THE PLACE OF LYDIAN IN THE ANATOLIAN FAMILY THROUGH THE LENS OF RECENT RESEARCH Текст научной статьи по специальности «Языкознание и литературоведение»

Ilya Yakubovich

University of Marburg, Germany; sogdiana783@gmail.com

The Place of Lydian in the Anatolian Family through the Lens of Recent Research

The Lydian language is traditionally classified as belonging to the Anatolian group of the Indo-European languages, but its further genetic and area! connections represent a subject of much debate. Recent advances in the synchronic interpretation of Lydian triggered the expression of radically opposing views on the filiation of this small-corpus language. While some scholars stress its isolated character within the Anatolian group, or even doubt its Anatolian character, others voice support for its membership in the Luwic subgroup (whose prominent members are Luwian, Lycian A, Lycian B, and Carian). As a rule, however, such claims are based on privileging particular aspects of the Lydian grammar. The goal of this paper is to provide a preliminary assessment of this controversy by way of summarizing the recent progress in Lydian linguistics and its phylogenetic implications. It is concluded that the hypothesis about the non-Anatolian origin of Lydian is rooted not in new empirical evidence but in attempts to dismiss some of the old etymologies without offering better alternatives. As for the new iso-glosses linking Lydian and the Luwic languages, some of them are cogent, but their interpretation in terms of areal diffusion appears preferable to treating Lydian as a Luwic language.

Keywords: Lydian language; Anatolian languages; Luwic languages; linguistic phylogeny; homoplasy; areal diffusion.

1. Introduction

The Lydian language is attested through the inscriptions excavated in Sardes and other locations in the western part of Asia Minor, which date back in their majority to the 7th-3rd centuries BCE (Gérard 2005: 20)1 As is clear from the numismatic evidence, this was the official language of the Lydian kingdom at the time when it was ruled by the Mermnad dynasty (c. 680547 BCE), even though the bulk of the written sources postdate its collapse and the incorporation of western Anatolia into the Achaemenid Empire. The interpretation of the Lydian language has been steadily progressing over the last hundred or so years, being in part facilitated by the availability of the short Lydian-Greek and Lydian-Aramaic bilinguals. The majority of the understood texts represent epitaphs, even though the Lydian corpus also includes votive texts, religious decrees, and ownership inscriptions.

There is a broad consensus among modern scholars, according to which Lydian belongs to the Anatolian group of the Indo-European (Indo-Anatolian) languages (Kassian & Yakubovich 2013, Melchert 2017, Rieken 2017, Kloekhorst 2022)2. As for the precise place of Lydian on

1 The work on the present paper was conducted in connection with teaching a course on the Lydian language at the University of Marburg in the summer semester of 2022. I am grateful to Christian Bruns, Jonas Doll, and Jens Goeckler for many insightful questions that helped me to sharpen my argumentation. H. Craig Melchert (Carrboro, NC) and Norbert Oettinger (Erlangen) were kind enough to read this manuscript and made useful comments, while Stephen Durnford (Brighton) and Diether Schurr (Hanau) did their best to improve its style on a very short notice. This said, I am alone responsible for any errors found below.

2 The difference between the terms "Indo-European" and "Indo-Anatolian" concerns the status of the Anatolian languages vis-à-vis the rest of the family. The reason for this debate is the incongruence between the gramJournal of Language Relationship • Вопросы языкового родства • 20/3 (2022) • Pp. 191-221 • © Ilya Yakubovich, 2022

the phylogenetic tree of the Anatolian languages, the mostly clearly articulated hypothesis goes back to Oettinger 1978. According to this article, the Lydian language was the second one after Hittite to split from Proto-Anatolian, thus predating the split between Palaic and what we nowadays call the languages of the Luwic group. Oettinger's view remains influential in present scholarship and was followed in a number of recent publications, e.g., Yakubovich 2010: 6, Kassian & Yakubovich 2013: 17, Kloekhorst 2022 passim). This is, however, merely one opinion, which now competes with two diametrically opposing points of view.

On the one hand, Lydian philology is objectively less integrated into the field of Anatolian studies when compared with the study of Hittite or the languages of the Luwic group. No ancestor of the Lydian language is attested in cuneiform transmission, and it is rarely studied as a part of the Hittitological curriculum. This relative isolation is naturally conducive to enquiries questioning the relationship between Lydian and the rest of the Anatolian family. Thus, Melchert (2003a: 267) considers three hypotheses regarding the genetic position of Lydian and formulates the third one as follows: "Lydian is not derived from the common prestage we define as Proto-Anatolian but is an independent IE branch in Anatolia (like the later Phrygian). What features it shares with Anatolian (most notably Luvian and Lycian) would be entirely due to convergence through contact once these languages spread into western Asia Minor". Oreshko (2019: 228) is bolder in his assertion, which is presented as his preferred viewpoint: "It is quite possible - or even probable - that Lydian is in its origin not an Anatolian language, but an Indo-European language belonging to a different branch which appeared in Anatolia somewhat later than Hittite and Luwian and subsequently absorbed some Anatolian features". While the points of view illustrated above never gained mainstream status in Lydian studies, they bear witness to the lingering anxiety regarding the genetic attribution of the Lydian language.

On the other hand, the 21st century saw an increasing attention toward the ethnic prehistory of the Lydians, in particular as regards the relationship between the ethnonyms Luwian and Lydian. Three papers published roughly about the same time, namely Beekes 2003, Gérard 2004 and Widmer 2004, concur in treating the ethnonym AuSioç attested in Greek transmission as an adaptation of Luwiya, the homeland of the Luwians attested in Hittite sources, even though the phonetic details of the proposed derivations exhibit minor differences. The former of these papers explicates its rationale as follows: "It is generally assumed that western Asia Minor was originally to a large extent - if not completely - Luwian. It is remarkable, however, that the name Luwian does not live on: Greek sources have not yielded a name that resembles the word Luwian" (Beekes 2003: 47). In the meanwhile, the dominance of Luwians in western Anatolia is no longer taken as an axiom, while an additional distinction is made between Luwian in the narrow sense, and the languages of the Luwic subgroup, such as Lycian A, Lycian B, and Carian, which were indeed predominantly spoken in the western part

matical profiles of the Anatolian languages, primarily Hittite, deciphered over the 20th century, and the Neo-Grammarian Indo-European reconstruction, which had emerged in the late 19th century. While the dominant trend in the 20th century consisted in accommodating Anatolian as one of the daughter languages of Proto-Indo-European (modifying the reconstruction of the proto-language if needed), an increasing majority of scholars in the 21st century view the Anatolian and Indo-European languages as two branches derived from the common ancestor but characterized by the innovations of their own. Although not everyone within the latter group insists on the necessity of assigning a separate name to this ancestor language, if one opts for such a name, then "Indo-Anatolian" emerges as a logical candidate. At least, this designation, advanced first in Yakubovich 2010: 3, has the advantage over "Indo-Hittite", which has a more impressive pedigree but unduly privileges one of the members of the Anatolian group. It must, however, be stressed that this debate, for all its importance for Indo-European taxonomy, has no direct bearing on the main topic of the present paper.

of Asia Minor (Melchert 2003b: 177, n. 7). While a degree of Luwian presence in western Anatolia in the late second millennium BCE is hardly to be doubted, and Luwian superstrate may indeed account for the ethnonym Lydian, this is obviously not the same as asserting that Lydian belongs to the Luwic group.

Nevertheless, claims have also been made that the Lydian language displays a particularly close relationship to the Luwic group on the genetic level. A recent example of such a proposal is found in Rieken 2017: 303, where a number of common innovations shared by Luwian, Lycian, and Lydian were postulated.3 Rieken (ibid.) concludes that "Lydian, in spite of its otherwise "strange looks", belongs to the Southern Anatolian group", where "South Anatolian" appears to represent something similar to what is called "Luwic" in other sources. The same hypothesis is formulated as a question in Sasseville 2020: 2: "While the issue of how Lu-wic the Lydian language is remains unsolved among Anatolianists, we aim at testing this hypothesis through the lens of their verbal stem formations". The data analysis in the same volume prompts its author to accept a number of common innovations in the Luwic and Lydian verbal derivation. On the strength of this evidence, he comes to the conclusion that "Lydian is, regarding the development of its verbal stem classes, a Luwic "dialect"" (Sasseville 2020: 551).

At the same time, there were recent attempts to demonstrate a horizontal transfer of both lexical and structural features from individual Luwic dialects to Lydian. Thus, Yakubovich (2010: 97) summarized the evidence for the Lydian proper names of Luwic origin gleaned from the earlier works of Onofrio Carruba, Roberto Gusmani and H. Craig Melchert. It includes the divine names Santa, Kufaw and Mariwda as well as personal names Walwet and Katowa4. As long as the proposed contact effects do not include the transfer of appellatives or grammatical interference, one could easily dismiss them as marginal phenomena, which bear witness to no more than superficial contacts (cf. Melchert 2020: 246-247). Nevertheless, the twenty-first century brought additional evidence for Luwic lexical borrowings in Lydian, e.g., masta- 'decoration' (Rieken & Yakubovich 2020) or qaAmu- 'king' (Valerio & Yakubovich 2022). Furthermore, Sasseville (2021: 647) pleads for a connection between the lenition of *-d- in Late Luwian and Lydian, arguing that it reflects "a diffusion area (or Sprachbund) between South-Central and West Anatolia, i.e., between the speech communities of the Neo-Hittite states and the predecessors of the Lydian Empire". A vaguer approach was advocated fifteen years earlier in connection with the non-trivial similarities between the Luwian and Lydian prefixes: "Future research will show whether these similarities reflect secondary contacts between Lydian and Lydian, betray their original dialectal isoglosses, or are pure coincidences" (Yaku-bovich 2005: 79).

It is obvious that both the treatment of Lydian as a non-Anatolian language and its attribution to the Luwic group are incompatible with the phylogenetic conclusions going back to Oettinger 1978. Although the presence of binary contacts encompassing Lydian and other lan-

3 The two alleged common innovations of Lydian and the Luwic languages regarded as significant in Rieken 2017: 303 are the 1sg. pres. *-wi and nom. pl. *-insi. Regarding the first innovation one can only say that it is non-Hittite, as the relevant verbal ending in Palaic remains unknown (cf. Kloekhorst 2022: 73). The assumption of the second innovation is based on the discussion in Gérard 2005: 80-82, where only select nominative plural endings of the common gender have been treated. Nom. pl. mAimns 'members of the mAimna-clan' and wanzas 'burial installations', with no palatalization of the final consonant, must be added to Gérard's dossier of the relevant Lydian forms. These examples suggest that the final consonant of nom. pl. -is is affected by the preceding vowel, and this Lydian nominal ending must be reconstructed as *-Vs rather than *-insi, the etymology that is also mentioned as a possibility in Gérard 2005: 81.

4 The third personal name, Tiwda, mentioned in Yakubovich 2010: 97, was reinterpreted as a title in Rieken & Yakubovich 2020: 216. On the name Walwet, previously read as Walwel, see now Dale 2015.

guages of western Anatolia, or even a western Anatolian linguistic area with the participation of Lydian, would not a priori contradict the same conclusions, the reinterpretation of commonly acknowledged genetic isoglosses in areal terms might lead to reconsidering the status of Lydian as an Anatolian language. Be it as it may, the availability of mutually contradictable claims advanced with regard to Lydian phylogeny in recent scholarship provides a justification for revisiting this issue. To this, one might add that at least some of these disagreements are likely to be rooted in the lack of consensus on several key diachronic issues of Lydian grammar.

Such a state of affairs determines the structure of the present essay: I will focus there on specific isoglosses that were used as arguments in recent debates on the position of Lydian, but also adduce a number of new parameters that have the potential to affect this discussion. As an approximate starting point for the survey, I have selected the presentation of Lydian phonology and morphology in Gérard 2005, which implies that the publications that precede this monograph are allotted only a sporadic mention. Although I am aware of the fact that the unconstrained selection of isoglosses makes my project open-ended, I flatter myself with the hope that even such a procedure might be conducive to drawing up an interim balance. In addition to its main purpose stated in the title, this essay also serves the goal of advancing our knowledge of Lydian historical grammar, since most sections combine the discussion of previous scholarship with the presentation of new hypotheses on the relevant topics.

2. Assibilation of

One of the phonetic isoglosses separating Hittite and the languages of the Luwic family concerns the development of palatalized velars. While in Hittite they routinely merge with the ordinary velar stops, showing thus the same development as the non-Anatolian Indo-European centum languages, in Luwian and its closest relatives, they undergo further palatalization (at least in the majority of environments). For example, the stem of the proximal deictic pronoun is ka- in Hittite, but za- (/tsa-/) in Luwian and sa(n)- in Carian (Adiego 2007: 410). In a similar fashion, the Hittite imperfectives in -ske/a-(mi) correspond to Luwian verbs with the suffix -zza-(i) the Lycian A verbs with the siffix -s-(ti), even though the mismatch between the Hittite and Luwian conjugation types complicates the comparison to an extent (Sasseville 2020: 439-461). In contrast, the destiny of palato-velars in Lydian remains a matter of debate due to the lack of assured etymologies illustrating the development of these sounds (Gérard 2005: 67-68). The earlier attempts to interpret some Lydian velars as reflexes of palato-velar stops did not stand close scrutiny. Pace Poetto 1979, Lyd. kof(u)- 'water' can be kept away from Classical Armenian cov 'sea', because Luwian (C) hap(i)- 'river' represents its closer match (Mouton & Yakubovich 2019: 222, n. 21).5 Contra Gusmani 1976-1977, there is no need to derive the Lydian economic term qasaa- from PIE kwas- in view of the attractive comparison between the derived verbs, Lydian qasani- and Lycian A qehnni- 'to rent', which pleads for the initial labiovelar stop (Sasseville 2020: 160).6

If searching for velar reflexes of the Indo-European palatalized velars in Lydian has not thus far yielded any tangible results, this is perhaps an indication that one has to search in a

5 Here and below, Luwian (C) and Luwian (H) refers to the Luwian forms attested in cuneiform and hieroglyphic transmission respectively. I prefer such an abbreviation to its more traditional alternative CLuwian and HLuwian, as a way of stressing that we are dealing with one Luwian language recorded in two different writing systems.

6 The Lydian transliteration conventions adopted in this paper incorporate the new values proposed in Schurr 1997: 201, n.1 (<p>, <s>, <s>, and <w> for the earlier <b>, <s>, <s>, and <v> respectively).

different direction.7 Sasseville (2020: 62-63) offered a connection between Lyd. ca-№, which he interprets as 'to give a share, to make a dedication', and Hitt. zaske/a-(mi), the imperfective of Hitt. dai-/ tiya-(mi) 'to put'. The implication of his analysis is the reconstruction of the Lydian verb as a derivative of *dhhiske/o-, where the "first laryngeal" presumably drops early on. While there are different interpretations of the Lydian consonant <c> (see section 5), most scholars assume its affricate character. If <c> represents a reflex of the earlier *dhsk-, this implies that the final consonant of the cluster underwent further palatalization, perhaps to the point of merging with the preceding sibilant. Sasseville (2020: 381) explicates his analysis by offering a typological parallel: he analyses the Lycian A verb ze-(ti), interpreted as 'to take in marriage' as a reflex of the imperfective *dh3ske/o- 'to take', which went through the Proto-Lycian stage *dsse-. Furthermore, he hypothesizes that the Lydian stem ce-si- 'to take' represents a cognate of Lycian A ze-(ti) extended by an additional suffix (Sasseville 2020: 506-507).

