CC BY-NC-ND
27
Discussion Articles / Дискуссионные статьи
H. Craig Melchert
Carrboro, North Carolina; melchert@humnet.ucla.edu
Initial *sp- in Hittite and sip(p)and- 'to libate'*
The Proto-Indo-European source of Hittite sip(p)and- 'to libate' has been the subject of much discussion, due to its implications for the treatment of initial clusters of sibilant plus stop in Hittite and potential implications for the much larger question of the status of the verbal category of the "perfect" in Anatolian: was the perfect, which in the oldest non-Anatolian IE languages expresses an attained state, inherited also in Anatolian and lost there, or is it an "Indo-Hittite" feature, i.e., a common innovation of "Core Indo-European"? Derivation of sip(p)and- from a PIE reduplicated perfect *s(p)e-spond- has justifiably been rejected on formal and functional grounds, but improvements in our understanding of the outcome of PIE *sp-in Hittite, as well as recent innovative proposals regarding the phonology of reduplication and its status in PIE verbal morphology call for a reconsideration of the issue.
Keywords: hi-conjugation, Indo-Hittite, Proto-Indo-European perfect, reduplication.
At the colloquium honoring Holger Pedersen in Copenhagen in 1992, Bernhard Forssman proposed that the Hittite stem sipand- 'libate; consecrate; offer' reflects a PIE reduplicated perfect stem *spe-spond-, while its rarer OH variant ispand- continues a root present (published as Forssman 1994). This account was not favorably received by the Anatolian specialists present upon its initial presentation, and it has subsequently with rare exceptions met mostly with rejection: e.g., Kassian and Yakubovich 2002: 34-5; Jasanoff 2003: 78, note 39; Tischler 2006: 1058 (with further literature); Kloekhorst 2008: 405; and Yakubovich 2009. Positive endorsements known to me are by Schulze-Thulin (2001: 384), LIV2: 577, and Hoffner and Melchert (2008: 27), the last of which elicited a renewed rejection by Yakubovich (2010a: 151).
All of those who have rejected Forssman's derivation of sipand- have explicitly or implicitly assumed that sipand- and ispand- represent alternate spellings of a preserved initial cluster /sp-/. This was also the interpretation I adopted in Melchert 1994: 31-2, although with considerable misgivings. We have learned a great deal more about the fate of initial *sp- in Hittite in the last twenty years, and I have for some time believed that the gist of Forssman's account of sipand- must be correct (hence the cautious reference in Hoffner and Melchert 2008: 27), but still outstanding formal and functional problems that I could not solve prevented me from asserting this in print. The time has now come for a complete review of the matter.
As has never been disputed, the development in Hittite of initial sequences of *st- and *sk-is consistently ist- and isk- respectively: istanta(i)- 'linger, be late' < *steh2- 'stand', istu(wa)- 'be-
* I am much indebted to Jay Jasanoff for making available to me the text of his forthcoming paper on the PIE perfect in advance of its formal publication and to Ryan Sandell and Sam Zukoff for extensive advice regarding the history of reduplication patterns. The standard disclaimer applies here with particular force, and I am solely responsible for any errors in the application of these authors' views to the case at hand.
Journal of Language Relationship • Вопросы языкового родства • 14/3 (2016) • Pp. 187 — 205 • © The authors, 2016
come known' < *steu-, iskalla- 'slit, tear' < *skelH-, iskar- 'prick, stick' < *sker-. This is also the most common result for *sp-: ispai- 'be satiated' < *spehi(i)-, ispant- 'night' < *(kw)sp-ént-, ispar-'spread out, strew' < *sper-, isparre- 'kick, trample' < *sperH- (on separation of the last two see Kloekhorst 2008: 406-9), ispart- 'escape' < *sperdh-.
However, we now have solid evidence for two additional though rare outcomes of *sp-. The first is preservation as /sp-/, where the presence of a synchronic cluster is crucially indicated by alternate spellings with sa-, se-, si-: sa/e/ipe/ikkusta- /spe/ikusta-/ 'pin, needle' (see now CHD S: 397 for attestations). As seen by Poetto (1986: 52-3), Neumann (1987: 282), and Kimball (1999: 108-9), this word clearly reflects a virtual *sp(e)ik-us-to- to the enlarged root *speig/k-'sharp, pointed' seen in English 'spike', Latin spica 'ear (of grain)', etc. The second rare result is anaptyxis of a vowel u: suppistuwara- 'adorned with appliqués, decorations', suppistuwari- 'applique, decoration'. The meaning is now assured by the occurrence of the i-stem noun in the Hurro-Hittite Bilingual, KBo 32.14 ii 43 (see Neu 1996: 81 and 146). However, the popular etymology (already Neu 1970: 68) as a compound 'brightly shining', allegedly consisting of suppi- 'ritually pure' and istu(wa)- 'become known' makes no sense whatsoever either semanti-cally or formally. Hittite suppi- means 'ritually pure', and there is no basis of any kind for a sense 'shining'. Nor is the role of the purported second member 'become known' in a compound allegedly meaning 'brightly shining' explained (see the justified doubts of Kloekhorst 2008: 791).1 This derivation also cannot account for the alternate form ispisduwaras in KUB 42.64 Vo 2, which cannot be dismissed as a scribal error, since is-pis-du-wa-ra- does not remotely resemble su-up-pí-is-tu-wa-ra- visually or aurally.
