Proto-Indo-European ‘fox’ and the reconstruction of an athematic ḱ -stem

Thispaperpresentsadetailedetymologicalanalysisof wordsfor‘fox’inIndo-European (ie) languages. We argue that most ie ‘fox’-words go back to two distinct pie stems: * h₂lō̆p-eḱ - ‘fox’ and * ulp-i - ‘wildcat, fox’. We provide a revised analysis of the etymology and relationship among the various Indo-Iranian ‘fox’-words, and we argue that Baltic preserves remnants of the ḱ -suffix found in Greek, Armenian, and Indo-Iranian. Additionally, we describe how * h₂lō̆p-eḱ - was borrowed from Indo-Iranian into Uralic and we outline the relationship among the reflexes of this word in various Uralic languages. Finally, we reconstruct the paradigm of * h₂lō̆p-eḱ - as a unique type of hysterodynamic stem, which nonetheless has close parallels in pie. We observe that a similar ḱ -suffix is found in pie adjectives and animal names.


Introduction
The Indo-European languages attest several words for 'fox ' , e.g., Skt. lopāśá-, Gr. ἀλώπηξ, Arm. ałowēs, Lith. lãpė, Lat. volpēs, Alb. dhelpër, which are similar enough to have justified hypotheses of a common origin, despite the fact that not all of them show regular sound correspondences. Throughout the history of Indo-European etymological research, these words have either been lumped together under a single etymon (e.g., iew: 1179) or split into several different roots (Schrijver 1998;De Vaan 2000). The aim of this article is to clarify the inner-Indo-European relationships between these stems, as well as their relationship to similar 'fox'-words in the Uralic languages. After discussing the evidence for pie *h₂lop-eḱ-'fox' in Section 1, in Section 2 we examine potential Uralic, North Germanic, and Iberian borrowings from an Indo-Iranian descendant of pie *h₂lop-eḱ-. In Section 3, we reconstruct pie *ulp-i-'wildcat, fox' based on Latin, Lithuanian, Persian, and Albanian evidence. Finally, in Section 4 we discuss the derivational history of pie *h₂lop-eḱ-'fox' and attempt to explain the ablaut preserved directly or indirectly in the branches of Indo-European.

1.1
Greek ἀλώπηξ 'fox' Gr. ἀλώπηξ, -εκος 'fox' shows an ablauting paradigm that is unique for Greek words with velar suffixes. De Vaan (2000: 287) assumes that the paradigm originally had short *-eḱ-throughout, which was lengthened in the nominative by analogy. However, as De Vaan himself mentions, other velar stems in Greek have the same vowel quantity in all case forms, which means that there is no plausible model for such an analogy. If the lengthened suffix in the nominative came about by analogy, one would have expected a long vowel in the oblique cases too, following the model of words like νάρθηξ, -ηκος 'giant fennel' (see also Beekes 2010: 79). Rather, ἀλώπηξ, -εκος reflects an ablauting stem nom. *h₂lōp-ēḱ-, obl. *h₂lōp-eḱ-.1 As for the accent, the columnar root accentuation of ἀλώπηξ can hardly be a result of the limitation rule, as an original accented *h̥ ₂ is excluded. Unless the accent changed through some other process, ἀλώπηξ Thus, assuming that the Greek ablaut of -ηκ-/-εκ-reflects an inherited pattern, the paradigm at one Proto-Armenian stage would have alternated between nom. *aluɸis-(V -)4 and obl. *aluɸes-(V -). At first, it would seem attractive to explain the Arm. oblique ałowes-as a direct reflection of this reconstructed oblique stem; however, this would not account for the aberrant nominative -ēs. Therefore, it appears more fruitful to start with the opposite development, namely the generalization of the strong ablaut variant, thus nom. *aluɸis-V -, obl. *aluɸis-V-. After the fixation of the accent and the loss of final syllables, we would expect analogy after the usual distribution of stressed ē vs. unstressed i, thus ałuwḗs, *ałuwis-V́-. After the noun had entered the u-stem class, obl. *ałuwisu-would yield ałowesow-through dissimilatory lowering of i due to a following u, i.e., the lezow-rule (cf. Olsen 1999: 187-88, 802). This scenario would sideline the rare nom.pl. ałoweskᶜ as a scribal error. Alternatively, one may consider this exact spelling more faithful to the original pronunciation, whereby a replacement with ⟨ē⟩ happened later, when the two vowels had merged in the spoken language (thus Clackson 1994: 95-96). Such an inconsistent replacement of ⟨e⟩ by ⟨ē⟩ does not seem to have sure parallels, however.5 In sum, it is likely that the Armenian paradigm reflects an inherited stem *h₂lōp-ēḱ-. However, unlike Greek, Armenian does not provide direct evidence for the lengthened grade of the suffix. Regarding the root vocalism, the underlying vowel *u could equally well reflect *ou, *ōu, or *u. However, there is no external evidence for *ōu or *u. On *ou, see the following section.

