The Sabellic accusative plural endingsand the outcome of the Indo-European sibilants in Italic

The Greek alphabet reflects no contrast in codas any longer, since the phoneme distribution was by that time contingent on the voice specification of the following consonant, as in aiCvim `bronze-' (Lu 5, Rossano di V.) vs. eoxouS `let him be' (Lu 62, Roccagloriosa), the loanword KFaiaxoQ (Lu 6, 7, Rossano di V.), and we no longer find the double spelling <oo> in any case.

The contrast between a voiced and a voiceless sibilant emerges in initial position only after *di- has been fricativised, as transpires from one single case in the Greek alphabet, the dative C^Fpi `to Jove' (Lu 35, Rossano di V., around 200 BC). By contrast, this name is rendered diuvei (Sa 1, Agnone) or iuvei (Cm 9, Cumae) in the national alphabet. As regards phonetics, all the synchronic contrasts are now likely to translate into [z] vs. [s] in onsets, that is to say in initial position and word-medially. This is suggestive of a reorganisation of the phonemic contrast, which has been abandoned in coda position but has arisen in word-initial position, where it was previously non-existent. I am agnostic as to whether this is a dialectal (cf. Zair 2016: 110-112) or a chronological problem, or both Concerning the date of the adoption of writing rather than that of the individual inscriptions, which in the case of the national alphabet can be situated in the second half of the 5^th C. BC at the latest..

f) We find exactly the same distinction in the Latin alphabet, but it is much more systematically spelt: cf. ZICOLOM vs. SVAE, CENSAZET vs. ESVF (Lu 1, Bantia), though in the Oscan dialects it is ignored in favour of a cover-symbol <S>, as in Paelignian and Marrucinian. There is no trace of the double spelling anymore, in spite of the same document containing <ll>, <dd>, and in codas the contrast has been given up wholesale, as in minstreis(cf. L. minister),meddis, etc. It cannot be put down to chance that both the Greek and the Latin alphabet have chosen these letters for the same contrast This is timidly conceded by Stuart-Smith (2004: 91, fn. 111): «This convention may have arisen because [z] from /-s-/ became identified with /z/ (< *dy-), which was represented with <Z>». Sometimes, this is described as the intervocalic allophone [z] of /s/ (cf. Zair 2016: 105-109; 129-130), which is contradictory, given the obvious contrast in initial position.. A subphonemic spelling system would be comparatively anomalous, but in this case we would find no fewer than two; this is hardly conceivable when it comes to the use of the Latin alphabet, in which <Z> played a very minor role. It follows from the abovesaid that the Greek and Latin alphabets are noting a phonemic contrast that is no longer one of tenseness, but of voice.

g) All this has some bearing on the problem of why Oscan borrowings from Greek or Latin show a geminate in unexpected slots, specifically in the sequence <sst>: cf. kvaisstur (Po 3, 4, Pompei), perisstyl[eis (Cm 1, Abella), passtata (Po 5, Pompei), as opposed to the well- integrated Latin loanword tristamentud (Po 3, Pompei), etc. Zair (2016: 163-164) notes that «the spelling of this sequence as <sst> is restricted to loan-words from Latin and Greek», and that «there is evidence of gemination (or ambisyllabicity) of /s/ before /t/ in both Greek and Latin» (fn. 27); he further observes that «we do not find <oo> in these environments in the Greek inscriptions: KpaioxoQ (Potentia 9/Lu 6, Potentia 10/Lu 7) [...] This could be a difference in the orthography of double letters between the Oscan and Greek alphabets ([sst] in borrowed words treated as any other geminate in the Greek alphabet, while being nearly always written in the Oscan alphabet)».

In my view, the use of <o> in borrowings attested in Oscan in the Greek alphabet is simply reflecting neutralisation of sibilants in coda position. By contrast, <ss> in the Oscan alphabet is not noting a geminate proper, but a tense sibilant that was still phonemic in that context at that time and place. <sst> cannot have been restricted to loanwords, even if no vernacular sequence [s:t] is attested ([s:k] in esskazsium is strictly equivalent, however). In fact, speakers of languages possessing the phonemic contrast /s/ -- /s:/ in intervocalic position, like Greek or Latin, tend to maximise it not only by lengthening the articulation of /s:/, but also by shortening /s/ as much as possible. In coda position, however, where it is usually non-contrastive, /s/ tends to have an allophonically longer duration, and preconsonantal resistance to aspiration is then greater.

This may be the ultimate reason for the sporadic writing <oox>, <ook>, etc. in Greek epigraphy: it is not likely to reflect the ambisyllabicity of /s/, as occasionally suggested, but is more plausibly related to the longer phonetic duration of /s/ in codas, which made it phonemically ambiguous for the speakers See Mendez Dosuna (1996) for these typological considerations and their application to the Greek case..

Conversely, as observed above, Oscan had a weak sibilant /s/ which came to stand in contrast with /s:/, the product of the simplification of some preconsonantal clusters, both in intervocalic and in coda position.

As a consequence, Latin or Greek sibilants, which were usually realised as tense in codas, were identified by the Oscans with their own tense phoneme, and systematically rendered <ss> in the national alphabet. In Oscan in the Greek alphabet, as we have seen, the contrast was often expressed by <Z> vs. <o> but had been neutralised in codas, where their use is contingent upon the voice specification of the following segment.

As a consequence, we do not a priori expect to find <oox>, <ook>, etc., whether in patrimonial forms or in loanwords from Greek (unless, of course, one could prove that blind imitation of Greek written models exhibiting this particular anomaly was at play).

