The Luwic inflection of proper names, the Hittite dative-locative of i- and ii̯a -stems, and the Proto-Anatolian allative

: The article establishes the inflection of proper names in Luwian and Lycian, which differs from appellative inflection in all oblique cases. It is argued that the locative, genitive and ablative were reshaped after the pattern of the ā- stems, which were the most frequent type in names. The characteristic dative * -Vi̯o , however, was generalised from the i -stems, whose type had become restricted to names, especially personal names, after the PD i -stems had been generalised in the appellatives. The origin in the i -stems appears from Hittite, which has a corresponding ending in i- and ii̯a- stems. In Hittite, the ending can be traced back further to the use of the allative in dat.-loc. function to circumvent the unfortunate combination of a stem in * ‑i‑ with the dat.-loc. ending * -i . The Luwic data can now be used to determine the character of the PAnat. allative, which must have been * -o on account of Lyc. -e . Since Anatolian shows a vigorous allative that is presupposed by petrified remnants such as * pr-o ‘forward’ in other IE languages, the allative provides an additional argument for the Indo-Anatolian hypothesis.


Introduction
The main topic of this article is the inflection of proper names in Luwic, which has so far not received much scholarly attention. I will outline the paradigms and offer explanations for the deviations from the appellative paradigms. The dative of this paradigm requires a treatment of the second topic, the Hittite dative-locative of i-and ii̯ a-stems. Finally, these matters have some implications for the exact reconstruction of the Proto-Anatolian allative. In the process, I will also make new proposals regarding the aberrant forms in the paradigm of HLuw. masani-'god', 1 the Luwian dative-locative of the genitival adjective -an, and the Lycian infinitive.

The Luwic inflection of proper names
While Luwic morphology has not received much attention in general, this is especially true for the inflection of proper names. The most comprehensive study so far is Meriggi (1980), which is restricted to synchronic Lycian. For Luwian, some details have occasionally been noted in passing. However, the special status of the onomastic paradigms is not always recognised, the details remain fuzzy, and a dedicated treatment or even overview is lacking. Here, I want to present the Luwian and Lycian onomastic stem types and their paradigms and compare them to the appellative paradigms (2.1-2.3), as well as to reconstruct their Proto-Luwic predecessors (2.4), providing explanations for their deviations from the appellative paradigms. The discussion of the origin of the dative will be concluded only after a treatment of the Hittite data that I propose to compare.

Personal names 2.1.1 Hieroglyphic Luwian
The most complete picture of Luwian onomastic declension is found in Hieroglyphic Luwian. I will first focus on the main inflection types of personal names, which are tabulated in Table 1 and illustrated with attestations in Table 2. The paradigms are also exemplified with divine names and toponyms, inasmuch as their inflection corresponds to that of personal names; the slight differences that these categories present will be discussed in 2.2 and 2.3. Forms with a following asterisk are not attested in any of these categories but are expected on the basis of parallelism with the other stems. The three paradigms all follow the same pattern, as summarised in Table 3 (p. 24), with V representing the respective stem vowels. For contrastive purposes, the corresponding regular appellative paradigms (restricted to the relevant common gender singular forms) are given in Table 4 (p. 24). 2 Diverging endings are given in bold.
restricted to adjectives and to be understood as a combination of the indicated stem types: -i-in the common gender and -V-(thematic) or zero (consonantal) in the neuter gender. The i(i̯ V)-stems are noted as -ii̯ V/i-. Further note that I have not used superscript or other transformations in the transcription of HLuw. plene writing. 2 For -a as the regular dative-locative of a-stems, cf. already Werner 1991: 27 and, more recently, Yakubovich 2015: §6.2. 3 Not attested in proper names, but cf. the potential testimony of DEUS-nidi below. nom.
-as -is -us acc.
The onomastic i-stems are analysed by Yakubovich (ACLT) not as i-stems, but as i(ya)-stems, i.e. iya/i-stems (cf. e.g. tadiya/i-'of father'). 4 The appellative iya/i-stems do have a similar inflection, as can be seen in Table 5 (p. 25). Crucially, however, their inflection differs in the oblique cases: here, iya/i-stems typically have -iyarather than -i-, whereas the onomastic i-stems never have forms with -iya-. 4 When only direct case forms are attested, however, they are analysed as "(i)-stems", i.e. the appellative i-stem type. The confusion in stem type assignment disappears with the recognition that names have their own i-stem paradigm of the shape presented above: neither appellative type is applicable. nom.
-as -is -us acc.
-asa, -asi -asa, -asi -uwasa, -uwasi* gen.adj. -asa/i--asa/i--uwasa/i-Indeed, in the iya/i-stems, the forms with -i-for -iya-are restricted to the southern part of the HLuw. area, meaning that the two declension types are always distinct in the north. 5 In addition, in CLuw. the direct cases are also distinct: the iya/i-stems show plene spellings (°Ci-i-iC), whereas the onomastic i-stems do not (°Ci-iC). These differences show that we are dealing with two different types. This is also expected given the origin of the iya/i-stem type, viz. the ii̯ o-stems (see Melchert 1990: 200;Norbruis 2018: 27-29), 6 whereas the onomastic i-stems are the onomastic counterpart of the appellative i-stems. Finally, there is also a genuine onomastic nom.
-iyasa, -iyasi I TONITRUS-hu-pi-ia-sa, I *447-nu-wa/i-ia-si The recognition of a distinct onomastic declension of the shapes presented above can also help explain some forms that have so far been enigmatic. In the paradigm of the noun DEUS-ni-(representing masani-) 'god', which usually inflects like a regular appellative i-stem (DEUS-nis -nin -ni -nadi -nasa/i--ninzi -nanz), we also find the forms gen.adj. DEUS-nisa/i-, abl. DEUS-nidi, dat.pl. DEUS-ninz, with unexpected -i-for -a-. These forms do, however, conform to the onomastic i-stem inflection, which has -i-throughout. This suggests that masani-was also sometimes conceived of as a name ('the Gods'), effecting a shift to the onomastic variant of the i-stem inflection. 7 Indeed, such shifts from the appellative variant to the onomastic counterpart of the stem class are the rule when a noun or adjective is used as a name. For example (cf.

Cuneiform Luwian
Although the limited Cuneiform Luwian corpus allows us to discern only hints of its basic onomastic inflection, the forms it displays generally correspond to those of Hieroglyphic Luwian. Thus, the acc.sg. ᵈi̯ a-ar-ri-in is accompanied by a gen.adj.nom.sg.c. ᵈi̯ a-ar-ri-iš-ši-iš, pointing to ᵈi̯ arri-with onomastic i-stem inflection (-i-throughout). The nom.sg. ḫa-ad-du-ša-aš 'Ḫattuša' occurs next to a dative URU ḫa-at-tu-ša-i̯ a, 8 with the dative ending -ai̯ a characteristic of the onomastic a-stems. These snippets show that the defining peculiarities of HLuw. onomastic inflection go back at least to Proto-Luwian. Due to the different nature of its corpus, CLuw. also has a few attestations of a case of which no certain instances are found in HLuw.: the vocative. An example of an a-stem vocative is ᵈkamrušepa, which shows a form identical to the stem. One potential attestation in HLuw. is (DEUS)ku+AVIS-pa-pa-a (KARKAMIŠ A6 § 21), which would show the same ending, but it is not excluded that this is rather a dat.sg., with Yakubovich (ACLT).

