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

The original structure of the word for `month' is much of a locus desperatus of Indo- European reconstruction. It is usually held to have been a holokinetic noun of the genus commune, with a nom. sg. *mehin-os, acc. *mehin-es-m, gen. *mhin-s-es (>>*mehin-s-es) `moon-cycle, month'⁵³, and then an original collective or simply an abstract derived from a -s-stem. Its locative would have been *mehin-es(-i), with early analogical spread of the /e/ grade of the root and regular /e/ grade of the suffix. While the Baltic languages have forms going back to *men-es-, like Lith. menuo `month, moon' gen. menesio, alternative nom. menesis, or Latv. menes(i)s, Proto- Slavic inherited a stem *men-s- (cf. Derksen 2015: 311-312). This points to a primitive paradigmatic alternation resolved by the daughter languages in different directions, and this may well have been the case with Sabellic *menes- vis a vis L. mens-i-. The Latin form could actually go back to the genitive *mehin-s-es, which became the base of a -i-stem (with spread of the /e/ grade of the root)⁵⁴.

I consequently assume that U. menzne has undergone medial vowel syncope and must be traced back to *men-es-n-ei, the dat.-loc. of a possessive adjective *men-es-o(n)⁵⁵. The following alternative derivational paths are conceivable: a) The stem *menes-, extracted from the accusative or locative form of the holokinetic formation, was thematicised into *men-es-o- and then received a nasal possessive suffix. b) A locative in -en, namely *men-es-en, gave rise to an exo- centric nasal stem derivative *men-es-o(n). This is reminiscent of Nussbaum's classic explanation of L. homo as a derivative of a «double full grade» locative *d^hg^hem-en, whether primitive or remodeled (1986: 189-190, 289). c) Finally, the bare stem *menes-, could have constituted the base for the construction of an adjectival derivative *menes-o(n)⁵⁶.

Original *mensnV- would hardly have become the attested Umbrian form, but probably *mefnV-, or possibly *mennV- or *meznV-, which would have contained stable consonant clusters. The traditional view equally falls short of explaining U. anter:menzaru `intermenstruarum; occurring between two moons' (gen. pl. of an -a-stem, Um 1 IIa). In fact, if the habitually reconstructed sequence *mensV- had undergone early stop epenthesis and become *men<sup>t</sup>sV-, nothing would have prevented it from becoming *mefV-, especially since inherited /ns/ would be intramorphemic. In any event, the reconstruction of a preform *menso-, with preservation and fortition of intervocalic *-ns-, is unconvincing, in spite of its close Avestan cognate antaro- Cf. Weiss (2009: 179) for other comparable sequences. See Beekes (1982, who starts from a -t-/s- alternation in the suffix, however), as well as Ruijgh (1967: 237) for Myc. me-no-e-ja /me:no:heia/, an adjective derived from *meh<sub>1</sub>n-os and meaning `decorated with moons'; more recent works in defense of the above holokinetic paradigm are, with slight discrepancies, e. g.MeiЯner (2006: 147-150), Vijьnas (2009: V.9), and Rasmussen (2016: 325). This would mean that Italic exclusively inherited the holokinetic paradigm, not, as often believed, that of OIr. mis, Gk. pqv, Skt. mas, in turn a homogeneous stem *meh₂n-s-, secondarily created from the genitive form. Syncope of the medial vowel in a closed syllable is equatable to that of U. MERSTO, from *med-es-to- (cf. L. modestus). In fact, short vowel syncope in closed syllables which terminate in /s/ is regular in Sabellic and perhaps related to the fact that the resulting, seemingly complex codas must be parsed as coda + extrasyllabic sibi-lant, e. g. -VR.s.CV-. On the extrasyllabicity of Latin sibilants in this position, cf. Cser (2012). Cf. Doric aiйз from *h₂em-es, probably the locative of an equally holokinetic *h₂em-os, as suggested by Pe-ters (1980: 79), and see MeiЯner (2006: 151) for aiei <*h₂ew-es-i. mapha- `between the moons'. It could be identical to the base of L. intermestris, if from *menes- ri-, and then both forms would go back to a hypostatic compound *hinter-+mehines-, in which the preposition governs an endingless locative (or alternatively an accusative, which would have an identical stem) As repeatedly asserted in the literature, we would expect an adjectival form in *-(i)io-, and it has been sug-gested that it is a noun for `Kalends' (cf. Weiss 2010: 28). It could consequently belong to a more archaic layer than the typical parasynthetic pattern found in Latin pomoerium, procastria, proverbium, suburbium, or the productive in- terdigitium, intermundium (see Bader 1962: 283). That is to say, the compound is of Proto-Italic date and belongs to the exocentric type originated in the thematicisation of a prepositional phrase attested in Gk. ev dAi ^ evdAiog `living in the sea', Lyvup `hollow at the back of the knee' (<*en-gnm-a, cf. Forssman 1964: 29), Skt. ati-ratr-a- `last-ing more than one night', adhas-pad-a- `found under the feet', and Av. antara-maph-a-..