The decorations attached to a copper cup (thus in the bilingual) and the gold and silver adornments added to clothing may well have been shiny (for the latter see refs. in Tischler 2006: 1198), but they were also more fundamentally stuck or stitched onto their respective objects.2 We are thus surely dealing with a derivative of a different form of the PIE root *spei-'pointed, sharp' seen already above in /spe/ikusta-/: the sense of /supistwara:-/ was 'appli-quéd', decorated with something 'stuck on' (for the semantics compare the history of English 'stick' and 'stitch' and German stechen). Note, however, that at least one Hittite speaker knew this word in a form with the regular treatment of *sp- as isp-.
I had already recognized the existence of these two examples in Melchert 1994: 32, but found them as exceptional and inexplicable as sipand-. It is now clear, at least to me, that these forms do fit into a well-known Hittite pattern: they show the two regular results of prehistoric *sm-: (1) preservation; (2) u-anaptyxis. The first treatment is shown by Hittite sa/e-me-en-zi, sa-am-na-an-zi 'withdraw; relinquish' where (pace Kassian and Yakubovich 2002: 12) the alternate spelling of the singular stem clearly shows synchronic /smen-/ (thus with Oettinger 1979: 104, Kimball 1999: 117, and CHD S: 120), in an ablauting root present *smén-, *smn-énti, even if the root etymology remains uncertain (thus also Kloekhorst 2008: 714-15).
There are now three examples for the treatment with anaptyctic -u-, which is quite real (contra Kloekhorst 2008: 782-5):
(1) summittant- 'axe' < virtual *smit-ént- '(the) cutting (one)' (already Knobloch 1956: 67, Kimball 1999: 199 et al.);
1 As per Kloekhorst (2008: 790), despite its clear behavior as an inherited word — an ablauting adjective — Hittite suppi- 'ritually pure' has no clear cognates or etymology. Unfortunately, the attractive comparison with Umbrian sopa/supa and interpretation as 'taboo' (Watkins 1975) is very doubtful: see the extended critique by Weiss (2010: 358-83).
2 I know of no basis for the meaning 'animal representation or icon (usually of metal)' adopted by Yakubovich (2009: 548, note 5). In any case, the word definitely does not contain suppi- 'ritually pure'.
(2) sum(m)um(m)ahh- 'unite, make one' < *sm- 'one' + -uman- 'belonging to' + factitive -ahh-(Rieken 2000: 174, modified by Hoffner and Melchert 2008: 60);3
(3) first plural enclitic possessive -summa/i- < *s-me- < aphaeresized *ns-me- (Rieken 2002: 414-15).4
As emphasized by Rieken, the change of initial *sm- > summ- with anaptyxis and gemination is a genuine Hittite sound law. She herself (2002: 408) left open the question of its precise conditioning versus that of the preservation as /sm-/. However, the contrast between sa/emen-< *smen- and enclitic possessive summa/i- < *sme- suggests that the different outcomes are conditioned by the accent: namely, that initial *sm- was preserved immediately before the accent but developed to *summ- when the following syllable was unaccented. We cannot be as certain about the accent in summittant- and sum(m)um(m)ahh-, but their morphological structure is more than compatible with supposing that the accent stood farther to the right than the original initial syllable.
Rieken (2002: 408) reasonably derives Hittite ismeri- 'bridle, rein' < *s(h2)mer-, but if the root etymology (to *seh2- 'bind') is correct, as it surely is, this example does not prove a development of *sm- > ism-, since it is more likely that it was the *sh2- that led to ish- (as in ishanittar-'relative by marriage, as per Rieken 1999: 283-4). The resulting unsyllabifiable *ishme- was then reduced to isme-. Pace Kloekhorst (2008: 394) nothing requires that the verb ishamai- 'sing' reflect a zero-grade *shim-; it may easily continue full-grade *sh2em-, as he himself assumes for the noun ishamai- 'song'.