Lycian
The Luwian state of affairs has a clear counterpart in Lycian, where the main personal name paradigms are those presented in Table 7. 9 In contrast with Luwian, ablatives and genitival adjectives 10 are not normally used with personal names 8 On this form and the slightly deviating inflection of toponyms in general, see 2.2. 9 Two further types that are not so well attested should also be mentioned here. We have a few cases of nominatives ending in a nasalised vowel: ati [bin]ẽ, xssbezẽ, xudalijẽ (rendered in Greek as Κυδαλιη[ς]), and, with -ã, ñturigaxã. Only xudalijẽ also attests a genitive, xudali[j]ẽh◊. We further have a type with a nominative in -ẽi : mutlẽi, pigrẽi, sbikezijẽi, tewinezẽi, uhetẽi, xerẽi. In accusative function we find huzetẽi, possibly also xerẽi. pttlezẽi and xuñnijẽi show the datives pttlezeje and xuñnijeje, respectively. The genitive is attested as xerẽh for xerẽi. Perhaps mutleh belongs to mutlẽi. It is not evident how we should interpret these types historically. In mechanical reconstruction, -ẽ and -ẽi point to PLuw. *-on and *-ontsi, respectively. Possibly, they are to be analysed as old n-stems, with the nom.sg. endings going back to *-ōn and *-ōn+is (Melchert 1994: 305). 10 Save a handful of exceptions, which regarding their stem vocalism behave like the genitive. in Lycian. The genitive, on the other hand, is restricted to proper names. 11 Of the allomorphs of the genitive, -Vhe is the oldest form, and -Vh and -Vhñ (no examples of the latter are included in the overviews below) are secondary forms created for nom.sg. and acc.sg. heads, respectively (see Adiego Lajara 2010 and 2.4.1 below). 12 -aje -eje -ije(*), -eje -uje gen.
-ah(e) -eh(e) (-ih(e)), -eh(e) -uh(e) 11 This is true for the genitive singular, with which we are concerned here. Possibly the badly attested genitive plural was also used with appellatives, as is currently usually assumed. The gen.adj. could also have a plural interpretation, however, cf. e.g. ẽni mahanahi 'mother of the gods'. 12 In a very small number of cases, the genitive appears without any ending (e.g. epñxuxa tideimi, mrexisa tideimi, wazzije kbatra). It has been speculated that these continue the old gen.sg. in *-s (cf. Adiego 1994: 13;Adiego Lajara 2010: 5;Melchert 2012: 276f.;Kloekhorst 2013: 141). I would be more inclined to regard them, with Neumann (1970: 62); Hajnal (1994: 203); Schürr (2010: 120f.), as secondary to -h, the regular nominative of the genitive, which resulted by analogy from -he < *-so (see Adiego Lajara 2010). As a typologically weak sound, in absolute auslaut, phonologically isolated within Lycian, the occasional loss of -h would not be very surprising. The survival of the genitive *-s would be. 13 The acc.sg. ending -ẽ does not occur in any name that is attested in multiple cases, so it is strictly speaking not certain whether this example belongs to the e-stems or perhaps to one of the types mentioned in note 9. However, the acc.sg. of the e-stems would certainly have been -ẽ. 14 Unfortunately, we do not have any attestation of a direct case to verify that the datives sxxulije and m̃mije belong to sxxuli-and m̃mi-, but this is the only option if these forms follow the regular morphological pattern of datives, viz. stem + -je. There is also a possibility that they are datives The a-stems, e-stems and u-stems are completely parallel to each other. 15 Also, note the existence of ije-and ija-stems corresponding to the Luwian iya-stems, inflecting like regular a-and e-stems, e.g. xssẽñzija, xssẽñzijaje, xssẽñzijah, and wazzije, wazzijeje. The only paradigm with deviant variants is that of the i-stems, which is clearly due to the encroachment on the onomastic i-stems of the appellative i-stem pattern, which has -i in the direct cases, but -e-rather than -i-in the oblique. Thus we find the old onomastic dat. -ije next to -eje, and in personal names the gen. -ih(e) has apparently completely given way to -eh(e). The original onomastic genitive is still regular in toponyms, however, e.g. telebehihe (telebehi 'Telmessos'), xadawãtihe (xadawãti 'Kadyanda'), xãkbihe (xãkbi 'Kandyba'). From these paradigms, we can abstract the pattern as presented in Table 10. -Ṽ dat.
- V-h(e) We may again compare the relevant cases of the appellative inflection, presented in Table 11. The genitive may be compared to the genitival adjective. The case forms that are different are indicated in bold. A first thing to notice is that, unlike appellatives, the onomastic inflection also features u-stems. As far as case forms are concerned, we see that, like in Luwian, the one case that formally differs from its appellative counterpart in all paradigms in -e, like uwiñte and tuhese (cf. the following note), but given that this type is much rarer, this should not be our default assumption. 15 A noteworthy deviation from the general pattern is that we occasionally also find datives of personal names without the characteristic -je; on these, see 2.5.2.

a-stems e-stems i-stems
nom.
-a -e -i acc. - In addition, the more original variant of the i-stems paradigm differs from its appellative counterpart by featuring -i-throughout, rather than having -ein the oblique cases.

Toponyms
The inflection of toponyms is generally identical to that of personal names, with the exception of one prominent aspect: the additional locatival functions, not found with personal names, are expressed with a separate locative case, which is identical to the stem. The functions of this case are not completely lexically complementary to datival function: toponyms also occasionally occur in datival function. In such cases, Luwian uses the separate dative ending as found in personal names, whereas Lycian uses the locative for this purpose as well.

Cuneiform Luwian
The distinction can also be seen in CLuw., where ḫattuša-occurs in locatival function (at least in our best current understanding) as ḫattuša and in datival function as ḫattušai̯ a (ex. 7f.).

Lycian
The Luwian locative also has a counterpart in Lycian, which adds the information that the vowel colour of the locative ending is usually identical to the stem vowel, i.e. -a in the a-stems and -e in the e-stems, and also -i in the less frequent i-stems.
Both -e and -a occur in the following passage (ex. 9).
In Lycian, however, this case is not only used in locatival but also in datival function; the PN dative case form -Vje is not used with toponyms, cf. ex. (10). Here, arñna, tlawa, pinale and xadawãti are clearly syntactically parallel, as is confirmed by the Greek version (which is phrased slightly differently in that the people of the cities rather than the cities themselves are mentioned). xadawãti therefore exemplifies the dat.-loc.sg. of a toponymic i-stem (cf. gen. xadawãtihe).