As regards Oscan and its dialects, the related Vestinian form mesene (VM 9, mesene flvsare `in the month of Flora') see segenni (2007) for this dialectal attribution. may have gone through the stages *menesnV->*meneznV->*me(n)znV->*me(n)zenV- through voicing, syncope, nasal loss (or possibly just defective spelling) and anaptyxis The new Oscan form minnaris is taken from *mens-n- by Poccetti (2013a: 213), who identifies it with the base of the Umbrian form. In view of the above considerations, his alternative, by which this is a reflection of *mena, possibly identical to L. Mena and Gk. ppvr| `moon' is more compelling (see more recently Poccetti 2016 in defense of the latter possibility). But this is not a usual context for the «famma»-rule, which has led Weiss (forthc.) to iden-tify it as a correlate of L. minuo `diminish'.. Oscan post-syncope -ms- became -m^ps- by epenthesis in the divine name vup^Soi papЈQTxoi (dat., Lu 28, Rossano di V., c. 300 BC), which goes back to *numasiio- According to some scholars, the epithet contains a further suffix *-id-(i)io- (Lejeune 1970). This is refuted by Poccetti (2009: 230) who compares the name Niupo5ir|ig (Me 1-2, Messana) and ingeniously treats this as a case of rendition of /z/ by means of <oS>, but these testimonies are too distant and it would be really remarkable that this spelling norm was exclusively applied to this name..

In sum, Oscan and Umbrian differ as regards the outcome of post-syncope -ns-: while Umbrian consistently blocked the nasal-to-fricative transition by stop epenthesis, giving rise to -n^fs-, the Oscan post-syncope cluster was spelt <ns>, and <vZ> in the Greek alphabet, since /z/ now enjoyed phonemic status.

The so-called «Opic» or «Pre-Samnite» dialect that is often linked with South-Picene in the wake of Helmut Rix, but seems to share a number of traits with its Oscan superstrate, equally shows <ns> and <ms> in the recently uncovered inscription of Niumsis Tanunis (see below 8.). Venetic and Latin have followed different paths, since medial vowels are not regularly syncopated: the Latin voiced sibilant has undergone rhotacism, as in umerus `shoulder'. In at least some Venetic areas, the voiced sibilant [z] was further weakened into an approximant, debuc- calised and eventually effaced, as in the personal name voltigenei (dat. sg., LV: 56, Este, 6<sup>th</sup>-5<sup>th</sup> C. BC) which must be traced back to *-genehei<*-genezei<*-genhi-es-ei `having a x-birth/ lineage' (cf. Prosper 2019a: 8-14).

6. Italic sequences of rhotics and sibilants

The above evolutions are partly paralleled by those of original vis a vis secondary clusters of rhotics and sibilants:

IE *-rs-, possibly with an early voiced realisation of /s/, is preserved in Umbrian, but assimilated in Latin, as transpires from unobjectionable cognates: cf. * torse- `chase' > U. tusetu, tvrsitv (fut. impv.), L. torreo, *torsa> divine name U. turse, tuse,tvrse (dat. sg.), tvrsa (voc.), and *fars-> U. farsiu, L. farreum `made of, or related to, wheat'. The Oscan outcome of Italic *-rs- is -Vr-, as in teer[um, terei, etc. (<*terso-, cf. L. terra; Cm 1, Abella). This probably means that this evolution aligned with that of Latin, and that at an intermediate stage the cluster -rr- was simplified with compensatory lengthening of the preceding vowel before medial vowel syncope.

IE *-rt^st- becomes Proto-Italic *-rtss- (cf. L. vorsus `turned'). This sequence survived into Proto-Sabellic. In Oscan, it evolved into -r(t)ss-, as in the agent noun Fepoopa `Tropaios', epithet of Iuppiter (<*uert^stor-; Lu 25, Vibo Valentia) and in Umbrian into *-rs- >*-rd->-rf- in the adverbial past part. trahuorfi `transverse' (<*-urt^sto-; Um 1 VIIa 25). See further details in the conclusions of this work.

IE *-Vrvs# yields -Vr in two nom. pl. forms going back to *-er-es: U. frater (Um 1 Va, etc.), frateer (Vb 16), O. *neer (recovered from a putative gen. pl. neer[um, cf. Utal. II: 674, Pompei 32). This indicates that final vowel syncope was followed by progressive assimilation, simplification of the geminate sound and compensatory lengthening of the preceding vowel: *-Vrs# > *-Vrr# > -Vr.

IE *-rVs- is unanimously held to show disparate outcomes in Sabellic: in Oscan, after voicing and syncope, *-rz- undergoes assimilation and surfaces as -rr-. The symmetry is destroyed by the alleged Umbrian outcome -rf- (or -rjS-). And yet, none of the examples adduced thus far in favour of the latter evolution is compelling: crucially, Weiss (2017) has recently disposed of the validity of the traditionally accepted equation of U. gerfe/SERFE(Um 1 Ib, VIb, etc.), Pael. CERFVM,⁶¹ O. kerri `Cereri' (dat. sg. <*kerez-ei, Sa 1, Agnone), etc., all of them traditionally traced to *kerVzV- and identified with the divine name Ceres or derivatives thereof. Weiss has favoured a connection of the first two forms with Skr. sardha- `(military) troop', usually applied to the Marutas. This is much more compelling since it furnishes an explanation for the apparent alignment of Paelignian with Umbrian, and spares us from accepting divergent outcomes within the Oscan family.

As a consequence, we are now left with a single instance of the evolution *-rVz->-rf/§-: the birdname U. parfam/PARFA(Um 1 Ib, VIa), purportedly identical to L. parra `small bird', and traditionally traced back to *parVzV-. Parfa and parra probably designate a sparrow or another sort of small bird involved in ritual divination in Umbrian and related to either ill or favourable omen in Latin, where it is attested since Plautus⁶².