We may now return to the matter of the Hittite treatment of initial *sp-. The observed vacillation is now explainable. Pre-Hittite language learners were faced with two models for how to treat *sp-: since it consisted of sibilant plus voiceless stop, they could follow the model of *st- and *sk- and add a prothetic i-; however, since *sp- also consisted of a sibilant plus labial stop, speakers could also follow the model of the other sequence of sibilant plus labial stop, namely *sm-, and according to the position of the accent, either preserve the sequence or insert an anaptyctic -u-. Although m generally behaves as a sonorant in older Indo-European languages (that is, as a continuant), one must not forget that in articulatory terms it is also a stop. It is thus not unreasonable that Hittite speakers did not show absolute consistency in their treatment of initial *sp-, where *p belonged both to the class of labial stops and to the class of voiceless stops.
The dominant practice for most lexemes was to follow the model of the other voiceless stops and add a prothetic vowel i-. Contra Melchert 1994: 32, Kimball 1999: 110-11, Kassian and Yakubovich 2002: 33-5, and Yakubovich 2009: 545-7, there is not the slightest justification to doubt the linguistic reality of the prothetic vowel in isT-, as assumed by Kronasser (1966: 48-9), Eichner (1975: 98), Oettinger (1979: 416-17), Kloekhorst (2008: 61), and others. First of all, the alternations in personal names from the Old Assyrian texts of the Colony period cited by Yakubovich (2009: 546) not only all involve *sp-, as he admits, but show exactly the same variation as we have seen in suppistuwara- ~ ispistuwara-: Su-pu-da-ah-su vs. Is-pu-da-ah-su, Su-pu-na-ah-su vs. Is-pu-na-ah-su, Su-pu-nu-ma-an vs. Is-pu-nu-ma-an. I emphasize that we find no spellings in these names of the type iSa-pu- or iSi-pu-, which is what we would expect were
3 Since the word is hapax, the objection of Kloekhorst (2008: 784) that the word does not show geminate spelling for either of the two -mm- is not compelling.
4 I am not persuaded by Rieken's two proposed examples of the change *-sm- > -summ- in morpheme-internal position. Hittite sumanzan- (sic!) means '(bul)rush' and has basic single -m- (see Melchert 2004: 129-31); CLuvian te/issumma/i- '(unfired) clay cup' contains the Luvian suffix -umma/i- also seen in annarumma/i- 'powerful'.
we facing alternate spellings for /spu-/.5 These names actually further confirm that the variation in the Hittite appellative is genuine: /sup-/ vs. /isp-/. Note that the scriptio plena of the stem vowel in suppistuwara- 'appliqué' suggests that the accent was not on the vowel following the initial *sp-, and therefore the treatment supp- beside isp- fits the pattern for summ- < *sm-. Unfortunately, there is no independent evidence for the position of the accent in the personal names or in sa/e/ippe/ikkusta- 'pin', but nothing stands in the way of supposing that the names reflect original accent beyond the first syllable, while the appellative was /spékusta-/ like /smén-/.6
Kimball (1999: 110) cites as "very convincing" my own argument (Melchert 1984: 110) that the Hittite adjective ishaskant- 'blood-shot, blood-stained' must reflect a compound *ishan-skant- with the participle of iske/a- 'anoint, smear', thus showing that the i- of iske/a- must be purely graphic. The argument is not at all compelling, however, since nothing precludes that the compound was formed in pre-Hittite before the addition of the prothetic i-. In any case, the overlooked new example i-is-ke-ez-[zi] in the fragment KBo 34.243:3 (Ritual of Zarpiya) now excludes both my etymology and that of Rieken (1999: 402), approved by Kloekhorst (2008: 402), which start from *(p)s-ske/o- and *sg(h)-yé/ô- respectively.7 The plene spelling (which would be entirely unparalleled for the prothetic vowel) appears to require a return to the etymology of Oettinger (1979: 327), despite the semantic difficulties associated with the root *(hî)eishi-.