Divine names
The most striking deviations from the inflection as outlined above are found in divine names. Most deviating of all are the name of the Storm-god, Luw. tarḫunt-, Lyc. trqqñt-, and that of the Sun-god, Luw. tiu̯ ad-. The deviant inflection of these names is related to the unique stem type they display, that of common gender consonant stems, which had been wiped out in appellatives due to a general conversion into i-stems. The type is clearly archaic. In the case of the Storm-god, we even find ablaut. We can establish the paradigms as outlined in Table 12 (Luwian) and Table 13 (Lycian; p. 32). The oblique stem can be reconstructed as *trHunt-(*trHʷnt-), and the dative ending is -i, as we would historically expect for consonant stems. In the nominative, the CLuw. form ᵈIŠKUR/U-anz agrees with Lyc. A trqqas, pointing to PLuw. *trHʷants. 17 An innovated form tarḫunz, resulting from levelling on the basis of the ᵈIŠKUR-u[n-t]i (DEUS)TONITRUS-hu-ti abl.
-(DEUS)TONITRUS-hu-ta-ti g.(a.) ᵈIŠKUR-aš-ša-°(DEUS)TONITRUS-hu-ta-sa(-°) oblique stem tarḫunt-, was present already in CLuw., and is the basis for the HLuw. forms tarhunzas and tarhunzan. These forms show that the unique declension of the direct cases was no longer understood and was therefore adapted to agree with the most common onomastic type, that of the a-stems.
A name with a similar inflection is tiu̯ ad-, the Sun-god, cf. Table 14. This lexeme is not found in our current Lycian corpus, but does survive in both versions of Luwian.
The acc. ᵈUTU-an is the only attestation of a consonant stem acc.sg.c. in all of Luwian. Again, we find a dative in -i, and in HLuw. a remade nom. -zas replacing the older nom. in -z. In this case, we also find another strategy to regularise the paradigm in the direct cases: the introduction of i-stem inflection. 18 In vocatival function, next to use of the nominative form, CLuw. also attests ti-u̯ a-ta and ᵈši-u̯ a-ta, 19 with an ending -a resulting from a reinterpretation of -a in the a-stems as an ending.
There may have been other remnants of this kind (cf. e.g. CLuw. dat. ᵈa-i̯ a-an-ti-i), but most other divine names inflect according to the more familiar vocalic stem types. However, these, too, behave slightly differently from regular personal names: like the consonantal stem type, their dative often matches that of appellatives rather than that of personal names. For examples from HLuw., see Table 15 and  Table 16.  (DEUS)hi-pa-tu (DEUS)sa₄+ra/i-ku gen. -- The ending -ya does sometimes occur as well, however, and both variants may be found with the same name. 20 The dative of kubaba-is attested both as (DEUS)ku+AVIS-pa-pa and as (DEUS)ku+AVIS-pa-ia, and likewise for tasku-we find both (DEUS)ta-sà-ku and (DEUS)ta-sà-ku-ia. The datives of álanzuwa-, iya-, karhuha-, tagamana-, and pahalati-are only attested with the ending -ya ((DEUS)á-la-zú-wa/i-ia, (DEUS)i-ia-ia, (DEUS)kar-hu-ha-ia, (DEUS)tá-ka-ma-na-ia, (DEUS)pa-ha-la-ti-ia). 21 The -a of the a-stems, to which the other forms without -ya are likely to be analogical (see 2.5.2), corresponds to the dative-locative also found in the appellative a-stems.
In CLuw., we find a peculiar dative of a unique shape: the dative of the deity kamrušepa-is attested as ka-am-ru-še-pa-i. This form does not have corresponding forms elsewhere in the nominal system: appellatives have -a, personal names -ai̯ a. Its ending is nevertheless morphologically transparent: it consists of the stem vowel -a-and the dative ending -i. It may in principle have been formed after other divine names (e.g. tarḫunt-s : tarḫunt-i = kamrusepa-s : X → kamrusepa-i), but the morphological deviations in divine names we have seen so far are archaic, and so the ending may also be an archaism. 22 In Lycian, the attested datives of vocalic stem divine names appear not to correspond to the general pattern of personal names either. The dative of malija-'Athena' is mali, with -i (i.e. *-iji) as in the appellative a-stems rather than with -aje as in the personal name inflection. Similarly, the datives of ertẽmi-'Artemis' and natri-'Apollo' are ertẽmi and (B) natri, respectively, rather than forms in -ije or -eje. 23 su-sa₅+ra/i-[i ? ] 'to [my] lord Karhuhas and to Kubaba, Queen of Karkamiš' (KARKAMIŠ A25a §6). Similarly (DEUS)CERVUS₃+ra/i-hu-ha-ia 1 BOS(ANIMAL)-sa OVIS-sa-ha (DEUS)ku+AVISpa-pa 1 BOS(ANIMAL)-sa 1 OVIS(ANIMAL)-wa/i-sa-ha (DEUS)sa₅+ra/i-ku OVIS-wa/i-sa ("*478")ku-tú-pi-li-sa-ha 'for Karhuhas, one ox and a sheep; for Kubaba one ox and one sheep; for the god Sarkus a sheep and a KUTUPILIS' (KARKAMIŠ A11b+c §18b-d). But the same combination of names is found as (DEUS)ka+ra/i-hu-ha-ia (DEUS)ku+AVIS-pa-ia-ha 'to Karhuhas and Kubaba' in KARKAMIŠ A13d §7. 21 One complicated case is runtiya-, the Stag-god. Next to the dative (DEUS)CERVUS₃-ia, which represents either the form in -a (runtiya) or that in -aya (runtiyaya), we also find (DEUS)CERVUS₃(-)‹ru?›-ti-i and (DEUS)CERVUS₃-ti=pa=wa/i=ta-a, with unexpected -i. A similar unexpected variation of the stem vowel is, however, seen in the nom.sg., where we also find (DEUS)CERVUS₃-ti-sá, an i-stem, and even (DEUS)CERVUS₃-za-sá, which recalls the old consonant stems tarhunzas and tiwazas. Its stem and endings may have been influenced by the latter two lexemes, with which it occurs in collocations. Indeed, (DEUS)CERVUS₃(-)‹ru?›-ti-i is immediately preceded by ‹(DEUS)›TONITRUS-hu-ti-i ‹(DEUS)SOL›, and likewise (DEUS)CERVUS₃-za-sá is immediately preceded by (DEUS)TONITRUS-hu-za-sá. 22 The same ending can be found in Hittite, e.g. ᵈḫašgalāi (ḫašgalā-), ᵈzinkuruu̯ āi (zinkuruu̯ ā-). In this case, too, it is unclear whether this is an archaism or an innovation. The match between CLuw. and Hitt. may, however, be taken to suggest that we are dealing with archaisms. 23 The appearance of the dative ending -i in zeusi 'Zeus' is probably rather related to the Greek origin of this name; cf. similarly e.g. mlejeusi (also probably with -eus-from Gr. -εύς, although the

Differences
The Luwian and Lycian onomastic paradigms are very well comparable but also show some differences. One noticeable difference is due to the introduction of the appellative vowel pattern (dir. i, obl. e) into the Lycian i-stems. The more vestigial type, which has -i-in the oblique cases, corresponds neatly to the one i-stem type found in Luwian. Another difference is that Lycian still differentiates between a-stems (< ā-stems) and e-stems (< o-stems), which have merged into a-stems in Luwian as a result of sound change.
Next to these two clear innovations, one on the part of each Luwic branch, there is the further difference that Lycian genitives and genitival adjectives are, as a rule, distributed complementarily: genitives are used with names, genitival adjectives with nouns and adjectives. In Luwian, there is no such distribution; CLuw., as far as we can tell, does not use the genitive, 24 and in HLuw. both forms occur with both types of lexeme. The existence of two morphologically different formations with the same function suggests the loss of an earlier distinction. Since Lycian shows a neat distinction by using the genitive with proper names and the genitival adjective with appellatives, I assume that this is the Proto-Luwic situation and that this distribution became blurred in Luwian. HLuw. developed a tendency towards a new distribution by which the genitival adjective was preferred in the oblique cases (Yakubovich 2008). Since the direct cases can be seen as the default, operating in the core of the sentence, the desire to inflect the preceding genitival element to bring out its dependency on a functionally more marked form was naturally highest in the oblique cases. A similar situation may have triggered the eventual removal of the genitive in pre-CLuw.
The various allomorphs of the genitive can in both Lycian and Luwian be shown to go back to a single form that was reinterpreted as an inflected form, triggering the creation of other inflected forms to establish agreement with the head noun: in Lycian, the oldest form is -Vhe < *-Vsso, on the basis of which the secondarily inflected forms nom. -Vh and acc. -Vhñ (B -Vs and -Vzñ) were created (see Adiego Lajara 2010). In a similar vein, in HLuw. the oldest form is -asa, which below the Taurus mountains obtained a pendant -asi for agreement with common gender head nouns in analogy to the pattern of the gen.adj., c. -asi-, n. -asa-(see Palmér 2021); in other words, -asa was adapted to -asa/i in analogy to -asa/i-. Note that this analogy proves that°a-sa spells -asa rather than **-as, as was already likely in view of Lyc. -Vhe.

