāc, ora for hōra). This small number of instances of <q> compares with more than 200 of <c> before <u> across the corpus

On the whole, those corpora which are more homogeneous, and produced in an environment which might favour uniformity of orthography, avoid <q> for /k/, in particular the scribes of the various Pompeian tablets, and texts from the army bases at Vindolanda, Vindonissa, Bu Njem, and Dura Europos as well as the Paedagogium. The single instance in the earliest Caecilius Jucundus tablet, and the Augustan dating of CEL 7 and 10, hint at a move away from its use in the first century AD (which we would expect on the basis of the inscriptional evidence). However, its use by P. Alfenus Varus in the Jucundus tablets, its appearance in a small number of the curse tablets, and in particular the quite remarkable cluster of instances in the Claudius Tiberianus letters, suggest that use of <q> maintained a somewhat underground existence in certain educational traditions. Given the assumption that the Claudius Tiberianus letters were written by military scribes, the use of <q> in some of them is particularly surprising. But, as already noted, these texts are remarkably heterogeneous in other features of their spelling, which might support the idea that their scribes had been educated in a less consistent fashion than in the other army corpora; I am not sure whether this means we should rethink the assumption that this education took place in the army.

There is not really enough evidence to discuss the lexical distribution of <q>; out of 20 tokens including <q>, 3 are pequnia, which may reflect the preponderance of this lexeme in the inscriptional evidence identified at the start of this section. However, we do not know how the scribe of CIL 4.3340.1 would have written other examples of /ku/ so cannot be sure that pecūnia had any special orthographic status for him; in Kropp 2.1.3/3 qicumqui does compare with pequnia.

Original *k^w was lost before back vowels in Latin, as in *sek^wondos > secundus ‘following, next’, *k^wolō > colō ‘I cultivate’, in the second half of the third century BC (Reference MeiserMeiser 1998: 92; Reference WeissWeiss 2020: 165).Footnote ¹ The spelling with <qu> was maintained (or reintroduced) in some words (e.g. equus ‘horse’, aequus ‘equal’, on the basis of parts of the paradigm where /k^w/ occurred before a non-back vowel; quottidiē ‘every day’, which is attested later than cottidiē,Footnote ² Reference Ernout and MeilletErnout and Meillet 1985: 146), notably in quom for cum ‘when, since’ < *k^wom (and by extension, also for cum ‘with’ < *kom). The spelling quom is found frequently in epigraphic texts of the first century BC and earlier, and there are still occasional examples in the first century AD and later.Footnote ³ An ‘original texts’ search on EDCS for ‘ cum ’ (date range ‘1’ to ‘400’, 29/09/2022) finds 4,489 inscriptions containing cum. Such a small number of instances suggests that the spelling was old-fashioned by the first century AD, and this is likely to be all the more true for other spellings with <qu>.

Velius Longus and Curtius Valerianus confirm the view that on the whole the <qu> spellings are old-fashioned:

However, Curtius Valerianus (in Cassiodorus, De orthographia 3.3–4) and Marius Victorinus make an exception for cum as a temporal adverb, recommending a (rather artificial) spelling quum. But the spelling quom (unmentioned by Curtius Valerius) is too old-fashioned for Marius Victorinus:

• Terentius Scaurus, however, does not rule out quom as old-fashioned:

Caesellius Vindex recommends not only quum but also quiusque for cuiusque:

In line with most of the writers on language, spellings with <qu> are infrequent in the corpora. Although cum is a common word across these texts, we find quom ‘when, since’ only in a letter written by a scribe at Vindolanda (Tab. Vindol. 248) and qu[u]m or qu[o]m in a Claudius Tiberianus letter (P. Mich. VIII 472/CEL 147).Footnote ⁵ At Vindolanda we also find q̣uụr (Tab. Vindol. 652) for cūr ‘why’ in a fragmentary letter whose writer is unidentifiable, although as they use an apex they are likely to be a scribe (see pp. 226–32).