To begin with, the acceptance of an Umbrian sound shift *-rVz->-rf- silently relies on the idea that Latin -rr- must inevitably go back to IE *-r(V)s-. What is more, the purported Umbrian evolution runs up against serious difficulties: as we have seen, Umbrian *-nVz- evolved into -n^ts-. In logic, we would expect *-rVz- to yield Umbrian -rs-, or to undergo progressive assimilation to -rr-, as in Latin and Oscan, or even to undergo epenthesis and surface as <rz> /rte/ (see more on this possibility in the conclusions). The acceptance of a sound change *-rVz->-rf- simply begs the question and takes advantage of the untenable assumption that /s/ in a number of contexts may straightforwardly become /0/ (and eventually /f/) in Umbrian without intermediate stages that look minimally plausible⁶³. I conclude that, since forms of a

⁶¹ In the renowned Herentas-inscription (Pg 9, Corfinium).

⁶² Plautus mentions parra as the herald of good omen: picus et cornix ab laeva, corvos, parra ab dextera consuadent (Asin., 260-261). Cf. Fisher (2014: 79-81) on the possibility that the couple U. peico/parfa vs. picus/parra continues an inherited list of augural birds belonging to a single archaic Italic ritual complex; the Iguvine Tables prescribe that the parfa should be seen from the right, too (parfam tesvam,PARFA DERSVA). This points to these forms being identical after all. In turn parus `tit' could be the inverse, hypercorrect product of the «flamma»-rule. Note that if parus were original and parra its alternative realisation, parra and parfa would be ipso facto unrelated.

⁶³ Bear in mind that this objection does not apply to the shift (-)sr- >(-)6/8r->(-)flfir-. (-)sr- is a comparatively unnatural or difficult cluster, resolved in western Indo-European by stop insertion (-)sr- >(-)str- (as in Germanic, Slavic, Old Prussian, Latvian and most Balkanic dialects), by assimilation /sr/ > /r/, or by loss of stridency (as in Italic and part of Celtic).

comparable structure in other languages are lacking, and the proposed evolution reveals a blatant internal asymmetry with that of -nVz-, an etymology containing an intervocalic sibilant is unreliable, and post-syncope *-rz- is unlikely to have evolved into -rf/fi- anywhere. It should be clear by now that a Sabellic sibilant only surfaces as Umbrian and South-Picene -f- in the neighbourhood of /n/ and /r/ when there is also a dentoalveolar sound at play, whether inherited or intrusive, but never in post-syncope environments. A birdname (itself a class of names very prone to borrowing) with ritual associations and far from transparent origins does not qualify as a candidate to fill the gaps of Italic sound-change reconstruction.

In addition, the word-formation presupposed by the traditional etymology is far from convincing: the root is usually held to be *(s)per- `fly' (see LIV: 579), but evidence for a -s-stem of this root is non-existent. Recently, Hofler (2017: 17-18) has ingeniously traced U. parfa, L. parra to an adjective *sprH-s-o- `feathered' (belonging to the russus-type), allegedly paralleled by the Hesychian gloss anapaciov `a bird resembling a sparrow', a diminutive formation which he ascribes to Magna Graecia. But, even if the last point were true (albeit there is not the slightest hint of it in our only extant source), late loss of word-initial s- in the attested Italic forms calls for an explanation Additionally, there is the minor inconvenience that the Italic form would be *(s)parazo- with a voiced sibi-lant; unless Hesychius' sources were very early, we would expect this phoneme to be spelt <Z>.. It has been suggested (Prosper 2020) that the so-called «palma»-rule should be formulated as #CRH.C->#CdRH.C->#CaR.C- with laryngeal loss and lowering of the central vowel under the accent. In fact, the traditionally posited outcome #CdR.Hd.C- is phonotac- tically suspect and unparalleled, except for the dubious case of Greek which has always constituted its model. This would make the intermediate structure *parazV- ipso facto unwarranted: we would expect either PItal. *praza (> U., L. *prara) or *parsa (> U. *parsa, L. parra) We could perhaps even expect L. *prassa if this was an originally rustic word preserving the original *praza reinterpreted in urban Latin as /Vss/, as per Vine (2016, to explain grossus, crassus). Given the Latin fate of IE /Vns/, we could alternatively expect *praza to be adopted as *prasa <pransa> which, given the expressive nuances associ-ated to these forms, could surface as prassa by the «flamma»-rule..

I alternatively reconstruct this birdname as *(s)pr(H)-g^hhiu-o/-eh.2`flying-goer', which straightforwardly accounts for U. parfa. It should have given L. \parva, but underwent expressive gemination (as in other animal names like peccus, cattus, vacca and probably vlverra `ferret'), which prevented homonymy with parva `small', and glide absorption. While this resyllabication of a heterosyllabic structure -Cu- is well attested in epigraphy, as in iannvarivs, fvttvere, acqva, and probably indicative of a lower phonostyle, the eventual loss of the glide (in fact very common in the Romance continuants of these forms), may be put down to the auditory difficulties inherent in the perception of a cluster *-rru-, owing to the minimum distance of its constituents in the scale of strength See Mendez Dosuna (1994) for this phenomenon in Greek.. This phenomenon also lurks behind the Umbrian evolution *-ru> -rr- in *katema>*katerra, as reflected in the denominative verb kat- eramu `arrange yourselves in catervae' (Um 1 Ib 20) One could of course assume that the early outcome of IE *-rg^hu- was *-rfiu-, which may have become Proto-Latin*-rru- by (context-bound!) assimilation of a glottal fricative to a preceding sonorant. Such an intermediate stage is phonetically plausible and may be reconstructed for Proto-Italic. In the mutatis mutandis comparable case of *h₂(e)ng^uh-i-> L. anguis `snake' we reckon with a PItal. stage*-ngu- with occlusivisation after a nasal segment. Since we lack any other instance of the evolution of this cluster in Latin, this argument cannot be pursued here..