Reconstruction of the paradigms
Apart from these differences, the paradigms match very closely. The overall pattern is entirely parallel and can therefore be straightforwardly reconstructed for Proto-Luwic, cf. Table 17. 25 The individual Proto-Luwic onomastic paradigms can be reconstructed as presented in Table 18. 26 -

Pre-Proto-Luwic: prehistory of the case forms
The nom. and acc. are always identical to their appellative counterparts. In the following, I will discuss the prehistories of the remaining cases, in increasing order of the length of the discussion: the genitive and the ablative (2.5.1), the locative (2.5.2) and the dative (2.5.3), the latter of which will turn out to require a more in-depth look at Hittite (3).
25 Note that I reconstruct the genitive with *-ss-rather than with *-s-only on the basis of the genitival adjective, which probably shares its ultimate origin with the genitive. 26 The length in the ā-stems is based only on etymological considerations and may be anachronistic.

The genitive and the ablative
With the disconnection of the Luwian onomastic i-stems from the appellative ii̯ a/i-stems (2.1.1) and the concomitant rejection of contraction as an explanation for the appearance of -i-, which is once more confirmed by the corresponding paradigm in Lycian, the inflection of the onomastic i-stems and the parallel u-stems, in particular their failure to show the vowel historically inherent to the genitival forms and the ablative, requires a different historical explanation. Fortunately, it is not difficult to find such an explanation. The various onomastic paradigms are completely parallel. Of these paradigms, the one corresponding most closely to its appellative counterpart is that of the ā-stems, which show a difference only in the PN dative singular. Similarly, the o-stems only differ from their appellative counterpart in the PN dative singular and the locative. Incidentally, unlike in appellatives, in names, the ā-stems are the most frequent stem class, followed by the o-stems, whose counterpart in appellatives was annihilated by the process of i-mutation (cf. Norbruis 2018). These facts suggest that the onomastic i-stem and u-stem gen. and abl. were reshaped analogically after the ā-stems and the o-stems: *-i-osso, *-i-odi were replaced with *-i-sso, *-i-di, and likewise *-u-osso, *-u-odi with *-u-sso, *-u-di, after *-ā-sso, *-ā-di and *-o-sso, *-o-di.

The locative
The history of the locative is not as straightforward. One complicating factor is the mismatch with the state of affairs in appellatives. This, in turn, is complicated in itself because Luwian and Lycian do not match and because Lycian appears to display a morphological asymmetry. In Luwian appellatives, i-stems have a dat.-loc. -i, and a-stems have a dat.-loc. -a. In Lycian appellatives, the dat.-loc. of i-stems is -i, but that of the a-stems comes in two allomorphs: -i and -a. These seem to be lexically distributed; there are no lexemes that show both endings. The distribution is largely semantic: -i is used with animates (e.g. hrppi ladi 'for(/on) the wife'), -a with inanimates (e.g. ebehi xupa 'in this tomb', ẽnẽ periklehe xñtawata 'under the kingship of Pericles'), although there are also a few inanimates with -i (e.g. prñnawi 'in the grave', ẽti sttali 'on the stele', sixli 'for a shekel'). The main question is whether this allomorphy goes back to a Proto-Luwic distinction between dative and locative, which would suggest that the onomastic locative likewise goes back to a separate locative formation, or whether it was innovated, through the introduction of a variant -i, from a situation as found in Luwian, which only has the one dat.-loc. -a with a-stems.
In itself, the Lycian allomorphy lends itself well to being analysed as a remnant of an earlier distinction between dative and locative: the form originally accompanying the most frequent function (the dative with animates, the locative with inanimates) would then also have come to be used in the less characteristic function, effectively merging the categories into a dative-locative with two allomorphs. We could therefore reconstruct a PLuw. dative *-i (or perhaps *-āi, in view of CLuw. ᵈkamrušepai) next to a locative *-ā. 27 There are, however, several facts that speak against this scenario. Although it can explain the Lycian data, it creates additional assumptions for Luwian, which would then independently have merged the dative and locative into a dativelocative -and have chosen to generalise the locative ending -a rather than the dative *-i or *-āi for the designation of the merged case in the ā-stems (in analogy to the i-stem pattern?).
Moreover, the locative would have been a separate appellative case only in the ā-stems. There is no indication that there ever was a separate locative in the i-stems. Even synchronically in Lycian, the i-stems do not have a separate locative, but only a unified dative-locative -i (cf. e.g. ebehi xupa 'in this tomb', not **ebehe xupa; ẽtri ñtata 'in the lower burial-chamber', not **ẽtre ñtata), and this agrees with the situation in Luwian and in Hittite.
It can furthermore be understood why a unified dat.-loc. -a would have been in need of some degree of replacement or recharacterisation in Lycian: the plural counterpart of this ending, *-ās (which was created in analogy to the o-stem dat.loc.pl. *-os, Hitt. -aš), had lost its final *-s by sound change and had thus become identical to the singular (e.g. hrppi lada epptehe 'for their wives'). This may well have triggered an importation of the ending -i from the other stem types. There was no similar motivation in Luwian, which still had a distinct dat.-loc.pl. ending (-anz). The peculiar restriction of Lyc. -i to animates may perhaps be explained by the same factor: the desire to be able to distinguish number may have been more acute with animate referents. The lexemes with inanimate referents but with the ending -i, among which relatively recent loanwords like sttala 'stele' and sixla 'shekel', confirm that this was the more productive ending and that -a may be a residue from an earlier stage. A replacement scenario (*-a >> -i) can also straightforwardly explain the lack of a functional opposition, i.e. the fact that only one ending per lexeme is found.
Thus, the Lycian appellative a-stem (dative-)locative -a may well be a remnant of a Proto-Luwic dative-locative *-ā, which on the way to attested Lycian was partly (namely in animates) replaced by the -i as found in the other stems. Similarly, the occasional Lycian dative -a in personal names and the Luwian dative -a found in divine names (e.g. (DEUS)ku+AVIS-pa-pa) can be regarded as archaisms reflecting the stage before the pre-Proto-Luwic recharacterisation of the dative of personal names through the addition of *-i̯ o (on which more below). The same can then be assumed for the locative of toponyms.
If we assume that the ā-stem locative *-ā is the old dative-locative, with the innovations of the PN dative *-ā-i̯ o and later Lyc. -i leaving it mainly in locatival function, the main remaining explanandum is the shape of the Lyc. loc. ending -e (e.g. mukale 'on Mycale', xbide 'at Kaunos'), which, like -a in the a-stems, also occasionally occurs in datival function with personal names instead of the more common ending in -je, e.g. hrppi ladi ehbi uwiñte xumetijeh zzimazi (TL 120, 2), hrppi ladi ehbi tuhese (TL 113,2). The dat.-loc. of e-stems is expected to be -i rather than -e, as indeed it is in appellatives (cf. e.g. isbazi, dat.-loc. of isbazije-n. 'bench, couch', esedeñnewi, dat.-loc. of esedeñnewe-c. 'offspring'). Since there appears to have been only one dat.-loc. case, and the ending -i corresponds to the Luwian and Hittite endings, the ending -e is likely to be the result of analogy. The most obvious source for analogy is the a-stem (dat.-)loc. -a: -a -ã -ahe -adi -a = -e -ẽ -ehe -edi X → -e. There are several factors that may have favoured such an analogy. First, the a-stems were the most frequent onomastic stem type and were therefore a more logical source for analogy than they were in the appellatives; cf. the adaptation of the onomastic genitive and ablative after the a-stem pattern discussed above (2.5.1). Second, common gender e-stems were all but restricted to names and were therefore much more closely associated with the neighbouring onomastic a-stems than with their almost non-existent appellative counterparts. The ending *-i for the onomastic e-stems may well have felt out of place in comparison with the more frequent a-stem pattern in which the ending matched the stem, and have been adapted accordingly.
It is not surprising to find that the much less frequent toponymic i-stems follow the same pattern, at least in Lycian (-i -i -ihe *-idi -X → -i). For Luwian, we do not even have any certain attestations of an i-stem locative. However, if the dative of divine names can indeed historically be equated with the locative, it suggests a loc. *-u for u-stems, and by extension *-i for i-stems. See the treatment of the dative of personal names below for the original shape of the dat.-loc. that this *-i probably replaced (*-ii̯ o).
A final difficulty is presented by the s-stems (e.g. nom. trm̃mis, acc. trm̃misñ 'Lycia'), which appear to show a dat.-loc. in -e (e.g. nom. arñnas, dat.-loc. arñnase 'Xanthos'). This is not the only difficulty of this type, whose entire prehistory is shrouded in uncertainty. There is no corresponding type in Luwian. 28 On account of the dat.-loc., Melchert (2004: xi) analyses this type as stems in *-se-with syncope of the -e-. Whatever the exact mechanism, 29 it is in any case probable that these stems have undergone some form of formal innovation, indeed perhaps with *-seas a starting point. If it is rather the consonantal type of the direct cases that is original, the ending -e may have spread from the e-stems so as to avoid having an endingless form, which we would expect as a parallel to the other stems. The choice for the e-stem form may be related to the default status of the forms with -ein the appellative system (e.g. -ehe/i-, -edi everywhere except in the a-stems).
In sum, we seem to be dealing with the following developments. Pre-PLuw. had a dat.-loc. *-ā in the ā-stems and a dat.-loc. *-i in the o-stems. In personal names, these endings were largely replaced with *-ā-i̯ o and *-o-i̯ o, respectively (see below). The older endings remained possible variants in names but were now mainly restricted to locatival function (i.e. to toponyms). After the common gender o-stems had been annihilated in appellatives, the (dat.-)loc. *-i was in the 28 As far as the suffix -(V)s-is concerned, this may perhaps be compared to the Luwian suffix -izza-< PLuw. *-itts-ā-that creates ethnicon adjectives, e.g. CLuw. URU taurišizzaš (dat. URU taurišizza) 'from Tauris', HLuw. karkamis-izas (dat. karkamis-iza) 'from Karkamisa', which has been connected to PIE *-isḱo-or *-iḱo-(cf. Melchert 1989: 29f.;differently Melchert 2012: 207). 29 The type could in principle also be analogical after the genitive (nom. -Vh, acc. -Vhñ, dat.-loc. -Vhe) rather than the other way around (as proposed by Adiego Lajara 2010, cf. 6 below), but the morphology of the genitive seems to be too much in flux to be a good model. onomastic o-stems adapted to *-o in analogy to the pattern of the more frequent ā-stems. In the Lycian appellative a-stems the dat.-loc.sg. and the dat.-loc.pl. had become homophonous (-a), and the singular was recharacterised with the ending -i from the other appellative types, with the older ending -a being left as a residue with inanimates.