There appears to be a hypercorrect use of <qu> in laqụonecoru (CEL 225), presumably for lacōnicōrum ‘steam baths’, in a papyrus letter of the fourth century AD from Karanis in Egypt, apparently a petition of some sort whose writer shows some substandard features.Footnote ⁶ Curiously, there are also hypercorrect examples before <a> in a pair of curse tablets from Baetica in the first century BC: omut[e]sq[ua]nt (Kropp 2.2.3/4), [om]utesquant, [omut]esquant (2.2.3/5) for obmūtescant ‘may they become dumb’.

This evidence suggests that while use of <qu> in these contexts was old-fashioned and uncommon, it did survive within some educational traditions for quite some time, although its restriction to words in which it was etymologically correct was not necessarily well learnt or taught.Footnote ⁷

Reference Mancini, Agostiniani and MarcheseMancini (2019) provides a useful summary of the history of the spelling <xs> for <x>. The earliest example in a Latin inscription is exstrad (twice) in the SC de Bacchanalibus (CIL 1².581) of 186 BC,Footnote ¹ although two instances of faxsit in testimonia of the Laws of the Twelve Tables are argued by Mancini to reflect an edition carried out by Sextus Aelius Paetus, curule aedile in 200 BC, consul 198, and censor in 194.Footnote ² In addition, the Marrucinian ‘Bronze of Rapino’, datable to the second half of the third century BC, has lixs ‘law’ < *lēg-s. In a corpus of inscriptions from this period until 30 BC, Mancini counts 135 occurrences of <xs> beside 1,310 of <x> (<xs> thus making up 9% of the total).

From the Augustan period it more or less dies out in ‘official’ inscriptions, with occasional archaising usages in juridical inscriptions in the first and, once, second century AD. However, it continues to be used in other inscriptions until a late period, although always making up a small minority compared to uses of <x>. Mancini compares the 655 examples of uixsit ‘(s)he lived’ with 62,946 cases of uixit (1%); likewise he finds 497 cases of uxsor beside 6,858 of uxor (7%).Footnote ³ To what extent these figures are reliable as to the rate at which <xs> was used is unclear. The EDCS with which he carried out these searches throws up plenty of false positives, and with such great numbers not much checking can have been carried out. As an additional contribution, I carried out searches for sexaginta and sexsaginta, and checked them for accuracy. The former appeared in 71 inscriptions, the latter in 14 (16%), giving a much larger minority for <xs>.Footnote ⁴ The variation may reflect the smaller numbers of sexaginta in inscriptions, or genuine lexical variation as to use of <xs> versus <x> (although no particular pattern arises on this front from my investigation of the corpora below).

The spelling <xs> is barely mentioned by the writers on language, except for a brief hint by Cornutus and more explicit statements by Caesellius and Terentius Scaurus that <xs> should only be used in compounds consisting of the preposition ex plus a word beginning with /s/. There is no suggestion that <xs> is old-fashioned, just incorrect:

Terentius Scaurus also mentions people who argue that words ending in <x> should have <xs>; again this is described as incorrect rather than old-fashioned:

The marginality of the spelling <xs> is confirmed in my corpora, although with some variation.Footnote ⁶ In the Vindolanda tablets (see Table 19) I find 10 certain instances of <xs> vs 66 of <x>.Footnote ⁷ Reference Mancini, Agostiniani and MarcheseMancini (2019: 27) reproves Reference AdamsAdams (1995: 90) for the statement that ‘xs is commonly written for x in the tablets’, but at 13%, <xs> does appear at a higher rate than uixsit and uxsor in Mancini’s calculations from the whole corpus of Latin inscriptions. Adams sees the use of <xs> as formal or archaising, while Reference Mancini, Agostiniani and MarcheseMancini (2019: 28) argues that it is ‘informal and bureaucratic’ (informale e cancelleresco).Footnote ⁸ This disagreement may be due to a different perspective on the status of the Vindolanda tablets. Mancini is comparing the presence of <xs> in ‘everyday documents’ such as the tablets from Vindolanda, London and those of Caecilius Jucundus from Pompeii, alongside papyrus letters, with its absence in public epigraphy. By comparison, Adams is more focussed on the usages of individuals in the Vindolanda tablets, and variation between genres within the corpus.