The first two arguments adduced by Kassian and Yakubovich (2002: 33) against the reality of the prothetic i- in isT- are also without foundation. Their statement that the prothetic vowel is always spelled i- is correct, but their claim that is-/es- alternations are frequent in cases with etymological *i- is patently false: Hitt. iskis- 'back', cognate with Grk. lxl(ov) 'loins' (a quite certain equation, pace Kloekhorst 2008: 402) is spelled exclusively with is-, while isha- 'owner; master, lord' < *hies-hi-ô- with regular raising of pretonic short *e to *i (see now on this word and its etymology Nussbaum 2014: 244-5) is also spelled exclusively with is-, with the single exception of the totally aberrant form es-hé in the NS copy KBo 3.34 i 25, a copyist's error that has no probative value.8 Their second point, that the prothetic vowel is never spelled with plene as i-is-, makes no sense, since we would expect the prothetic vowel to be unaccented and thus never lengthened.9 The further argument adduced by Yakubovich (2009: 546, note 3) is also less than compelling. He claims that the HLuvian form sà-ma-ra/i-ka-wa/i-ni (URBS) for the city appearing in Hittite cuneiform as URUIs-mi/e-ri-ka- shows that the Luvians learned this city name through the Hittites with /sm-/, since Luvian had eliminated all cases of initial *sC- in their own language. There are two problems here: first, to my knowledge we know only that Luvian eliminated initial *s+stop by deletion of the sibilant (e.g., HLuvian (*261)tapai vs. Hittite
5 One could, of course, argue that the empty vowel used in the spelling for /sp-/ merely copied the following real /u/ vowel, but the evidence from Hittite appellatives for the reality of u-anaptyxis argues decisively against this.
6 The spelling of the "ethnic" suffix -uma(n)- with plene, as in Lthi-is-tu-u-ma-as (KBo 23.99 i 19), provides some indirect support for an accent */Spunoman-/ at least in the one personal name.
7 Contra Kloekhorst (2008: 402), the inflection iskezzi, iskanzi must be older than that of iskiyazzi, since the inflectional type in -e-/-a- in base verbs is recessive in Hittite, while that in -ye-l-ya- is notoriously productive. Thus Rieken's etymology is excluded also on this basis.
8 Contra Kloekhorst (2008: 390) the form e-es-ha-as-si-is is very unlikely to belong to this word (see Otten 1961: 130-1) and is irrelevant. There is thus no basis for appealing to the sporadic New Hittite change of is- to es- (see further below.)
9 The claim of Kloekhorst (2008: 61) that the prothetic vowel cannot be identified with the Hittite phoneme /i/ because it fails to undergo the New Hittite lowering to -e- is also false, since Yakubovich (2010b: 309-15) has made compelling arguments that the very sporadic change of e > i in New Hittite is not a regular sound change.
istapi 'blocks up'). I am not aware of any evidence that tells us the fate of initial *s+sonorant. Second, even if Luvian had no native words with initial *sR-, the argument is not probative. There is no way to exclude that the Hittites adapted the name *Sme/iriga- in their fashion with prothetic i-, while the Luvians dealt with the initial *sm- by anaptyxis of an -a-. The Luvian form may easily be read as /Samariga-/.
We are thus left with sipand- alternating with ispand- as the only basis for doubting the reality of the prothetic i- in isT-. But we have now seen that this orthographic alternation cannot possibly be interpreted to stand for /spand-/, despite the assertions of Kassian and Yakubovich (2002: 33-5) and Yakubovich (2009: 547-8). We now know how a preserved initial /sp-/ was written where it existed, and as we would predict, it is expressed by alternation between sa-pV-, se-pV- and si-pV- in sa/e/ipe/ikkusta- 'pin, needle'. Given that sip(p)and- is spelled several hundred times with absolute consistency as si-(ip)-pa-an-t/d°, it is not credible that this spelling stands for /sp-/. The first syllable of the word must be read as /si-/.
Possible additional evidence for the reality of a stem /sipand-/ comes from HLuvian and Lycian. Yakubovich (2009: 555) cites the suggestion of Hajnal (1995: 133-4) that HLuvian (CAELUM.*286.x)sa-pa-tara/i-i-sa (KARKAMiS A 2+3, §17a) might mean 'libation priest' and reflect an earlier */sVpentero/i-/ also continued by Lycian hppnterus, which is a professional title or institution.10 It is now clear that Lycian hpp- must be derived from a prehistoric *sVp-(contra Melchert 1994: 304-5), and the HLuvian may be read /sapandaris/. For Yakubovich (2009: 556) these forms attest a hybrid Luvo-Hittite creation *sapantalli- 'pertaining to a libation' that underwent rhoticism in Luvian and was then borrowed into Lycian. The last step is pure speculation, and the very different morphology of hppnterus- argues rather for a native Lycian word that is at best a root cognate with the Luvian. That the verbal stem is not attested in Luvian or Lycian (thus far!) is not a compelling argument against a Proto-Anatolian stem *sepond- that led by regular phonological developments to sipand-, */sapand-/, and *hppnt-. I must emphasize, however, that I place no weight on this argument, since the meaning of the Luvian is not fully assured, and that of the Lycian is based entirely on the putative etymology.