The proposed etymologies represent progress over the preceding treatments of the Lydian verbs under discussion: for the first time they are accounted for with reference to the regular sound changes. Although the Lydian outcome of the *-ske/o- suffix is never directly spelled out in Sasseville 2020, the above discussion implies that it undergoes the same type of development as in Luwian or Lycian A. Yet, as long as the proposed development of PIE *k remains isolated in Lydian, the etymologies under discussion can only be regarded as provisional, in particular given the limited number of attestations of the relevant verbs and the absence of a strict combinatorial proof for most of the relevant contexts. An etymology that can arguably yield more support for the assibilation of *k is the likely connection between Lyd. te-sast(i)- 'right' (1x) and the Indo-European root deks- 'id.' (cf. already Gusmani 1964: 212). Since the consonant cluster -ks- is otherwise attested in Lydian, its absence in tesast(i)- can be most easily explained by a specific reflex of the palatalized *k. Unfortunately, this case cannot be regarded as full proof either: the matching adjective srfast(i)- (1x), which could contextually mean 'left', does not have an etymology that could support such a meaning.

In what follows, I would like to offer an additional Lydian lexeme in line with the proposed derivation of *k into a fricative. This is the demonstrative pronoun es- 'this', one of the most frequent lexemes found in the Lydian corpus, whose meaning is established with an absolute certainty. Its attested forms are nom.sg.c. es(s), acc.sg.c. esv, esn, nom./acc.sg.n. est, dat.sg. esA, nom.pl.c.(?) es, nom./acc.sg.n. es dat./gen.pl. esvav. It is traditionally compared with the Hittite demonstrative pronoun asi 'yon' (see, e.g., Gérard 2005: 94). There are, however, three difficulties that undermine this comparison. First, Lydian es- functions as the proximal deictic pronoun in most of the available contexts, whereas Hittite asi is a distal deictic pronoun and a marked element of the system, being considerably less frequent in texts that either proximal or medial pronominal stems. Second, contra Gérard, Hitt. asi is strictly a nominative form of the common gender, whereas other members of the paradigm, for example acc.sg.c. uni or dat.sg. edi do not contain the sibilant. While the paradigm of asi does show some instances of analogical restructuring in the history of Hittite, this never leads to the extension of

7 An attempt to assign a centum substrate in Greek to the Lydian language was recently undertaken in Garnier and Sagot 2020, as implied by the title of their paper. Yet, if one considers its substance, one can see how little is done there to promote such a goal. Section 1 of this paper is devoted to the analysis Lydian toponyms and personal names in Greek transmission, but none of the proposed etymologies is based on the centum reflexes of the known Proto-Indo-European roots. In contrast, such reflexes appear in section 2, devoted to "possible Greek loanwords from Lydian", but the Lydian origin of the Greek lexemes treated in this section is not philologically substantiated. If anything, this discrepancy should yield support to the hypothesis that the centum substrate under discussion is not Lydian in origin.

asi to other slots (Kloekhorst 2008a: 220, Goedegebuure 2014: 115-117). Third, the assimilated nom.sg.c. es(s) can only be explained as a result of syncope of *esas, as opposed to *esis.8

In fact, there is a better Lydian match to the Hittite paradigm of asi: this is the demonstrative stem ed- (of unknown deictic properties), represented by nom.sg.c. eds nom./acc.sg.n. edt, dat.sg. edA. Given that ed- represents the generalized oblique stem of asi in Hittite, it is easy to see how the matching stem could spread across the paradigm in Lydian (cf. Gérard 2005: 95). At the same time, there are grounds to argue that its initial e-vocalism became abstracted as a deictic prefix, which spread to other Lydian demonstrative stems. One case in point is the adverb epad < *apad 'here (?)', which can hardly be separated from Hitt. apadda 'there' (Gérard 2005: 99), while a closer formal match of the Lydian form can be found in Luw. (H) abadi 'there' (for which see Goedegebuure 2010: 80-83). This adverbial form can be contrasted with the Lydian anaphoric pronoun p(i)- '(s)he, it'; in this case the lack of the e- prefix correlates with the absence of the deictic function. If e- ended up functioning as a deictic marker in Lydian, I submit that the same element could also have been added to the proximal demonstrative pronoun, triggering a syncope in its root.9 In this connection, it is appropriate to mention a direct contrast between e-s- and the rare demonstrative stem o-s- in LW 70 es sadmës osek anlola 'this inscription and that burial complex'. If both e- and o- are taken as deictic particles, it becomes possible to reconstruct the earlier paradigm of the main Lydian demonstrative pronoun as nom.sg.c. *sas, acc.sg.c. *sav, nom./acc.sg.n. *sad, dat.sg. *saA, dat./gen.pl. *sav. This lexeme would in turn be amenable to a straightforward analysis as a lexical cognate of Hittite ka- and Luwian za-, which function as proximal deictic pronouns. The implied sound change *k > s in the history of Lydian would then be compatible with *ks > s and *sk > s, accommodating the proposed etymologies of tesast(i)- 'right', ca-(t) 'to give a share, make a dedication', and cesi- 'to take'.

One must acknowledge that the assumption of reanalysis operating across the semantic field of the demonstrative pronouns complicates the proposed etymology. At the same time, the hypotheses offered in this section transform the interpretation of the Lydian pronominal system, making its distinctly more Anatolian in comparison with the picture presented in Mel-chert 2009. If one believes, on independent grounds, that Lydian represents a member of the Anatolian group, then the reconstruction offered here has the advantage of aligning the system of Lydian pronouns with the rest of the Lydian grammar.

3. Hardening of the "laryngeals"

The Anatolian languages are notoriously distinct from all the other members of the Indo-European family in that they preserve segmental reflexes of post-velar phonemes, which are traditionally called "laryngeals" in the field of Indo-European Studies. No less than four dif-

8 *s > <s> is the regular palatalization after i, at least in the inflectional endings (Gérard 2005: 71, although the transliteration used in this account is now dated). Since the palatalization of *s represents the most common source of Lydian <s>, its phonetic interpretation as alveopalatal /ç/ rather than palate-alveolar /J"/ appears likely (cf. the discussion in section 5).

9 The addition of deictic particles to demonstrative pronouns is a common phenomenon cross-linguistically. As more or less random examples from other Indo-European languages, one could mention Russian э-тот 'this' vs. mom 'that', Vedic esâ- 'this' vs. sâ- 'this, that', and classical Armenian ays 'this (adj.)' vs. sa 'this (n.)'. The last example is of particular interest because the Armenian proximal deictic stem *s(a/o)- represents an Indo-European cognate of Lyd. es- if the hypothesis advanced in this section is to be believed.

ferent "laryngeal" reflexes, /x/, /y/, /xw/ and /yw/, have been identified in Hittite and Luwian (Yakubovich 2020: 227, Melchert 2020: 247), even though it is normally assumed that all of them, except for /x/, evolved first within Anatolian. Despite the conventionally used signs for velar fricatives, the mainstream hypothesis is their place of articulation was uvular in the second millennium BCE. The situation must have, however, changed in the first millennium, since the foreign transmission of the "laryngeal" reflexes in the Lycian and Carian languages rather speaks for their velar place of articulation, as argued in detail in Kloekhorst 2008b.

For a long time, Lydian was separated from the other Anatolian languages, in scholarly opinion, with regard to the assumed development of the "laryngeals". Throughout the twentieth century it was believed that they had disappeared in Lydian without a trace, and such a view still claimed the adherence of Gérard (2005: 71). This consensus was shaken with the publication of Melchert 2004, where a number of examples were offered as an illustration of the Indo-European "second laryngeal" h) developing into k in Lydian in intervocalic position. All of them concern abstract nouns with the reconstructed suffix -eh2- > *ax- > -ok-, e.g. saroka- 'protection'. Judging by the other Anatolian languages, such a combinatorial restriction imposed on the segmental "laryngeal" reflexes would be highly unusual, as the word-initial position before a vowel tends to be particularly favourable for their preservation. Furthermore, the disappearance of the initial h2 in this position in Lydian was essentially backed by a single example, namely esa- 'child' < h20nso- 'grandson' (Melchert 2004: 142).

In recent years, however, examples supporting the development *x- > k- in Lydian began to pile up. Thus Lyd. kof(u)- 'water' was compared with Luw. (C) hap(i)- 'river' (see section 2 above), while Lyd. kast(V)- 'bone' and kastaAc(i)- 'remains (vel sim.)', both occurring in the same burial inscription, were matched with Hitt. hastai- 'bone' and Luw. (C) has(t)- 'id.' (Yakubovich 2019a: 402). Furthermore, doubts were raised whether the name of Gyges, the first Lydian king of the Mermnad dynasty, must necessarily represent a loanword (Oettinger 2020a: 119-120). According to the scholarly consensus, the name of this king, which is attested on Lydian coins through the possessive adjective kukal(i)- 'of Gyges' is related to Hitt. huhha-, Luw. (C) huha-, and Lyc. (A and B) xuga- 'grandfather'. As long as Lydian was treated as a "la-ryngeal-free zone", the name under discussion was taken as a Luwian or Carian loanword, in conformity with the respective origins of several other Lydian personal names (cf. Yakubovich 2010: 94-95; Dale 2015: 157). If one accepts the development *x- > k- in the history of Lydian, this linguistic hypothesis prompts revisiting the origin of the name under discussion. This said, the etymology of the name of Gyges remains a complex problem, which cannot be resolved by linguistic means alone.10

The assessment of the new correspondence must naturally depend not only on amassing the examples in its support but also on evaluating potential counterexamples. In this connection, it seems appropriate to consider the passage that provided the basis for the interpretation of esa- 'child'. The protasis of a curse formula in the burial inscription LW 5 ak na-qis ëmA kanaA

10 The origin of the royal name Gyges must be discussed apart from the origin of its transmission with the voiced stops in both Greek (ruyqg) and Neo-Assyrian (Gu(g)gu). There is no doubt about the existence of a Carian name quq-, which appears in Greek sources as Tuyog (see Adiego 2007: 408 with cross-references). The rendering of the Anatolian velar stops as <y> probably implies specific acts of transfer, which were unlikely to be replicated. Given the known international mobility of the Carians, it seems possible that the name of King Gyges was transmitted abroad via Carian envoys at his service, and then the Greeks adopted the same pattern for rendering the genuine Carian names. Although no legend about Gyges would directly link him with Caria, one can propose Carian etymologies for the Mermnad clan that he brought to power (Hajnal apud Yakubovich 2017a: 288-289) and the Lydian word for 'king' (Valerio & Yakubovich 2022).

kileA buk eminav esav citalad fadint can be interpreted as 'whoever causes harm to my wife Kile or to my esa-s', while the apodosis of the same curse formula invokes the retribution of Artemis on the perpetrator and his property (qira-)11. While the combination 'wife and children' is certainly possible on contextual grounds, this is not the only possibility. A curse against an evil-doer who attempts to damage or steal the property of the deceased appears quite logical in the context of a burial inscription, in particular given the following mention of the evildoer's own possessions. On the etymological side, since the notions of 'being' and 'belonging' were expressed by the same verb es-(mi) in Hittite, and presumably by its archetype in Proto-Anatolian, there is nothing strange about the assumption that the Lydian word for 'belongings' is derived from the same root12. Against such a background, one can simply discard the interpretation of esa- 'child' if it contradicts the mounting evidence for *h2- > k- in Lydian.

The clause under discussion contains one more item that is potentially relevant to the present section, namely the personal name kile-. The latter can hardly be separated from the indigenous Anatolian name attested as KtA^^ in Greek transmission (Zgusta 1964: 230), which in turn probably represents a retrenchment of Anatolian compound names, such as KtAapag or KtAwpaaig. In Pre-Achaemenid Anatolian sources, however, we find a number of names with the initial element /xilV-/, such as Hilamaddu, Hiliya, and Hilaruada, all attested in cuneiform transmission (Laroche 1966: 67). Scholars tend to agree that the names in /xilV-/ contain the reference to Hitt. %ila- 'courtyard' or its Luwian cognate (Zehnder 2010: 153-154), and it seems reasonable to extend the same conclusion to the Hellenistic Anatolian names in KiA(V)-. Lydian kile- bridges the chronological gap between the two groups and is likely to represent the earliest attestation of the change *h2- > k- in the Anatolian root under discussion. Unfortunately, as in the instance of kukal(i)-, there is no proof that the name is genuinely Lydian, as opposed to a loanword from another Anatolian language.

One more potential case illustrating the same sound change *h2- > k- in the history of Lydian is kAida- 'earth, soil', whose meaning is assured by the Lydian and Aramaic bilingual LW 1. The traditional comparison of this noun was with Greek yAia 'glue', and English clay, made formally possible by Melchert's discovery of the change *y > /5/ in the history of Lydian (Melchert 1994: 136, cf. the next section), suffers from the lack of cognates within the Anatolian subgroup. The alternative cognate is Hitt. halma- 'clay': while it is genetically closer, the comparison is more complicated from the formal perspective, since one has to account for the n-suffix.13 The tertium comparationis that facilitates the formal account is provided by the Latin verb lino 'to smear, rub' and the rare Greek verb dAivw 'id.', which are reconstructed as *h2li-nH- in Beekes 2010: 68-69. One can argue that the Hittite noun is deverbative in origin and re-

11 Here and below, the abbreviation LW refers to the inscriptions published in the Lydisches Wörterbuch (Gusmani 1964). This does not, however, mean that their reproduction in this essay corresponds exactly to the one given in that source; whenever appropriate, changes in the transliteration have been implemented (cf., e.g., n. 6 above).

12 H. Craig Melchert informs me that he has independently arrived at the same conclusion. A typological support to the proposed reconstruction comes from the Persian noun hasti, a derivative of Proto-Iranian *as-/*ah-'to be', which has the semantic range 'being, existence; property'. See further Nussbaum 2014 for the likely etymological connection between Hitt. ässu- 'goods, property' from es-(mi) 'to be'. I believe, however, that the derivation of 'goods < belongings' via the concept of belonging (cf. Latin tres servi tibi sunt 'you have three slaves' and the matching constructions in the other ancient Indo-European languages) is a simpler semantic path that the one outlined in Nussbaum 2014: 248-251.

13 According to Gérard (2005: 73), Lyd. kAida- is to be connected with Hittite kulei- /kwlei-/ 'fallow land', but this etymology appears to be formally impossible, since *kw- yields its regular reflex q- in Lydian consonant clusters, cf. 1sg.pres. qAastänu, 3sg.pres. etqratad, or 3sg.pres. qrifrit.

fleets the fusion of the root and the etymological verbal suffix.14 Nevertheless, the complexity of such an explanation relegates the comparison between Lyd. kAida- and Hitt. halma- to the category of possibilities, while the match between Lyd. kAida- and Greek yAia remains a valid alternative, and the two etymologies are mutually incompatible.