5. The Indo-European sequences *-ns-, *-ns# and *-nt^st- in Italic

The changes associated to the unstable cluster /ns/ in Italic are far from well known. Contrary to common belief, the phonotactic side is not irrelevant, and it is necessary to distinguish carefully between the prevocalic and preconsonantal outcome.

The IE sequence *-Vn.sV- may have passed to -VsV- in all the dialects, possibly at an early date. In fact, loss of the heterosyllabic nasal segment in this context is a widespread phenomenon in the Indo-European family: it is a very early change in Celtic, and occurred in several Germanic branches (specifically Anglo-Frisian and Scandinavian) and Balto-Slavic. The poverty of reliable information coming from Sabellic stops us from venturing any far-reaching assumptions. While some scholars (notably Buck 1904: 72) have proposed an early Sabellic evolution -Vns- >-Vn^ts- on the strength of forms like U. menzne, uze, O. keenzstur, kenzsur, these are unreliable pieces of evidence: the whole lexical family of the Oscan form is now considered to be a borrowing from L. censor (WOU: 382-386). On the Umbrian forms see below 5.4. Since the forms containing this inherited sequence are comparatively few, it may have been the case that the resulting intervocalic sibilant in -VsV- was phonemicised as a tense/voiceless sibilant, which possibly merged in Sabellic with the one going back to geminates and clusters. In other words, loss of the nasal segment, however early, necessarily followed Proto-Italic voicing of intervocalic /s/, and consequently stands in a counterfeeding relation to it. IE /s/ never became voiced by the action of a preceding nasal, and the cross-linguistic generalisation obtains that fricatives, as opposed to stops, usually fail to undergo voicing after nasals According to one theory (Vaux 1998), voiceless fricatives are specified as [+spread] and voiced fricatives are specified as [-spread]. Since the feature combination [+nasal][+spread] is not allowed, nasals cannot spread voicing to a fricative.. Let us now examine our extant evidence:

U. isegeles (abl. pl. fem., Um 1 IV 7, held to stand for correctUsegetes) is said to go back to a past part. *en-sek-eto- `cut' with loss of /n/ and lengthening of /e/ across morpheme boundaries. It is immediately preceded by asegetes (abl. pl. fem., Um 1 IV 7), and earlier by asegeta (abl. sg. fem., Um 1 Ila 29), in turn possibly from a very similar compound *an-sek-eto- `not yet cut'. This could point to early loss of the identity of the preverbs, which would indicate that -VnsV- generally gave -VsV- in Umbrian. Couples like snata vs. asnata (Um 1 IIa 19), which are equally held to contain the same privative prefix an- but systematically omit the nasal in writing, are also likely to have undergone early nasal loss and probably lengthening even when the sibilant preceded another consonant, at least across a morpheme boundary (see below 5.4 for Sa- bellic -n#s-> -n^ts-). We have no information about the evolution of *-VnsV- in Oscan Poccetti (2016: 357) argues that -ns- is generally preserved in Oscan, but none of his examples reflects ety-mological -ns-: patensins is the product of samprasarana, censaumis a loanword, and ^evaep/fensernum on coins (nCm 6a, b) goes back either to a *-tV- derivative of *g^uhen-d^h(h₁)- `attack, kill' or to *d<sup>h</sup>ehines- `yield', as in L. fлnus `interest'..

The arguably early date of this phenomenon as a whole is seemingly belied by the regularity of the Latin notation of /n/ in its original loci. The first attestation of forms without the nasal is the epitaph of Scipio Barbatus' son (CIL I²: 8, 9; 3^rd C. BC, Rome), where we find cesor and COSOL (borrowed in Oscan as kusul, Po 54, Pompei, cf. Poccetti 2006). Relative chronology is of little help here. In point of fact, a number of examples have been invoked in favour of the idea that the change deleting nasals before fricatives was not early, since it was preceded by vowel reduction in closed syllables and raising of -enK-, in turn later than vowel reduction (see Weiss 2009: 120). But the validity of the first obstacle is contingent upon the chosen etymology (if anhelus is from *anaslo-, not *an-anslo-, the problem vanishes) As for the second, Weiss (fn. 19) invokes *k^uenk^u-noi>*k^uinx-noi>quпnп`five at a time'. But, beside the fact that these two changes are in principle independent, vowel raising between two velar sounds must be quite early since the best term of comparison is *kenk-to->L. cпnctus, U. sihitu. I would then start from *k^ipx-to-/no->*k^п(x)tos>>quпntus, *hп(x)noi>quпnп vs. a cardinal *k^uoqk^ue..

