The Lydian nominal paradigm of i-mutation

This paper offers an analysis of the Lydian nominal paradigm of i-mutation. The main goal is to uncover the accentual type of this paradigm. This is achieved, first, by analyzing the phonological outcome of the endings and, second, by identifying the position of the accent within each given lexeme. In the end, a barytone character is posited for this paradigm. This observation has consequences regarding the original distribution of i-mutation in Anatolian and its chronological spread.


Introduction
The following paper deals with the grammar of Lydian, a poorly attested language belonging to the Anatolian branch of the Indo-European family. Although most of the nominal endings of the Lydian language are known, an extensive and comprehensive classification of all nominal stem classes is lacking so far in the literature. Recent works on Lydian grammar such as Gérard's Phonétique et morphologie de la langue lydienne make mention of nominal Indo-European Linguistics 5 (2017) 130-146 stems (Gérard 2005:79-91), but provide no synchronic paradigm of the various nominal stem classes. Hajnal (2004), on the other hand, attempts an analysis of nominal paradigms, but deals with three large nominal stem classes within a single article. A precise classification of these classes is relevant not only for the grammatical annotation of Lydian lexemes, but also for Anatolian historical grammar in general. Thus, the present paper provides a synchronic and diachronic analysis of one nominal stem class as well as its corresponding paradigm. The nominal stem class selected for this paper is the one with imutation, which is left out by Hajnal (2004) in his study on Lydian a-stems. An additional goal of this paper is to shed light on the chronology of the spread of the Luwic i-mutation to various nominal stems.1

Determining the position of the accent
Due to the pioneering work of Eichner (1986aEichner ( , 1986b on Lydian accentuation, we are now aware that this language more often than not provides precise information regarding the position of the accent within a word. For example, the Lydian vowels ẽ, ã, e and o always carry the accent, except in preverbverb combinations and compounds, where more than one of these vowels may appear, and thus one of them is accented or carries only a secondary accent. However, the metrical inscriptions scanned by Eichner (1986a:16-21) demonstrate that preverbs are regularly unaccented. The vowels a, i and u can be either accented or unaccented, but the occasional plene spelling in -aaand -ii-is a device used by the Lydians to indicate that the vowel is accented. On the other hand, vocalic liquids (λ, r) and vocalic nasals (ν, n) are normally unaccented. Another means for determining the position of the accent was provided by Eichner (1986a), as he, in our opinion, has deciphered the Lydian poetic meter.2 By properly scanning a verse, the position of the accent in a given word is revealed. In this respect, Lydian has a great advantage over the other Anatolian languages.

The paradigm of i-mutation
The paradigm of the i-mutated stems represents an innovation shared by Luwian, Lycian and Lydian (cf. Starke 1990:56-93, Zeilfelder 2001:215-228, Rieken 2005. The process of i-mutation consists on the one hand of replacing the thematic vowel *-o-with the vowel -i-in the nominative and accusative, singular and plural of the common gender. Thus, this paradigm represents on the inflectional level a syncretism between the Proto-Indo-European i-stems and thematic o-stems. Furthermore, the morpheme -i-is added to a stem-final consonant in the same cases, numbers and gender as for the thematic o-stems, e.g. -nt-+ -i-→ -nt(i)-, thus causing another paradigmatic merger. As shown by Eichner (1986b:212-217), a barytone nominal stem may show a different set of stem-finals than an oxytone one. This can be observed among a-stems, where the vowel -a-in front of a nasal will show a different outcome depending on whether it is originally long or short, accented or unaccented, i.e. acc. sg. -ẽν (< *-ó-m), -ãν (< *-a-m) versus -aν (< *-V-m) (cf. Hajnal 2004). What is missing so far in the literature is the accentual pattern of the paradigm of imutation. In the next section, this paradigm will be presented and its endings will be phonologically evaluated. Afterwards, the lexical data constituting this paradigm will be examined and the position of the accent determined for each given lexeme.