Clearly, the letters found at Vindolanda are ‘informal’ relative to public epigraphy, but we do have a hint that they could be marked out from other genres by the tendency for apices to be used preferentially in letters as opposed to other types of text (see pp. 235–6). And in fact, apices and <xs> co-occur in Tab. Vindol. 628. The sequence <xs> also tends to appear in letters, which provide 8 out of 9 instances in which the genre of the text is recognisable.Footnote ⁹ This compares with <x>, of which 37/65 instances appear in letters (one document is of uncertain genre). The numbers are too small, however, to be sure that <xs> does correlate with letters.Footnote ¹⁰ The use of <xs> is also not necessarily consistent within a text: 301 has explices beside sexs,Footnote ¹¹ and in the letter of Octavius (343), apart from uexsaṛe, there are 6 instances of <x>, consisting of dixi and 5 examples of the preverb ex-.

Adams also observes that the three examples of <xs> in 309 appear alongside the spelling of mīsī ‘I sent’ as missi, ṃissi, in a text whose spelling is otherwise standard. And in fact there are further connections between use of <xs> and <ss>. The same hand that writes 181, which contains ụexṣịllari, also writes 180, another account, and 344, a letter which enables the author to be identified as a civilian trader. 180 (which has <x> in ex) also has <ss> in ussus for ūsūs ‘uses’. 344 has no instances of <x(s)> but writes comississem for comīsissem ‘I had committed’. Given the civilian status of the author, it may be that the spelling he or his scribe uses is the result of different training from that of the scribes in the army. In 343, the spelling <ss> is also attested indirectly in the form nissi for nisi ‘if not’ (see p. 185). It is striking that 5 of the 22 instances of <ss> in the Vindolanda tablets occur in documents which also have <xs>. In 343, apart from using <xs> and <ss>, the writer is also characterised by the rare use of <k> in a word other than k(alendae) and cārus ‘dear’: karrum and ḳarro for carrum ‘wagon’.

The spelling <xs>, therefore, appears in texts which use other spellings which might be considered old-fashioned. It is reasonable to suppose that <xs> may have had a similar value. From a sociolinguistic perspective, <xs> appears in letters from a range of backgrounds. In 301, the writer Severus is a slave, writing to a slave of the prefect Flavius Genialis in his own hand. The author of 284 is probably a decurion, writing to the prefect Flavius Cerialis, and 628 is also a letter to Cerialis from a decurion called Masclus (but both are probably using scribes; note the use of apices in the latter). The author of 309 (Metto?) is probably a civilian trader, though most of the letter is written in another hand. Very little remains of 662 or 735. All of these show otherwise standard spelling, as far as we can tell (other than Masclus for Masculus in 628, with a ‘vulgar’ syncope; but since this is the author’s name this does not necessarily suggest a lower educational standard on the part of the writer).Footnote ¹²

On the other hand, the writer of 343, whose author was Octavius, who may have been ‘a civilian entrepreneur and merchant, or a military officer responsible for organising supplies for the Vindolanda unit’, according to the editors, combines use of <xs>, <ss> and <k> with the substandard spellings <e> for <ae> in illec for illaec ‘those things’, arre for arrae ‘pledge’, que for quae ‘which’, male for malae ‘bad’, <ae> for <e> in ṃae for mē ‘me’, and <i> for <ii> in necessari for necessariī ‘necessary’. The letter 344 contains only standard spelling, but the accounts 180 and 181, by the same writer, do include a few substandard spellings: bubulcaris for bubulcāriīs ‘ox-herds’, turṭas for tortās ‘twisted loaves’ (both 180), emtis for emptīs, balniatore for balneātōre, and Ingenus for Ingenuus (all 181).