The conceivable cognacy of Italic and other European bird names may now ne explored: OPr. spurglis `sparrow' reflects a diminutive of the same form. Gk. auopyLAog `sparrow' (Ar., Av. 300), which can equally be derived from our reconstructed *(s)pr(H)-g^hhiuo- or to *(s)por(H)- g^hhiuo-, is particularly interesting because it has undergone all the sound laws proposed by

Vladimir Georgiev for Pelasgian, an unidentified Indo-European substrate language which is only known through a number of Greek borrowings, like xuqpog the match of xa^oc; `tomb', or uupyoc; `tower', the match of Goth. baurgs, Celtic brig- `elevated village', etc. Most recently, Mihaylova (2016) has vindicated the validity of Georgiev's tenets and has enumerated the features of this lost IE dialect: Grassmann's law, post-Grassmann consonant mutation, delabialisation of labiove- lars and [u], [o] as the outcome of vocalic sonorants, which probably means that a sound [o] was variously parsed according to context. Under these premises, our form may have undergone the following changes: [r] > [or], de-labialisation *-g<sup>h</sup>u-> *-g<sup>h</sup>-, and «Grimm»`s mutation *-g^h- > -g-. The word-initial cluster *sp- is an exception to the rule because, as in Germanic (E. stand, etc.), the sibilant blocks the fricativisation of the voiceless stop in order to avoid an undesirable sequence of fricatives. In fairness, there are variants of this very form which do not conform to the pattern in that they show the full grade of the first member, like the Greek gloss onepyouAoc; `a bird' (Hesychius) or OPr. sperglawanag `hawk'. While, mostly on the strength of the Germanic forms, some forms are said to come from a variant form *spVr-uo-, the examples adduced to this effect are very doubtful: OE. spearwa, etc. `sparrow' can be traced to *(s)por(H)-g<sup>h</sup>hiu-o-.<sup>B</sup>Toch.spara-yakre`kind of bird', related to ^AToch. spar `sparrow?', is traced to earlier *spara, and ultimately to *sperueh2, by DTB: 730, but the reconstruction of a suffix -uo- is ungrounded The very idea that IE *-ru- gives PToch. -r- relies on the prejudice that an inherited suffix *-uo- must have been there because other cognates have it: thus, <sup>B</sup>Toch. smare `oily, sleek' may be a match of PGerm. *smerwa-; needless to say, this is strictly hypothetical, and fatally gainsaid by the preservation of the approximant in ^BToch. serwe `hunter', which contains the same sequence.. Gk. onapdorov is also taken from *auapF-dorov by EDG: 1375 on the strength of the Germanic forms. But, again, the reconstructed *-u- has left no trace, and consequently there is neither comparative nor internal basis for introducing it.

As conceded in passing by LIV: 579, the anit nature of the root *(s)per- is disputable. Koch (1990 I: 431) has drawn attention to its set character in Slavic. Both Albanian fier `fern' and OCS.pero `feather' reflect *(s)per(H)om Cf. Schumacher et al. (2013: 218), Derksen (2007: 396).. The Tocharian forms ^Aspar, ^Bspara- can consequently be traced back to a possessive vrddhi-derivative *(s)perHo- `endowed with feathers' > `bird'. If the Greek form onapdorov is correctly transmitted It is, in fact, misplaced, since the gloss occurs as nr. 1398, which would be fitting if the reading were tonaAaoiov., it could contain the regular outcome of an earlier *sprH-o-, which saves us from resorting to the Lindeman effect. What is more, it might not even contain a diminutive suffix -doio-. We could analyse it as a diminutive derivative of the root participle *sprH-nt- `flying', which would make this form the only remainder of the verbal inflection of this root outside Slavic. A new suffix -doio- may in fact have been metanalysed from original *-nt-iio - once this structure became opaque, as already contended by Solmsen (1907).

Along the same lines, Zair (2012: 98) has tentatively suggested that MB. frau, B. frav (m.) `crow, jackdaw', from *sprauo-, is similar to Goth. sparwa, etc. and perhaps L. parra, U. parfa, and has posited «a root *(s)perH-, in which case the Greek and Germanic forms would represent o-grade *sporH-u-on- and *sporH-g- respectively (with loss of laryngeal in Greek by the Saussure effect), Old Prussian the zero-grade (though with a formation extremely close to that of Greek), and Italic a derivative of an old s-stem, hence *prH-es-eh2. The Celtic forms would then point to *sprH-uo->*sprauo-».