Inflectional endings and phonology
The paradigm of i-mutation is given above. The lexical data constituting this paradigm are presented and analyzed from section 4.1 to 4.9.
A striking feature of this paradigm is the consistent syncope found in the accusative singular ending -ν or -n (e.g. adẽnν, alarmlν, šiwraλmn) and dative Indo-European Linguistics 5 (2017) 130-146 singular ending -λ (e.g. niwislλ, qaašlλ, sfardẽtλ). This speaks a priori for a barytone paradigm. It has already been established that unaccented vowels in final syllables are syncopated in Lydian (Melchert 1994:373f.). However, only vowels that were originally short are expected to show syncope. The originally long ones are preserved (cf. Melchert 1994:379).
This syncope points towards the fact that the accusative and dative singular endings must have been short originally, i.e. -ν < *-ĭ-n ← *-ŏ-m and -λ < *-ăλ. The fate of the accusative singular ending demonstrates that the vowel -i-of the mutation was short. This observation is in contradiction with the assumption that this vowel in Luwian is long, as advanced in Melchert 2003:188 andRieken 2005:67.3 On the other hand, endings containing an original long vowel do not undergo syncope. For example, the dative/genitive plural ending of the paradigm under discussion is not **-ν, but -aν < *-ām < *-o-om (e.g. fẽntašẽnaν, sfardẽtaν, ištaminlaν). The absence of the vowel -ẽ-or -ã-before the nasal -ν brings support to the unaccented status of the dative/genitive plural ending.
The neuter singular ending -id (e.g. wiswid, tesastid, manelid), like its nonmutated counterpart -ad, reflects an extension of the stem vowel with the Indo-European Linguistics 5 (2017) 130-146 pronominal neuter ending -d (Gérard 2005:80). This is an innovation vis-àvis e.g. Hitt. -an, Lyc. -ẽ (< pie *-om), which can be dated to Pre-Lydian. No syncope occurs in this ending, although the vowel in the ending -id (-ĭ-d ← -ăd < *-ŏd) is certainly short. Therefore, one could posit that the ending -Vd was created after the Pre-Lydian syncope took place, thus replacing an expected *-ν /n̥ /. Alternatively, if the analogical extension had already taken place, one could claim that the absence of syncope is due to the fact that Lydian will not tolerate certain consonant clusters, e.g. lid, not **-ld (cf. Melchert 1994:353, 374). An anonymous reviewer makes me aware that the second option is superior, if one assumes ciward comes from *ciwarid. In this case, a final cluster -rd would be tolerated in Lydian as opposed to e.g. **-ld. The latter solution seems to us preferable, especially if one considers the similar reasons for the lack of syncope in the nominative singular ending -iš discussed below.
As opposed to Luwian and Lycian, the Lydian language has extended the i-mutation to some neuter endings. The extension appears to be restricted to adjectival neuter stems (see 4.2 and 4.4). The reason behind this might be that the adjectival neuter stems share a paradigm with their common gender counterpart, which motivates the extension process; the nominative/accusative neuter singular takes over the i-mutation by paradigmatic levelling. Nonmutated adjectival a-stems (< *-ó-) on the other hand will not show i-mutation in the neuter, just as it is the case in the common gender; cf. aλa-'other' (< *ali̯ ó-), aλas : aλad. The neuter plural in -a (< *-eh2) (e.g. niwiswa), which was originally long for thematic stems, is rarely attested with adjectives. The paucity of its attestations might be due to the fact that collectives or neuter plurals seem to be agreeing with adjectives in the neuter singular (e.g. lw 5.2-3) or with the endingless case marker (e.g. lw 54.1-2).
The nominative singular common gender ending -iš < *-ĭ-s ← *-ŏ-s (wissiš, qaašliš, maneliš), although short in origin, did not undergo syncope. The reason for this must be a prohibition on the word-final consonant sequences just like the more probable explanation of the neuter singular ending -id. A very small number of sequences -Cš# reflecting nominative singulars can be found in Lydian, e.g. edš (cf. edλ), kaτ fλadš (suffix -ad(i)-(?), cf. awladiš, awladλ) and patnẽdš, all showing the sequence -dš. Because of the palatalization of the ending -š, these words must have a mutated stem, and one can conclude that the -i-of the nominative singular ending can only be dropped following -d-.
Based on this observation, one could argue that the endings of the nominative and accusative plural -iš and -is follow the same restrictive rule as the one of the nominative singular. However, since the syllable was long originally and we do not expect their vowel to drop like in the dative plural -aν, it is better to retain the explanation with the length of the vowel (see above). If Indo-European Linguistics 5 (2017) 130-146 nominative plural forms in -diš were to be found beside singulars in -dš, it would vindicate our hypothesis.7 Based on these observations, one can affirm that short vowels in the endings of this paradigm underwent syncope in Pre-Lydian, which points towards a barytone paradigm. An investigation of lexemes belonging to this paradigm is necessary before drawing conclusions about the accentual type of the entire nominal class. The investigated lexemes will be divided according to their respective suffix and stem formation.