Overall, Adams’ view that <xs> is formal or archaising, within the context of the Vindolanda tablets, receives some support from its association with other old-fashioned spellings, in the form of <ss> for <s> by three different writers, and with <k> in one of them. However, we cannot be sure that its greater frequency in letters is due to the relatively more formal status of these than other types of document. The writers who include <xs> in their texts all probably belong to the sub-elite, consisting of slaves, scribes and perhaps civilian traders. It is found in texts which demonstrate both standard and substandard spelling. It is conceivable that <xs> is not actually a major part of the scribal tradition of the army itself, since at least 4 of the instances come from letters whose authors were civilians (5 if Octavius, the author of 343, was also a civilian), and only 284 (1 example) and 628 (1 example) seem to have definitely been written by military personnel. But of course, military scribes, and/or education in writing, may have been available also to non-military personnel.

Two of the other corpora are particularly noteworthy in terms of use of <xs>. One is the London tablets, which contain 4 examples of <xs> and only 3 of <x> (see Table 20). The spelling of WT 44 and 45 is standard; WT 55 is substandard (see p. 264), and also uses another old-fashioned spelling, <ss> after a long vowel in u]s{s}uras and promis{ṣ}it; the spelling of WT 67 is also substandard (see p. 264). As for the tablets which have <x>, WT 29 has substandard features (see p. 264), along with 2 instances of <ss> in [o]cassionem for occāsiōnem ‘occasion’ and (hypercorrect) messibus for mēnsibus ‘months’. WT 31 has standard spelling except for Aticus for Atticus, which may simply be a haplography. WT 72 has Butu for Butum, but the reading is difficult and the word is at the end of a line anyway so may reflect lack of space. What other text there is has standard spelling (n.b. Ianuarium) and a hypercorrect use of <ss> in ceruessam. It seems that in these tablets <xs> can correlate with both standard and substandard spelling, and with <ss>, while <x> is found with substandard spelling and <ss>, but there is hardly enough evidence to draw particular conclusions from this other than that <xs> is remarkably common.

The other corpus is the tablets of Jucundus, in which <xs> is characteristic of the scribes, who use it 35 times to 11 instances of <x>, whereas the other writers have 2 examples of <xs> and 15 of <x> (see Table 21 and Table 22). In fact, there seem to be three important factors which apply to the use of <xs>. 25 of the examples of <xs> occur in the word dixsit (and dixserunt) in tablets concerning auctiones, which contain the formulas habere se dixsit … ‘(s)he said that (s)he has [a certain amount of money]’ and accepisse se dixit/dixserunt … ‘(s)he/they said that (s)he/they has/have received [a certain amount of money]’, which are always written by scribes. The difference between use of <xs> in dīxit/dīxē̆runt and in other words by the scribes is statistically significant.Footnote ¹³ An explanation for this might be that the spelling with <xs> was felt to be particularly appropriate for this word because it appears in a formulaic context.Footnote ¹⁴ However, even if we leave dīxit/dīxē̆runt out of the equation (and not including one uncertain case), there is still a statistically significant difference between the rates of use of <xs> and <x> in other words by scribes (10:11) and other writers (2:15); see Table 23.Footnote ¹⁵ Tablet 1 is the earliest of the tablets, and perhaps reflects a slightly different orthographic training: as well as using <x> in dixit, it also uses the spelling pequnia versus the pecunia found uniformly in the other tablets.

It is difficult to identify a cohesive pattern in the use of <xs> across the corpora. On the one hand, the scribes of the Caecilius Jucundus tablets heavily favour <xs> at a rate of 69%, or 53% if we assume that dix(s)it is a special case, which compares significantly with the usage of the other writers, who use <xs> only 12% of the time. By comparison, the typologically, geographically and chronologically similar corpora TPSulp. and TH² demonstrate an avoidance of <xs>, on the part of both scribes and others. The former has a single use of <xs> in sexsṭum (TPSulp. 46, scribe), compared to 87 other examples of <x> (and one case of <cs> in Ạḷẹcsì, TPSulp. 90). The latter has 32 instances of <x> and none of <xs>.

In Kropp’s corpus of curse tablets there are 19 instances of <xs> overall, and 131 of <x>, giving a rate of 13%. 7 of these are in texts dated to the second and first centuries BC; Table 24 gives all examples from the first century AD onwards. All of these tablets except 3.2/26 feature substandard spellings; 3.2/24 and 3.22/3 also have (hypercorrect) <ss> in nissi for nisi ‘if not’.