While the potential Lydian counterparts of Hittite and Luwian /x/ were frequently addressed in the recent years, the Lydian reflexes of other "laryngeal" segments have received less attention. The reason for such a state of affairs is the paucity of Lydian lexical items where the relevant phonemes have been reconstructed. Thus, the only potential context for tracing the reflex of the Anatolian voiced/lenis "laryngeal" is the above-mentioned possessive adjective kukal(i)- 'of Gyges'. The Lycian (A and B) stems xuga- 'grandfather' speak directly for the voiced/lenis character of the second consonant in the root under discussion, while Luw. (C) hu-u-ha-ti (instr, 1x) supports the same conclusion. Yet, the lenition of the second consonant in this root was apparently not a pan-Anatolian phenomenon: Hitt. huhha- 'grandfather' contains two fortis consonants. If Lyd. *kuka- 'grandfather' was an inherited lexeme, the historical interpretation of its second consonant depends on whether it shared the morphological pattern with Hittite or with the Luwic group; in the latter case one can safely posit the merger of voiceless/fortis and voiced/lenis "laryngeals" in Lydian.

But even aside from the case discussed above, one could argue that the Lydian language did not know the distinction comparable to that between /x/ and /y/ in Hittite or Luwian. Indeed, if such a contrast existed, and the fortis "laryngeal" turned into <k>, one might expect the development of its lenis counterpart into <g>. Yet, the counterpart of the Greek gamma is attested only four times in the Lydian inscriptions, and the two better-understood attestations plead for a free variation between <k> and <g> (Gérard 2005: 58). Furthermore, the Lydian alphabet features no signs for velar or post-velar fricatives. Under such conditions, one may entertain two hypotheses: either the merger of the two "laryngeals" or the development of the lenis one into zero. What makes the second alternative less likely is the lack of evidence for the disappearance of any other lenis consonants in the history of Lydian, while the merger between the fortis and lenis stops can be illustrated by the development of the additive particle, PIE *fkwe 'and'. The reflex of this morpheme predictably undergoes lenition in postvocalic position in Luwian (Mouton & Yakubovich 2021: 42-43), but the Lydian clause-initial particles ak and fak, which are commonly analysed as reflecting merger with fk < *fkwe 'and', show no traces of lenition (cf. section 10). Summing up, although direct evidence is lacking, one may cautiously hypothesize that the Lydian language either failed to implement the lenition of the "laryngeals" or reversed its results.

The final question to be addressed in this section is the reconstruction of the labialized "laryngeal" in the prehistory of Lydian. If such a phoneme existed, given *h2- > *x- > k-, one might expect the parallel process *xw- > *kw-. PIE *kw merges with k in Lydian before the labialized vowels and in word-final position, but otherwise its regular reflex is q (cf. Gérard 2005: 67). Accordingly, one may wonder if there are lexemes featuring Lydian q of likely "laryngeal" origin. The most plausible candidate is the verb qisre-(d) 'to take care (vel sim.)' (Gusmani 1964: 187), an ostensive denominative derived from the reconstructed verbal noun *qisr-. I am not aware of an Anatolian primary verb in *kw- that could provide a base for such a verbal noun, but Hitt. huwai-/huiya-() 'to move' and Luw. (H) hwiya- 'id.' are formally suitable as cognates of Lyd. *qi-sr-. The comparison may appear surprising on the semantic side, but one should

14 As a typological parallel, one may consider the case of Russian глина 'clay, loam', a cognate of English clay whose nasal suffix also must have had a verbal origin, cf. Old Irish glenim and Old High German klenan 'to stick, smear' (Beekes 2010: 276-277).

keep in mind that the Hittite phrase peran huwai-/huya-(r>, literally, 'to move in front (of someone)' was lexicalized with the idiomatic meaning 'to help' (García-Ramón 2016: 72-73, Oet-tinger 2022: 62-63). The extension of this secondary meaning to the nominal derivatives without a prefix also appears possible: thus, the phrase huwayalli dUTU-i 'to the Sun-god the assistant can be compared with kutrui dUTU-i 'to the Sun-god the witness' in the parallel version of the same Hittite text (Yakubovich 2010: 378).15 Consequently, one can reconstruct Lyd. *xwi-sr-> *qi-sr- 'assistance, help' as yet another derivative of the Anatolian verb for 'moving'. The etymology under discussion provides minimal evidence for the presence of *xw- in Pre-Lydian but obviously remains in need of further confirmation.

4. Pronunciation of <d>

The Lydian grapheme that is traditionally transcribed as <d> poses problems of phonetic interpretation. On the one hand, the underlying sound must be phonetically close to [d], because the Lydian capital, known as LápSeig in Greek transmission and the name of the Achaemenid satrapy of Lydia (Old Persian s-p-r-d) cannot be separated from the Lydian noun sfarda/e-'Sardian' and the derived adjective sfardet(i)- 'Sardian'. On the other hand, it cannot be actually [d], because the name of the Greek theonym Demeter (Ionic a^^^t^q, Doric Aa^áx^p) and the royal name Alexander (AAé^avSpog) are rendered as lametru- and aAiksantru- respectively in Lydian transliteration (Gérard 2005: 57-58). Diachronically, the phoneme recorded as <d> has a dual origin: it can continue either the Anatolian lenis stop /d/, as in tad < *tod 'it' or the Anatolian glide /j/, as in pidv < *pij-om 'I gave' (Gérard 2005: 66-67, 69). The interdental fricative [5] was tentatively proposed as the phonetic interpretation of Lydian <d> in Gusmani 1965: 209, and this suggestion is cited as a viable option in Gérard 2005: 25, 58.

A rather different interpretation is defended at some length in Oreshko 2019. Observing that the frontal fortition of *j to [5] is cross-linguistically rare, if at all existent, he suggests as an alternative that *j and *d merged into /j/ in the history of Lydian, and the grapheme <d> might accordingly be re-written as <j>. The author acknowledges that the reading "sfarja-" 'Sardian' is not helpful for explaining Gk. EápSag or Persian s-p-r-d, but addresses this lexeme as "a special case reflecting fortition of j in the cluster -rj- > -rd- (> -rd-)" (p. 225).16 This argument is in itself rather problematic: if the phoneme /j/ had allophones [j] and [5] in Lydian, one would expect to find some counterpart of this distinction in the writing system. The Lydian alphabet was not strictly based on phonological principles, but involved experimentation with new signs for allophones, such as <g> or <y> (Gérard 2005: 26, 58). The sounds [j] and [5] are distinct enough for the Lydian scribes to attempt to reflect this distinction, at least sporadically.

Naturally, this consideration alone is not sufficient to refute Oreshko's hypothesis, as long as other combinatorial data would support it. This does not, however, appear to be the case. If a phoneme hidden behind <d> were a glide, one would expect it to participate in the same type of synchronic alternations as the labial glide <w>. In the case of the latter, its combinatorial behaviour is sufficiently outlined in the paper under discussion: it includes phonologically

15 The proposed interpretation of huwayalli- derives support from the divine epithet KUB 2.1 ii 25 [SA LA-BAR]NA peran huiyawas dLAMMA-ri 'to the protective god of moving forward the [Labar]na' = 'to the protective god of helping the [Labar]na'.

16 Here and below in this section, the forms reflecting the alternative reading of <d> are enclosed within double quotation marks without being italicized. This is done in order to avoid any confusion with the readings accepted in this paper.

conditioned alternation between <u> and <w> in the same morpheme, e.g., 1sg.pret. ko-w vs. kantor-u, instances of synchronic contraction, e.g., mruwa- vs. mru- (Oreshko 2019: 210-211). No parallel changes can be observed in the case of <d>: this we find nom.-acc.sg.n. mru-d and ci-war-d (not **ciwari) and not a single dropped <d> in a fairly common verbal ending -i-d (3sg.pres.).17 What we do have is the frequent alternation between <t> and <d> as allomorphs of the same morpheme, e.g., nom.sg.n. in LW 1.1 es-t mru-d 'this stele' or 3sg.pres. in LW 24.9 pasvsakvaki-d 'causes impurity(?)' and LW 24.11 in-t 'does' occurring in parallel disjunctive clauses. Oreshko (2019) is aware of this difficulty but attempts to obviate it by treating the nominal endings -t and -d as synchronically suppletive (p. 197) and arguing that the verbal 3sg.prs. endings -t and -d have different historical origins (p. 227-228). The doubts that Lydian belongs to the Anatolian language group are expressed in context of discussing the origin of the verbal ending -d ("-j" according to Oreshko).

Within the context of the present essay, it is important to stress the last point: Oreshko's phylogenetic stance is not based on the currently accepted decipherment of Lydian but represents a price that he is willing to pay for its revision. The co-occurrence of the 3sg.pres. allo-morphs -t and -d is not only phonologically natural but also derives support from the cognate allomorph pairs -ti/ -di in Lycian and -tti/ -ti in Luwian cuneiform texts. The considerable efforts to prove this point were invested in Melchert 1992, while the full documentation of derivational verbal classes that do or do not trigger the lenition of 3sg.pres. verbal endings in Luwian, Lycian, and Lydian is now available in Sasseville 2020. Furthermore, I recently attempted to demonstrate the presence of the matching allomorphs -taA and -daA functioning as the mediopassive ending of 3sg.pres (Yakubovich 2019a). While the current reconstruction of Lydian verbal morphology is not cast in stone, and its individual elements will most certainly continue to undergo revisions, I doubt that a controversial new reading of <d> represents a sufficient reason to give up its most basic premises. Quite to the contrary, the more challenge the proposed new reading poses for the mainstream morphological reconstruction and phy-logenetic assumptions, the more stringent its evaluation should be.18

A separate problem concerns the reinterpretation of word-initial d- as "j-". While the etymological *d- is devoiced into t- in the history of Lydian, the development *nd- > d- was proposed for a number of morphemes, such as the verbal prefix da-, dav 'in, cheZ and dum(v) 'furthermore' (Yakubovich 2005: 79-83). Thus, Lyd. da- can be directly compared with the pre-verb nte < *endo in Lycian A, while dav and dum(v) probably represent etymological compounds with da-.19 The development of prenasalized stops into plain voiced stops, as in the

17 For fairness sake, one should mention the contrast between in-t 'does' and fa-din-t 'causes' addressed below in this section. One may, however, doubt that the alternation between the root variant in and din was perceived in this case as synchronic.

18 To illustrate this general point, it might be appropriate to contrast the proposal in Oreshko 2019 with the New Readings of Anatolian hieroglyphs advanced in Hawkins, Morpurgo-Davies & Neumann 1974. While the initial impetus for the breakthrough in the decipherment of Luwian was the discovery of short biscriptal graffiti in the Urartian fortress of Altin-Tepe, the main effort of the three scholars consisted in demonstrating that the new values obtained by combinatorial means actually strengthen the previously assumed connection between Luwian and "Hieroglyphic Hittite", and this demonstration played a decisive role for the positive reception of the New Readings.

19 The etymology of this group of forms helps explain why the Greek theonyms beginning with A- were borrowed into Lydian with the initial l- (e.g., Ar||xr|Tr|Q vs. lametru-). Presumably, at the time when the Greek divine names entered the Lydian language, the archetypes of Lyd. da- and the related forms were still pronounced with the initial *nd-, and the voiced plosives were still proscribed word-initially, which made l- the optimal substitute of Greek d- at the time (cf. section 8 on another likely example of lambdacism in Lydian).

history of Modern Greek, is typologically trivial, and the use of the same grapheme <d> for the allophones [d-] and [-5-], as in Spanish, does not strain common sense either. Even the lenition *nd- > *d- > [5-] would be a conceivable development, whether it reflected a synchronic constraint on word-initial voiced plosives the elimination of allophonic variation, but the change *nd- > /j-/ would be unusual indeed. Oreshko is naturally aware of this problem, but his reaction is simply claiming that "the origin and exact meaning of Lydian da- and dav remain ... unclear" (Oreshko 2019: 203). Once again, we face the situation when the assumption of a new phonetic value leads to the reduction of the postulated cognate set.

Against such a background, it is natural to wonder what the positive arguments that support the revised reading of <d> are. Most interesting, in my opinion, is the hypothesis that this letter renders /j/ in personal names of foreign origin. Thus, the Iranian name mitridasta- attested in Lydian transmission obtains a cogent explanation as a reflex of *midra-yasta- 'Mithra-worshipper' (Oreshko 2019: 219), while the same Phrygian(?) name is arguably attested as sa-karda- and Eaydpiog in Lydian and Greek transmission respectively (Oreshko 2019: 223-224). I doubt, however, that these insightful comparisons preclude the phonetic realization of <d> as an interdental fricative. As long as the glide [j] was absent from the sound pattern of Lydian, the fricative [5] may have functioned as its regular substitute in the lexicon of borrowed origin.20

Another argument is the rarity of the fortition *j > /5/, which was already mentioned in this section. There is in principle no doubt about the possibility of such a sound change: indeed, Oreshko (2019: 225) explicitly acknowledges it in the instance of "sfarja-" = *sfarda- > [sfarda-]. What he doubts is the availability of cross-linguistic parallels for the change *j > /5/ across the board. It is worth, however, pointing out that Lydian does not implement such a fortition across the board either. As an example, let us consider the paradigm of the stem tro(d)-(d) 'hand over (vel sim.)', a cognate of Luwian (C) tarawi- 'id.'. The 1sg.pret. form trod-v contrasts with the prefixed forms fa-kan-tro-w (1sg.pres.), fa-tro-d (3sg.pres.) fa-tro-s (2sg.pret.), and fa-tro-l (3sg.pret.), with the implication that the derivational suffix *-je/o- drops without a trace in most members of the paradigm. Another instructive example is 3sg.pres. in-t 'does' contrasted with the prefixed derivative fa-din-t 'one causes'. The Hittite root cognate of this verb is ye-(mi) 'to do, make', but the contraction *jV- > *ji- > i- was apparently implemented in the history of Lydian, except for the prefixed form *fa-jin- > fa-din-, where the glide was preserved or restored on a morpheme boundary (cf. Oreshko 2019: 207). These examples, the list of which can easily be continued, suggest that the fortition of *j was in fact conditional, even if the precise description of its licensing conditions remains a task for the future.21 This validates the parallel with a similar conditioned change *j > /5/, which occurs in several environments in the history of Brittonic languages and is duly mentioned in Oreshko 2019: 194. Eska (2018/2019: 19-26) offers a sociolinguistic scenario accounting for such a change and adduces its additional instantiation in "most varieties of Fijian".

20 Incidentally, the perceived functional equivalence between the foreign glide /j/ and the Lydian fricative /6/ may be responsible for the shape of the Lydian letter <d>, graphically <i> with an additional stroke (for which see Oreshko 2019: 216-217). While this argument is admittedly speculative, so are the attempts to base the reading of <d> on the graphic etymology of this sign, consisting of only two strokes.

21 I personally find not a single convincing example that could illustrate the change *j- > /6-/ in word-initial position in Lydian. The LW 24.20 detv 'mobile property (acc.sg.)', frequently cited as a reflex of the participle *jVnt-'moving', need in fact not be separated from the indeclinable form det (9x) < *endo-ndo 'inside (vel sim.)' (cf. Yakubovich 2017a: 277-278). At the very least, the proponents of the participial origin of detv should find additional examples illustrating the sound change *j- > /6-/. An example that in my view supports the disappearance of *j- in Lydian is ora- 'month', to be compared with Luw. ar(i)- 'time', Greek шра 'season, time period', as well as Church Slavic яра 'spring', English year and German Jahr 'year' (differently Oettinger 2021b).

Summing up, although the change *j > /0/ is typologically uncommon, an attempt to restrict it to the environment after /r/ in the history of Lydian creates more problems than it solves. In particular, the attempts to dismiss the available Anatolian etymologies without suggesting better alternatives do not represent a convincing way of arguing for a new value of Lydian <d>.