The dative of personal names
This leaves the dative in *-i̯ o, whose shape is completely unlike that of its appellative counterpart. There is only one possible comparandum within Luwic. The dative-locative of the Luwian appellative ii̯ a/i-stems (as in tadii̯ a/i-'of father') usually has the morphologically expected shape -i (tadi), but possibly there also exists a variant -ii̯ a (tadii̯ a, see section 5). Yakubovich (2015: §6.2), who was only aware of the onomastic ending -i̯ a for a-stems, proposed that the onomastic ending might be analogical after this ii̯ a/i-stem dative-locative variant -ii̯ a. The analogy would then have to be -is : -in : -ii̯ a = -as : -an : X → -ai̯ a. Even if we adjust this by replacing -awith -V-to include the other stems, in accordance with the paradigms as established above, this proposed analogy runs into various problems. First, within Luwic, this is quite an obscure ending, restricted to the ii̯ a/i-stems, and all but ousted by the productive ending -i -indeed its very existence is not completely certain (see 5). It would in any case not have been a powerful model for an analogy. This is even more acute considering that it would have to have induced an apparently unmotivated analogy. Most tellingly, in this scenario it would not be understandable why the spread of the ending was restricted to personal names, whereas the appellative system, which even harbours the purported source of the analogy, remained unaffected. Therefore, I reject the (potential) ii̯ a/i-stem dative-locative variant ending -ii̯ a as a possible source of the onomastic dative.
The lack of other comparanda within Luwic impels us to look beyond its borders. In Hittite, the inflected shapes of names are often concealed due to the common practice of akkadographic writing, which results in writing only the bare stem, in the dative typically preceded by ANA, rather than the full form. There are exceptions, however, which allow us to discern the paradigm presented in Table  19 (  The inflection of these i-stem names is strikingly similar to that of the Luwic onomastic i-stems (*-is, *-in, *-ii̯ o, *-is°<< *-ii̯ os°), likewise featuring -i-throughout, and, promisingly, a dative of the exact same shape. Therefore, I propose equating the two paradigms, including their peculiar datives, historically. Fortunately, within Hittite, this dative ending is not isolated, and we can put it into context and trace its origin. This is what I will do in the next section.

The Hittite dative-locative of i-stems and ii̯ a-stems
In Hittite, unlike in Luwic, names and appellatives have similar inflections. The ending -ii̯ a appears in the paradigms of the names tabulated in Table 19 because these are non-ablauting i-stems. The non-ablauting i-stems are among the main loci of the ending -ii̯ a, together with ai/i-stems and ii̯ a-stems. The paradigms of these types (restricted to the singular) are given in Table 20, where I also include the i/ai-stems, a similar stem type in which the dat.-loc. in -ii̯ a is conspicuously absent (more on this in 3.2). The ending is used both in datival and in locatival functions.
-ii̯ aš -ii̯ aš -ii̯ aš -aš, -ai̯ aš instr. - It is apparent from the overview that the dat.-loc. -ii̯ a, in all stem types in which it occurs, is in competition with -ī and -i, which are morphologically transparent: they result from the combination of the -i-of the stem and the dat.-loc. ending -i. We also notice that the alternative dat.-loc. ending -ii̯ a is identical to the allative ending. For the allative, the form -ii̯ a is morphologically expected: it results from a combination of the -i-of the stem and the allative ending -a. This suggests, as is also commonly thought, that the dat.-loc. ending variant -ii̯ a is originally the allative ending, whose function was extended to the domain of the dative-locative at the expense of the dat.-loc. ending -i (cf. Laroche 1970: 33). A reason for this replacement that has been put forward is that the latter ending had become blurred due to its identical shape to the preceding stem vowel. This scenario has recently been contested by Frantíková (2016). Also, the exact distribution of the various forms has been the subject of some confusion. These issues will be discussed in the following sections.