Let us suppose that /n/ had been effaced in this context before the creation of the Latin alphabet, but later than the Proto-Italic voicing *-VsV- > -VzV-. As regards etymological *-Vn.sV-, it would be feasible to recover the lost nasal segment (even if the preceding vowel had already been denasalised, which is debatable), from synchronic morphophonemic alternations, id est from sequences going back to *-nt^st-, which may have been simplified and lost the nasal in Latin some time later, but possibly before or around the time at which the alphabetic writing was adopted. In these sequences, the existence of an underlying nasal was recoverable in view of such morphophonemic alternations as fendo - fensus, in which according to this hypothesis the nasal had been effaced in the past participle. In imitation of these cases, which were comparatively frequent, such forms containing the inherited heterosyllabic sequence *-n.s- as ansa [a:sa:] `grip', ensis [e:sis] `sword' could be rightly interpreted as underlying /ansa:/, /ensis/ and were accorded a distinctive spelling which prevented confusion and promoted orthographic uniformity, and vowel length in a sequence -Vs- was correctly parsed as derived from the underlying nasal (it should be clear by now that, at any rate, the upper or cultivated classes might have preserved vowel nasalisation longer). By that time, intervocalic sibilants had been voiced and progressively weakened into an approximant, as in /a:sa:/ [a:za:] `altar' >ara (where the underlying /s/ could be positively recovered from alternations like *flos, floz-es) To judge from cases like L. censeo<*kns-eh₁-, this change presupposes [n] > [*n], which in turn must have been quite early. One could, for instance, object that a two-stage evolution [n] > [an] > [*n] is more plausible and brings Latin into line with Venetic and Sabellic. But then, forms like grandis, scando or sancio must have missed the last shift for some reason. At any rate, forms like inscensio are likely to be the product of an analogical proportion on the present indicative, in this case inscendo (vis a vis scando; scansus) and not the direct outcome of *-skansso- through reduction [a] > [e] followed by loss of /n/..

In short, Latin might have used morphophonemic writing from the beginning, a well- known means of keeping the visual uniformity of the paradigm. Once the spelling of the main allomorph, in this case the one synchronically preserving the nasal, as in fendo,condo,was chosen and assigned to the morphologically derived forms like the past participles and the action nouns, the spelling <NS> naturally spread to the rendition of forms that showed identical phonetics but lacked the requisite inner-paradigmatic alternations. The «correct» usage in forms which were not individually subject to these alternations fell out from this practice straightforwardly: in fact they were a minority, and they could often be parsed as past participles, as in ansa, densus,or nominal derivatives, as in mensis,as well as the origonyms in -ensis,which may have contained *-t^st-.The erstwhile existence of a nasal could be additionally retrieved in compounds like consol, consulere, which would definitely have favoured morphological spelling. On the other hand, the overgeneralisation of the pattern to forms without alternations of either origin would have had a distortion effect if there had been more than one source for these forms. In other words, if there had been any other IE sequence giving -Vs-in nonalternating contexts, hypercorrection leading to insertion of <n> would probably have occurred. This implies that degemination in the context -Vss-, as in casus, causa, took place posterior to the emergence of Latin orthographic conventions, and that at that time there were contrasting pairs like /ansa:/ `handgrip', /a:sa:/ `altar', /a:s:a:/ `burnt'.

Later on, the distinctions were largely obscured by rhotacism, geminates were simplified after long vowels, and for most speakers, especially when they were illiterate, ansa and asa `burnt' were simply homonymous, and there was no reason to accord them a different phonemic status anymore. Consequently, spellings like formonsvs constitute the late,hypercorrect side effect of this merger. Only the literate, educated individuals persisted in the distinction and even made a point of pronouncing the nasal as a token of high status and distinction. More crudely put, the pronunciation of a nasal in this context is entirely artificial and based on the psychological tendency to establish a biunivocal one-to-one correspondence between alphabetic signs and phonemes that makes Latin orthography so easy and English orthography so tricky to learn. Classical testimonies about the alternation of the realisations [V:s], [V:ns] or [Vns] simply reflect the perplexity of the upper literate classes, who had nothing close to the modern notion of diachrony, but suspected that what was in fact only spelling pronunciation must have been more «correct» or elevated than everyday usage. Specifically, the occasionally reported realisation [V:ns] introduces the effect of nasal loss without nasal loss, because it unduly conflates written and spoken word, and testifies to the distance between spelling and phonetics When Cicero reports in his Orator a pronunciation in-doctus but m-sanus he is accordingly testifying to the morphological restoration of the nasal in compounds, not to its preservation, since he fails to eliminate the vowel length resulting from nasal effacement. That he (or the high class as a whole) opted for partial restoration of the prefix (undoubtedly aided by the ironclad stability of the Roman orthography) is immaterial to the fact that he was said to pronounce foresia, hortesia, Megalesia by Velius Longus (incidentally, -ensis is probably the best example of a non-lexical intramorphemic cluster [V:s], and the only one in which omission of <N> was by that time widely tolerated in writing). The isolated Latin sequences MENSSES (Rome), COMITATENSSI (Picenum, 4^th C. AD) are most unlikely to be rendering a non-existent phonetic sequence [ns:]. they reflect a compromise solution between or-thography and the usual hesitation [V:C]/[VCC], usually termed «Iuppiter-» or «famma»-rule, which has given rise to the uncommon spelling MESSES (Transpadana, Britannia, Proconsularis). Poccetti (2006) believes all the variants to be somehow phonetic, which I deem very unlikely. See Adams (2013: 178-182) for the opinions of the ancient grammarians on this matter..

If, as I contend, the designers of the earliest Latin writing system felt it was convenient to introduce nasals for the sake of homogeneity and economy in places in which they were being progressively given up or no longer actually realised by at least lower social layers, their decision amounted to deduction. In spite of much of current scholarship, it is idle to invoke the authority of Greek or Oscan borrowings: they are often taken from the written language, and are accordingly as likely to be diagnostic of the preservation of the nasal segment at that time as contemporary Sp. defensa, consul or infierno to bear witness to the Latin pronunciation of, possibly, any period after the writing tradition set in In spite of this, most educated native speakers would swear these forms are patrimonial and have always contained [n], and handbooks often misleadingly speak of «preservation» of the nasal in learned words.. The fact that the effects of nasal loss and the ensuing compensatory lengthening surface in the Romance languages transparenly cannot mislead us into thinking that Late Latin is the only conceivable terminus ante quem, and that, as universally assumed, this must be a comparatively late phenomenon.