Kassian and Yakubovich (2002: 33) and Yakubovich (2009: 547) argue that one cannot interpret the first vowel of the Old Hittite/Old Script spelling si-pa-an-t/d-° as real, because this could only imply a reading /siband-/, and voicing of the stop in this environment cannot be motivated by any known Hittite sound change. This argument reflects a fundamental methodological fallacy and a profound misunderstanding of how orthographies devised by and for native speakers work. Such orthographies cannot be compared to the International Phonetic Alphabet. Native speakers know how the words of their language are pronounced and also the grammar that predicts where they will occur, and writing systems (especially those used by a small elite) need only give just enough clues for another native speaker reader to successfully identify the word intended. Examples like the Anatolian hieroglyphs for Luvian and Linear B for Mycenaean Greek show just how much information can be omitted! Many factors determine spelling practices in a given tradition: aesthetics (important in the Anatolian hieroglyphs used for public inscriptions), convention, convenience, and above all simply imitation of one's teachers.
The Hittites knew that /sipand-/ contained a voiceless labial stop; there was no compulsion to indicate this in a word that occurred hundreds of times in Old Hittite ritual texts. Since the first vowel of si-pa-an-t/d-° has to have been linguistically real, Yakubovich's attempt (2009: 550-55) to motivate a Luvian-influenced anaptyxis into the non-existent /spand-/ is beside the point, but he does raise the legitimate question of why, beginning in Middle Hittite, the spell-
10 For a similar independent interpretation of the HLuvian word and comparison with the Hittite hapax sa-pa-an-ta-al-la (KBo 31.8+ i 7) see Giusfredi 2010: 123-4.
ing si-ip-pa-an-t/d°- was introduced and in fact became the dominant orthography. Here the increasing role of Luvian native speakers among the Hittite scribes may well be the responsible factor. The Luvian-speaking scribes surely learned fairly quickly the general Hittite scribal practice of distinguishing intervocalic voiceless from voiced stops by -VC-CV- versus -V-CV spellings. It would be entirely natural if they chose to apply this to what seemed the unmotivated exception of si-pa-an-t/d-°. I stress, however, that this scenario is by no means necessary. Since, I must insist, the word was pronounced /sipand-/ from the beginning of attested Hittite, a senior scribe could have decided at any time that the exception should be eliminated and a new standard spelling be adopted. A number of changes were made in Hittite spelling practices from Old to New Hittite, and this is merely one of them.
I may cite as a parallel for the non-writing of a geminate stop in Old Hittite versus its expression in later manuscripts the example of /tarsikke-/, the older iterative of tar- 'say'. In Old Script we find only tar-si-kan-zi and tar-si-ke-ez-zi in KBo 22.2 Ro 8 and Vo 4, but in Middle Script tar-si-ik-ke-mi (HKM 46:27) and tar-si-ik-ke-si (KUB 14.1 Ro 34), and in New Script copies of Old Hittite texts tar-si-ik-kan-zi (KBo 3.1 ii 33 and 3.16 iii 14).
Whatever the motivation may have been for the introduction of the spelling si-ip-pa-an-t/d°-, the absolutely fixed spelling with initial si- excludes the reading /spand-/ for Old Hittite, and since there is indeed no way to motivate a voicing of the labial stop, si-(ip)-pa-an-t/d°- must be interpreted as /sipand-/, while the rarer variant is-pa-an-t/d-° stands for regular /ispand-/. The problem then becomes: how do we account for the existence of these two stems and explain their attested shape and use?
The source of the stem ispant- is straightforward: it may continue a PIE root present of the fee-conjugation *spond-ei, *spend-nti 'libate', yielding regularly attested ispanti, ispantanzi (Jasanoff 2003: 86) — but see below for an alternative account. An ablauting root present *spend-, *spnd- (Forssman 1994: 102) would also lead to ispant- phonologically, but such a reconstruction is morphologically incompatible with a Hittite hi-verb root present. That the hi-inflection of ispand- is secondary after sipand- (LIV2: 577) is unlikely. Other Hittite root mi-presents standing beside reduplicated hi-presents show no such influence: wekzi beside wewakki 'demands'.
Forssman (1994: 103) proposed to derive sipand- from a reduplicated stem *spe-spond-, *spe-spnd-, assuming a full reduplication of the initial *sp- of the root and differing simplifications leading to Hittite sipand- and Old Latin spepondi. The need to assume a complicated double dissimilation for Hittite whereby the first *p but the second *s was lost has undoubtedly been one of the reasons for the widespread rejection of Forssman's account.