The distribution of the dat.-loc. -ii̯ a
In their grammar of Hittite, Hoffner & Melchert (2008: 87) Frantíková (2016: 188-191) adds: ubatii̯ a 'on the land' (ubati-), utnii̯ a 'in the country' (utne-), ḫuu̯ ašii̯ a 'at the ḫ.-pillar' (ḫuu̯ aši-). 32 Frantíková (2016: 188f.) concludes that "the locatival -a is found in a number of instances" in OH. The impression remains that this is a marginal phenomenon. Indeed, Frantíková (2016: 193) explicitly states that "the -a ending is used only in a few dozen i-stem lexemes (the overall number of i-stem nouns and adjectives exceeds a thousand)." She also speaks of "the scarcity of its occurrences and its even distribution throughout the recorded history of Hittite" (Frantíková 2016: 195).
A more systematic approach leads to a different picture. Table 21 is intended to be an exhaustive collection of attested dat.-loc.sg. forms (NB not including -ii̯ a in allatival function) of the relevant stem types in OS and OH/MS, whether of the shape -ii̯ a, -ī or -i.  generally regarded as loanwords (for GIŠ ḫalpūti-, the source is also identifiable as Hattic). All other instances of -ī are from a later period. This suggests that the ending -ii̯ a received some competition from the paradigmatically expected form -ī in later Hittite, when the lack of an overt ending was apparently less universally regarded as problematic. 46 The fact that many i-stem lexemes do not exhibit the ending -ii̯ a, then, is not because of lexical restrictions, but due to the limitations of our corpus: the overview suggests that these i-stems, too, had (or would have had) a dat.-loc.
-ii̯ a in OH.

The origin of the dat.-loc.-ii̯ a
The origin of the dat.-loc. -ii̯ a is transparent. As was mentioned above (section 3), it is identical to the allative, -ii̯ a, where this shape is morphologically expected. Therefore, the straightforward scenario is that the allative form was used instead of the expected dative-locative form in the relevant stem types to express the dative-locative function. This is semantically unproblematic, as the domains of the allative and the dative-locative are very close. The motivation for this slight semantic stretch of the allative is also clear. The use of the allative form in dativelocative function is restricted to stems in -i-, -ai/i-, -ē/i-and -ii̯ a-. These share the formal feature that the oblique case endings attach immediately to a stem-inherent -i-. This formal distribution shows that the motivation behind the existence of the dat.-loc. -ii̯ a must be related to this formal feature, and it is not difficult to find it: the morphologically expected combination of the stem-inherent -i-and the dative-locative ending -i leads to a clash of identical phonemes. This was apparently so undesired that speakers preferred an alternative, which they found in the semantically close allative. This analysis is confirmed by the fact that the use of the allative form to express dat.-loc. function is conspicuously absent from the i/ai-stems (see Table 20): the oblique stem of this type does not have -i-, but -a(i̯ )-, and thus it features a characterised dat.-loc. -ai.
Frantíková's objections to such a scenario and her consequent aporia about the origin of the dat.-loc. -ii̯ a are unwarranted. She predicts that if the motivation behind the use of -ii̯ a instead of -i was to disambiguate, neuters should exhibit -ii̯ a more often because they also have an identical nom.-acc.sg. in -i, which adds to the ambiguity. However, in the scenario above, the only ambiguity that is being removed by the use of the allative form is that resulting from the clash of a stem vowel -i-with the dative-locative ending -i. The allative is used in order to have a dative-locative marker at all, rather than one that has disappeared due to the previous vowel. No disambiguation with other forms in the paradigm is implied in this explanation, so Frantíková's expectation that neuters would have shown the ending -ii̯ a more often does not apply. Neither is it a counterargument that OH already has examples of -ii̯ a in dat.-loc. function. Indeed, the allative could only be extended in function at a point in which it was still alive. Finally, the supposition that the dat.-loc. ending -a would have spread to other stems (Frantíková 2016: 191) is not justified, because these did not have the same formal problem which this form was created to solve.
The use of the dat.-loc. in -ii̯ a is at its peak in the oldest stage of Hittite and only decreases with time. This means that the functional extension of the allative by which it arose must be placed in prehistory: in pre-Hittite.

The origin of the Luwic onomastic dative
From the investigation into the status of the Hittite dat.-loc. ending -ii̯ a in the previous section, it is apparent that this ending must have come into being before our earliest records, meaning that it may be compared with Luwic data to see if it may be of Proto-Anatolian date. Since the i-stem type corresponding to the Hittite i/ai-stems was generalised in the Luwic appellative system, the main Luwic comparandum for the Hittite stems with -i-in the oblique stem, the locus of the dative-locative in -ii̯ a, are the onomastic i-stems. This leads us back to the identification in 2.5.3. The fact that we find exactly the ending *-ii̯ o (Luwian -ii̯ a, Lycian -ije) shows that it was there already in Proto-Anatolian, cf. Table 22.
For Luwic, the identification suggests that the dative of the personal name declension was inherited as such in the i-stems. 47 On the basis of Hittite (3.2), we now know that it was originally restricted to the i-stems, where it was borrowed from the semantically neighbouring allative to remedy the clash of the -i-of the stem and the normal dative-locative ending -i. This suggests that the other Luwic onomastic stems received the ending *-i̯ o analogically. Specifically, *-is : *-in : *-isso : *-ii̯ o = *-Vs : *-Vn : *-Vsso : X, which resolves into the reconstructable forms *-āi̯ o, *-oi̯ o and *-ui̯ o. After the generalisation of the ablauting i-stems in nouns and adjectives, the non-ablauting i-stems survived only in the onomastic system, especially personal names, and their isolated dative in *-ii̯ o had become one of their characteristics. Its spread to the other PN stem types, showing the embracing of this characteristic, created parallelism in what had probably been a mixed bag of forms (*-ā, *-i, *-ii̯ o, *-ui), leading to the unification of the PN declension pattern, which was realised in conjunction with the generalisation of the ā-stem pattern in the other oblique cases (2.5). That *-ii̯ o became characteristic of personal names, but not of toponyms, which would originally have had the same dat.-loc., may be understood from the much higher frequency of i-stems in personal names. In toponyms, *-ii̯ o was itself replaced with the ā-stem pattern, leading to *-i. Of course, the morphological analysis had originally been *-i-o, with *-i̯ -appearing only as an automatic glide, resulting in *-ii̯ o. The analogy suggests that this was reanalysed as *-i-i̯ o. 48 This reanalysis could easily happen in Luwic, where the form was no longer associated with an allative, causing the morphological boundary to become opaque. The analogy neatly explains the exceptional occurrence of intervocalic *-i̯ -after other vowels than *-i-. It suggests that the *i̯ was phonemic, unlike in Proto-Anatolian. For Proto-Luwic, we can indeed reconstruct a contrast between *i and *i̯ . 49 For example, the dative ending *-Vi̯ o contrasts with *-Vii̯ o-, which resulted from the addition of the appurtenance suffix *-ii̯ o/i-to vocalic bases, as for example in Lyc. adaije-(to ade-, a unit of money), contrasting with the onomastic a-stem dative -aje. The *i̯ had probably been phonemicised through the development *ǵ (h) > *i̯ (> ∅), e.g. *ǵʰes-r-'hand' (Hitt. keššar) > CLuw. i-iš-sa-ri-(does i-still spell i̯ -?), HLuw. istri-, Lyc. izri-. 50 47 Cf. in essence already Laroche 1970: 32 andHajnal 1995: 93f. 48 For such a reanalysis cf. e.g. the Spanish 1-3sg.poss.pron. mío, tuyo, suyo, and similarly Neapolitan mio, tuio, suio, from an ancestral state as still found in Italian mio, tuo, suo, with generalisation of the automatic glide after i in mio. 49 Contra Kloekhorst's (2008b: 123f.) analysis of Lycian j as an allophone of i. 50 Cf. also CLuw. ku-um-ma-i-in-zi = kummai̯ inzi. Unless orthographic limitations hide a continued differentiation, sequences of the shape *Vii̯ V were apparently simplified to *Vi̯ V in Luwian. Cf. Lyc. ebeija (virtual *h 1 obʰo-ii̯ eh₂) vs. HLuw. ápaya (and likewise zaya < *ḱo-ii̯ eh₂).