Only in absolute final position (or, more generally, in complex codas) was the original nasal segment impossible to restore in Latin, as in the acc. pl. ending *-ons > -os, since the outcome of -Vs was identical. In these endings, the resulting form eventually merged with nasalless forms, e. g.,the fem. sg. -as, in contrast to such proclitics and prefixes as trans, which often appear preceding a vowel, and in such cases of final vowel syncope as mons, montis or present participles in -ens, where it was also restored in writing for morphophonemic reasons.

To recap, the shift *-Vn.sV- > -V.sV- is likely to be shared by all the Italic languages, even if it may have been an independent process taking place at different dates and language stages, and we can at most vaguely speak of a Common-Italic tendency. The change may have been gradual and socially stratified, and the date(s) in which nasal effacement was completed cannot be ascertained, but in any event it may have preceded the period of our earliest documents.

*-ns(-), *-ns(-) in coda position including auslaut

The sequence -Vns.C- becomes -VssC- in Oscan, -VfC- in South-Picene and probably Umbrian, and -V/VsC- in Latin and probably Venetic (note that in the extant examples the cluster is additionally intramorphemic).

L. monstruum and SP. mufqlum `monument' are habitually traced to *mones-tr/lo- (Marinetti 1985: 118, 127; cf. WOU: 480), but Vine's alternative reconstruction *mons-tlo- (1993: 127-130) is more compelling (since, to begin with, post-syncope -ns- would not be certain to yield -f- in South-Picene). The erstwhile existence of a nasal segment could be deduced in monstrum from the verb moneo, but was ignored in the less transparent mostellaria, as in menstruum vs. trimestris. But the regular, inherited forms could have been *mostrom and *mestruom. In that case, mostellaria and semestris would not only have regularly lost the nasal segment, but would additionally have contained short vowels all along.

As for SP. mufqlum, the first vowel is reflective of raising of /o/ before the nasal, and the vowel may consequently never have been long <um> and <om> then reflect the respective outcomes of IE *-om# and *-om#. According to Weiss (1998), the former has been raised to -om# (-<um>) before the latter was shortened and lowered into -om# (-<om>).. Interestingly, if we assume a homogeneous behaviour of nasal + fricative clusters in codas, this is suggestive of compensatory lengthening in the Proto-Italic sequences *-nx.n- and *-nx.t- having taken place only after the progressively lenited (post-)velar sound had become debuccalised into /h/ This is fully compatible with the idea that this segment has undergone j-prevocalisation in Venetic, see Prosper (2018b)..

It follows that <u>> stands for /o/ in O. kwc(t)it (?3^rd pers. pl., Lu 5, Rossano di V.), to judge from examples like nwpnovic, (in the same document), or the acc. sg. ending -u>q. The nasal segment was trivially omitted in writing but probably never lost in a transparent compound *kom-st-. Consequently, eestint (3^rd pers. pl., Sa 1, B1, Agnone), if from *en-st-, would be the only case of lengthening in codas after nasal loss. In fact, as often remarked, it is more likely to continue the Sabellic prefix e- from *ek- attested as O. eh-, U. ehe- (Meiser 1986: 167) and eestint can then be translated as `exist', or possibly `outstand, be remarkable'.

Conversely, O. esskazsium `disembarkation (point)' (Cm 2, Campania), which WOU: 235, in the footsteps of Mario Russo, compares to L. escensio and traces back to *esskansseom (<*eks- skand-tei-o-), is much more likely to go back to *en-skand-tei-o-, a full match of L. mscensio `embarking'. This noun may have had a specialised meaning `embarkment place'; alternatively, it simply designated the staircase access that ran upwards from the sea to the temple of Minerva. As observed above, an ablative prefix would be expected to give *e-, and the attested form would have been feeskazsium.

The word-final sequence *-ns#, attested only in the accusative plural endings, shows the same outcomes in Sabellic as in coda position, and a preceding short vowel regularly undergoes syncope in athematic inflections

In the consonant stems, *-ns# is attested in O. usurs `?', fakinss `actions', aginss `actions, rituals', U. capif `offering tray' (<*kapid-ns), frif `fruits' (<*frug-ns), U., SP. nerf `(noble)men' (<*ner-ns, respectively Um 1 Via and SP TE 6, Penna Sta. Andrea).

In the -i-stems, the acc. pl. *-i-ns# surfaces in Umbrian as avif, aveif, avef, which reflects -ef(without compensatory lengthening), and not -if (from *-f or *-zs). Restoration of the stem vowel -i- may have been favoured by the undesirability of an outcome *auf>*of, however. In SEVACNE(f) `sollemn', vowel loss would have resulted in samprasarana, creating a skewed paradigm. In turn, TREiF/tref `three', straightforwardly renders /tref/ (from *trins) and may have partly served as the model for the other forms (and one does not see why it should have been refashioned from earlier, allegedly regular *trif or *fns).

There is additionally no reason to believe that U. manf `hands' is not a -u- stem like the abl. sg. mani or the loc. sg. manuve, and in that case it must have undergone regular syncope without subsequent restoration of the stem vowel, since, in contrast to the above -i-stem examples, it caused no real disturbance to either the uniformity or transparency of the para- digm In spite of virtually all previous scholarship, summarised by WOU 450. Klingenschmitt (1992) simply ig-nores the obvious solution when he states that this acc. pl. has been remodeled on that of *ped-f `feet', for which there is no reason whatsoever. For the stem *manu- in Oscan manim and a new reading pavo[up] of Lu 62, B7, Roccagloriosa, cf. now Zair (2016: 207). On the origins of *manu-, cf. Neri (2012: 185, fn. 6)..