However, there is now a growing consensus that the history of reduplication in Indo-European should be understood very differently, namely as an inherited synchronic process whose operation is subject to renewal (whatever theoretical approach one takes to its description): see the extensive argumentation of Keydana 2006, followed by Byrd 2015: 118-21 and others. Furthermore, one should in reconstructing the PIE state of affairs follow the standard procedure of giving most weight to isolated archaisms that cannot easily be motivated as innovations. On this basis, following already Brugmann 1897: 40-41 (!), Keydana (2006: 107), Byrd (2015: 120) and others argue on the basis of non-productive forms like Latin present sisto '(cause to) stand; stop', Grk. (brn^i 'stand' plus Avestan hi-staiti 'stands' and OIrish se-scaind 'jumped' that the PIE reduplication pattern with roots in initial *sT- was *sV-sT-.11
11 Hittite sish(a)- 'order, decide' may also be a relic reflecting *si-shr- to the root *seh2- 'bind' (thus Kloekhorst 2008: 758-9; cf. tentatively already Melchert 1984: 153, note 125). For the original stem as sish(a)- see the MH/MS attestations cited by Kloekhorst and the CHD S: 450-51.
This means that we may suppose that the PIE reduplicated stem behind Hittite sipand-was *se-spónd-, *se-spnd- (also considered as an alternative by Schulze-Thulin 2001: 384). These preforms will in terms of vocalism lead regularly to attested sipanti, sipant/danzi, with regular raising of pretonic short *e to i (see Melchert 1994: 101) and lengthening of the accented short *ó to Hittite a in the strong stem (spelled plene a few times, as in KBo 17.11 iv 4&14, OH/OS).
What remains to be accounted for is the deletion of the second *s of the preform *sespVnd-. Once we regard changes in productive reduplication patterns as reflecting renewal of a syn-chronic process, there are (at least) two ways to account for the loss of *s in this context. The first may be formulated in terms of pre-Hittite constraints on the syllabification of consonants. Synchronically, an [s] in contact with another consonant at a syllable boundary appears to be treated as ambisyllabic in attested Hittite: note spellings such as ti-is-sa-kán-zi 'they (usually) step' (IBoT 1.36 iv 30) beside usual ti-is-kán-zi for [tis.skan.tsi] or wa-as-sa-pa-an 'garment' beside wa-as-pa-an for [was.span] (see Bernabé Pajares 1973: 446-7 and passim; Melchert 1994: 150-52). However, we have compelling reasons to think that at an earlier prestage of Hittite there was a constraint against [s]+stop as a syllable onset.
For word-initial position, of course, the evidence is the development of the prothetic i- before *sT-. As argued above, this was undeniably the regular treatment of such initial clusters. The (thus far) unique exception of /spekusta-/ 'pin' was "licensed" only by the pressure of preserved /sm-/ with [s] plus labial nasal stop. Addition of the prothetic vowel naturally enabled a prehistoric syllabification *[is.TV-]. Evidence for the same prehistoric constraint on [sT] in medial onsets is furnished by the pattern of anaptyxis in marked imperfectives with the suffix *-ske/o-, where a vowel was inserted between a preceding consonant and the *s or in the case of coronals between the *s and the *k: appiske- 'take', akkiske- 'die', but tarsikke- 'say' (see Melchert 2012: 179-80). Once again, the anaptyxis solved the prehistoric synchronic syllabification problem, permitting *[ap.pis.kV-], *[ak.kis.kV-] and *[tar.si.kV-].12 I emphasize that the forms with anaptyxis became underlying representations by the time of attested Hittite, leading by then surely to phonetic realizations [ap.pis.skV-] etc.
We may therefore assume that likewise there was a stage at which pre-Hittite (arguably Common Anatolian) *sespVnd(V)- could no longer be syllabified as *[se.spVn.d(V)-], just as the word-initial *[spó/én.d(V)-] of the nominal stem (DUG^ispanduzzi- 'libation' and its derivatives could not be syllabified (likewise in the fee-present if it existed at this point). In this case, solving the problem in the former by anaptyxis, producing *[se.sV.pVn.d(V)-] beside the new [is.pó/én.d(V)-] with prothesis, would have seriously disrupted the formal relationship of words that were in semantic terms transparently related. A simpler alternative solution was to resyllabify *[se.spVn.d(V)-] as *[ses.pVn.d(V)-].