4.2
Adjectivization in -(i)-The adjective wiss(i)-/wisw(i)-'good, pious' represents an adjectivization of the substantive *h1ósu-/*h1u̯ ésu-'goods' by addition of a morpheme subject to imutation. Since u-stems do not take i-mutation in neither Luwian nor Lydian, the reconstruction of an additional morpheme is necessary. Moreover, the raising of a short -e-to -i-in the sequence w_s (or any coronal) in Lydian is made difficult by the counterexample wesfa-. Therefore, I assume an adjectivization in *-o-with a vr̥ ddhi-ablaut of the root, i.e. *h1u̯ ēsu̯ -ó-, and retraction of the accent followed by i-mutation, i.e. *u̯ ísu̯ -o/i-.8 Its antonym is built by means of adding the prefix ni-, i.e. niwiss(i)-'bad, unpious' . They are attested as nom. sg. c. wissiš (lw 6.5; 8.11; 10.23; 11.12), dat. sg. niwislλ (lw 24.10), nom./acc. sg. n. wiswid (lw 22.3, 6), nom./acc. pl. n. niwiswa (lw 44a.17), adv. niwiscν (lw 6.5; 7.6; 8.11; 10.24; 23.3).9 There are two indications of its barytone character. The first one is drawn from a metrical inscription, where the accent occurs on the first syllable: The third foot must be scanned this way, since an anaclasis of the type | ⏑ --| would be unparalleled in the Lydian poetic meter (cf. Eichner 1986a:18f.). The second one is the attestation in the dative singular niwislλ (< *niwissλ + assimilation), which shows syncope of the vowel in the ending. This syncope excludes the possibility of an oxytone type and, since the accent is not expected to be on the prefix, it must be on the root, i.e. *niwíswaλ > *niwíssλ > niwislλ.10
nom./acc. sg. n. wa/i-su (Rieken 2010:658f.), but a simple assimilation of the cluster -sw-> -ss-in the nominative singular, which is motivated by the following -š in the ending, would be sufficient. 10 As pointed out in section 2, in Eichner's metrical system preverbs or prefixes do not carry the accent. 11 Cf.