The Luwian dat.-loc. of the genitival adjective
With the identification of the Hittite and Luwic i-stem paradigms above, the practice of using the allative ending in dative-locative function in stems in -i-reveals itself 51 Yakubovich (2015: §6.2) analyses the ii̯ a/i-stems as partly contracting a-stems, and accordingly, the dative -ii̯ a as containing the a-stem dative ending -a. This is certainly not correct: the a-stems (< *ā-stems, Lyc. a-stems) should be kept separate from the ii̯ a/i-stems (< *ii̯ o/i-stems, Lyc. ije/i-stems). 52 The regular ending -i is sometimes seen as a contraction of -ii̯ a (CHLI: 120; Yakubovich 2015: §6.2). However, it can hardly be a coincidence that -i is also the morphologically expected form, resulting from a combination of the stem -i(i̯ )-and the normal dat.-loc. ending -i. Indeed, the CLuw. spelling°Ci-i points directly to a preform *-ii̯ i. The ending -i therefore rather results from morphological regularisation: like in Hittite, the use of the morphologically aberrant form *-ii̯ o was at some point no longer preferred over the use of the morphologically expected form.
to be Proto-Anatolian. One unexpected side-effect of this is that it provides us with an explanation for the enigmatic Luwian dative-locative of the genitival adjective, marked in bold in Table 23. nom.
-ass-in -ass-inz dat.-loc. -ass-an -ass-anz abl. -ass-adi The Luwian genitival adjective suffix -assa/i-is a regular a/i-stem in all respects except the dat.-loc. singular, which has the completely unexpected shape -an rather than -i. It was explained by Morpurgo Davies (1980: 135-137) as resulting from an analogy with the accusative and the plural: *-ass-inz : *-ass-in = *-ass-anz : X → *-ass-an. While this is plausible in itself, it remains unclear why this analogy happened only in the genitival adjective, and not also in all other (a/)i-stems, and what triggered the analogy. Morpurgo Davies' assumption that it disambiguated the dat-loc.sg. of the gen.adj. from the genitive in -asi can no longer be upheld in view of the secondary, dialectal character of -asi (Palmér 2021), whereas -an goes back to Proto-Luwian. A consensus is emerging that the only formally and etymologically plausible reconstruction of the genitival adjective is *-osio-, an inflecting pendant to the IE gen. *-osio (see e.g. EDHIL: s.v. -ašša-;Melchert 2012: 282;Sasseville 2018: 315). If we reconstruct the expected Proto-Anatolian paradigm of this suffix, crucially with a dative-locative *-o after *-i-in line with the analysis above, we end up with the paradigm presented in Table 24. nom. *-osios acc.
*-osiodi After *-si-> *-ss-and the spread of the i-stem direct case endings, we get the picture presented in Table 25. nom. *-ossis acc.
*-ossin dat.-loc. *-osso abl. *-ossodi At this point, the *-i-had been swallowed by the preceding *-s-, leaving the remaining dative-locative ending *-o isolated. Now the analogy proposed by Morpurgo Davies can be understood as an attempt to make sense of this *-o. The dat.-loc. *-osso was partly identical to its plural counterpart *-ossonts (a Luwian adaptation of *-ossos) but missed a final *-n in comparison to the similar accusative pair *-ossin : *-ossints, which followed a familiar pattern. This scenario provides a motivation for the analogy and explains its restriction to just this suffix. If the connection between the Luwian ending -an and the alternative dative-locative ending *-o is accepted, its implication of a preceding *-i-definitively settles the reconstruction of the suffix on *-osio-. 53