As we have seen, however, there is no basis for reconstructing a suffix *-uo-. The inclusion of the Celtic form in this group is accordingly contingent upon the assumption that the Brit- tonic (and Gaulish) outcome of IE /g^uh/ and /g^hu/ or /g^hu/ in intervocalic position is -u- As contended by Koch (1992) in the footsteps of Cowgill (1980), who first put forward the idea that IE /gW was preserved as a voiced labiovelar in Proto-Celtic. This is based on the equation OIr. guidid = Gaul. uediluml

(Chamalieres) `pray', later confirmed by the root participle *g<sup>uh</sup>ed<sup>h</sup>-ont- in the Celtib. family name kuezontikum (gen. pl., Botorrita III, see Beltran et al. 1996: 144), and also on the equation of OIr. gonaid, W. gwanu, confirmed by the Celtib. personal name GVANDOS, in my view from the past part. *g<sup>uh</sup>n-to- (Penalba de Villastar), which is identical to OIr.goite (gen. pl., Milan Gl. 77a19).. Note, additionally, that original *sprH-uo-would be expected to become *s$>rauo-,not *s^rauo- As per Schumacher (2004: 136), partly on the strength of *prH-ti- `fern' > PCelt.*<prati-. Cf. also Lith. papartis `fern'.. Conversely, the short vowel would probably be regular if we start from Proto-Celtic *styrag<sup>u</sup>o- Note that the set of names presumably meaning `shining' like Baedro, Baedronenses, the personal name BAE- DARI (gen., CIL III: 1585, Dacia), the ethnonym Baedul, which can be traced back to *g^uheh₂id- (Prosper 2019a: 8) on the strength of Gk. ^aiSpog `bright', Lith. giedras, gaidus, gaisas `glow' (*g^uheh₂id-s-o-; cf. *4>aL5og in the Hesychian gloss 4>ai5Јi: opei) must have undergone the influence of *b^heh₂- `shine'..

To recap: there is not a single indication that there have ever been any -s-derivatives of the root *sper(H)-`fly'; all the conceivably related compounds or suffixal derivatives are in principle compatible with a sequence *-g<sup>h</sup>u-,not -u-.We may consequently reconstruct a secondarily thematicised compound *(s)prH-g<sup>h</sup>hiu-which evolves into the expected forms in Italic (under acceptance of the «palma»-rule),Celtic and Baltic, and possibly the Greek substrate borrowing, whereas Germanic certainly reflects an /o/ grade and Baltic and Greek have variant forms with the /e/ grade of the first member, which show no trace of a laryngeal. This leads me to suspect that they may be due to a crossing of *(s)prH-g<sup>h</sup>hiu-with the external derivative of an acrostatic noun*spe/org<sup>h</sup>u-`speedy thing'. In fact, the root *sperg^h-`hurry up' (LIV:581) is likely to go back to a univerbation of *sper(H)-+ g<sup>h</sup>eH-`move' (cf. LIV:153) For both roots, only the Avestan testimony speaks in favour of the reconstruction of a palatal sound. Given the existence of Skt. vanar-gu- `traversing the wilderness', etc., compared with Lith. zmogus `man', from *d^hg^hmeh₂-g^uh₂u- `walking on earth' by Garcia-Ramon (1985: 56; the appurtenance of the second member is ambiguous in Baltic), we can start from *(s)prH-g<sup>u</sup>h<sub>2</sub>u-. This explains the Greek form straightforwardly (a similar compound npeopug has a dialectal variant npeayug `old man, ambassador'), but not U. parfa..

Let us say, as a corollary, that there is at least one form in which *-rVs-may be held to have eventually become -rz- (or -rs-, since rhotacism never occurred): gersiaru is an adjective in the gen. pl. fem., or more probably a month name (see Weiss 2010: 28), accompanying anter- menzarum (Um 1 Ila). While WOU:391 provides a number of reconstructions, the most promising of which presuppose a base *kerts-or *kers-,the identification of this form with an adjective *keres-iio-is satisfactory in phonetic, formational, semantic and cultural terms. This preform is then identical to Pael. cerria, O. kerriiai,etc. (L. cerealis may be dissimilatory or haplological for *cerer-alis,or even replaces a syncopated *cerr-alis)and only the problems concomitant with the equation of U. parfa and L. parra and their stubborn attribution to a non-existent -s-stem have presented an obstacle to the obvious solution Burroni (2016) definitely abandons the connection of gersiaru with Ceres and ingeniously proposes a `har-vest month', either from a root *kers- `reap, cut', via an action noun *kerso-, or from an (unparalleled) sibilant stem, and then from *ker-s-o- `having a cutting'. In his footsteps, Weiss (2017: 373) reconstructs *kers-iio- `pertaining to cutting' and compares Gk. d-Kepoe-Kopnc; `with unshorn hair'. Of course, this only works under the assumption that *-rVz- must become -rf-.. If we rule out U. gerfe,etc. as allusions to the goddess Ceres in the wake of Weiss (2017), the rich Umbrian pantheon is deprived of one of the most important Italic divinities, unless we call other flawed reconstructions into question. A very ancient adjective *kerhmes-6-,based on the early but innovative double-full-grade oblique stem, is probably attested in the L. divine name Cerus,only attested by Paulus ex F. 109 Lindsay «in Carmine Saliari Cerus Manus intellegitur creator bonus», and duonus Cerus in Varro. Either this form or more probably its nominal base *kerhmes-`Ceres' was enlarged by a relational suffix *-iio-(as in Venus, venerius)and became Proto-Italic *kerVz-iio-.This form was as likely a candidate to acquire a specialised meaning as a month's name as many other adjectives derived from divine names in Italy by means of the same suffix. In other words, IE *-rs- and IE *-rVs- simply merged in Umbrian after medial vowel syncope Note that the Latin adjectives sin-cerus `sound, whole' possibly < `of one growth', and especiallyprocerus `tall' < `having a high growth', usually taken from *kerh₁o- or *kreh₁-ro- (see EDLIL: 491 for details), could alterna-tively go back to this very form *kerh_1/3es6-, which would phonetically give *kerro- but could have undergone hy-percorrection (undoing the action of the «flamma»-rule and perhaps thereby favouring a less «rustic-flavoured» se-quence), and it could merit a separate research to find out whether this is more likely to happen in compounds. It should be borne in mind that the divine name Cerus may have undergone the same process, and the testimonies of Paulus and Varro may thus have contained a long vowel /e:/ instead of a long /r:/ that was not noted in their early Latin sources. In fact, these forms could be thematicised residues of the ancient compositional schema X-R(e)-S(e)- exemplified by IE *h₁su-genh₁-es, etc. See more on the involved root and the reconstruction of L. crassus as *krh₃-s-6- in Vine (2016: 138-139)..