4.4
The suffix -l(i)-The adjectival suffix -l(i)-(< *-(é)-lo-with i-mutation; see Starke 1990:85), which has been grammaticalized as an adjectival genitive, is also found in substantivized and lexicalized derivatives. This suffix corresponds to Luwian -alla/i-, Lyc. -ele/i-and Hitt. -ēl-(< *-é-lo-).12 Derivatives in *-é-lo-were originally barytone and they are mutated in Lydian. We note here that the connecting vowel *-é-before the suffix *-lo-has lost its accented status in Lydian and has been dropped. Consequently, the suffix was reanalyzed as -l(i)-.13 Due to its productivity, most Lydian formations in -l(i)-are surely late and one should not posit a change in the accent to have occurred within each available lexeme. The first example treated here is the adjective arlill(i)-'one's own' attested as nom. sg. arlylliš (lw 23.13), dat. sg. arlilλ (lw 24.13), arlyllλ (lw 23.11), which cannot be oxytone because of the syncope in the dative singular ending. Since the vacillation -i-/-y-in the second syllable reflects the lack of accent (Melchert 1994:342f.), the lexeme arlill(i)-must be accented on the first syllable (Melchert 1999b(Melchert :2432. The double -ll-must reflect an assimilation, perhaps from a cluster -dl-, i.e. *arl-id-l(i)-> arlill(i)-.14 The next examples represent substantivized adjectives in -l(i)-, e.g. šerl(i)c. 'high authority' (< *sér-lo-)15 attested as nom. sg. šerliš (lw 24.4, 24.16), ištaminl(i)-c. 'family (?)' attested as nom. sg. ištaminliš (lw 13.2) and as gen. pl. ištaminlaν (lw 13.7). The last lexeme is attested in a metrical inscription and the accent falls on the penultimate syllable: For the origin of the Hittite suffix, see Rieken 2008. 13 Accent shift followed by syncope of the previously accented vowel is very common in Lydian, especially if one takes the comparative data of Luwian and Lycian into consideration, e.g. Lyd.
The amount of data for genitival adjectives built from appellatives is meagre. Nevertheless, one can observe that when the genitival adjective in -l(i)-is derived from an oxytone a-stem, the accent shifts to the beginning of the word and then the previously accented vowel is immediately syncopated, e.g. qašaa-c. /kwaçá-/ (lw 10.13) → qaašl(i)-/kwáçl(i)-/ (lw 12.2, 12.5)19 and širma-c. 16 For the publication of the inscription lw 115, see Gusmani & Akkan 2004. For the numbering of the latest Lydian inscriptions, see Payne & Wintjes 2016:7780. 17 Pace Schürr (1997Schürr ( :2023 and Gérard (2005:81f.), karolas cannot be the accusative plural of the mutated stem karol(i)-(part of a grave), because the accusative plural ending should be mutated. Therefore, it must be taken with Melchert (1991:132f.) as karola(ν)=s, the possessive adjective of the personal name karo-. 18 Based on one attestation in a metrical inscription, we can posit that the divine name Artemis in Lydian is accented on the first syllable, i.e. /ártimu-/ (Eichner 1986a:18). 19 See also the antonym niqašllλ /nikwáçlλ̥ / in lw 12.9, whose place of accentuation is also shown by the poetic meter. The word niqaašlad (lw 14.6; 80.6) must be a denominative verb in the 3rd sg. pres. derived from niqašl(i)-, since the neuter singular ending of the adjective in -l(i)-would be *niqašlid.
Indo-European Linguistics 5 (2017) 130-146 /çirmá-/→ šrml(i)-/çr̥ ml(i)-/.20 This also means that the vocalic resonant in šrml(i)-is secondarily accented by analogy, in which case the resonant does not vocalize to -i-/-y-, pace Eichner 1986a:2126and Melchert 1994 Further evidence for its accented status can be adduced from the rest of the data on stems in -l(i)-, which point towards a barytone type. The lexeme alarm-c. '(one)self' , which is barytone, has one attestation of a genitival adjective: The accent is on the second syllable and the ending shows syncope, i.e. /alármln̥ /, thus the genitival adjective is barytone as well. Pace Melchert (2006: 11639), Lyd. alarm-c. '(one)self' is synchronically a consonantal stem and not a barytone a-stem. No restitution of the vowel -a-occurs in the suffix. The case of alarmas (lw 13.1) instead of regular alarms, which has been wrongly used as an argument for a general restitution of the vowel -a-in barytone a-stems, serves a metrical purpose (cf. Eichner 1985a:17). The suffix -m-c. probably goes back to the neuter suffix *-mn̥ -, which was remade into a common gender consonantal stem. This etymological approach is supported by Lyd. ẽtam-c., from which the denominative verb in *-i̯ e/o-preserved traces of the original shape of the suffix, i.e. f-ẽtamν-i-(contra Gusmani 1964:126, 172).
In sum, the examples shown above suggest not only through the presence of the vowel ẽ, ã, o, e or -aa-in a syllable other than the ultimate, but also through the syncope in the accusative and dative ending and through the lack of accentual indication, i.e. -ẽ-or -ã-, in the dative plural -aν, that derivatives in -l(i)-are barytone. 20 Further evidence for the oxytone property of širma-(< *sr̥ méh2, cf. Lyc. hrm̃ma-c.) is the vacillation -i-/-y-in the first syllable, cf. šyrmas (lw 23.1). The unaccented original vocalic *-r̥ -developed to *-ir-/-yr-(cf. Melchert 1994:363), e.g. pira-c. 'house' (< *pr̥ (n)-ó-), contra Melchert 1994:367. For the spread of the root vowel quality of the weak stems to the strong stems in the southwest Anatolian languages, cf. CLuw. zārt-n. (*kēr-/*kr̥ d-). 21 For a possible parallel for a secondarily accented resonant, see srfast(i)-in section 4.7. 22 No letter is missing from the space (pace Gusmani 1964:256). Traces of the initial a-in alarmlν can be seen on the other side of the space occupied by the bas-relief (see Buckler 1924:30). This is also supported by the meter.