5. Palatalization of dentals

The Lydian language is assumed to feature two affricate sounds, corresponding to the graphemes <t> and <c>, but their exact pronunciation remains sub judice. Gérard (2005: 75), reconstructs an asymmetrical pair of the voiceless alveolar-palatal affricate <t> = /tç/ and the voiced alveolar affricate <c> = /dz/. Yakubovich (2005: 77, n. 11) doubts that voicing constituted a distinctive feature of the Lydian affricates and tentatively suggests the interpretation of <c> as <ts> or <tj>. Oreshko (2019: 208) interprets the contrast between <c> and <t> as that between the dental affricate /ts/ and the palatal affricate /cç/, each of them having voiced allo-phones. Finally, Kloekhorst (forthcoming) offers a new interpretation of <c> as the palatal stop /c/.

There is more consensus with regard to the historical origin of <t> and <c>. On the one hand, the Lydian nominal/adjectival suffix -Ta- was analysed as a cognate of the Hittite adjectival suffix -zzi- < *-tjo- (Gérard 2005: 87-88), while the cognate Luwian adjectival suffix -zza-represents a closer formal match, and the Lycian A nominal suffix -za- offers a typological parallel for the nominalization of the relevant derivatives (cf. Yakubovich 2013: 165, n. 16 et passim). In particular, it is worth noting Lyd. paA-za-, presumably derived from p(a)A- 'fore-' and thus arguably meaning 'foremost',22 as well as a similarity between the Lydian title arm-za-and the Lycian A title armanaza, both presumably derived from Arma, the Anatolian Moon-god (Sasseville 2018: 130, n. 2). On the other hand, <c> apparently reflects palatalization of dental stops before -i and -u in several Lydian morphemes. Thus, Lyd. tac(i)- 'offering (vel sim.)' has been explained as a reflex of PIE *dhehi-ti- (derived from *dhehi 'to put, place'), while the stem da-cuwe- 'to erect (vel sim.)' was compared with Luw. (C) duwa-(i) 'to place, erect' and Lyc. A tuwe-m 'to erect' (Gérard 2005: 59-60).23

The above examples suffice to show the contrast between the two kinds of palatalization. In the instance of *-tj- yielding <t>, which can also be called iotation, we are dealing with a sound change that finds close counterparts in Hittite, Luwian, and Lycian. In contrast, the palatalization yielding <c> appears to be restricted to Lydian in the morphemes constituting its scope. To be sure, the palatalization of dental plosives before -i is known in Hittite, witness the 3sg.pres. verbal ending -zi < *-ti. Yet, the counterpart of this 3sg.pres. ending in Lydian is -t / -d, which presumably reflects the relative ordering of the Lydian second palatalization after the apocope. The palatalization of dental plosives before -u finds a formal parallel in Palaic, where the stem tazzu- 'to take' was tentatively accounted for as a late reduplicated form of *daw-/du-'to take' < PIE *dehsw- (Melchert apud Sasseville 2019: 25, n. 13). I am not, however, aware of any Palaic and Lydian cognate pairs that would share palatalization before u, while Palaic a-as-du 'let it be' and Lyd. tutra- 'daughter' suggests that the palatalization *tu > cu was subject to

22 For the Lydian prefix p(a)A- as a cognate of Luw. pari 'away, forward', see Yakubovich 2019a: 400.

23 Another plausible instance of palatalization before u in Lydian is the noun cuAdale- attested as dat.sg. cuAdaleA in LW 27.3', which can be formally compared with Lyc. B tulijele- 'assembly'. Yet, given the fragmentary character of LW 27, this match naturally remains tentative.

additional yet unknown licencing conditions in both languages.24 The natural implication of the discrepancies outlined above is that the origins of palatalization before /i/ and /u/ in Hittite, Palaic, and Lydian are mutually independent. This implies in turn a relative chronology, according to which the palatalization yielding <t> and shared by Lydian and the other Anatolian languages is to be ordered before its language-specific counterpart yielding <c>.

Yet, not all the occurrences <t> or <c> reflect palatalization or iotation. The inflectional forms sfardënz 'Sardians (nom.sg)', presumably a syncopated form of *sfardënt-is, kaz-zadmë-'decree', presumably a noun with the prefix kat- derived from sadmë- 'inscription, seal', and kaz-zirs 'they prescribed, decreed', presumably a prefixed variant of sers with a similar meaning, all bear witness to the secondary <t> arising through contraction on a morpheme boundary. All three pairs confirm the status of <t> as an affricate, but it is remarkable that the merged fricative has post-alveolar articulation in each case.25 In contrast, LW 3.2 ardëc appears to represent a singular subject on contextual grounds, and therefore can be interpreted as *ardënt-s.26 This analysis speaks in favour of <c> being another affricate but differing from <t> through its place of articulation.27 If one accepts the phonetic transcriptions <s> = [s] and <s> = [ç], then the phonetic transcriptions <c> = [ts] and <t> = [tç] appear to be the most natural solution for the matching affricate sounds.

There are several other considerations that yield indirect support to this new phonetic interpretation, or at least are compatible with it. Thus, the prefixed forms si-zënit and ën-zawAo-can be contrasted with the base forms cënit and cawAo- respectively (Sasseville 2020: 520). The prefix si- contains the final i-vowel, while ën- can be reconstructed as en(i) at the Proto-Anatolian level (Boroday & Yakubovich 2018: 8-11). This is an argument for treating the conditioned development of <c> into <t> in a parallel fashion to the development of <s> into <s> after i-, e.g., in the nom.sg. endings of the common gender. The lexical etymologies that are compatible with the affricate interpretation of <c> were already addressed in section 2: this is ca- < *dhhiske/o- and ce-si- < *dfes£e/o-. Although neither of the two can be regarded as assured, the simplification of consonant clusters yielding <c> remains the default assumption in both cases, since <c> cannot be explained via palatalization in either of them.

It is easy to see that the proposed solution (<c> = [ts], <t> = [tç]) is situated between Gérard's suggestion (<c> = /dz/, <t> = /tç/) and Oreshko's alternative (<c> = [ts], <t> = [cç]). Gérard's predilection for the voiced interpretation of <c> has to do with a number of etymologies where this phoneme goes back to the voiced stop, notably Lyd. ciw- 'god' < PIE *djew-/diw- (Gérard 2005: 79). Yet, given that the Lydian palatalization yielding <c> is a language-specific development, nothing precludes dating it later than the devoicing (fortition) of wordinitial stops, which represents a common Anatolian, albeit areal phenomenon. In fact, in the

24 For the identification of the Lydian word for 'daughter', which presumably goes back to Proto-Anatolian *dugater, see Schurr 2006: 1570-1572.

25 The morphological analysis of sfardenx is provided in Yakubovich 2017a: 272, where the pre-syncopated form is, however, reconstructed as *sfardent-is. While acceptable for some early period, such a reconstruction is unlikely for the stage immediately before the syncope occurred, because the nominal ending *-is automatically develops into -is in Lydian (cf. section 2 above).

26 While the stem ardent- may represent a derivative of dent- 'mobile property', the only morphological assumption that constitutes the prerequisite of the present analysis is that ardent- is a genuine consonantal stem, not the result of a secondary syncope.

27 To this one may add that the shape of Lyd. <c>, graphically an upward arrow, is very similar to that of the Anatolian glyph L 376, which was originally use with the value <za/i>, i.e., for a syllable consisting of the affricate /ts/ followed by the vowels other than /u/. The same arrow-shaped sign is used for the letter <t> in the Carian alphabet, where it is likewise interpreted as an affricate (Adiego 2007: 251).

instance of cuwe- 'to erect (vel sim.)', one probably has to reckon with an etymological t-: a se-mantically cogent derivation of Lycian A tuwe-(ti) 'to erect, place vertically' from PIE *(s)teh2u-(García Ramón 2015: 120-124, cf. Kloekhorst 2022: 73, n. 41) must naturally be extended to the derivation of its Lydian cognate.28 Against this background, Oreshko's transcription <c> = [ts] clearly represents a step forward. As for Oreshko's reconstruction of the palatal affricate <t> = [cg], it is arguably prompted by perceived system pressure: the table in Oreshko 2019: 213 contains four other palatal consonants, but no palato-alveolars. I hope, however, to have shown above that in some cases <t> reflects a merger of a coronal stop with the following fricative, which supports the traditional interpretation <t> = [tg].

Lydian stands alone among the Anatolian languages with regard to the presence of two graphemes corresponding to two distinct affricate phonemes. It is, however, unclear whether this peculiarity as phonological or merely graphic. Thus, claims were made that the <za>/<zu>/<zi> series in Hittite may stand for the syllables with two different affricate sounds [ts] and [tj] (lately Patri 2019: 229-238). Whether or not there is substance behind such a hypothesis, the well-known facts of Slavic language evolution ostensibly show the potential for interplay between iotation and palatalizations proper for the extension of phonological inventory, including the rise of multiple affricates. Therefore, the development of affricates in Lydian cannot be regarded as typologically uncommon.

6. Luwic umlaut vs. Lydian syncope

The phenomenon that triggered paradigmatic vowel alternations in Lycian is variously called "umlaut" and "vowel harmony" in academic literature (cf. Hajnal 1995: 78-90, Sasseville 2014/2015: 111). Within the scope of a phonetic word, the Lycian front vowels e and i trigger the anticipatory fronting a > e, the back vowel a triggers the anticipatory backing e > a, while u underlyingly a back vowel but remains transparent to the palatal umlaut, which arguably implies the existence of two allophones, the back vowel [u] and the front vowel [y]. For example, Lyc. A tese/i- 'oath, vow' has the collective plural tasa (nom.-acc.), while Lyc. A atra-, atla- 'person, self' has dative singular forms atli and etli. The same processes play a role in the derivation, contrast the personal names Sbikaza and Sbikezijei, and in the adaptation of loanwords: thus, the Greek personal name AG^vayoQag is adapted into Lycian as Tenagure or Tenegure.29 Anticipatory backing appears to be more ancient than anticipatory fronting and is implemented more consistently (Hajnal 1995: 90). The scope of anticipatory changes is potentially the whole phonetic word: thus, the possessive adjective ehetehe/i- is derived from ahata-'peace', while the allomorph nta 'in' is used instead of nte in the phrasal verb nta ta- 'to put in'. The last peculiarity renders the Lycian changes under discussion more similar to the prototypical vowel harmony, as in Turkish, that to the prototypical umlaut, as in German. On the

28 An independent combinatorial argument against the inherently voiced character of <c> in Lydian concerns the apparent accusative forms of the common gender endowed with the suffix -c-, namely nivis(s)cv, qardoAcv, kastaAcv, laafcv, kafoAcv, pitocv, tacv, intistcv, mursucv, porfcv (cf. the discussion of them in the following section). Not all of these forms are provided with contextual translations or etymologies, but it is clear that they form a natural class. It seems highly that unlikely the suffix -c- was pronounced as a voiced affricate /dz/, on all of the above forms, including those where it is embedded in consonant clusters.

29 The last example is of particular interest, since anticipatory fronting a > e is apparently implemented here before [y], the fronted variant of u. Another Lycian personal name that supports the fronting of u is Xertubi (the vowel e is not expected after x, unless triggered by umlaut). The analysis of Tenagure as a secondary variant of Tenegure in Hajnal 1995: 80 is hard to follow: other things being equal, Greek a is borrowed as a in Lycian.

other hand, many scholars reserve the term "vowel harmony" for the perseverative longdistance vowel assimilation, which prompts them to use "umlaut" for the matching anticipatory change. The latter are in the majority, and the term "umlaut" will accordingly be adopted for the Anatolian languages in the rest of this paper.

Although palatal umlaut has been primarily described with reference to Lycian A, there is no doubt that its traces are also preserved in the other Luwic languages. Thus, Lycian B uwedre/i-'all' features nom.-acc.pl.n. uwadra, while the divine epithet esenemla has variants esanamla and asanamla in the Lycian B language. The Carian nouns en < *eni < *annis 'mother' and ted < *tedi < *tadis 'father' supply a clear indication of anticipatory fronting in the prehistoric period, while otono < *atono- in Carian otonosn 'Athenians' apparently reflects anticipatory backing (Adiego 2007: 259). The situation in Luwian is more complicated. The difference between the front and back low vowels is not reflected in cuneiform orthography, but the earlier Anatolian hieroglyphic inscriptions (up to the mid-ninth century BCE) deploy the signs <a> and <a> in a way that is consistent with the distinction between the word-initial e- and a- in the cognate Lycian A forms, which suggests the distinction between Luwian /^-/ and /a-/. In the majority of cases, the distribution between <a> and <a> is lexical, but the paradigmatic alternation is clearly attested in the first-person possessive pronoun, e.g., nom.sg.c. *a-mi-i-sa, instr. *a-mi-ia-ti but nom.-acc.pl.n. a-ma (Burgin 2016: 8-9). It seems reasonable to assume, as also does Burgin, that the origin of this variation has to do with the vowel quality of the second syllable, which is tantamount to considering this case together with that of Lyc. A tese/i- 'oath, vow' vs. tasa (coll.pl.).30

In contrast, the Lydian language does not show any vowel alternations that can be described in terms of umlaut.31 The factor that plays a defining role for the evolution of Lydian vocalism is strong dynamic stress. As shown in Eichner 1987 and a number of other publications by Heiner Eichner, a number of Lydian vowels, notably e, o, e, and a, normally do not occur in a phonetic word more than once, and therefore their attestations are likely to correlate with the stressed syllables. The frequent correlates of unstressed morphemes are syncope and vowel reduction. As an example of the former phenomenon, one may consider the accusative singular morpheme -v, which frequently occurs outside the consonantal stems. Thus, tac-v 'stele', kastaAc-v 'remains', and similar stems ending in <c> require a historical vowel that caused the palatalization, which implies the reconstruction *tac-iv, *kastaAc-iv etc. for pre-Lydian. Unstressed vowel reduction can be illustrated with the variation among the endings -ev, -av, and -av in the genitive-dative plural. The first two endings characterize stressed morphemes, which correlates with the absence of stressed vowels in the respective stems. A tell-tale case is dat.sg. porlA vs. dat.pl. prwav 'year': the stress shift in the plural form causes the syncope of the root vowel (Yakubovich 2019b: 306-310). In contrast, the gen.-dat.pl. ending -av occurs in barytonic nouns and adjectives, as suggested in many cases by their stem vowels, e.g., esav, sfardetav, sfenav.32

30 This brief discussion obviously does not exhaust the topic of the Luwian umlaut, which should and will be considered elsewhere in more detail. Although it is quite unlikely that the distinction between and /a-/ was limited to the word-initial position in Luwian, I did not attempt to reflect it in my Luwian transcription outside this paragraph, since such an unusual convention is likely to cause confusion unless backed up by detailed argumentation. At the same time, it is appropriate to mention that the proposed account is not quite new: some vowel assimilation rules resembling the umlaut have already been proposed for Luwian on the basis of hieroglyphic evidence in Melchert 2010: 155, albeit without adducing the Lycian comparative data.

31 It particular, this observation is relevant for section 2, since there were attempts to derive the Lydian pronoun es- from the earlier *as(s)i in terms of palatal umlaut.