The Oscan thematic acc. pl. forms feihuss and bravus[s] (Cm 1, Abella) are hardly diagnostic, since for probably dialectal reasons peculiar to Samnium and some parts of Campania, these forms show a short vowel /o/ in line with the rest of the endings in which a long vowel would be expected, like abl. sg. -ud, nom. pl. -us. By contrast, Umbrian shows <u>/<v>, which points to /o:/> /u:/ (turuf/TORV). Consequently, the synchronic endings are O. -oss/-uss, U. -uf. In my present view, these forms have reinserted the thematic vowel by analogy with the nom. pl. *-os after final vowel syncope, since all the plural forms contained a thematic vowel -o/o-.For all we know, this may have happened independently in the two branches of Sabellic.

If the feminine correlate of *-ons was Proto-Italic *-as, and not *-ans, with Indo-European loss of the nasal, the attested Sabellic endings, O. -ass (viass, ekass), U. -af (vitlaf, eaf), must have come into being by analogy with the masculine. By contrast, the South-Picene acc. pl. fem. ending -as in qoras `stone monument (vel sim.)' (Sp TE 7, Penna Sta. Andrea), otherwise identical to O. kurass (Sa 10, Pietrabbondante) may be the only Sabellic ending to continue the inherited *-as unchanged.

Latin shows -Vs in all the inflections: cf. L. eas, deos, patres. Venetic is ambiguous: our best example thus far is te.r.mon.io.s de.i.vo.s. `deos terminales' (LV: 125, Vicenza). But one early example reads dona/.s.to .a..i.su.s `donavit deos' (LV: 243, Gurina), where the acc. pl. still shows the symbol <s> for the tense sibilant that later fell into disuse, probably because of the progressive weakening and eventual effacement of the lax sibilant in a number of contexts. This leads me to suspect that Venetic may have teamed up with Oscan in this respect, and that *-ns# > *-n^ts# may have been an areal feature covering Sabellic and Venetic (cf. Prosper 2018a for the similar northward expansion of other innovations).

To recap, Proto-Sabellic *-ens, *-ins and *-uns regularly lost their stem vowel by syncope somewhere down the line. The undesired grammatical consequences of this change were counteracted by partial restoration of the stem vowel where needed. Contrary to one widespread idea, it is illogical to take syncopation to have exclusively happened in the consonant stems, in which *-ens goes back to *-ns. If the nasal remained vocalic long enough, it could even have escaped syncope This would make the actual outcome ipso facto analogical, a path tentatively followed by Rix (1986: 584-585) which I find rather speculative, and that he later on (588) replaced by an allegedly regular, Ursabellisches end-ing -(e)f.. There is also every reason to cast doubts on the assumption that compensatory lengthening must have taken place, and that the allegedly Proto-Italic thematic ending -off-os is regular and the rest of the stems, failing to show the expected outcome -ef, -If and -uf, are the product of analogical processes.

A phonetic explanation

The divergent outcomes of -ns in coda position have never been properly accounted for beyond mere description. This has always included a number of versions of an unwarranted shift -s > -9, which for some reason only took place when a nasal or rhotic immediately preceded the sibilant. In point of fact, in the world's languages it is usually the nasal that adapts its articulation to the following obstruent. This is the norm with stops but not with fricatives, however, the resulting clusters often remaining non-homorganic (see Repetti 2002). Accordingly, it is unlikely that the whole process was set in motion only because the apico-alveolar sibilant had spontaneously advanced its place of articulation to the point of becoming [0], and exclusively in this particular context.

In my view, the first stage, common to all the Sabellic languages, can be described as blocking of the difficult nasal-fricative transition by an epenthetic homorganic stop, a phenomenon attested in Hittite, Tocharian, Basque, a sizable number of Italian dialects, a number of Modern Greek dialects, optional realisations of British English words, like mince [min(t)s], and most American English dialects. It is due to mistiming in the coordination of several articulators: see Ohala (1997), Warner et al. (2001). The velum is raised before the oral cavity is opened and the air is released. This suggests, firstly, that the Sabellic outcome of -Vns# reflects the preconsonantal sandhi variant, and that it has undergone excrescence of a subphonemic homorganic dentoalveolar stop and became -n^ts³⁷. It should be noted that, while the excrescence of a homorganic stop to block the nasal-fricative transition generates highly marked codas, some dialects, notably American English, only show this phenomenon in coda position (as in false vs. falsity). Thereupon, the nasal segment was lost, probably leaving a trace in the nasalisation of the preceding vowel, and -ts became phonemic. This was possible because there were no morphophonemic alternations that allowed the speaker to recover the underlying -ns, and perhaps also because the epenthetic stop was no longer linked to the presence of a nasal in the speaker's conscience, though this is not strictly necessary: as we are going to see, the intervocalic sequence -ns- no longer existed and -ntss- had an underlying stop and evolved in Sabellic in exactly the same direction as -n^ts³⁸.

Thereupon, -ts was weakened to a fricative. In turn, this was probably favoured by the comparatively high tendency of coronal sounds to assimilate to following sounds. In Oscan, the resulting sound merged with the tense sibilant (> -ss). In Umbrian and South-Picene, the new sibilant *-(V)s preserved its dentoalveolar articulation, subsequently became an interdental voiceless fricative *-(V)9³⁹, and eventually merged with /f/ by acoustic similarity, since these sounds are easily confusable. In both cases, the difference with original -V/Vs was preserved, and consequently syncretism of the acc. pl. with the nom. pl. never took place. In fact, this is an inescapable conclusion given the similar, earlier change involving an inherited sequence *-nts#, which became *-(n)d# and eventually -f# in all the Sabellic dialects in the nom. sg. of -nt-stems.