4.7
The suffix -ast(i)-There are two examples in Lydian of an adjectival suffix -ast(i)-, which goes back to *-os-to- (Melchert 1999a:365f.). The first is tesast(i)-'right' , attested as nom./acc. sg. n. tesastid (lw 5.2), and the second is srfast(i)-'left' , attested as nom./acc. sg. n. srfastid (lw 5.3). The accentual type of the first example is clearly barytone based on the preserved vowel -e-(< *é). The second example, on the other hand, is ambiguous. The accent could be on the vowel -a-, or on the vocalic -r-in the root srf-. Even though vocalic resonants are generally unaccented, it could have received accent secondarily by analogy to other formations in -ast(i)-such as tes-ast(i)-. Because of the paucity of evidence, this last assumption must be considered tentative.

4.8
The suffix -ẽt(i)-The suffix -ẽt(i)-(< *-Vnt-with i-mutation) goes back to a consonantal stem. There is the lexeme sfardẽt(i)-'Sardian' attested as nom. sg. c. sfardẽti(š) (lw 22.13), dat. sg. c. sfardẽtλ (lw 22.2), dat. pl. c. sfardẽtaν (lw 22.8), sfardẽta(ν) (lw 22.4). The substantivized nominative plural common gender form sfardẽnτ (lw 22.1)25 is strange, as one would expect sfardẽtiš. Perhaps, it is an endingless collective (cf. dẽt-below) followed by the emphatic particle =s, i.e. sfardẽt + =s → sfardẽnτ.26 The position of the accent is indicated by the vowel -ẽ-and, thus, this lexeme is barytone. 25 The word mλimnas at the end of the line is interpreted here as accusative plural dependent on the postposition wicν; cf. Yakubovich forthcoming. 26 Another possibility is to assume syncope of the plural ending -iš, which is not expected to occur if the vowel is originally long. However, non-mutated consonantal stems seem to have a nominative plural ending -š, which probably goes back to a short vowel, *-ĕs, e.g. ciwš 'the gods' (< *díu̯ -es < pie *di̯ eu̯ -es) (lw 44a.17). The hypothesis that this ending was beginning to extend to the paradigm of i-mutation is possible, but unprovable due to the lack of independent confirmation.