32 There are also derivational pairs that serve to illustrate unstressed vowel reduction in Lydian: contrast ceqra- '(an installation)' vs. the derivative caqr-la- 'inhabitant of ceqra- (vel sim.)' (Schurr 2011: 76). The rarity of such pairs has mainly to do with our poor knowledge of the Lydian lexicon.

There are reasons to think that the vocalic alternations of the sort addressed above represent a fairly recent and abrupt phenomenon in the history of the Lydian language. On the one hand, the double syncope in Lydian qaAm(A)u- 'king' postdates the borrowing of this title from Luwic (pre-Carian?) *qwalija-muwa-, lit. "(having) the strength of the army" (Valério & Yakubovich 2022). On the other hand, the syncope of stem-forming vowels sometimes results in the mechanical renewal of inflectional endings in the Lydian nominal system. A case that has been known long ago is that of gen.-dat.pl. esvav, which represents an analogical formation disambiguating the earlier *e-sv < *e-sav from the stem *e-s(a)- 'this' (cf. Gérard 2005: 91 and the discussion in section 2 above).33 Since the syncopated *esv 'of/to these' would be indistinguishable from the form of acc.sg.c., the productive gen.-dat.pl. ending -av was secondarily appended to the ambiguous form.

The recent work on the Lydian dating formulae was conducive to identifying many more cases of iterated inflection, which ultimately came into being as a compensation strategy for syncope in inflectional endings. Thus, in the instance of dat.sg. *poruA > *porl > porlA 'in the year', the syncope of the stem-forming vowel *-u- triggered the phonetic change of the dative ending *-A into *-l, while the resulting lack of morphological transparency was remedied through the addition of another *-A (Yakubovich 2019b: 309).34 The same type of iterated dative endings is attested in a number of additional cases, e.g., niwislA, ipsimlA, mAimlA, prafrlA, qelA-k, which bear witness to the semi-productive character of paradigmatic restructuring driven by syncope (Yakubovich 2019b: 307). Additional instances of iterated inflection in the nominative forms will be addressed in the following section. While the reduction of unstressed vowels, including the stem-forming ones, is cross-linguistically common, the compensation strategy consisting in the iteration of inflectional endings is markedly less so. In particular, one may contrast the situations in Lycian A, where syncope, unlike umlaut, almost never leads to inner-paradigmatic variation (van de Kasteelen 2015: 25), and its usual compensation strategy is analogical levelling (van de Kasteelen 2015: 24-25, 34).

7. i-mutation

A salient peculiarity of the Luwic languages is the progressive merger between the thematic stems and i-stems, known as "i-mutation". For example, the paradigm of the Late Luwian pronominal adjective tanim(a/i) 'all, every' features nom.sg.c. tanimis, nom.-acc.sg.n. tanimanza, nom.acc.pl.c. taniminzi, nom.-acc.pl.n. tanima and instr. tanimadi. This paradigm fragment illustrates the basic distributional pattern of the i-mutation in Luwic: the "mutated" stems are declined like i-stems in the nominative and accusative cases of the common gender (both singular and plural) but behave like a-stems elsewhere. To this one must add that the Luwian language eliminated the thematic stems (reflexes of Proto-Indo-European o-stems) as a class, and all of them have possibly passed to the mutation stems.

33 The syncope *esv < *esav is irregular on face value, since in other cases the gen.-dat.pl. ending -av < *-on is preserved in unstressed syllables (cf. the following section). A way to account for this exceptional syncope is to assume that the deictic prefix e- was attached to gen.-dat.pl. *sav when vowel length was no longer a contrastive feature of Lydian phonology (cf. section 2).

34 I follow the traditional connection between dat.sg. in -A and the adverbial suffix -li, which is seen e.g. in Lyc. A ebeli 'there' and teli 'where'. The grammaticization of this suffix in Lydian may have represented a response to the early apocope of the inherited datives in -i. I cannot follow Sasseville 2021 in his attempt to treat this morpheme as a cognate of the Luwian instrumental ending -adi, since the 3sg.pres. verbal forms ending in *-adi show very different reflexes in the Lydian language. I am likewise skeptical with regard to the development *-j > -A postulated in Kloekhorst 2012 (see Yakubovich 2019a: 399-400).

The situation with "i-mutation" in Lydian is apparently not identical. While some of its nuances were addressed in recent publications, this evidence has not yet been summarized. First, the Lydian language preserves the inherited class of thematic stems in the common gender, which was lost in Luwic through the spread of mutation stems. The paradigm under discussion is restricted to the oxytonic stems, and its peculiar endings are acc.sg.c. and gen.-dat.pl. in -ëv, e.g. nom.sg.c. tawsas vs. acc.sg.c. tawsëv 'powerful' or nom.sg.c. aAas vs. dat.pl. aAëv 'other' (Hajnal 2004: 189).35 While acc.sg.c. in -ëv can be explained as a regular reflex of *-om, gen.-dat.pl. in -ëv is presumably analogical in origin (cf. the regular sound change *-om > -av in prwav). It is unlikely that -ëv could be extended from acc.sg.c. to gen.-dat.pl. via simple analogical levelling, but given a class featuring both acc.sg.c. and gen.-dat.pl. -av (presumably the reflexes of oxytonic fl-stems), the extension of -ëv could be conceived as a proportional analogy. Yet another paradigm supporting this analogical scenario consisted of the reflexes of barytonic o- and fl-stems and featured acc.sg.c. and gen.-dat.pl. in -av.36

The second peculiarity, which to an extent complements the first one, is revealed in Sas-seville 2017. All mutation stems in Lydian are synchronically barytonic, judging by the gen.-dat.pl. ending -av and the syncopes affecting this group. For example, nom.sg. sfardëti(s) 'Sardian' (LW 22.13 sfardëtiek) correlates with syncopated nom.pl. sfardëti(s) > sfardënz and dat.pl. sfardëtav. This implies a synchronic complementary distribution between the thematic stems and mutation stems.

Third, as argued in Yakubovich 2019b, the endings that are typical of the mutated stems are secondarily appended to the syncopated endings in Lydian as a means of their morphological renewal. This strategy, which results in the surface doubling of inflectional morphemes, appears to have no parallel in the other Anatolian languages. For example, wissis < *wiss < *wisu-s 'just (vel sim.)' can be contrasted with the inherited nom.-acc.pl.n. ni-wisw-a 'injustice (vel sim.)', where the stem-forming vowel of the stem *wisu- is still preserved as a glide. This example illustrates how the iteration of inflectional endings can encroach upon the domain of barytonic u-stems, even though this does not lead to the complete elimination of this class (contrast the loanword lamëtru- 'Demeter). The same strategy, however, is not limited to a particular stem type. Thus, the adjective ipsimva- 'Ephesian' (acc.sg.c. ipsimvav) apparently underwent syncope ipsimvas > ipsims in nom.sg.c., which was followed by morphological renewal yielding *ipsimsis and then assimilation to *ipsimsis (Yakubovich 2019b: 308 pace Gérard 2005: 87). The sole limitation of the mechanism under discussion is that it normally affected the barytonic stems (only unstressed vowels can be syncopated).

Forth, the Lydian adjectival stems with "i-mutation" are also declined like i-stems in nom.-acc.sg.n. (cf. Sasseville 2017: 133-134). The largest group of forms illustrating this phenomenon are the possessive adjectives in -l(i)-, e.g., LW 2.3-4 akead karolid sapAalid 'it belongs

35 The distribution of stressed and syncopated vowels across Lydian nominal paradigms suggests that most of them had column accent, e.g., can be characterized as either oxytonic or barytonic, although the contrast between dat.sg. porlA vs. dat.pl. prwav reveals a rare case of a mobile paradigm in the Lydian word for 'year' (cf. the preceding section).

36 A different scenario, advocated in Hajnal 2004: 192-194, assumes the neutralization of long and short vowels in closed syllables in Lydian, which would imply the derivation of both acc.sg.c. and gen.-dat.pl. in -ev by sound law. The price of such a reconstruction is high: one has to assume that the nominal ending -av represents a result of secondary contraction, and the whole class characterized by this ending is innovative (i.e. there were no *eh2-stems in proto-Anatolian, while the ¿-stems in the Luwic languages and Lydian represent independent innovations). A counterexample of which Ivo Hajnal could yet not be aware is dat.pl. prwav, which belongs to the archaic mobile paradigm of a u-stem (cf. the previous footnote) and is therefore quite unlikely to have been influenced by the ¿-stems.

to Karo son of SapAa'. Yet, this morphological pattern is also attested elsewhere, e.g., LW 5.2-3 ak tesastid siwamlid mAola 'the right (one) is the mAola-object of Siwami-' or LW 22.5-6 akms aAidad wiswid kazziwv 'we approved for them a good aAida-'.37 The contrast between the noun mAola and the dependent adjective tesastid, or the noun aAidad and the dependent adjective wiswid, should suffice to illustrate the scope of this innovation. The ending -id was presumably introduced through the proportional analogy -as/-ad = -is/x ^ x = -id) in lieu of the inherited zero ending.

Summing up the previous observations, one could argue that the phenomena that are commonly subsumed under Lydian "i-mutation" cannot be reduced to a single morphological change but reflect a set of formally related changes, which varied in scope and motivation. Given that the short unstressed vowels are in principle all liable to syncope in Lydian, one could alternatively argue that the similarities to Luwic i-mutation are restricted to the secondary restitution of Lydian stem-forming vowels. If so, then the Luwian influence on Lydian mutation stems could be reinterpreted as a contact-linguistic phenomenon (see further the concluding section).38

8. Verbal inflection

Oettinger (1978) conjectured the position of Lydian within the Anatolian family based, to a large extent, on several characteristic verbal endings. The Lydian 1sg.pres. ending -u/-w matches the Luwian ending -wi and the Lycian A ending -u. This is a clear common innovation of Lydian and the Luwic languages vis-a-vis the state of affairs in Hittite, where two distinct conjugation patterns are characterized by the 1sg.pres. endings -mi and -(h)hi. The uniform "non-Hittite" 1sg.pres. ending *-wi presumably represents an analogical formation, built on the model of the 1pl.pres. ending *-weni. In the instance of 1sg.pret., Lydian and the Luwic languages innovated in two opposite directions, even though they share the trend toward eliminating the formal opposition between the two conjugations. The Lydian ending -v generalized the inflectional marker of the mi-conjugation (cf. Hitt. 1sg.pret -un), while Luwian (C) -(h)ha and Lycian A -xa continue the respective marker of the M-conjugation. The last isogloss represents a clear dividing line between Lydian and the Luwic languages (as conventionally understood), which reflects two innovations in the opposite directions.

The purpose of this section is to address two Lydian endings that can be regarded as more archaic than their counterparts in the Luwic group. The first one is -ris, -rs, which had long been regarded as a participial suffix but was identified as the marker of 3pl.pret. in Melchert 2004: 147, n. 14. This suggestion is accepted on contextual grounds in Gérard 2005: 104, but the forms in -ris, -rs are still considered there as etymological participles with the copulae omitted.39

37 The neuter noun mAola '(a type of memorial object)' comes as a surprise against the background of a-stems normally associated with the common gender in the Lydian language. Given that the ë-vowel is reduced to a in unstressed position, the simplest solution is to analyze mAola is a neuter barytonic n-stem (the Indo-European morphological class exemplified by Gk. of|ja 'mark, sign'). Contrast Lyd. sadmë- 'inscription, seal' as an oxytonic n-stem of the common gender. Pace Gérard 2005: 83, the parallel between mAola and anlola 'burial complex (vel sim.)' would be deceptive, since mAola is accompanied by singular attributes and thus cannot represent a collective plural noun.

38 As such, it could be compared to the similarity between Luwic and Lydian possessive adjectives (Gérard 2005: 84), which precludes a genetic explanation in view of the different origins of the respective possessive suffixes but probably reflects a tendency toward double case marking in the Anatolian linguistic area (Luraghi 2008).

39 Gérard (2005: 104) also postulates the existence of a related ending -rst but comes short of assigning a semantic interpretation to this morpheme. In my opinion, none of its two alleged attestations are compelling.

A new way to the historical understanding of this morpheme was opened in Gusmani 2010, where it was compared with the Gathic Avestan 3pl. pluperfect marker (Y. 32.11 cikoit-aras 'they had appeared') as well as the third plural ending -ras of the Younger Avestan aorist optative, namely 3pl. daiS-iia-ras from da 'to give, put', 3pl. hu-iia-ras from hu 'to press' (?), 3pl. jam-iia-ras (besides jam-iiq-n) from gam 'to go', 3pl. bu-iia-ras besides bu-iiq-n from baw 'to be(come)', and possibly 3pl. h-iia-ra<s> from ah 'to be', if the emendation is correct. It is worth noting that the Avestan language preserves the contrast between the 3pl. perfect ending *-r or *-er as in ad-ara 'they have said', caxr-ara 'they have made' and the 3pl. pluperfect/precative ending *-rs, as in the forms cited above in this paragraph.

The situation in Hittite requires special comments. The Hittite endings of the hi-conjugation are generally acknowledged to represent the historical counterparts of the Indo-Iranian inflectional morphemes appearing in the perfect system. On functional grounds, we expect a match between the Avestan 3pl. pluperfect ending and the Hittite 3pl.pret., but the relevant Hittite ending is -er /-er. One might be tempted to entertain a formal comparison with Avestan 3pl.perf. of the type ad-ara, but an alternative possibility would be assuming compensatory lengthening in the original ending *-rs/*-ers.40 The last solution, offered in Jasanoff 2003, § 24, received a direct confirmation through the identification of the Lydian 3pl.pret. ending -ris, -rs. According to the tentative suggestion of Gérard 2005: 103-104, the Lydian allomorph -ris reflects a secondary anaptyxis in the word-final consonant cluster; the comparison with Avestan and Hittite fleshes out this hypothesis. A distributional peculiarity of Hittite -er /-er is that this ending is not restricted to the verbs of the hi-conjugation, where it is etymologically at home, but generalized to all the verbal forms of 3pl.pret. This development reflects, of course, the mutual influence of the two Hittite conjugations, which lost whatever functional distinctions they had shown at the Indo-Anatolian stage. Its reverse is attested in 3pl.pres., where the ending -anzi spread from the mi-conjugation across the board: this is why the Avestan inflected forms of the type ad-ara simply have no formal matches in Hittite. Crucially for our discussion, Hittite shows traces of both nt-endings and r-endings in the third person plural, and the situation in Lydian was presumably the same (cf. the 3pl.pres. form LW 22.13 taqtul-at, Gérard 2005: 100-101).

Turning to the Luwic languages, we observe here a degree of formal merger between the two Anatolian conjugations, which is stronger than in Hittite. The only paradigmatic form where the two conjugations are contrasted in Luwian (and possibly in Lycian) is 3sg.pres., while the endings of 3pl.pret. show a uniform pattern: -anta in Luwian and -ëte in Lycian A. For a possible origin of these two endings, which are obviously cognate with one another, see Yoshida 1993: 30 with ref; for our present purposes, it would suffice to acknowledge that they continue in some way the 3pl.pret. inflectional ending of the mi-conjugation. Nowhere in the Luwic branch does one find traces of 3pl.pret. in *-p/*-ers, which clearly represents a negative innovation of this subgroup vis-à-vis the situation in Lydian or Hittite (cf. Kloekhorst 2022: 72, n. 36).