Conclusions

As repeatedly observed by previous scholarship, Proto-Sabellic is especially difficult to reconstruct as a homogeneous language previous to differentiation is Oscan, Umbrian and South- Picene, because it has undergone such a plethora of innovations crossing the territory from different points and at different times (and often shared by Latin, Venetic and Etruscan) that a genealogical tree cannot be drawn with any confidence.

As implied above, one of the battlehorses of the Sabellic controversy is the date of medial and final vowel syncope. Benediktsson's assertion that the final vowel syncope took place after the shift -ns> -f/-ss, and for this reason the new, post-syncope cluster -ns was never affected by further changes, Cf. Benediktsson (1957: 258). is quite plausible. The preconception that syncope must have occurred prior to the dialectalisation of Sabellic has no basis, and the respective outcomes of Pre-Samnite and Oscan on the one hand, and Umbrian and South-Picene on the other, often differ simply because they are independent. Rix's ideas are further conditioned by the prejudice, already to be found in Buck (1904: 72), that the evolution of *-nt-s (> -f presupposes an intermediate stage -ns and is therefore posterior to that of original *-ns (O. -ss, U. -f in the acc. pl.). This makes the homogeneous outcome of *-nt-s impossible to come to grips with and generates a complex battery of unwarranted steps, essentially intended to account for the disparate outcomes of «primary» *-ns.

Languages respond differently to the problems inherent in the nasal + fricative transition. Sometimes, the nasal segment is effaced, often, but not necessarily, with concomitant lengthening of the preceding vowel. As we have seen, this change has been recently explained as auditory in nature. In my view, this is what actually happened at one or more stages to the Proto-Italic sequence -Vn.sV-. By contrast, -ns-> -ss- is a less likely evolution (see the arguments against an assimilation /ns/ > /s:/ in Recasens 2018: 165-167). For instance, the fact that the primary sequence *-VnsV- yields a tense sibilant in Hittite, as in *densu-><da-as-su-> `powerful' need not be indicative of regressive assimilation, but of a process very similar to the one depicted for Italic in this work: loss of /n/ created a secondary contrast between the new intervocalic tense sibilant (in turn lenited in Hittite in some contexts) and a lax sibilant that continued IE intervocalic /s/, while the contrast between *-Vns# and *-Vs# may have been abandoned early on The idea that we have to start from a Proto-Anatolian tense sibilant /s:/ in the context -VnssV-, -VmssV-> -VssV- is not contradicted by examples hinting at a tense sound, as in IE *-VsnV- > Hitt. -VssnV- and -VsrV- > -VssrV- vis a vis the lax outcome of /s/ in -VsmV-, -VslV- and the assimilation in -VrsV-> -VrrV-. There is no other case of the alleged regressive assimilation of /s/ in any consonant cluster, even those which constitute a cross-linguistically favourable context, like -ts-, -ks-: they preserve the stop and undergo fortition of the sibilant, like ClsV-, CrsV- (cf. EDHIL: 71-73). A homogeneous treatment of the Hittite internal evidence runs counter to the suggested as-similation. On the other hand, Simon (2020) has vindicated the contrast between voiceless and voiced phonemes in Hittite, which can be extended to sibilants..

Occasionally, however, progressive assimilation takes place. As a result, a sequence -Vn.nV- emerges. This, however, is favoured by a weak realisation of the sibilant, which may have evolved into an approximant-like [s] or a glottalic [h]. This was never the case in Italic, though it might explain the Oscan and Latin assimilation *-Vr.sV-> *-VrrV- For Greek * -Vr/n.hV->-VrrV-, -VnnV-, see Mendez Dosuna (1994: 111)..

Finally, an excrescent homorganic stop may block the contact between the sonorant and the sibilant. This is an articulatory phenomenon, but the reason why the epenthetic sound is reproduced intentionally by the listener lies in his failure to understand the hearer's intention and discount it; as a consequence, the excrescent sound is phonemicised and reflected in writing (see Ohala 1997). In their classical study, Fourakis et al. (1986) observe that the non- etymological dentoalveolar sound is slightly shorter in duration than the inherited one. Recent experiments have shown that conscious perception of unintentionally produced excrescent sounds is comparatively high (but far less frequent than that of intentionally produced sounds). It is comparatively unusual that excrescent stops are parsed as underlying segments and trigger historical sound change over time, which may be partly explained by the influence of orthography besides production variability (see Warner et al. 2001: 81-82). In this work, I have tried to show that it is the previous existence of an underlying segment /t/, but also, in some contexts, the excrescence of a dentoalveolar sound, that constitutes the ultimate explanation for a number of Sabellic sequences at a pre-documentary stage.