4.9
Miscellaneous The following lexeme is hard to classify. The substantive aν(i)-c. is attested as acc. pl. aνis27 (lw 11.3) in a metrical inscription: -⏑ -| ⏑ ⏑ -| ⏑ ⏑ -| ⏑ ⏑ -|| aνis qid=ad šawν wratuλ arsτãnšrs kašnod28 lw 11.3 Pace Eichner (1986a:18), the initial vowel in aνis is surely accented. Scanning the foot as | -⏑ -| also makes the entire poem more homogeneous with respect to the first foot of each verse (cf. lw 11.1, 2, 5, 6, 9, 10, 11, 12). Besides, to scan the foot as | ⏑ --| is not an option; cf. section 4.2. The accented status of the vowel -aplaced before a nasal without the expected outcome -ẽ-or -ã-is rare, but not unparalleled, e.g. tamν 'I built' /támn̥ / (Eichner 1986a:11). These anomalies have not yet been explained satisfactorily, although one should notice here that both examples concern the root vowel. From a historical perspective, it is difficult to decide whether the substantive aν(i)-reflects a primary i-stem or a primary thematic o-stem, i.e. *(H)Vn-i-or *(H)Vn-o-, since both are well-established word formations. In any case, the preform must have been barytone. 27 I interpret iškos aνis as the direct object of kaτared and not as the subject. The subject of kaτared is mentioned in the previous sentence: šarištrosλ astrkoλ areλ 'Ares, the šarištros-, the astrko-' . 28 The unaccented status of šawν supports the hypothesis that it is a pre-/postposition (Boroday & Yakubovich forthcoming).

Conclusions
In spite of the small number of examples available for each type of nominal formation, the fact that all of them are uniform with regard to the accentual type allows us to conclude that the paradigm of i-mutation in Lydian is barytone.
On the other hand, the original thematic o-stems that were accented on the suffix survive in Lydian as a separate class (Hajnal 2004:189-192) and show no i-mutation, e.g. dat./gen. pl. aλẽν 'other' < *ali̯ -ó-om, acc. sg. tawsẽν 'powerful' < *teu̯ H-s-ó-m. If Lydian was consistent in this pattern, i.e. all oxytone o-stems escaped i-mutation, then it is expected that all barytone stems in *-o-were affected by i-mutation. That would mean that barytone o-stems were the first ones to undergo i-mutation in the prehistory of Luwian, Lycian and Lydian.29 While Luwian and Lycian went on to eliminate the oxytone o-stems, Lydian preserved them intact.30 This observation calls into question the origin of the Lydian barytone astems, which never show syncope of the stem vowel -a-, e.g. wãna-c. 'grave' (cf. footnote 10). On phonological grounds, the stem vowel must have been originally long, in order to fully resist syncope. Therefore, an origin in *-eh2 should be posited for the Lydian barytone a-stems, i.e. wãna-c. < *u̯ ónā-< *u̯ ón-eh2 (cf. HLuw. /wanid-/ n. 'stele' < *u̯ ón-id-).31 However, a full analysis of this nominal stem class cannot be presented here. Therefore, this conclusion must be regarded as provisional.

Summary
In this paper, the Lydian nominal paradigm of i-mutation has been extensively researched. After its endings were phonologically analysed, a barytone character for the paradigm was proposed. In order to verify this hypothesis, sub- stantives and adjectives belonging to this paradigm were classified according to their respective stem formation and, then, their accentuation type was identified. All verifiable lexemes turned out to be barytone, which was confirmed either by indication of an accented vowel in the penultimate or antepenultimate syllable, or by determining the position of the accent by scanning the metrical verse. In sum, the Lydian nominal paradigm of i-mutation is barytone. This paper will hopefully contribute to the ongoing research on the phenomenon of i-mutation in Anatolian, in which Lydian has so far received little attention.