The Proto-Anatolian allative
The analysis above not only sheds light on the origins of the Luwic dative(-locative) in the onomastic inflection and in the appellative suffixes *-ii̯ o/i-and *-osso/i-, but also has consequences for our reconstruction of Proto-Anatolian, specifically for the reconstruction of the allative. Anatolian *-i-o (*-ii̯ o), with -o on account of Lycian -e. Since this is originally the allative of stems in *-i-, it follows that the Proto-Anatolian allative ending was *-o.
Reconstructions of the allative have taken all shapes that Hittite -a, -ā could theoretically go back to (and even some to which it could not), most notably *-o, *-eh₂ and *-h₂e, all of which still feature prominently in the literature, with *-eh₂ topping the list. The most recent cases were made by Melchert (2017), for *-eh₂, and Villanueva Svensson (2018), for *-h₂e. Both regard the Lycian infinitive as the only inner-Anatolian evidence that has any bearing on the vowel quality of the allative, which they identify as a (Melchert 2017: 535;Villanueva Svensson 2018: 147).
Unfortunately, the infinitive ending cannot carry the weight it has been given. Problematically, according to the current communis opinio, this ending comes in two shapes: -ne and -na. 54 Although it is indeed quite likely that the allative ending is continued in the vowel of one of these formations, 55 it is on the basis of the infinitive data alone absolutely unclear whether it should be the one in -ne or the one in -na. Melchert (2017: 535;cf. already 1994: 325) speculates that -na continues the 'genuine' consonant stem ending, i.e. *-eh₂, while -e was reshaped after the supposed o-stem ending, *-o-h₂. This scenario is extremely problematic. Since the grammaticalisation into an infinitive must have happened before Proto-Luwic, we expect it to have been crystallised as such by Lycian times and not to have undergone any analogy on the basis of a continued analysis as an allative. Indeed, since Proto-Luwic, never mind Lycian, no longer featured the allative case, an innovation based on the allative is quite impossible at these stages. If the spread is supposed to have happened in pre-Proto-Luwic, some two millennia later we should expect any free variation to have been ironed out.
A priori, a much more likely scenario is that -ne and -na were made with different morphemes. This idea is confirmed by the remarkable fact that almost all attestations of -na occur beside an occurrence of -ne in the same inscription, which strongly suggests that there was a synchronic distribution. Since there does not seem to be a phonetic distribution, it is likely that the distribution was functional. Unfortunately, our scarce data do not allow us to grasp the syntactic and semantic 54 For an overview of the forms and their contexts, see Serangeli 2019. A marginal third variant in -ni has traditionally been assumed, but the only alleged example in a non-broken context is rather interpreted as a mediopassive form by Sasseville (2020a: 58f.). 55 The Luwic infinitive is based on the Proto-Anatolian verbal noun suffix continued in Hitt. -u̯ ar, -u̯ aš < *-ur, *-uen-s; in Luwian it has the shape -una, e.g. CLuw. karš-una 'to cut', HLuw. ád-una 'to eat'. On the basis of the parallel that Hittite offers (inf. -anna < *-ot-n-+ all., based on the verbal noun suffix -ātar, -annaš < *-ót-r, *-ot-n-os), and the general typological likelihood of the development of an infinitive from a form with allatival function (cf. e.g. Eng. to …; see Kuteva et al. 2019: 49, 351f.), an analysis as *-un-plus the allative ending is quite plausible. details. We cannot pretend to understand all details of TL 44a, which contains all cases of -na in unbroken context. At most, the restricted distribution of -na is itself noteworthy. Six out of seven attestations of -na occur in only two inscriptions, TL 44a (4×) and TL 29 (2×), which are also exceptional for containing a (military) narrative. This may not be coincidental. The function of -na may have been more in the realm of a participle or a verbal noun, perhaps comparable to the English ing-forms. This would make sense for a formation in -a, a suffix which, among other things, is used to make abstract nouns, cf. e.g. xñtawati-'king'~xñtawata-'kingship'. 56 I would therefore tentatively interpret -na historically as *-un-plus the suffix *-eh₂-, used in the dative-locative ('in (the process of) …-ing'). Perhaps the form in -a was even directly based on the infinitive.
The upshot is that one simply cannot use -na to infer that the allative had a-character. If anything, the regular infinitive is that in -ne, which points to o-character. More importantly, however, since the morphological and syntactic details behind the variation in the shape of the infinitive are essentially unclear, both synchronically and certainly diachronically, we should let any conclusion based on the infinitive be overruled by the unambiguous evidence for the shape *-o provided by the onomastic dative. Indeed, we may use this evidence to conclude that the infinitive in -ne is the one that goes back to the allative.
The situation with the alleged extra-Anatolian comparanda is comparable. Many mutually exclusive putative remnants have been identified in other Indo-European branches. They cannot all be correct. The analysis above is clear evidence that the reconstruction must be *-o, and that reconstructions with a-character are incorrect. Villanueva Svensson's (2018: 148) assertion that "potential extra-Anatolian cognates come as "*-ai" (…), "*-a" (…), and "*-ō" (…)" which "seems to rule out reconstructions involving only *-o (…) or only *-a (…)" is a non sequitur: this would only be the case if the extra-Anatolian cognates pointing to a-character were compelling rather than only potential, and if better available evidence, namely in favour of *-o, which is somehow left out of the equation here, were not incompatible with a-character.
I will briefly discuss some of the main motivations for reconstructing a-character for the Proto-Anatolian allative. One of the most popular is Gr. χαμαί 'on/to the ground' (Melchert 2017: 535). This is clearly not a form in -η or -α, but in -αι, with an -ι that has been analysed as an additional locative ending. While the assumed accumulation of endings is not obvious to begin with, more importantly, this analysis means that the locatival semantics could be entirely due to the added -ι. 57 The same is true for the Greek infinitive in -ναι, which must also contain the locative ending -ι, attached to an ᾱ-stem abstract noun (see Rix 1992: 238). Greek adverbs in -α such as ἀνά 'up along', ἅμα 'together', ἄντα 'over against, face to face', ἔνθα 'there', κατά 'down(wards) from', παρά 'from the side of', 58 etc., not only often do not have allatival meaning at all, but can also not be formally united with the Anatolian allative: in terms of reconstructions with *h₂, Gr. -α could only go back to *-h₂ or *-h₂e, whereas Hitt. -a, -ā would require *-eh₂ or *-oh₂. This can hardly be justified morphologically. 59 Moreover, a more straightforward and plausible interpretation is that Gr. -α goes back to the accusative ending *-m̥ (cf. e.g. ἄντα 'over against'~ἔναντα 'opposite, over against', ἄντην 'against, over against'; κατά 'downwards'~Hitt. kattan 'downwards' < *ḱmt-m). Even more tenuous is the contention that the allative can be distilled from Hitt. menaḫḫanda 'against, opposite, before, facing' "< *menaḫ anda 'in(to) the face' " and Lith. žmogùs 'man' "< *dʰǵʰm-eh₂-gʷ(h₂)u-'one who walks on the earth' " (Kim 2013: 122f. with lit.), or < "*dʰǵʰm-oh₂a-gʷh₂u-" (Villanueva Svensson 2017: 135). The implied univerbation with an intact case form is a rarely seen process, and more straightforward explanations should be preferred. Hitt. menaḫḫanda is rather to be analysed as a compound of mēna-'face' and ḫant-'face, forehead' (see EDHIL: s.v.; for ḫanda cf. also Kloekhorst 2010: 223-225). The formation of Lith. žmogùs 'man' is unclear, and even in the unlikely univerbation scenario, the -o-element does not have an allatival meaning. The -o-also occurs in žmónės 'people', and may have a completely different origin (see EDBIL: s.vv.). 57 It is in fact quite possible that the whole sequence -αι in χαμαί is analogical. An unexpected -α-also shows up in χαμᾶζε 'to the ground', the actual functional equivalent of the allative. The allative in -δε is typically built to the accusative, with -ζε resulting from the combination with the -ς of the accusative plural. However, an acc.pl. **χαμᾱ́ς does not exist. It is therefore likely that the element -αζε was taken over in its entirety from a source in which it was at home, such as the type of θύραζε and Ἀθήναζε (DELG²: s.v. χαμαί, EDG: s.v. χαμαί), or the other archaic word for 'earth', which made it to the historical period chiefly in the shape of the petrified allative ἔραζε 'to the ground'. The expected locative of the latter lexeme is *ἔραι, which may similarly have contributed to the creation of χαμαί. Whatever the correct scenario, it is clear that no sound argument regarding the allative can be based on χαμαί. 58 Specifically, in order of frequency, '(+ gen.) from (the side of); (+ dat.) by the side of, at; (+ acc.) beside, along, past' (see LSJ⁹: s.v.). Note that the meaning is not allatival. 59 Note that the idea that Hitt. -ā would represent an o-stem variant "*-oh₂" is furthermore contradicted by the data: we only find -ā in consonant stems, whereas the o-stems only attest -a. It is highly improbable that such archaic paradigms as that of keššar 'hand' (allative kišrā) and tēkan 'earth' (allative taknā), much less petrified allatives such as parā 'forward', took their allative endings from the o-stems (and this idea is indeed shown to be incorrect by the clear correspondences of parā < *pró). the allatival adverb Hitt. anda, CLuw. ānta, Lyc. ñte < *h 1 ndo, and by the regular Lyc. infinitive in -ne < *-un-o.
Traditionally, the allative is not reconstructed for PIE, but this seems to be changing (cf. e.g. Fortson 2010: 117; Ringe 2017: 25f.; Kloekhorst & Pronk 2019b: 4;Bauhaus 2019: 24f.). As an argument against an archaism, one could object that the accusative seems to be an older device for expressing allatival function, as in Lat. eō domum 'I go home', a construction that may well be taken to suggest that the accusative originated from the grammaticalisation of an allative to a direct object marker (cf. Sp. veo a Juan 'I see Juan', with use of the allatival preposition a 'to'). However, this is not necessarily the right scenario. Although grammaticalisation from an allative to a direct object marker is indeed a plausible development, the opposite is as well. The development from a direct object marker to an allatival marker is completely natural with verbs of going: as a direct object marker normally expresses what an action is directed towards, the combination with a verb of going naturally leads to a goal interpretation. Such a development happened, for example, in Modern Greek, cf. e.g. πάω σπίτι 'I go home', πάω Ελλάδα 'I go to Greece', πάω σουπερμάρκετ 'I go to the supermarket', etc. (see e.g. Holton et al. 2012: 335). Like in Greek, where direction is more usually expressed with the preposition σε 'to; in', the PIE accusative of direction, which is also marginally attested in Hittite (Hoffner & Melchert 2008: 248f.), may always have been a marginal phenomenon. 62 In my view, the PIE formations with petrified allatives such as *pr-o, *h₂p-o, *up-o, etc., can only have been formed when the creation of such allatives was productive. The state of affairs in non-Anatolian IE therefore already suggests that there once was a more vigorous allative. Since no non-Anatolian language shows any evidence for this case except for remnants in petrified adverbs, the stage in which the allative was a regular case must predate their common ancestor, in which it had been lost as such. The fact that we find a vigorous allative of exactly the right shape in Hittite can hardly be interpreted in any other way than that Anatolian descends from this earlier stage in which the allative still was a vigorous case. Therefore, the allative is an argument in favour of the Indo-Anatolian hypothesis.
62 Another critical thought could be that spatial cases can easily be secondary, as for example in Baltic. While the allative could indeed in principle have been secondary and must of course have come into being at some point in time, the remnants in non-Anatolian IE clearly favour a scenario in which the allative did already exist in PIE but was lost on the way to the common ancestor of non-Anatolian IE. Baltic also offers a parallel for the opposite development, by which an allative case was lost as such and only survived in scattered remnants. For example, the Old Lithuanian allative in -p survives only in a few petrified expressions in Modern Lithuanian, such as the adverb vakarop 'towards the evening'.