However, there is now reason to believe that the syllabification *[ses.pVn.d(V)-] might itself have been problematic. Zukoff (2014: 272-5) has argued for a context-sensitive version of the well-known Obligatory Contour Principle that prohibits identical adjacent segments. Zukoff proposes that there was also operative in early Indo-European an OCP-SYLLABLE (OCP-ct) constraint: "Assign one violation mark * for every syllable that contains identical segments." 13 If we assume that this constraint also applied at some stage of pre-Hittite (or Common Anatolian), then it would have prohibited the syllabification *[ses.pVn.d(V)-], which
12 For the assumption that intervocalic voiceless stops spelled double were geminates that closed the preceding syllable see Melchert 1994: 18 with references and also Kloekhorst 2014: 545-6 (with a different phonological analysis).
13 For an extensive discussion of OCP effects in PIE and its descendants (including but not confined to OCP-a) see Sandell 2016, who also duly notes (2016: 146) the notorious exceptionality of PIE *ses- 'sleep' and its reflexes.
would have been solved by deletion of the s in the syllable coda.14 If loss of the coda consonant led as expected to compensatory lengthening, producing a virtual *[se:.pVn.d(V)-], the pre-tonic long vowel could have been shortened in time to undergo the specific pre-Hittite change of pretonic short *e to i. Compare Hittite hippara- 'serf' (or sim.) < *hiepor-o- (Eichner 1973: 72).15
Hittite sipand- may thus be derived by regular phonological developments from a reduplicated stem *se-spond-, *se-spnd-, and I stress again that its absolutely fixed i-vocalism cannot be plausibly explained by any other means. There remains, however, the question of whether such a reduplicated stem is a viable source for the Hittite verb in its attested use. One of the few supporters of Forssman's original proposal, expresses doubts: "Ist ein altes Zustandsper-fekt semantisch sinnvoll?" (Kümmel in LIV2: 577, note 5). Yakubovich (2009: 547) also reasonably protests that there is no discernible functional difference between attested sipand- and ispand- (cf. also Kloekhorst 2008: 406). I myself previously looked in vain for any such contrast in usage.
I now believe that such a venture failed because we based our search on false premises. A perfect with the standardly assumed value of an "attained state" hardly fits the usage of the Hittite verb, which is clearly eventive: 'libate', secondarily 'consecrate' (by pouring a libation over), then by metonymy 'offer X (to a deity)' and by syntactic change 'worship (a deity) with X': see CHD S: 384-95. I had supposed that the reduplicated stem belonged to what I regarded as the small class of iterative-durative perfects, such as *we-w(o)rt- 'roll, revolve' (on such a meaning for at least some instances of Vedic vavart- see Kümmel 2000: 462ff.). But I could find no clear traces of an iterative-durative or even processual value for sipand-.
Jasanoff (forthcoming) has now argued that the "attained state" value of the perfect in Core Indo-European is an innovation and that the classical "perfect" originates in a reduplicated h2e-aorist of the shape *Ce-CoC-, *Ce-CC-, whereas the few "perfects" that show iterative semantics reflect rather reduplicated h2e-presents of the form *Ce-CoC-, *Ce-CC-.16 Hittite wewakk- 'request' (repeatedly) and mema/i- 'speak' are direct reflexes of the latter category. By this scenario, *se-spond-, *se-spnd- would have been a reduplicated h2e-aorist and should have referred to the act of libating not as an activity (which would have been expressed by the h2e-present), but as a single telic act.17
If one examines all thirty-plus instances of sipand- in Old Hittite/Old Script, one finds that it is consistently used in such a fashion. It is used to refer to the act of libating once at a particular "station", such as in front of the window (KBo 17.11+ iv 23) or to the hearth (KBo 17.19
14 One may compare typologically for a similar "repair" the Sanskrit weak perfect stem sed- 'sit' < *se-sd- and more broadly other Sanskrit weak perfect stems of the shape CeC- as well as long-vowel preterite formations in Germanic and Celtic: see Schumacher 2005: 601-5, Zukoff 2014: 274, and Sandell: 2016: esp. 142-3 and 156-7.
15 Zukoff (2015) has now refined his account of Indo-European reduplication patterns in terms of what he labels the Poorly-Cued Repetition Principle: "A CVC sequence containing identical consonants (CaVC„) is dis-preferred, due to repetition blindness; it is especially dispreferred if one or both of the consonants lack phonetic cues which are important for the perception of its presence (in contrast to zero) in the speech signal." For reasons he sets forth, this principle applies especially to the second fricative [s] in a sequence #sVsT-. Since this newer formulation will also handle the case of sipand- < *se-spond-, I forego extensive discussion here and refer interested readers to Zukoff's own presentation, available online.
16 While verbs of the latter class have mostly been assimilated to the true "attained state" perfects in the attested languages, Jasanoff stresses that in the oldest Greek their separate origin is still betrayed by a different pluperfect inflection.