The Isola Sacra inscriptions contain a few instances of <xs>, with 5 compared to 105 of <x>. 1 example of uixsit (IS 258) compares with 43 instances of the perfect stem of uīuō with <x>, and the 1 example of uxsori (IS 98) with 7 of uxor (though this does give rates of 2% and 14% respectively, both twice as frequent as the rates found by Mancini in the epigraphic evidence more generally). Strikingly, the word most frequently spelt with <xs> is the cognomen Fēlix, with 3 instances of <xs> (IS 44, 225, 312) versus 4 of <x>. Only one of the inscriptions containing <xs> also contains a substandard spelling, in the form of comparaberunt for comparāuē̆runt (IS 312). The same inscription also has <x> in Maxima.

The other corpora mostly show no or little use of <xs>. The only instance of <xs> in the Bu Njem ostraca is sexsagi[nta (O. BuNjem 78), in a letter written by a soldier called Aemilius Aemilianus, whose spelling is not as bad as in some of the other texts, but does include some substandard features (see p. 263). They also include the non–old-fashioned transmisi, which appears in all the letters, but this spelling probably comes from the template that Aemilianus was using (Reference AdamsAdams 1994: 92–4). There are 24 instances of <x> in other ostraca. At Dura Europos <xs> is entirely absent, and there are more than a hundred cases of <x>. Vindonissa has no examples of <xs>, but only 3 of <x>. The graffiti from the Paedagogium have 25 instances of <x> and none of <xs>.

Within the corpus of letters, <xs> is interestingly absent from those of Claudius Tiberianus, despite the preponderance of both old-fashioned and substandard spellings (although there are only 8 instances of <x>, 4 each in P. Mich VIII 467/CEL 141 and 472/147). The letters definitely attributed to Rustius Barbarus also have 9 instances of <x> (CEL 73, 74, 77, 78) and none of <xs>, although CEL 80, which belongs to the same cache but may not have been written by Rustius, has exsigas for exigās ‘you should take out’. Of the other letters, the private letter of the slave Suneros (CEL 10), of Augustan date, has 3 instances of <xs> (on Suneros’ spelling, see pp. 10–11). There is then 1 in CEL 88 (probably first century AD), and CEL 140, a papyrus copy of an official letter of probatio from Oxyrhynchus (AD 103), which also contains three examples of <x>, and which has otherwise standard spelling (including <k> in karissiṃ[e]).

A number of different changes took place to reduce original geminate consonants in Latin. In addition, there was another rule (or rules) which produced geminates out of original single consonants. Since these changes did not take place at the same time, and were not necessarily reflected in spelling at the same rate, I will discuss them here separately.

<ss> and <s>

Double /ss/ was degeminated after a long vowel or diphthong around the start of the first century BC (Reference MeiserMeiser 1998: 125; Reference WeissWeiss 2020: 66, 170), for example caussa > causa. A search for caussa finds 23 inscriptions from the first four centuries AD, compared to 269 for causa (a frequency of 8%), although the spelling with <ss> is rather higher in the first century AD (18 or 19 inscriptions containing caussa to 60 inscriptions containing causa = 23 or 24%),Footnote ¹ including in official inscriptions such as the Res Gestae Diui Augusti (Reference ScheidScheid 2007; CIL 3, pp. 769–99, AD 14),Footnote ² the SC de Cn. Pisone patri (9 instances of causa to 3 of caussa in the B copy; Eck et al. 1968, AD 20), and CIL 14.85 (AD 46, EDR094023). By comparison, a search for (-)missit finds 4 instances in the first four centuries AD compared to 192 of (-)misit (a frequency of 2%).Footnote ³

Most of the writers on language clearly considered the <ss> spelling old-fashioned:

Although Terentius Scaurus states that there are ‘many’ who use the double <ss> spelling in causa:

At Vindolanda the 21 instances of etymologically correct <ss> compare with 24 of <s>, giving a total of 47% (see Table 25).Footnote ⁴ The frequency with which the <ss> spelling is found in mīs-, the perfect stem of mittō ‘I send’, is out of kilter with the uncommon spelling of this lexeme with <ss> in the epigraphic evidence as a whole.