The above assumption allows us to bridge the distance between original /s/ and Umbrian /f/. The comparative verisimilitude of a Sabellic language having merged the IE voiced aspirates /g^uh/, /b^h/, /d^h/ and the fricative /s/ into /f/ is contingent upon the situation of the cluster in question in the following hierarchy: the (Indo-European) presence of an underlying dentoal- veolar segment > the (Sabellic) emergence of an excrescent dentoalveolar sound > the (postsyncope, exclusively Umbrian) emergence of an excrescent dentoalveolar sound.

a) The presence of an inherited segment /t/ generates Oscan, Umbrian and probably South-Picene /f/ in word-final position.

Only original *-nts# became *-(n)d# and eventually -f# in all the Sabellic dialects, probably at a very primitive, Proto-Sabellic stage. These dialects show -f in the nominative singular masculine of the active participles, which results from the fusion of the suffix -nt- and the nominative ending of the common gender -s. Later on, Oscan, or perhaps only its Campanian variety, transferred -f to the nom. sg. if the -n-stems. This transfer probably never took place in South-Picene and cannot be claimed for Umbrian, as has been previously done, on the meager grounds that the -f is not reflected because of late Umbrian weakening or defective spelling. As a consequence, Oscan iaf cannot be the acc. pl. of a feminine anaphoric pronoun *eia, but only the nom. sg. masc. of the root participle *i-n.t-s (<<*hii-ent-s by paradigm leveling), preserved in L. iens and Ven. *iant-. The word final sequence *-(V)nts# merged with *-(V)ns# in Latin and eventually surfaced as -Us.

As in the case of U. traf `trans' <*trants, it is unclear whether iaf has undergone compensatory lengthening or not. The Umbrian forms in -ef have undergone Osthoff's Law at the stage *-e-nt-s and are invariably spelt with <e>. This means that the Oscan present participles statif and kunsif, whether they ultimately go back to *-ent-s or *-e-nt-s, should in principle have been rendered dkunsef and <z> (phonemic /e/) must be analogical. This is unclear in view of our paltry evidence.

b) The inherited sequence *-Vt^stV- became Proto-Italic -VssV-, but *-Vn/rt^stV- became -Vn/r.tssV-.

Consequently, /t/ was an underlying segment in this context. Later, -n.tss- became -f- and -r.tss- became -rf- in Umbrian and possibly South-Picene, but remained -tss- in Oscan, at least in the neighbourhood of nasals. Given the post-syncope emergence of an affricate phoneme /te/, its phonemic status may have become ambiguous and it is rendered <ss> but also <zs>⁸¹.

c) IE *-ns.C-, *-ns# and *-ns# in coda and word-final position result in Umbrian and South- Picene -f, but Oscan (and probably Venetic) -ss, and Rix's postulation of a common stage -f, followed by a cascade of analogical processes, has nothing to recommend itself. What is more, the evolution -ns> -f tacitly demands a leap of faith. It can hardly be explained in phonetic terms except by admitting the intrusion of a (subphonemic) dentoalveolar sound that blocked the nasal-to-fricative transition and yielded Sabellic (and probably Venetic) *-n^ts. This turns the scales in favour of a radically different relative chronology from the one assumed so far.

After the nasal was lost, compensatory lengthening of the preceding vowel did not take place, either because the syllable stayed closed or because the epenthetic dental stop was still there; the Sabellic thematic endings are consequently analogical. The Latin outcome -Us is suggestive of the generalisation of the prevocalic outcome, but one could go so far as to posit an evolution *-Vn^ts> *-Vss > *-Vss > -Us (where compensatory lengthening would be due to simplification of the geminate and not to nasal loss). Final -s and -ss never contrasted in Latin, and -Us eventually merged with the nom. sg. *-Vnt-s anyway. Under a completely unitary picture, we could place the Italic languages at different stages of an evolution *-Vn^ts> *-Vts (> -Vs >-(V)d> -f in Umbrian) > -Vss > -Vss (Venetic; -ss by final syncope in Oscan) > -Us (Latin). Even if the first stage had been shared by all the dialects, it can only have been reached after Proto- Sabellic *-nts evolved into *-(n)Q.

As we have seen above, the hypercharacterisation of the nom. sg. of nasal stems can at present be held to have affected Oscan (or at least its central dialects), and cannot be described as a process *-ns# > -f, but as a mere transference of the synchronic ending -f from the -nt- stems to most or all nasal stems⁸².