The second ending that is relevant for this discussion is 3sg.pret. -l. For a long time, it was analysed as a former suffix of a periphrastic formation, comparable to the participial suffixes

LW 23.1 dacuwerst, arguably occurring in clause-initial position, can be interpreted as 3pl.pret. in -rs and the Wackernagel clitic =t (e.g., dacuwers=t). The case of LW 23.22 piferst is simpler: this form belongs to the apodosis of a curse formula and therefore must feature the 3sg.pres. ending -t attached to the stem pifers-. Differently Euler and Sasseville 2019: 131, n. 22.

40 One reason that can be advanced for such a reconstruction is the matching 3sg.pret ending -s, typical of the Hittite te'-conjugation. Cf. Kümmel 2018: 239: "Both Anatolian and Indo-Iranian appear to presuppose an ending set sg. *-s / pl. *-rs/*-er besides sg. *-e / pl. *-(e)r, the former being connected to non-present usage".

attested in Classical Armenian (-eal) and Old Church Slavic (-At>). Although still treated as mainstream in Gérard 2005: 102, this hypothesis lost much of its allure after the demonstration that the matching 3pl.pret. ending -ris, -rs has nothing to do with a periphrastic construction. I am not aware of any parallels for a finite verbal form giving way to periphrasis in the third person singular but not in the third person plural. Yet, an obvious difficulty to tracing -l back to a Proto-Anatolian verbal ending is the absence of comparable third person singular endings in the other Anatolian languages.

An attempt to resolve this difficulty was undertaken in Yakubovich 2005: 86-87, where 3sg.pret. -l was derived from the Anatolian verbal ending -t having the same function on the assumption of a lambdacism that occurred in the history of Lydian. A "functional necessity to differentiate Lyd. 3 sg./pl. prs. -t / -d from Lyd. 3 sg./pl. prt. -l" was supposed to play a role in this process. Unfortunately, the licensing conditions of Lydian lambdacism were not precisely defined in that paper, while the other parallels adduced there for the same phenomenon in Lydian are all open to doubt.41 The next attempt to account for the same Lydian ending through lambdacism (via the intermediate stage of flapping) was undertaken in Sasseville 2021: 642-644, where one spots a renewed effort to describe the history of the relevant morphemes in terms of functional differentiation. In particular, it is suggested that "the preterit tense is recharacterized by generalizing the flap from the lenited mi-conjugation across all stem classes". In contrast, in the present tense "the flap did not go through but ... the dental fricative has in fact been restored".

The evaluation of this reconstruction requires some additional comments. The Lydian 3sg.pres. ending -t/-d features two allomorphs that are distributed according to the rules of Proto-Anatolian lenition (Gérard 2005: 99-100). The Lydian 3sg.pres. ending -l shows no comparable contrasts but presumably reflects lambdacism, which can be described as an advanced stage of lenition. In contrast, both Luwian and Lycian A feature show the contrast between the fortis and lenis endings in both the presence and the preterit of the mi-conjugation (Luw. 3sg.pret. -tti/-di, 3sg.pret. -tta/-da; Lyc. A 3sg.pres. -ti/-di, 3sg.pret. -te/-de). Since all the wordfinal short vowels drop in the Lydian language, assuming a Luwic-type system for pre-Lydian entails a potential for the merger between 3sg.pres. and 3sg.pret. in further history of the language. Therefore, one could argue that a sequence of two separate changes was implemented in the history of Lydian in order to avoid undesirable ambiguity. On the one hand, the advanced stage of the lenition was generalized in the preterit, while the more archaic pronunciation of the lenited consonant was retained in the present tense.42 On the other hand, the morphological distribution between the fortis and lenis variants was levelled in 3sg.pret. but not in 3sg.pres. While such an analogical conspiracy is not out of the question, a historical scenario that does not feature any preventive analogies should be given preference ceteris paribus.

It is indeed possible to account for the Lydian changes in terms of sound laws if one starts with a different system, namely 3sg.pres. *-ti/-di and 3sg.pret. *-d. The third person preterit ending without a final vowel is indeed attested in Hittite, and judging by the Indo-European comparative data, it is more archaic than its Luwic counterpart. There is no indication that the laws of Proto-Anatolian lenition were applied to word-final consonants, while the Lydian

41 The comparison between Lyd. arlil(i)- 'household member (vel sim.)' and Luw. (H) atr(i)- 'person, self' was explicitly withdrawn in Yakubovich 2017b: 16, n. 15. For antola as a rare dissimilated variant of anlola 'memorial', see Yakubovich 2019a: 405.

42 Technically speaking, such a change must be conceived as a result of interaction between the two dialects or registers. For example, one could claim that the standard pronunciation of 3sg.pret. was borrowed at some point from the low register (basilect) of Lydian.

nominal inflexion, e.g., qid 'what', mrud 'stele', suggests that the default realization of the inherited word-final coronal stop was voiced/lenis. The only additional step that is required to arrive at the attested distribution is the advanced lenition (lambdacism) *-d > -l in clause-final position, which predates the apocope. The standard reconstruction of the Proto-Anatolian syntax implies the unmarked SOV word order, and despite many instances of right dislocation, the verb-final word order remains the most common syntactic pattern in the Lydian language (Rizza 2013: 93).43 Therefore, it is only natural that the Lydian verbal ending of 3sg.pret. generalized the variant with lambdacism, while the nominal endings of nom.-acc.sg.n. retained the variant without it.44

The proposed alternative comes virtually at a zero cost. Since the 3pl.pret. ending -ris, -rs represents a Lydian archaism that is not shared by the Luwic languages, it seems only natural to extend the same analysis to the 3sg.pret. ending -l. Although the Lydian lambdacism bears a degree of resemblance to rhotacism/flapping in Late Luwian, in that both phonetic processes reflect further weakening of the lenis dental obstruents, the trivial character of both changes is compatible with treating them as independent innovations, while the restricted licensing conditions of the Lydian lambdacism advocated in this section support the same conclusion.45

9. Verbal derivation

A recent dissertation devoted to the study of verbal derivation in Luwian, Lycian, and Lydian is now published as Sasseville 2020. It reflects a significant progress in the synchronic study of the languages involved, since most of the relevant verbal forms are now assigned to the derivational classes. Furthermore, Sasseville has demonstrated that the majority of verbal classes reconstructed for the Luwic languages and Lydian correspond to one another, while additional matches to the respective derivational types are found in Hittite. This analysis vindicates the distribution between the lenited and non-lenited verbal endings as an archaism that is common to Lydian and the languages of the Luwic branch (cf. section 4). More controversial, is my opinion, is the author's contention that the morphology of verbal derivation in Luwic and Lydian displays non-trivial common innovations vis-à-vis Hittite. In what follows I intend to go over the evidence for such innovations and highlight its inconclusive character.

The summary of the author's claims that are potentially relevant for refining Anatolian phylogeny can be found in table 82 (Sasseville 2020: 552). First, it is observed that the Hittite ^¿-conjugation verbs in -axx- correspond to the verbs in *-fl- in both Luwic and Lydian, which are both endowed with the endings of the historical mi-conjugation. In order to evaluate the significance of this parallel, it is important to keep in mind that although both the Luwic lan-

43 As an illustration of this syntactic generalization, see the analysis of LW 22 in Yakubovich 2017a: 267. If one subtracts the obvious instances of right dislocation, only 3 out of 16 clauses of this inscription deviate from verbfinal word order.

44 For an additional instance of lambdacism on Lydian loanwords from Greek, compare section 4. In this case the change is limited to word-initial position, where *d- was proscribed in pre-Lydian.

45 See n. 34 above for my criticism of Sasseville's attempt to explain the Lydian dat.sg. ending -A as a result of lambdacism. Given that -n- is attested many times in intervocalic position in Lydian, I am equally skeptical regarding the comparison between the Luwian infinitive in -una and the Lydian infinitive in -l, offered in Sasseville 2021: 644. More promising, in my opinion, would be the comparison between the Lydian infinitive marker -l and the Luwian abstract suffix -ar (as in wassar (H) 'favour', dupadupar-sa (C) 'striking, affliction'), although given a limited number of the Luwian forms endowed with such a suffix, this suggestion remains very tentative. Yet another possibility is mentioned in Yakubovich 2017a: 274-275, n. 10.

guages and Lydian display the tendency toward the merger between mi-and M-conjugations, they manifest it to varying extents. In Lydian the merger is completed, while Luwian still shows difference between the "mi-conjugation" endings -tti, -di and the "foi-conjugation" ending -i in 3sg.pres. (the situation in Lycian is less clear). Therefore, only in the instance of Hittite vs. Luwian can one be sure that the derivational class under discussion underwent a conjugation shift, whereas the situation in Lydian can be described as a mechanical result of the merger between the two conjugations. As for the development *-ax-ti > *-a-ti, it presumably reflects a sound law: the laryngeal disappears before a dental stop (Sasseville 2020: 71). There is no exact way of knowing where this sound change took place in the common ancestor of Lu-wian and Lydian or separately in both languages, but the fact that it was not followed by the Anatolian lenition *-a-ti > **-a-di tips the scales in favour of the latter possibility. Summing up, although the evolution of the morphological class under discussion indeed yields the same result in Lydian and the Luwic group, there are arguments for a parallel innovation.

Second, the author suggests that Proto-Anatolian verbs in *-ax-je/o- yield verbs in -aizzi/ -anzi in Hittite and verbs in *-a-di/*-ai-nti in the ancestor dialect of both Luwic and Lydian.46 I submit that the synchronic Lydian data fail to provide sufficient support for the proposed interpretation. While the Lydian corpus contains several 3sg.pres. forms in -ad (fa-pad, fa-si-tawad, is-follad, pitad, si-lawad, possibly also et-qratad, tarpzad, kipzad), there is not a single 3pl.pres. in **-ait attested in the same corpus. Sasseville (2020: 79) attributes this gap to chance, but the matching 3pl.pres. forms may have simply ended in -at, in which case they were formally indistinguishable from the 3pl.pres. forms of the ax-class verbs. In the absence of paradigmatic pairs, we are simply not sure where a given 3pl.pres. form in -at reflects the -at/-at class (suffix *-ax-) or the -ad/-at class (suffix *-ax-je/o-). The existence of the -ad/-at class derives support from the isomorphic Lydian stems in -id/-it and -od/-ot. If this analysis holds, then the Lydian verbs in -ad / -at occupy an intermediate position between the matching Hittite class in aizzi/-anzi and the matching Luwian class in -adi/-ainti.

Third, David Sasseville stresses the functional difference between the Hittite reflexes of the Indo-European *-eje/o-stems, which are few in number and restricted to the deverbatives, and their counterparts in Luwic and Lydian, which are far more frequent and can also function as denominatives. This is indeed a significant distinction, but it is by no means clear that the Hittite distribution reflects a more archaic state of affairs, because the reflexes of the denominative transitive verbs in *-eje/o- are also productive in several Indo-European branches, for example, in Indo-Iranian, Greek, Germanic, and Balto-Slavic. If one assumes that these are parallel innovations (cf. the discussion in Sasseville 2020: 253-254), it becomes very easy to claim that Luwic and Lydian have likewise independently extended the function of their *eje/o-verbs.

Fourth, the same scholar maintains that Anatolian stems in *-je/jo- have given up their ablaut in Proto-Luwic and generalized the e-grade, i.e., *-je-. He adds that "the Lydian data is less clear, so we cannot prove that it had the same innovation as Luwian and Lycian, but so far nothing speaks against it" (Sasseville 2020: 551). Using such an argument for phylogenetic purposes would naturally render the discussion circular.

Finally, the class of Lydian verbs in -od/-ot (e.g., tro(d)-(d) 'to hand over') is tentatively equated with the Luwian verbs in /-au-/ or /-u-/ in Sasseville 2020: 551 (even though this consideration is introduced as carrying less weight than the previous ones). The author could not yet have been aware of the segmentation of the Luwian 1pl.pret. ending /-unta/ and 1pl.impv.

46 Here and below in this section, the forms divided by slashes refer to the 3sg.pres and 3pl.pres endings (+ thematic vowels) belonging to the same paradigm.

ending /-untu/, the evidence for which was presented in Melchert & Yakubovich 2022. As a result of this morphological discovery, the majority of forms assembled in the problematic class of the Luwian verbs in /-au-/ or /-u-/ can be re-assigned to other stems. For example, the Lu-wian form na-ak-ku-us-sa-a-u-un-ta (C), interpreted as 3pl.pret. /nakkussau-nta/ in Sasseville 2020: 194, can be segmented as 3pl.pret. /nakkussa-unta/ under the new analysis, which is conducive to assigning the relevant verb to the class of a(i)-stems (Proto-Anatolian suffix chain *-dx-je/o-).47 The elimination of the Luwian stems in /-au-/ or /-u-/ admittedly leaves a large group of Lydian stems in -od/-ot in need of a historical explanation: whatever it is, the rise of this productive class apparently represents an innovation that is not shared by the Luwic group.

The critical remarks of this section need not be taken as an indication that the Lydian language lacks any innovations in verbal stem formation with suggestive counterparts in the Lu-wic languages. The Luwic and Lydian verbal morphology undoubtably shared the general vector of development, as can be ostensibly seen from their similar but not identical handling of the inherited opposition between the mi- and ^¿-conjugations. I submit, however, that none of the observed convergent developments in this domain are specific enough to warrant the intermediate Luwic-Lydian stage in the reconstruction of the Anatolian verbal system.

10. Clause-initial particles

A common feature of many Anatolian languages is the presence of clause-initial particles, to which the Wackernagel clitics are typically appended. Such particles serve the goal of textual cohesion, functioning as either strong sequencers (markers of subsequent or consequent events) or weak sequencers (indicators that the events follow each other in a natural way). In the instance of Hittite, the most common clause-initial particle is nu, even though ta and su are also widely attested in Old Hittite, and the semantic difference between the three particles is understood only partially (Luraghi and Inglese 2018).48 In the Palaic language, the clause-initial particle nu, which appears to represent a strong sequencer with the additive nuance 'then, furthermore', co-occurs with the clause-initial particle a, which functions as a weak sequencer.

A non-trivial similarity between the Luwian and Lydian languages consists in the cognate sets of clause-initial particles a, pa (Luwian) and ak, fak (Lydian).49 The relevant Luwian evidence is addressed in considerable detail in Sadykova 2019; it is limited to the cuneiform texts, because the particle pa disappeared in Late Luwian. The particle a appears as weak sequencer:

47 The remainder of the Luwian verbs booked under the stems in /-au-/ or /-u-/ are likewise amenable to alternative interpretations. For example, malhu- (C) 'to crush' and talku- (C) 'to flatten' are both amenable to interpretations as root stems ending in labiovelars, even though /-w-/ may represent a historical suffix in both cases. As for the sequence nu-u-tu-, nu-du- in KUB 35.113, this is probably not at all a verbal stem but a sequence of Palaic particles nu" dw (there is no proof that KUB 35.113 contains a Luwian text as opposed to Palaic).

48 One of the typical functions of Hittite nu is marking the boundary between topic and comment in pragmatic configurations that span more than one clause (Widmer 2009). This appears to be a paradigmatic case of nu functioning as a strong sequencer.