In 225, <ss> is used in the draft of a letter probably written in the hand of Flavius Cerialis, prefect of the Ninth Cohort of Batavians himself, a man apparently of some education (on which, see Reference AdamsAdams 1995: 129, and p. 1), who also uses <uo> for /wu/. It is also found in 255, from Clodius Super to Cerialis; the editors suggest that though a centurion, Clodius may have been an equestrian (but there is no evidence he wrote it himself). In 256, a letter to Cerialis from a certain Genialis, <uo> is also used for /wu/ in siluolas; there are no substandard spellings. In the case of 312, a letter from Tullio to a duplicarius whose gentilicium is Cessaucius, the editors note that ‘[t]he hand is rather crude and sprawling’, which may suggest a lower level of education in the writer, although no substandard spellings are found.Footnote ⁵

Metto (the author of 309) and the anonymous author of 180 and 344 were probably civilians, and therefore not necessarily using military scribes. The writer of 309 also uses <xs> for <x>, as does the writer of 180 and 344 (who also writes 181: ụexṣịllari), who also includes substandard spellings in 180 (bubulcaris for bubulcāriīs ‘ox-herds’, turṭas for tortās ‘twisted loaves’ and 181 (emtis for emptīs, balniatore for balneātōre, and Ingenus for Ingenuus). Substandard spellings are also found in 892, a letter from the decurion Masclus to Julius Verecundus, prefect of the First Cohort of Tungrians, which has commiatum for commeātum and Reti and Retorum for Raetī, -ōrum. Since the final greeting is in a different hand, presumably that of Masclus himself, the writer of the rest of the text was probably a scribe.

Tab. Vindol. 343, whose author, Octavius, could have been a civilian or in the military, also contains a number of substandard spelling features (see p. 262), but also <k> for /k/ before /a/, and <xs>. The single example of <ss> in nissi is interesting because there was never an etymological *-ss- in nisi, which comes from the univerbation of *ne sei̯. However, since this univerbation must have occurred after rhotacism, nisi presumably contained an intervocalic voiceless /s/, a feature shared almost exclusively with forms like mīsī < mīssī, where it was the result of degemination of original /ss/. The writer of 343 must have learnt the spelling with <ss> and mistakenly overgeneralised it to nisi.

We can conclude that the spelling <ss> for /s/ after a long vowel or diphthong is common at Vindolanda (nearly half the examples). It correlates with other old-fashioned spellings such as <uo> for /wu/, <xs> for <x>, and <k> for /k/ before /a/. However, it does not correlate with quality of spelling: although it is used by the well-educated Cerialis, it also appears in texts which also feature substandard spellings, and in texts which are not necessarily written by military scribes.

Reference Cotugno, Marotta and MolinelliCotugno and Marotta (2017) argue against <ss> at Vindolanda being an old-fashioned feature, on the basis that since <ss> is found in accounts as well as letters, it cannot have been used as a stylistic marker, as might be the case in letters, and consequently that its use should not be considered an archaism. They suggest that instead it arose as a way of marking a voiceless tense /s/ among Batavian speakers of Latin (North-Western Germanic languages having, like Latin, turned original voiceless *s into /r/ by rhotacism); in this view, therefore, the use of <ss> would reflect Germanic interference in the Latin spoken by the Batavians at Vindolanda. But this is unlikely for several reasons. Firstly, given that (almost) all examples of <ss> are etymologically correct, Occam’s razor would lead us to prefer old-fashioned spelling as an explanation; secondly, spellings with double <ss> are found in other corpora where Germanic influence is not to be suspected (albeit mostly at lower rates); thirdly, other old-fashioned features, such as use of <xs> (Chapter 14) and <uo> for /wu/ (Chapter 8) are also found in documents other than letters; fourthly, it is implausible that the highly educated Cerialis, who otherwise spells in a completely standard manner and uses other old-fashioned features (<uo>), should have used a non-standard spelling solely in the use of <ss>; fifthly, at least three of the documents containing <ss> originate from civilian authors, who were therefore probably not Germanic speakers; these may of course have been written by military scribes but they might well not have been. The argument also rests on the implicit assumption that old-fashioned spelling is a variable that differs according to the register of text in which it is found. This may, but need not, be true, and requires demonstration rather than being a premise.