The course followed by Umbrian is reminiscent of the change -ti->-ts-> -s- > -9- that took place in Spanish in the 16^th C. AD (cf. martio>margo>marzo; sperantia>esperanga>esperanza), and was probably due to the need of maximising the contrast between sibilants: the dental voiceless /s/, the apicoalveolar voiceless /s/ and the palatoalveolar voiceless ///. By contrast, the Spanish variety spoken in America and vast areas of southern Spain simply abandoned the (cross-linguistically infrequent) phonemic contrast between the inherited /s/ and /s/, which are usually realised as a dental sibilant [s]. The phoneme /// is also known to have existed (at least after syncope eliminated the phonetic conditioning) in Umbrian and South-Picene, whose national alphabets have a special symbol for ///, but did not exist, or not so early, in Oscan. This suggests that Umbrian already had /s/ in its phonemic inventory by the time it became /0/ because the threefold contrast tended to be maximised and eventually merged with /f/, reducing the number of phonemic contrasts. In Oscan, our sibilant was smoothly integrated in the system and simply fed the preexistent contrast /s:/ - /s/.

Consequently, this goes some steps towards explaining the limited distribution of the change [ts] > [s] > [0] > [f] in Italic. It is difficult to say whether there was a last common Sabellic phonetic stage before final vowel syncope (-n^ts never merged with post-syncope -ts anyway) or whether syncope affected two already differentiated sequences, e. g. O. -Vss and U. *-Vs/*-V9.

A different question is whether compensatory lengthening is expected after loss of the nasal segment, and then, which inflections show analogical leveling. In my view, compensatory lengthening is a priori not expected in codas, since the nasality was reassigned to the From now on, the symbol <^l> will be used to note the product of the rule by which a subphonemic [t] is op-tionally inserted, and not a specific phonetic content; accordingly, it will not be used inside square brackets (cf. on this notational problem Akamatsu 2011). While the subphonemic status of the epenthetic stop raises some questions, Akamatsu (2011) has argued that [nt], in which [t] is epenthetic, is an allophone of /n/. By neutralisation of stridency as per Hamp (1972), who introduces this step to account for the Proto-Italic shift (-)sr- >(-)9/Sr->(-)ffpr-. vowel and the nasal segment may have disappeared before the excrescent stop became phonemic and the resulting -ts underwent weakening. As claimed above, the Sabellic evolution, in order to be credible, must be envisioned as a many stage process *-ns >*-n^ts> *-s >*-Q/-ss> -f/-ss, whose second stage would actually have prevented lengthening.

In the last years, a number of approaches have pointed to a perceptual basis for compensatory lengthening. According to Ohala et al. (1995), the margins of vowels preceding fricatives have an acoustic effect that mimics nasalisation. If there is a nasal segment, listeners could regard it as illusory and discount it. Along the same lines, Kavitskaya (2002: 58-60) has further argued that nasal deletion before fricatives triggers lengthening «since nasalized vowels are phonetically longer than oral ones and thus can be reinterpreted as long with loss of the nasal.» She, however, adds that some exceptions shown by a number of Greek dialects (Cyre- nean, Thessalian, Arcadian) which have an acc. pl. form in -os, fail to show lengthening because, in her own words, «word-final syllables stay closed regardless of the n-loss. Thus, it can be argued that, even if nasalized vowels in the last syllable in the word are as long phonetically as nasalized vowels word-medially, they do not necessarily get reanalyzed as long» See also Alonso-Deniz (2011) for nasal loss without lengthening in the Greek sequence *-VnsC-, which he puts down to the fact that the mora count stays stable after nasal loss. This means that, even if the nasal segment had survived until a (Common Sabellic?) stage *-ns- had been reached, it is reasonable to assume that this se-quence would have lost the nasal without lengthening in preconsonantal and final position. Like, for instance, -ns- was artificially transferred from written Latin to spoken Romance, as revealed by doublets like the Spanish abstract defensa vs. the inherited concrete dehesa `pasture' < `enclosure for cattle'. Cf. Meiser (1986: 70, 77), WOU: 463-464, 688-689, etc.. The same loss, without consequences for the length of the vowel, which is only expected if the language already possesses a length contrast in the vowel system, is also reported to be occurring in contemporary Romance languages: see Recasens (2018: 165) for dialectal Catalan tens `you have' >tes, etc.

Indo-European *-nt^st - in Sabellic

The outcome of *-nt^st- merged with the outcome of intervocalic *-ns- in Latin, but yielded Oscan -ntss-, Umbrian -f-, and Venetic -nss-.

In Oscan, this transpires from the context-bound use of <zs> in esskazsium [es:kants:iom] and in kenzsur `censor'. Oscan borrowed the family of kenzsur, keenzstur (redone as *kens-tor-), etc. from the Latin written language and interpreted the cluster at face value: it must have sounded similar to their own agent nouns and past participles built from roots ending up in a dentoalveolar sound. In other words, the Oscan scribes overestimated the phonetic reality of the Latin spelling <NS><sup>41</sup>. In such forms as hurz `garden' (<*g^hortos),<z> is rendering an affricate phoneme /te/. Accordingly, the Oscan tense sibilant /s:/ had a post-nasal variant [ts:] which was spelt <ss> in kenssurineis, but may have been more or less regularly spelt <zs>. As for Umbrian, *-nt^st- yielded -f- with nasal loss: cf. mefa (if identical to L. mensa), possibly spefa, the form that follows it in some passages (in Um 1 VIa/b, VIIa), and spafu (if <*spnd-to-, Va 20)⁴². Compensatory lengthening cannot be ascertained.