49 The relationship between Luw. a and Lyd. ak is commonly acknowledged. The comparison between pa and fak was offered in Carruba 1959: 34 but not universally followed. I am not aware of any attempts to compare the particles under discussion that would be backed by their semantic study. A step backward was an attempt to plead for the etymological relationship between the clause-initial particle pa and the contrastive clitic * pa in Luwian cuneiform texts (cf. Yakubovich 2010: 62).

it links the Luwian clauses within larger textual units but does not order or organize the underlying events and therefore is usually left without translation. In contrast, the particle pa may mark the sequence of events, e.g., KUB 35.103 iii 4-6 'Let them lift and remove this child. Then (pa) let them put him to his mother's breast', or the cause-and-effect relationship, e.g., KBo 12.100 obv. 12-13 '(Goddess Kamrusepa [looked from the] ...ed [sky]. Then (pa) she saw it'. This functional range is compatible with defining pa as a strong sequencer. In many instances, however, the effect of a situation is expressed by a modal sentence, e.g., KUB 35.54 ii 42-45 'Here are heaven and earth. Then (pa), as heaven does not become earth and earth does not become heaven, may this offering likewise not become [...]'. The last passage is to be understood in its ritual context: the presentation of the models of heaven and earth provides a pretext for a homeopathic magic incantation. Yet, the correlation between the inferences and modal statements need not be culturally specific: in ordinary life, many of our attempts to reason logically represent motivated requests, e.g., "It's cold here, close the window!".

Turning to the situation in Lydian, there is a consensus that the final consonant of ak and fak goes back to the clitic particle tk 'and', which is also attested in Lydian (Gusmani 1964: 54, 119, cf. section 3).50 This naturally facilitates the comparison between the Lydian particles and their proposed Luwian cognates. As for the distribution between Lyd. ak and fak, it has been acknowledged that the occurrences of the latter particle cluster in the apodoses of the curse formulae (Gusmani 1964: 118). Thus, LW 4b 2-5 features the following sentence: aktmA4 qis fensAipid fakemAet qAdans artymuk wcpaqent 'Whoever causes damage to it, may (the god) QAdans and Artemis trample on him!'. This pattern cannot be called accidental, since there are no curse formulae in the Lydian corpus where fak would appear at the beginning of the protasis or ak would head the apodosis. In comparative terms, it appears to be perfectly compatible with the Luwian distribution: by definition, the apodosis of a conditional sentence addresses the consequences of the events mentioned in its protasis, and therefore the particle at the beginning of such clauses could be categorized as a strong sequencer.

Yet, the direct projection of the functions of Luwian particles upon Lydian is likely to be misleading.51 One of the consequences of the Lydian apocope (cf. the previous section) is the loss of formal distinctions between the present indicative and imperative forms in the third person. Thus, wcpaqent of the previous citation formally continues the indicative form 'they strike' as well as its imperative counterpart 'may they strike'. One may, however, doubt that the Lydian language lost any ways of expressing modal distinctions: if morphology was no longer capable of it, the functional load was probably transferred to syntax. The only element that could convey the optative modality in the Lydian curse formulae was the same particle fak, while its obligatory use in this context strengthens the hypothesis that it was grammaticized as a modal marker and can be translated into English as 'may'. This assumption does not in any way contradict either the comparison between Luw. pa and Lyd. fak or the semantic reconstruction of fak as a strong sequencer: its reanalysis within Lydian was presumably facilitated by the frequency of modal clauses in the apodoses of conditional sentences and other contexts favouring the use of strong sequencers.

The evaluation of this hypothesis naturally depends on the interpretation of the contexts other than curse formulae featuring the use of fak. Although our poor knowledge of Lydian

50 One reason why fak cannot be synchronically analyzed as a morpheme combination **fa* k is the presence of the clause-initial particle fa in Lydian, which is etymologically unrelated to fak and must be analyzed as a pre-verb fa- detached from the verb in certain dependent clauses (cf. Carruba 1959: 34-35).

51 The final part of this section summarizes a part of the presentation "The Lydian Particle fak in Curse Formulae and Beyond", held jointly with Anna Sadykova at the conference Beyond all Boundaries: Anatolia in the 1st Millennium BC (Ascona, Switzerland, June 2018).

lexicon complicates the understanding of many passages, the modal interpretation imposes itself in LW 10.16-18 fakeum an is-lodaA alarmn kastaAcv kudenak enas amas 'May one honour my remains/grave just as [one does] the ancestors of mine' (see Yakubovich 2019a for the detailed interpretation of this context). Another noteworthy passage is LW 3.3 ak qis qisred fakeas si-lawad 'Whoever acts as a caretaker, may he be reverent' (Yakubovich 2019a: 405), where the modal interpretation is also beyond question, but the connection between the two clauses hardly qualifies as strong sequence.

A challenge to the translation of fak as 'may' comes from those passages that contain verbal forms in -l, to be formally interpreted as either infinitives or 3pl.pret. The best-understood among them is LW 22.3-4 mAimns aAidad wiswid kazzirs fak sfardetaek an katwvel anead fentasenav 'The mAimna-clan (Mermnads?) accepted the aAida- as just, (it is) for the Sardeans and fentasena-group to acknowledge it, too'. Since the predicate katwve- 'to acknowledge (vel sim.)' cannot be linked to an animate singular noun capable of acting as its nominative subject, one must accept the infinitival interpretation of katwvel (Yakubovich 2017a: 274). The translation given above implies that the infinitives accompanied by fak can be further combined with oblique-case agents to form final clauses or independent modal clauses. The function of fak as head of a final clause would be well compatible with its reconstruction as a strong sequencer, while the extension of a modal particle fak to infinitival clauses would also be a trivial matter. Further speculations on this topic should probably be postponed to the point when other constructions of 'fak + infinitive' are adequately understood, but one can already now say that this group of examples is unlikely to undermine the comparison between the Luwian and Lydian clause-initial particles.

11. Conclusions

The new findings addressed in the previous sections naturally bear upon the discussion about the genetic and areal connections of the Lydian language. To begin with the negative result, they do not appear to add any weight to the hypothesis that Lydian represents an outlier within the Anatolian group of languages or does not belong to this group. Any support for such a claim would have to offer new isoglosses that unite all the Anatolian languages except for Lydian or to provide exclusive links between Lydian and non-Anatolian Indo-European languages. One result of recent research was rather the opposite: the Lydian reflexes of the "laryngeals" turn out to be similar to those found in the other Anatolian languages, and distinct from what is attested elsewhere in Indo-European (section 3). The etymology linking the Lydian proximal deictic pronoun to its Anatolian counterpart removes one more obstacle to treating Lydian as a typical Anatolian language (section 2). In contrast, questioning the participation of Lydian in Proto-Anatolian lenition is not based on an alternative proposal about the distribution of fortis and lenis stops in the history of Lydian. Rather, these doubts are introduced as a background for narrowing down the empirical base in the discussion of a new controversial reading of the Lydian letter <d> (section 4). Therefore, from the purely methodological perspective, they are more relevant for evaluating this new reading than for refining Lydian phylogeny.

Nor did recent research yield arguments for clustering Lydian and Hittite within the Anatolian group. The isoglosses shared between these two languages but not observed in the Lu-wic group must either represent common archaisms (e.g., 3pl.pret. in *-rs/*-ers, section 8) or must be evaluated as homoplasies (e.g., palatalization *ti > /tsi/, section 5). Now, as before, it seems necessary to treat Hittite as an outlier within the Anatolian group of languages. This

does not, however, preclude the secondary changes that are common to all the members of this group and reflect areal convergence. Here one may mention the devoicing of stops in word-initial position, which constitutes a well-known feature of the Anatolian linguistic area and is briefly addressed in section 5.

The new evidence pertaining to the relationship between Lydian and the Luwic group is less straightforward. To be sure, some of these changes can also be explained as homoplasies: this holds, in particular, for the merger between the mi-and foi-conjugations, which yielded different results in Luwo-Lycian and Lydian, or the Luwian rhotacism/flapping vs. Lydian lambdacism (section 6). In other instances, one might suspect common archaisms: this is, for example, the case of the cognate word-initial particles functioning as strong and week sequencers in Luwian and Lydian (section 10). Just as the strong sequencer pa disappeared in Late Luwian, so this may have been the situation in both Hittite and Palaic too, while Hittite bears witness to the later generalization of nu at the expense of a. At the same time, the Lydian development *k > s (section 2), if correctly identified, is certainly an innovation, and there are no structural reasons to treat it as homoplastic vis-a-vis a similar assibilation of palato-velar stops in the history of the Luwic languages.

Given that the Lydian-speaking area was geographically adjacent to the territories populated by the Luwic speech communities, and sometimes overlapped with them, one might legitimately wonder whether some of the Luwic-Lydian isoglosses might be due to secondary contacts. Such an explanation appears to impose itself in the instance of i-mutation (section 7). On the one hand, the proliferation of i-stems in select groups of cases is peculiar enough to argue against the hypothesis of accidental similarity, on the other hand, its lexical distribution in Lydian and the Luwic languages does not match. Thus, Luw. tad(i)-, Lyc. A tede/i-, Car. ted 'father' contrasts with Lyd. taada- 'id.' (Kloekhorst 2022: 74), while Luw. ann(i)-, Lyc. A ëne/i-, Car. en 'mother' contrasts with Luw. ëna- 'id.'. In fact, the majority of the innovative "mutated" endings attested in Lydian represent secondary accretions, as a result of either doubling the syncopated endings or deriving possessive adjectives. If so, one can treat the relevant inflectional morphemes as Luwic calques, which must have arisen at the time when the Luwic influence upon Lydian was particularly strong.

Another instance where one should probably assume areal diffusion is aphaeresis in pre-verbs and postpositions, as was mentioned in section 4. Thus, the Lydian preverb da- represents a counterpart of nte in Lycian A and both go back to Proto-Anatolian *endo, while the Lydian postposition dav can be compared with Carian postposition Sen, and both continue Anatolian *endo-en (Yakubovich 2020b: 304). Yet, the Luwian forms a-an-ta, a-an-da (C) plead for the absence of aphaeresis in the matching preverb. Therefore, the initial vowel probably disappeared in the morpheme under discussion only in those Anatolian languages that were spoken in the western part of Asia Minor, perhaps in connection with the fusion between the preverb and the verbal forms.52

These steps made, several other changes in the history of Lydian can also be reinterpreted as areal features. This holds, for example, for the assibilation *k > s vis-à-vis *k > /ts/ in Luwian and *k > s in Lycian and Carian (section 2). One may adduce here a parallel with the Indo-European satdm languages: while the satdm/centum distinction is usually not used nowadays for defining subgroups, it is still appropriate to regard it as an areal isogloss. Furthermore, although the similarities in verbal derivation treated in section 9 are not specific enough to de-

52 This areal configuration does not really change if one follows Kloekhorst 2008a: 185 in adopting the reconstruction *h1ndo instead of *endo. One could assume that *n was processed as a syllabic consonant among the Lu-wians but failed to syllabify in the west of Asia Minor, but this difference would still cut across the Luwic group.

fine a genetic node, some of them may well bear witness to convergent evolution. The relevant contact-induced changes do not contradict the existence of the categorical innovations that separate the Luwic verbal system from its Lydian counterpart, such as the generalization of particular endings of lsg.pret, 3sg.pret, and 3pl.pret in the Luwic languages (section 8).

It would be, however, unfair to state that the observed difference between Lydian and the Luwic languages is mainly due to the evolution of the latter group. The two innovations of Lydian treated in this paper are not only unusual for the languages of Ancient Anatolia but also typologically rare in general. This is the fortition *j > [ö] (section 4) and the strong syncope leading to the iteration of the syncopated inflectional endings (section 6). Both changes appear to have taken place relatively recently in the history of Lydian: if one follows the traditional readings, /j/ is still absent in the language of Lydian inscriptions, while the morphological system created by the iteration of inflectional endings is baroque to the point of being inherently unstable (cf. section 6). One possibility to account for both changes is to assume an abrupt language shift in the history of Lydian. If /j/ was absent in the source language and the shift was rapid, this could explain the substitution of this phoneme by its closest available equivalent in the course of imperfect language learning (cf. the substitution of /j/ by [ö] in recent loanwords into Lydian). The renewal of inflectional endings would likewise make sense in the context of imperfect language learning, when the syncopated case markers were no longer perceived as such by the next generation of Lydian speakers. Although the scenario of language shift would remain no more than theory unless substantiated by the relevant historical evidence, it seems important to stress the availability of a hypothesis that provides a uniform explanation for both postulated changes whose only common feature is their typological oddity.

As stated in the introduction, the set of topics highlighted in this paper is rather random, being essentially based on the domains of recent advances in Lydian linguistics. Nevertheless, I hope to have demonstrated that even such a limited dataset is conducive to progressing the discussion of the place of Lydian among the Anatolian languages.

References

Adiego, Ignacio X. 2007. The Carian Language. Leiden: Brill. Beekes, Robert S. 2003. Luwians and Lydians. Kadmos 42: 47-49.

Beekes, Robert S. 2010. Etymological Dictionary of Greek (with the assistance of Lucien van Beek). Leiden: Brill. Boroday, Sergey, Ilya Yakubovich. 2018. Hittite local adverbs in comparative perspective. In: Elisabeth Rieken (ed.). 100 Jahre Entzifferung des Hethitischen: Morphosyntaktische Kategorien in Sprachgeschichte und Forschung. Akten der Arbeitstagung der Indogermanischen Gesellschaft vom 21. bis 23. September 2015 in Marburg: 1-22. Wiesbaden: Reichert. Burgin, James. 2016. Graphical Origins of 'Initial-A-Final' in Hieroglyphic Luwian. Studi Micenei ed Egeo-Anatolici 2, 2016: 7-32.

Carruba, Onofrio. 1959. Studi sul nome, sui preverbi e sulle particelle in lidio. Quaderni dell'Istituto di glottologia

dell'Università di Bologna 4: 13-43. Dale, Alexander. 2015. Walwet and Kukalim. Kadmos 54: 151-166.

Eichner, Heiner. 1987. Die Entdeckung des lydischen Akzents. Bibliotheca Orientalis 44: 79-88. Eska, Joseph F. 2018/2019. Grounding Celtic Diachronic Phonology I. Die Sprache 53(1): 17-32. Euler, Katrin, David Sasseville. 2019. Die Identität des lydischen QAdäns und seine kulturgeschichtlichtlichen Folgen. Kadmos 58: 125-156.

García Ramón, José Luis. 2015. Licio, Griego, Indoeuropeo: I. Lic. epñnene/i- 'hermano menor', lat. opiter, aaa. aftero, IE *h1op(i)- 'después, detrás'. II. Lic. tuue- 'poner (en pie)', IE *(s)teh2u-. III. Lic. Malija hrixuuama- 'Malia supervisora' (: Atena enioxonoç, enL^Qavoç, enÎKOUQOç), hit. ser huuai-, hom. £ql-ouvloç. In: Emmanuel Dupraz, Wojciech Sowa (eds.). Genres épigraphiques et langues d'attestation fragmentaire dans l'espace méditéranéen. Acte du colloque de Rouen, 25.-27.06.2012: 165-176. Rouen: Presses universitaires du Rouen et du Havre.

García Ramón, José Luis. 2016. Vedic indrotá- in the Ancient Near East and the shift of PIE *h2euh1- 'run' ^ Core IE 'help, favor'. In: Dieter Gunkel et al. (eds.). Sahasram Ati Srajas: Indo-Iranian and Indo-European Studies in Honor of Stephanie W. Jamison: 64-81. Ann Arbor: Beech Stave.