In the tablets of the Sulpicii, apart from in the sections written by C. Novius Eunus, which I consider separately below, spellings with <ss> are outnumbered by those with <s>: there are 4 instances, all of caussa, and 12 of <s> (25%); however, two of the instances of <ss> belong to a single writer, Lucius Faenius Eumenes, and another is found in the scribal portion of the same tablet (one wonders if the scribe, who also uses <s> in causá, could have been influenced by the spelling of Faenius). The clustering of examples of <ss> in causa and not in other lexemes seems to fit with the usage of the epigraphic evidence as a whole (see Table 26).

Eunus shows a consistent double writing of intervocalic /s/, regardless of whether it results from original /ss/ or not. Once again, this will be an overgeneralisation of the rule that <ss> is to be written for /s/ in many words after a diphthong or long vowel to apply to all instances of /s/ (Reference AdamsAdams 1990: 239–40; Reference Seidl and RosénSeidl 1996: 107–8).Footnote ⁶ Thus, in addition to promissi (TPSulp. 68), where <ss> is etymologically correct, he consistently spells the name Caesar with <ss> (51; 52, 3 times; 67, twice; 68, 3 times), generally does so for the name Hesychus (51, twice, 52, twice, 68 once, but twice with <s>), and also uses double <ss> in writing Asinius (67) and positus (51; 52, twice).

In the curse tablets, all instances of etymological <ss> are spelt with single <s> (7 examples, 3 of amisit,Footnote ⁷ 4 of causaFootnote ⁸), but Britain, and in particular Uley, provides a large number of instances of non-etymological <ss>, particularly in the word nisi (see Table 27).Footnote ⁹ Should we explain double <ss> in nissi/nessi as the result of failure to learn (or teach) the rule whereby some words with /s/ are written with <ss> due to degemination after a long vowel, as with Eunus? Or should we posit some other local development, whether that be an educational tradition or influence on pronunciation from a second language (presumably Celtic)?

The former seems more likely: it may seem remarkable that (mis)use of <ss> should cluster around this word in particular, but its frequency is probably just the result of the formulaic nature of the curse tablets: in the curse tablets from Britain it is common for the curse to threaten a thief with unpleasant punishments unless (nisi) the property is returned either to the owner (thus Kropp 3.22/2, Kropp 3.22/29) or to a temple (Kropp 3.2/24, Kropp 3.18/1, Kropp 3.22/3, Kropp 3.22/5). An alternative formula is that the thief is given as a gift to the god, and ‘may not redeem this gift except (nisi) with his own blood’ (3.2/79, 3.22/32). And in most of the tablets there are no other examples of single /s/, so we cannot say that it is only nisi which receives this treatment, while in 3.22/4, the only example is missericordia for misericordia ‘pity’, which is also spelt with a geminate.

However, there are two cases where /s/ is spelt singly in tablets which also have <ss> after a short vowel; in 3.22/3 there is also amisit, which has <ss> etymologically, and in 3.22./34 there is thesaurus, which does not have etymological <ss>, but which might be expected to be spelt with <ss> if the writer had generalised the rule that all instances of /s/ were to be spelt <ss>. But it is also possible that the writers of these tablets were simply inconsistent in their spelling.

The use of <ss> correlates with <xs> in 3.2/24 (paxsam ‘tunic’, but [3]xe[3]), 3.33/3 (exsigat ‘may (s)he hound’ twice, but laxetur); in both the spelling is not far from the standard, although the former has Minerue for Mineruae and the latter lintia for lintea. Most of the tablets have some substandard features in addition to <ss> after a short vowel:Footnote ¹⁰ Minerue for Mineruae, serus for seruus, redemat for redimat, nessi for nisi (3.2/79), [di]mediam for dīmidiam, nessi for nisi (3.18/1), coscientiam for cōnscientiam (3.22/5),Footnote ¹¹ tuui for tuī, praecibus for precibus, pareat for pariat (3.22/29), redemere for redimere (3.22/32).