All this points to the existence of a common Sabellic stage -Vn.tssV-, which apparently remained unchanged in Oscan, but underwent the expected changes into *-Vs- >*-V/Vd->-Vf- in Umbrian. The reasons for positing an underlying, not excrescent homorganic dentoalveolar sound in this context will become clear at the end of this work. This contrasts with the pan- Italic evolution *-Vt^stV- > -VssV-: cf. U. fiso (divine name, Um 1 VIb, <*b^hid^h-to-), and also *-ts-, *-dz-> -ss-, as in O. nessimass `closest, nearest' (Cp 24, Capua), U. nesimei (Um 1 Via) <*nedz- izVmo-<*nezd-ismHo-.

Scribes may have hesitated as to whether it should be written or not, which explains -<(n)zs> vs. -<nss>-. The (in most cases) post-syncope sequence -ts(-) carrying no morphological information became a new affricate phoneme /te/ in Sabellic. This is the case with Oscan forms like the conjunction puz `that' (Po 39), the preposition az (if <*ads or *atVs, Sa 1, Agnone), or the dative vezkei (<*uetesk-, Sa 1, Agnone). The digraph <zs> accordingly looks like a compromise spelling between<z> and <ss>. But it may have been designed to distinguish /nte:/ from a postsyncope cluster /nte/. That is to say, the former may have become phonemically ambiguous if speakers of higher registers started to parse /te:/ as a distinct tense dentoalveolar affricate phoneme. Lower registers may have tended to merge it with -ss-. The case of kvaizstur (Po 8, Pompei) for expected fkvaisstur may consequently be put down to hypercorrection.

The Venetic appellative a.n.sore.s. (nom. pl., LV: 203, Calalzo) occurs in a fragmentary context. It goes back to an Italic agent noun *ant^stor-, ultimately (as if) from *fen~ + d^hehi-tor- `constructors; officials?' with zero-grade of the root by virtue of the «Italic» rule, according to which agent nouns are derived from synchronic past participles (see Prosper 2018b). The symbol <s> notes a tense sibilant. A much later, indirect instance of the preservation of this sequence is Ven. TATSORIAE in Emona/Smarata (personal name, CIL III: 10722, Pannonia Superior), the full match of L. tonsor (as if <*tond-tor). In this case, however, the excrescent den- taoalveolar stop may be the product of misperception of foreign names by Gaulish scribes who did not have that sequence in their native language, and the actual realisation of the stem may have been either [t3ns:o:r]- or [t3nts:o:r]- (see Prosper 2019b: 52)⁴³.

The post-syncope cluster -n(V)z-

The cluster going back to Italic *-nVz- (< IE *-nVs-) evolved in different directions in Sabellic, too. As observed in 5.1., there is every reason to believe that post-nasal voiced fricatives are cross-linguistically dispreferred.

In Oscan, the weak sibilant, probably realised as [z] when it was still intervocalic, was spelt in the national alphabet as <ns>, and the verb form patensins<*pdt-na-se- `aperirent' (Cm 1b 24-25, Abella) is our best example. As for Oscan in the Greek alphabet, the content of <Z> in pЈCЈL9 (Iltal. III: 1485, Caulonia 2)<sup>44</sup>, pЈvCqi <*uen(H)-es- `Venus' (Lu 31, Rossano di V.), is not quite certain: while Lejeune (1970), followed by Stuart-Smith (2004: 97) believes it to be rendering an affricate, itself the product of the insertion of an excrescent dentoalveolar sound, it is more likely to represent a phoneme /z/⁴⁵. Finally, minstreis (Lu 1, Bantia) goes back to *ministero-. Across compound boundaries, -n#s- was probably realised as [ns:], as in O. Kwa(x)iT (Lu 5, Rossano di V) and perhaps kunsif deivuz<*kon-sent-s demot-s, the divine name corresponding to the del consentes, according to Poccetti (2013b: 35)⁴⁶. Cf. Iltal. II: 1183, Pentri/Terventum 22, c. 150-100 BC⁴⁷. The Celtic outcome of *-Vnt^st- was -Vss-, a sequence preserved in Eastern Gaulish and variously rendered attested in the Venetic record as <s>, <ss> or <s>. For instance, the personal names le.s.sa (LV 208, Cadore), leso (LV 93, Este) are Gaulish derivatives from *splend-tu- `splendor' (in OIr. lйs`light'). Cf. Prosper (2019a: 151). Omission of the nasal in coda position is trivial across languages and writing systems and is not necessarily caused by phonetic weakening and loss (see Mйndez Dosuna 2007). As observed above, by this time, or in the southern region, the contrast had become one of voice, and /z/ was phonemic in every position except probably in auslaut and in codas, where the contrast had been abandoned and the sibilant acquired the voice specification of the following consonant. To my mind, however, deоvыz,in view of its final letter, can hardly be a singular form, but a nom. pl. *demot-es, mutatis mutandis a match of the Latin expression. This leaves us with an insoluble problem that cannot

be addressed here given our lack of evidence: kыnsоfwould have to be traced back to post-syncope *-V(:)nts, either regularly or by analogy with the singular form.