Garnier, Romain, Benoît Sagot. 2020. New results on a centum substratum in Greek: The Lydian connection. In: Romain Garnier (ed.). Loanwords and Substrata: Proceedings of the Colloquium Held in Limoges (5th-7th June, 2018): 169-200. Innsbruck: Institut für Sprachwissenschaft der Universität Innsbruck.

Goedegebuure, Petra. 2010. The Luwian Adverbs of Place and Manner. In: Itamar Singer (ed.). Ipamati kistamati pari tumatimis: Luwian and Hittite Studies presented to J. David Hawkins on the occasion of his 70th birthday: 76-94. Tel-Aviv: Institute of Archaeology.

Goedegebuure, Petra. 2014. The Hittite Demonstratives: Studies in Deixis, Topic, and Focus. Wiesbaden: Harrassowitz.

Gérard, Raphaël. 2004. Quelques remarques autour de *y > lydien d. Res Antiquae 1: 125-131.

Gérard, Raphaël. 2005. Phonétique et morphologie de la langue lydienne. Louvain-la-Neuve: Peeters.

Gusmani, Roberto. 1964. Lydisches Wörterbuch, mit grammatischer Skizze und Inschriftensammlung. Heidelberg: Winter.

Gusmani, Roberto. 1965. Sulle consonanti del lidio. Oriens Antiquus 4: 203-210.

Gusmani, Roberto. 1976-1977. Greco nEnAMAI. Incontri Linguistici 3(2): 167-168.

Gusmani, Roberto. 2010. Tracce anatoliche di una desinenza verbale indoeuropea. In: Ronald I. Kim et al. (eds.).

Ex Anatolia Lux. Anatolian and Indo-European Studies in honor of H. Craig Melchert on the occasion of his sixty-fifth birthday: 68-84. Ann Arbor: Beech Stave.

Hajnal, Ivo. 1995. Der lykische Vokalismus. Graz: Leykam.

Hajnal, Ivo. 2004. Die lydischen a-Stämme. In: Adam Hyllested et al. (eds.). Per Aspera ad Asteriskos. Studia Indogermanica in honorem Jens Elmegard Rasmussen sexagenarii Idibus Martiis anno MMIV: 187-205. Innsbruck: Institut für Sprachen und Literaturen der Universität Innsbruck.

Hawkins, J. David, Anna Morpurgo Davies, Günter Neumann. 1974. Hittite Hieroglyphs and Luwian: New Evidence for the Connection. Nachrichten der Akademie der Wissenschaften in Göttingen (Philologisch-historische Klasse) 6: 145-197.

Jasanoff, Jay. 2003. Hittite and the Indo-European Verb. Oxford University Press.

Kassian, Alexei, Ilya Yakubovich. 2013. Anatolijskie jazyki. In: Andrej A. Kibrik, Jurij B. Koriakov (eds.). Reliktovye indoevropejskie jazyki Perednej i Centralnoj Azii: 15-26. Moscow: Academia [in Russian with an English abstract].

van de Kasteelen, Frank. 2015. Syncope in Lycian. MA Thesis, Leiden University.

Kloekhorst, Alwin. 2008a. Etymological Dictionary of the Hittite Inherited Lexicon. Leiden: Brill.

Kloekhorst, Alwin. 2008b. Studies in Lydian and Carian Phonology and Morphology. Kadmos 47: 117-146.

Kloekhorst, Alwin. 2012. The origin of the Lydian dat.sg ending -A. Kadmos 51: 165-173.

Kloekhorst, Alwin. 2022. Anatolian. In: Thomas Olander (ed.). The Indo-European Language Family: 63-82. Cambridge University Press.

Kloekhorst, Alwin. Forthcoming. New interpretations in Lydian phonology. To appear in the proceedings of the 7th workshop Luwic Dialects: Inheritance and Diffusion held in Santiago de Compostela in February 2022.

Kümmel, Martin J. 2018. Anatolisches und indogermanisches Verbum: Erbe und Neuerung. In: Elisabeth Rieken (ed.).

100 Jahre Entzifferung des Hethitischen: Morphosyntaktische Kategorien in Sprachgeschichte und Forschung. Akten der Arbeitstagung der Indogermanischen Gesellschaft vom 21. bis 23. September 2015 in Marburg: 239-257. Wiesbaden: Reichert.

Laroche, Emmanuel. 1966. Les Noms des Hittites. Paris: Klincksieck.

Luraghi, Silvia. 2008. Possessive Constructions in Anatolian, Hurrian, Urartian, and Armenian as evidence for Language Contact. In: Billie Jean Collins et al. (eds.). Anatolian Interfaces: Hittites, Greeks, and Their Neighbours: 143-151. Oxford: Oxbow.

Luraghi, Silvia, Guglielmo Inglese. 2018. Trends in the development of sentence connectives in Hittite: Evidence from subordination. In: Elisabeth Rieken (ed.). 100 Jahre Entzifferung des Hethitischen. Morphosyntaktische Kategorien in Sprachgeschichte und Forschung. Akten der Arbeitstagung der Indogermanischen Gesellschaft vom 21. bis 23. September 2015 in Marburg: 259-275. Wiesbaden: Reichert.

Melchert, H. Craig. 1992. The Third Person Present in Lydian. Indogermanische Forschungen 97: 31-54.

Melchert, H. Craig. 1994. PIE *y > Lydian d. In: Petr Vavrousek (ed.). Iranian and Indo-European Studies. Memorial Volume of Otakar Klima: 181-187. Prague: Enigma Corporation.

Melchert, H. Craig. 2003a. The Dialectal Position of Lycian and Lydian with Anatolian. In: Mauro Giorgieri et al. (eds.). Licia e Lidia prima dell'ellenizzazione. Atti del convegno internazionale (Roma, 11-12 ottobre 1999): 265-272. Rome: Consiglio Nazionale delle Ricerche.

Melchert, H. Craig. 2003b. Language. In: H. Craig Melchert (ed.). The Luwians: 170-210. Leiden: Brill.

Melchert, H. Craig. 2004. Second Thoughts on PIE *y and *h2 in Lydian. In: Michel Mazoyer and Olivier Casabonne (eds.). Mélanges offerts à Professeur René Lebrun, vol. 2, Studia Anatolica et Varia: 139-150. Paris: Harmattan.

Melchert, H. Craig. 2006. Medio-Passive Forms in Lydian? In: Raffaella Bombi et al. (eds.). Studi Linguistici in onore di Roberto Gusmani: 1161-1166. Alessandria: Orso.

Melchert, H. Craig. 2009. Deictic Pronouns in Anatolian. In: Kazuhiko Yoshida, Brent Vine (eds.). East and West: Papers in Indo-European Studies: 151-161. Bremen: Hempen.

Melchert, H. Craig. 2010. Spelling of Initial /A-/ in Hieroglyphic Luwian. In: Itamar Singer (ed.). Ipamati kistamati pari tumatimis: Luwian and Hittite Studies presented to J. David Hawkins on the occasion of his 70th birthday: 147157. Tel-Aviv: Institute of Archaeology.

Melchert, H. Craig. 2017. Anatolian. In: Mate Kapovic (ed.). The Indo-European Languages: 171-201. London: Routledge.

Melchert, H. Craig. 2020. Luwian. In: Rebecca Hasselbach-Andee (ed.). A Companion to Ancient Near Eastern Languages: 239-256. Hoboken, NJ: Wiley Blackwell.

Melchert, H. Craig, Ilya Yakubovich. 2022. New Luwian Verbal Endings of the First Person Plural. Incontri Linguistici 45: 11-30.

Mouton, Alice, Ilya Yakubovich. 2019. Internal or External Evil: a Merism in Luwian Incantations. Bulletin of the School of Oriental and African Studies 82/2: 209-31.

Mouton, Alice, Ilya Yakubovich. 2021. Where did one speak luwili? Geographic and linguistic diversity of Luwian cuneiform texts. Journal of Language Relationship 19/1: 25-53.

Nussbaum, Alan J. 2014. The PIE Proprietor and His Goods. In: H. Craig Melchert et al. (eds.). Munus amicitiae Norbert Oettinger a collegis et amicis dicatum: 228-254. Ann Arbor: Beech Stave.

Oettinger, Norbert. 1978. Die Gliederung des anatolischen Sprachgebietes. Zeitschrift für vergleichende Sprachforschung 92: 74-92.

Oettinger, Norbert. 2021a. Language Contact between Lydian and Greek or The Origin of Lydian k. In: Michele Bi-anconi (ed.). In Search of the Golden Fleece. Linguistic and Cultural Interactions between Greece and the Ancient Near East: 116-130. Leiden: Brill.

Oettinger, Norbert. 2021b. Überlegungen zum lydisch ora ,Monat' und dem Vokal o. In: Annick Payne et al. (eds.). Beyond All Boundaries: Anatolia in the First Millennium BC: 467-475. Leuven: Peeters.

Oettinger, Norbert. 2022. OaoywÀÀLç als ,Zeus Stratios' in Karien, lyk. ese und heth. huwai-i/ hui- ,sich dahin-bewegen'. Münchener Studien zur Sprachwissenschaft 74(2): 61-72.

Oreshko, Rostislav. 2019. Phonetic value of Lydian letter <d> revisited and development of PIE dentals in Lydian. Wekwos 4: 191-262.

Poetto, Massimo. 1979. Lidio kofu-. Incontri Linguistici 5: 198-200.

Rieken, Elisabeth. 2017. The dialectology of Anatolian. In: Matthias Fritz et al. (eds.). Comparative Indo-European Linguistics. An International Handbook of Language Comparison and the Reconstruction of Indo-European, vol. 1: 298-308. Berlin: De Gruyter.

Rieken, Elisabeth, Ilya Yakubovich. 2020. Ein lydisches Schmuckstück. In: Matthias Fritz et al. (eds.). Maiores Phi-lologiae Pontes: Festschrift für Michael Meier-Brügger zum 70. Geburtstag: 215-223. Ann Arbor: Beech Stave Press.

Rizza, Alfredo. 2013. Lidijskij jazyk. In: Andrej A. Kibrik, Jurij B. Koriakov (eds.). Reliktovye indoevropejskie jazyki Perednej i Centralnoj Azii: 75-97. Moscow: Academia [In Russian with an English abstract].

Sadykova, Anna. 2019. The particle pa* in contrast with other sentence particles in Luwian cuneiform texts. MA Thesis: Marburg University.

Sasseville, David. 2014/2015. Luwian and Lycian Agent Nouns in *-é-leh2. Die Sprache 51(1): 105-124.

Sasseville, David. 2017. The Lydian nominal paradigm of i-mutation. Indo-European Linguistics 5: 130-146.

Sasseville, David. 2018. The Lydian word for 'prosecutor'. Kadmos 57: 129-135.

Sasseville, David. 2019. "To show" in Hittite and Palaic Rituals. Altorientalische Forschungen 46: 22-32.

Sasseville, David. 2020. Anatolian Verbal Stem Formation: Luwian, Lycian, and Lydian. Leiden: Brill.

Sasseville, David. 2021. Rhotacism in first-millennium Anatolia: Comparative Luwian and Lydian Phonology. In: Annick Payne et al. (eds.). Beyond All Boundaries: Anatolia in the First Millennium BC: 636-650. Leuven: Peeters.

Schürr, Diether. 1997. Lydisches IV: Zur Grammatik der Inschrift Nr. 22 (Sardis). Sprache 39(2): 201-212.

Schürr, Diether. 2006. Elf lydische Etymologien. In: Raffaella Bombi et al. (eds.). Studi linguistici in onore di Roberto Gusmani: 1569-1587. Alessandria: Dell'Orso.

Schürr, Diether. 2011. Zwei lydische Götterbezeichnungen. Incontri Linguistici 34: 71-80.

Valerio, Miguel, Ilya Yakubovich. 2022. From 'Foreman' to 'Warlord': Royal Titles in Iron Age Western Anatolia. Aula Orientalis 40: 345-353.

Widmer, Paul. 2004. Ли5[а: Ein Toponym zwischen Orient und Okzident. Historische Sprachforschung 117(2): 197-203.

Widmer, Paul. 2009. Hethitisch nu als Mittel der informationsstrukturellen und syntaktischen Verknüpfung. In: Elisabeth Rieken, Paul Widmer (eds.). Pragmatische Kategorien: 323-335. Wiesbaden: Reichert.

Yakubovich, Ilya. 2005. Lydian Etymological Notes. Historische Sprachforschung 118: 75-91.

Yakubovich, Ilya. 2010. Sociolinguistics of the Luwian Language. Leiden: Brill.

Yakubovich, Ilya. 2013. The Degree of Comparison in Luwian. Indogermanische Forschungen 118: 155-168.

Yakubovich, Ilya. 2017a. An Agreement between the Sardians and the Mermnads in the Lydian Language? Indogermanische Forschungen 122: 265-294.

Yakubovich, Ilya. 2017b. The Luwian word for 'place' and its cognates. Kadmos 56: 1-27.

Yakubovich, Ilya. 2019a. Showing Reverence in Lydian. In: Adam A. Catt et al. (eds.). QAZZU warrai: Anatolian and Indo-European Studies in Honor of Kazuhiko Yoshida: 399-409. Ann Arbor: Beech Stave Press.

Yakubovich, Ilya. 2019b. The Lydian Dating Formulae. In: Ignasi-Xavier Adiego et al. (eds.). Luwic Dialects and Anatolian: Inheritance and Diffusion: 299-316. Barcelona: Universitat de Barcelona.

Yakubovich, Ilya. 2020. Hittite. In: Rebecca Hasselbach-Andee (ed.). A Companion to Ancient Near Eastern Languages: 221-237. Hoboken, NJ: Wiley Blackwell.

Yoshida, Kazuhiko. 1993. Notes on the Prehistory of Preterit Verbal Endings in Anatolian. Historische Sprachforschung 106: 26-35.

Zehnder, Thomas. 2010. Die hethitischen Frauennamen: Katalog und Interpretation. Wiesbaden: Harrassowitz.

Zgusta, Ladislav. 1964. Kleinasiatische Personennamen. Prague: Czechoslovak Academy of Sciences.

И. С. Якубович. Статус лидийского среди анатолийских языков под призмой недавних исследований

Лидийский язык традиционно относят к анатолийской группе индоевропейской языковой семьи, но его дальнейшие генетические и ареальные связи остаются дискуссионными. В недавних работах, посвященных новым интерпретациям отдельных аспектов лидийской грамматики, содержатся противоречащие друг другу гипотезы относительно классификации данного языка с ограниченным корпусом. Одни ученые подчеркивают его изолированный характер в анатолийской группе, или даже сомневаются в его принадлежности к данной группе, тогда как другие рассматривают возможность отнести его к лувической подгруппе анатолийских языков, что сделало бы его ближайшим родственником лувийского, лидийского, милийского и карийского. Целью настоящего исследования является оценка данных гипотез на основе интегрированного подхода к последним исследованиям по структуре лидийского языка. Произведенный обзор позволяет заключить, что попытки отделить лувийский от анатолийской группы основаны не на новых изоглоссах, а на исключительно на попытках подвергнуть сомнению некоторые из ранее предложенных сопоставлений. Напротив, некоторые из предложенных изоглосс, связывающих лидийский с лувическими языками, являются вполне убедительными, однако это не означает принадлежности лидийского языка к лувической группе, поскольку интерпретация данных изоглосс как результата ареальной диффузии остается предпочтительной.

Ключевые слова: лидийский язык; анатолийские языки; лувические языки; филогенез; гомоплазия; ареальная диффузия.