In the London tablets (see Table 28), <ss> shows a remarkably high distribution, including in tablets relatively late in the first century AD; WT 56 includes two spellings with <ss> (promissit, ussurae) and one with <s> (causae). In addition there is mistaken use of <ss>, in messibus (WT 29) for mēnsibus ‘months’, which would have been pronounced [mɛ̃ːsibus] and hence appeared to be a case of single /s/ after a long vowel, where only one <s> is found in the 4 other instances of the same word in this tablet. The word ceruesa ‘beer’ is generally supposed to have been borrowed from Gaulish, and there is no evidence that it ever contained double /ss/. Four other instances in this tablet are spelt with single <s>. If the reading is correct, this would be an example of use of <ss> for /s/ after a short vowel. Once again, this is a corpus which has high frequency of the spelling <xs>.

In the tablets from Herculaneum, geminate <ss> is only found in the name Nassius (TH² A3, D13, A16, 4),Footnote ¹² where the spelling change may have been retarded in a name (cf. causam TH² 89, proṃisi A10, repromisisse 4, all 60s AD). In the letters, the only possible instances of <ss> being used after a long vowel or diphthong is bessem ‘two thirds (of an as)’ in CEL (72) in a papyrus letter of AD 48–49 from Egypt. I think the preceding vowel was probably long, but cannot be certain.Footnote ¹³ Otherwise, 31 other instances show <s>.Footnote ¹⁴

In the Isola Sacra inscriptions there are instances of causa (IS 57), laesit (IS 10), manumiserit (IS 320), permisit (IS 142 and 179) and, with <ss>, the word crissasse (IS 46) for crīsāsse ‘(of a woman) to move the haunches as in coitus’ (in a graffito written on a tomb; not earlier than the reign of Antoninus). The word is otherwise found with a single <s> at AE 2005.633 (second half of the second or early third century AD) and Reference Solin and KarivieriSolin (2020, no. 24a). An original geminate is implied by the absence of rhotacism, and is found in Martial and in the grammarians (TLL 1206, s.v. crīsō).Footnote ¹⁵

At Bu Njem there is no sign of the <ss> spelling, but 17 examples of original /ss/ with <s>. At Dura Europos there are 5 examples of (a)misit; there are no examples of a double spelling for an old geminate.

<ll> and <l>

Double /ll/ was degeminated after a diphthong, as in paulus < paullus ‘little’, caelum ‘sky’ < *kai̯d-(s)lo- (Reference Weiss, Jamison, Melchert and VineWeiss 2012: 161–70) and between [iː] and [i], as in uīlicus ‘estate overseer’ beside uīlla ‘estate’, mīlle ‘thousand’ beside mīlia ‘thousands’ (Reference MeiserMeiser 1998: 125).Footnote ¹⁶ The latter change had taken place by the second half of the first century BC.Footnote ¹⁷ The standard spelling for mīlia and mīlibus retained the double <ll> until late in the first century AD. Not including the TPSulp. tablets, I find 27 inscriptions containing these spellings dated to between AD 14 and 100 (many of which would be characterised as official),Footnote ¹⁸ and only 11 in this period with the spelling milia, miliarius, milibus.Footnote ¹⁹

The only reference to the geminate spelling in this context in the writers on language which I have found is by Terentius Scaurus, who actually recommends the double spelling:

The corpus with the greatest number of relevant forms is the tablets of the Sulpicii, all in the word mīlia, mīlibus ‘thousands’.Footnote ²⁰ By comparison to the use of <ss>, where <s> is favoured by both scribes and writers other than Eumenes and Eunus, <ll> appears to be the standard for milia and milibus in the tablets, in agreement with the rest of the epigraphic evidence.Footnote ²¹ In Table 29 there are 30 instances of these words being spelt with <ll>, by both scribes and others; none of the 11 instances of spelling with <l> are by scribes; 9 of them are by C. Novius Eunus, whose spelling is highly substandard (see p. 262).