17 I do follow LIV2: 577 and others, against Jasanoff forthcoming, in supposing that the concrete meaning 'libate' of Greek and Hittite is original, from which already in PIE developed the secondary sense 'pledge, dedicate' (in the middle 'pledge, dedicate oneself').
ii 11). It alone (never ispand-) is used with specification of how many discrete times one performs the act of libating: 'once' (KBo 17.11+ iv 33, KUB 43.30 ii 11&15 and often), 'twice' (KBo 20.10 i 9), 'three times' (KUB 43.30 ii 14), 'seven times' (KBo 25.127 ii 25). It alone is attested in the telic sense 'consecrate' a sacrificial animal or other object (KBo 17.36+ iii 9 and 17.33+ i 14). Finally, it may be used of worshipping a deity (in the accusative) by libating into a bowl (KBo 25.61 Vo 9).
Trying to determine whether the stem ispand- has a synchronically distinct sense and whether its absence in the contexts just cited for sipand- is systematic or merely due to chance is made extremely difficult by the very small number of examples, especially of examples with full context. Aside from the "Ritual for the Royal Couple", which uses only ispand- in its attested portions (see Otten and Soucek 1969: 97), there are a mere handful of other attestations, either in Old Script or later copies. However, the examples in KBo 20 ii 5&6 (OH/OS), where ispanti 'performs a libation' occurs in the immediately context of hinga 'bows' is strongly reminiscent of that of KBo 25.104 ii 12-13 (OH/OS?), where we read LUGAL-us dKuwassas USKE[N...] sipanti. Similarly, the phraseology [... ]* 2 ekuzi [...hu]ppari ispant[i] 'drinks two [...] libates into a bowl' (KBo 25.51 i 18-19; OH/OS) hardly differs from that of hüppari sipanti (KBo 25.61 Vo 9; OH/OS) cited above.
It therefore seems extremely unlikely that the stem ispant- has any different sense syn-chronically from that of sipant-. Both refer to libating conceived as a single telic act and to the other telic meanings derived from that. By the oldest attested Hittite ispant- survives only as a marginal variant of sipant-. In fact, one may reasonably ask: does the very rare verbal stem ispant- continue a genuine prehistoric present stem at all, or is it merely an analogical creation based on the nominal forms (DUG)ispanduzzi- 'libation', DUGispanduzzi(y)assar- 'libation vessel', and (DUG^ispantuwa- 'libation vessel'? Of course, if one opts for the latter interpretation, then one must ask in turn what the basis was for the nominal stems, which appear to be deverbative.
As to DUGispanduzzi- (from which DUGispanduzzi(y)assar- obviously is further derived), if one looks at the class of Hittite nouns in -uzzi-, some are indeed undeniably deverbative, formed to synchronically existing verbal stems: e.g., KU^annanuzzi- '(part of a) harness' < annanu- 'train', kuruzzi- 'cutting tool' < ku(e)r- 'cut'. Others, however, appear to be rather deradical, being derived from forms of the respective roots whose existence in pre-Hittite as verbal stems is dubious: e.g., ishuzzi- 'belt, chain' < *s(e)h2- 'bind' (but all verbal forms are based on ishi- < *sh2ei-), tuzzi- 'camp; army' < *dh(e)hi- 'place' (whereas the present stem of the verb is dai- with an *-i- suffix).18 The nominal stems DUGispanduzzi- and DUGispanduzzi(y)assar-are thus not probative evidence for a genuine pre-Hittite verbal stem ispand-. The stem ispanduzzi- may be a primary derivative from the root *spend-. It is true that (DVG)ispanduwa- is hypostasized from the verbal noun (thus with Carruba 1966: 23, note 35), but precisely in this case there are also a number of spellings as (DUG)sipanduwa- (see CHD S: 396). In this noun, then, the variant ispanduwa- may be analogical, just as in the other verbal forms.
I therefore must conclude that evidence for a pre-Hittite present stem of any kind is less than compelling. A h2e-conjugation present *spónd-ei, *spénd-nti may well have existed, but its existence must be based on other evidence (see Jasanoff 2003: 78 on Greek cttc£v5w 'pour, libate' and Latin spondeo 'vow'). The fundamentally telic senses of the Hittite verb sipand- are in any case fully compatible with the proposal that it continues a reduplicated h2e-aorist. With due revisions, then, the much maligned derivation suggested by Forssman more than twenty
18 The primary meaning of tuzzi- is 'camp', as shown by the derived verb tuzziya- 'encamp'. One must with Kloekhorst (2008: 908) insist on this etymology of Carruba (1966: 23, note 35). There is no connection with western Indo-European *teuta-.