In Umbrian, medial vowel syncope gives rise to an undesired sequence which, again, became stabilised by the intrusion of an epenthetic sound that blocked the nasal-fricative transition, yielding *-n^ts-,which was subsequently phonemicised: cf. uze/ONSE `shoulder' (<*omeso-, Um 1 Ilb 27-28, etc.), the full match of L. umerus<sup>48</sup>. At any rate, we must assume an evolution *-nVz- > *-ns- >-n<sup>t</sup>s->-nts-<sup>49</sup>.The sequence -n#s-across synchronically recognisable morpheme boundaries underwent the same stop excrescence and eventually became -nts-,as in U. anze- riatu, anseriato `let him watch'⁵⁰. This is also the case with sequences containing an underlying dentoalveolar segment, like -nt/d#s-(as in the 3^rd pl. dep. fut. ostensendi, Um 1 Via 20, from *op(i)-stend-s-).Syncope in forms containing an inherited segmental /t/ gave rise to a new Umbrian phoneme /te/ spelt <z>, as in kazi`kettle' (<*katesim,Um 1 III 16, 18), ezariaf`food' (<*edesasiio-,Um 1 IVa 27), pihaz`piatus' (<*pwtos,Um 1 Ib 7). As a consequence, the secondary, postnasal cluster was interpreted as phonemic.

The form menzne (loc. sg., Um 1 Ila) is habitually traced back to *men-s-(e)n-i`moon', an adjectival derivative of *mens- `month' (cf. WOU:471)⁵¹. Accordingly, the assumption that After finishing this work, I have become aware of a very recent one by Fortson and Weiss (2019). They trace kыnsоfto *kons-eie- an iterative present of *kens- in Skr. a-samsaya `give hope', MW. dan-gos- `show', and recon-struct PItal. *konsк-and *kens-e-, later conflated into *kens-к-.The Oscan form would mean `decreeing, approving'. This is a brilliant idea, if an early generalisation of the stem *kons-eie- as in Latin is accepted (the original sequence is *kons-eio-nt-). As they say (fn. 14) <i>has been reintroduced in analogy to forms containing -к-, which cannot be the case with -sents. On the other hand, given the plethora of Latin loanwords of this root, kыnsоf,if its immediate predecessor was /ko:se:/-, could have been influenced by them as regards preservation of <ns> in this lexical field, or, like L. consentes,could have been reanalysed as a compound, especially if the legal term only attested in PRAE- SENTID (Lu 1, Bantia) had already been borrowed. Most recently, Hofler (2018) has reconstructed a preform *h<sub>1</sub>emHso- for this form and L. umerus `shoulder', but the difference is in principle immaterial to the present argument. Even if a cluster [nz] had survived unchanged, epenthesis is unusual between two voiced segments, since continuous voicing causes less air pressure build-up and the stop burst is hardly audible, at least in the sequence [lz] vis а vis [ls] (Ohala 1974: 359; cf. also Akamatsu 2011: 91 on [nz] > [ndz]). The sibilant was probably devoiced after syncope and before stop excrescence. Apparent exceptions, like the Yiddish forms Gandz `goose' and Haldz `throat' may not presuppose a former stage with word-final voicing [ns] > [nz], [ls] > [lz], as often assumed, but epenthesis ['gans] >['gants], ['hals] > ['halts] and subsequent voicing of the affricate (probably not before the dental sound was perceived as phonemic), as opposed to Fentster `window' from ['fenster], in which voicing is blocked. The same explanation may apply to [ndza'lata] `insalata' in Neapolitan, cf. [kan'dzone] `canzone' (Rohlfs 1968: 363). Note that epenthesis is very rare, for instance, in English plurals like ten-s ['tenz], in which -z is not the im-probable product of voicing after a nasal, but the basic inherited morph vis а vis secondary -s and -az. By contrast, it is the usual realisation of American English tense ['tents]. Excrescence is absent in English compounds, e. g. gunshot ['gAnJbt] vs. mansion['mжntf(s)n], because in the first case the speaker is aware of a virtual pause (see Akamatsu 2011: 109). But in the Umbrian case, an- is a recog-nisable but not especially productive bound morpheme; accordingly the boundary is ignored and excrescence oc-curs. Note, however, that excrescence does not occur or is dissimilated in anstiplatu `instipulato' (Um 1 Via). Note that the consistent Umbrian spelling <NS> in the Latin alphabet probably means that the sequence had been fricativised by the time the Tables were transliterated. It is the same in pihaz/PIHOS `piatus' and consequently does not mean that we have two different outcomes in Umbrian and its dialects, one in which the sibilant has been voiced «qui ne peut кtre occasionйe que par la proximitй de [n]» and another with effacement of the nasal and no voicing that would be similar to Latin and supported by MESENE and ASERIATO, pace Poccetti (2006: 34), who be-lieves Etruscan <z> to be employed in Umbrian in this context because it was noting a voiced sound [z]. This is implausible, since the alphabetic usage is essentially phonemic and the allophonic voicing when a nasal precedes is unlikely to be consciously perceived as relevant and reflected in writing. Additionally, as we have seen, an evo-lution [ns] > [nz] is hardly possible.original intervocalic -n.s- yielded U. /nte/ favoured the reconstruction of a single form *mens- for all the Italic languages. In view of the arguments displayed above in 5.1., however, one cannot attach much credence to the idea that primary *-ns- would have undergone preservation of the nasal segment and, in addition, epenthesis. Note, by way of comparison, that the Latin cluster -nsn- is simplified even across (different) morpheme boundaries, as in *trans#sno `I swim across' >trano and *kon#snoud/fi-iio->conubium⁵².