d) In Umbrian, the post-syncope cluster -n.s- and the cluster -n#s- across a compound boundary underwent excrescence of a dentoalveolar stop, yielding -n.^ts-; this onset was pho- nemicised as /te/ and spelt <z> and experienced no further changes. In Oscan, the cluster -n.s- suffered no visible changes. The originally intervocalic sibilant was parsed as a lax phoneme, and later (or dialectally) as a voiced phoneme judging by the orthographic conventions, but, This is why the Italic word for `dinner' is unlikely to go back to *kert-sna (EDLIL: 106). The rendition of the sibilant in this context shows some hesitation in O. kerssnais, kersnu, kerssnasiais, kersna<t>iais. It is the full match of U. gesna `food' (Um 1 Vb 9), gersnatur `having had dinner' (Um 1 Va 22, past part., nom. pl.) and L. cena. According to WOU: 393, this form ultimately goes back to *kertesna, but this is unwarranted, since it would have become L. **certena. This form may be traced to *kers-sna (LIV: 355-356 <sup>2</sup>*kers- `(ab)schneiden'). The extension of this idea to the -r stems, whose nom. sg. would have been remodeled as *-rs in Proto- Sabellic according to Rix (1986: 592-593) fares no better, since it is exclusively predicated on the high spelling fre-quency vis a vis omission of final -r in Umbrian. In Rix's words, *-rs, like the syncopated nom. pl. -res, has pre-served a «strong» -r, the product of progressive assimilation. The question actually revolves around the analysis of two forms in -tor, affertur,AR(S)FERTVR and kvestur, in which -r is more consistently spelt than in the passive end-ings in -r. Nominal stems have been strongly influenced by the rest of the paradigm, however. If the nom. pl. U. frateer, O. *neer bear witness to compensatory lengthening of the preceding vowel, the reasons for the preservation of a «strong» -r, which additionally would play no visible grammatical role, escape me. needless to say, there are no reliable data about its actual phonetic realisation in this particular context, in which voicedness is disfavoured This change has a parallel in Hittite, where only particular sequences of the form -ns-, -ms-, -rs- undergo epen-thesis: *nsos><an-za-a-as> `us' (gen.), *h<sub>2</sub>msosio- ><a-an-za-a-as-sa-> `descendant', besides forms in which these segments were separated by a laryngeal, as in *genh₂su-><ge-en-zu> `lap, womb', or in the nom. sg. in -r- + -s, as in *h<sub>2</sub>ster-s <ha-as- te-er-za> `star' (see EDHIL: 70-74). In Tocharian, this change might have occurred in intervocalic position only after syn-cope, judging by ^BToch. antse `shoulder' (if <*h<sub>2</sub>omVso-) but <sup>A</sup>Toch. es, kentse `rust' (if <*koniso-), intsu `which, what kind of' (<*h<sub>2</sub>eno- + sV-), ontsoytte `insatiable' (a compound of privative *n- + sV-), kwants `firm' (if <*kuneso-, cf. DTB: passim)..

e) The sequence *-r.s-, whether original or secondary, tends to assimilation, regressive or progressive, or deletion of the rhotic.

In Umbrian, a post-syncope cluster -r.z- did not undergo rhotacism. It merged with the outcome of IE *-r.s-, which in turn shows a tendency to simplification to judge by the spelling hesitation in *torsa> divine name U. turse, tuse, etc. This is in fact the most common evolution for this cluster across languages: see Recasens (2018: 177-178).In sum, it never underwent excrescence, since no factor favoured it. In point of fact, stop excrescence in this context, by which *-r.s- yields *-r^ts-, is very uncommon, although sporadic cases are reported Epenthesis [rs] > [rts] is reported to happen in the Romansch dialect of Bergun (Switzerland) and in central and southern Italian dialects (Old Romanesco perzona `persona', Calabrian vurza `borsa', etc.). Cf. Rohlfs (1968: 381). It is sporadically attested elsewhere in particularly difficult clusters, for instance in ^AToch. kursar `mile, vehicle', obl. pl. kursarwa/kurtsru; in a couple of cases in Old Languedocian (the singular forms chartz, cavalliertz, cf. Cha- baneau 1879, who considers these two forms as scribal errors, as opposed to comparatively more widespread -<ntz>, -<ltz>); and in some Modern Greek dialects like Pontic (cf. Moutsos 1976). Even sporadic epenthesis in this particu-lar context always presupposes comparatively regular epenthesis in /ns/ and /ls/, suggesting it is more unnatural. In the conclusions of his study of epenthesis in /ns/ and /ls/ in a Catalan dialect, Recasens (2012: 88), observes that «the failure for /rs/ to undergo stop insertion and to exhibit a well-defined burst may be attributed to a weak clo-sure resulting from the antagonistic manner requirements involved in the performance of the tongue tip vibration for a syllable final trill and the generation of audible turbulence for the lingual fricative».. Conversely, while an evolution *-sn- >-s<sup>t</sup>n- is less common than *-ns- >-n<sup>t</sup>s-, every Indo-Europeanist knows a sizable number of cases of *-sr- >-s<sup>t</sup>r-, given the stronger realisation of the rhotic in -s.r- (with a trill [r] unless epenthesis occurs, e. g. -s<sup>d/t</sup>r- or -s<sup>d</sup>c-) than in -r.s- (with a tap [r]). This indicates that our only plausible example of this phenomenon, U. trahuorfi `transverse' (< *-urt^sto-; Um 1 VIIa 25), had a segmental /t/, which would in turn mean that *-n/rt^st- directly yielded Sabellic -n/rtss-, as contended immediately above (5.3., 6.).

Accordingly, since *-r.s- >*-r^ts- is infrequent (and *-r.z- > *-r^dz- nonexistent, see fn. 49 above), the last argument that could be reasonably invoked to rescue the validity of the change *-rVs- >*-r.s/z-> *-rf- vanishes. In a nutshell, the traditional etymologies of parfa and qerfe have been underpinning each other, but both lack comparative support and phonetic plausibility. What is more, they have been instrumental in sacrificing the self-evident cognacy of the outcomes of the Italic divine epithet *keres-iio-.

In Oscan, as in Latin, *-r.z- underwent progressive assimilation and became -rr-. This allows us to place our three major dialects at two different stages of the natural evolution *-rVz- (Italic)> *-rz- (> -rs-, Umbrian) > -rr- (Oscan, Latin If and only if the regular phonetic outcome presupposes syncope, that is to say if L. ferre `bring' is from *b^heresi and Cererl, by contrast, has reinserted the medial vowel.).