1.3
Indo-Iranian 'fox'-words Various Indo-Iranian words for 'fox' have been treated by De Vaan (2000), who attempts to derive them all from a single pie stem *ulp-i-. In the following, the evidence will be reviewed, and a different scenario will be presented (see 1.3.5) that assumes two pie stems and that better accounts for the attested material. 4 Note that the nominative with final -s cannot directly reflect a consonant stem like the Greek one since the outcome of *-ḱs would be -cᶜ. It can thus be assumed that the stem had joined a vocalic class before the simplification of this cluster. Of course, it remains possible that -s was later generalized from the oblique cases. 5 When this replacement occurs there is usually a clear motivation, such as in paradigms where ē, originally reflecting a diphthong [ei], had been reduced to e before vowels and was later reintroduced orthographically, e.g., in the imperfect endings 1/2sg. -ei, -eir, later written -ēi, -ēir based on 3sg. -ēr (Godel 1975: 11). It is far easier to understand the sporadic replacement of ⟨ē⟩ by ⟨e⟩ in ałoweskᶜ, since ⟨e⟩ is the more frequent grapheme, certainly in stems, and is found in the oblique cases of this word.
Recent accounts (Schrijver 1998: 431;De Vaan 2000: 290) have separated pii *(H)raupaća-from Gr. ἀλώπηξ 'fox' , Arm. ałowēs 'fox' due to the difference in root structure. Instead, the former is seen as being derived within Indo-Iranian from *(H)raupi-(see 1.3.3). However, while the pii *-u-is not reflected in Greek or Armenian,7 the fact that the remaining root consonants are identical and that all words reflect the rare ḱ-suffix puts it beyond reasonable doubt that the words are cognate. As noted by De Vaan (2000: 286), this etymon is the most plausible origin for the 'animal suffix' pii *-āća-, which spread to three or four other Iranian (but notably not Indic) words.
The only reliable evidence for the placement of the accent is Skt. lopāśá-. Whether this preserves the old oxytone accentuation of the oblique cases of an original athematic stem (cf. Section 4) or is analogical based on the suffix -śá-(cf. yuvaśá-'youthful' , babhruśá-, 'brownish') is unclear.
Regardless of whether the thematicization was a pii or post-pii development, we reconstruct a pre-pii *(H)rāpać-< *h₂lop-eḱ-, which corresponds to the Greek and Armenian cognates.

1.4
Proto-Baltic *lapeš-'fox'? Within Baltic, Lithuanian and Prussian share an ē-stem, cf. Lith. lãpė, Pr. (Elbing Voc.) lape · vochz. While these forms resemble those found in Greek, Armenian, and Indo-Iranian, they lack any trace of the ḱ-stem. A trace of the ḱ-suffix, however, may well be found in Latv. lapsa 'fox' ,15 which could show the zero-grade of the suffix (pie *h₂lop-ḱ-; cf. Frisk i: 83, iew: 1179). At first sight, it may seem like rather a bold claim to assume that an ablauting ḱ-stem would have been preserved all the way into Proto-East-Baltic. Nevertheless, the forms lãpė and lapsa can be compared directly to two East Baltic reflexes of the word for 'daughter': The two Lithuanian words for 'daughter' must have split from a hysterodynamic paradigm, which had remained more or less intact in East Baltic, viz., *duktḗ(r), 15 A similar form is also found in the Zinov Vocabulary, viz., łaps · lisa 'fox' . However, leaving the question of the authenticity of this glossary aside, it cannot be decided whether łaps represents *lapsV with apocope (cf. zem 'earth'~Lith. žẽmė) or a masculine *lapVs with syncope (cf. gars 'stork'~Lith. garnỹs).
acc.sg. *dúkterin, gen.sg. *duk(t)rés. The word duktėis still declined as an rstem to this day (cf. acc.sg. dùkterį); however, the nominative duktėhas also dialectally served as the basis for a regular ē-stem (acc.sg. dùktę, gen.sg. dukteṡ, see lkž s.v. duktė). The same development was also found in Lith. obs. mótė 'wife' (acc.sg. mótę beside móterį, alew 771), as well as in its cognates Latv. mãte (acc.sg. mãti), Pr. (iii Catechism) acc.sg. mūtien 'mother' . If the inherited nominative form was *lápē, the same development could theoretically be assumed for Lith. lãpė. The only problem here is the development *lápēš >> *lápē in the nominative singular, which has no exact parallels; however, a close parallel is the s-stem meṅuo 'moon' (acc.sg. meṅesį), where the nominative form probably reflects an older *mḗ'nōs.16 Since *s was not regularly dropped after a long vowel (cf. Lith. jus 'you (pl.)' < *iuH-s; gen.sg. f. -õs < *-eh₂-es), the nom.sg. ending *-ō in meṅuo must be analogical after other paradigms showing apophonic alternations, such as the n-stems (cf. ãkmuo, acc.sg. ãkmenį 'stone'). It is therefore justified to assume that a similar analogical development took place in a nominative form *lápēš (>> *lápē), which might have been encouraged by an association with other feminine animal names, e.g., vìlkė 'she-wolf' (Bammesberger 1970). Latv. s is admittedly ambiguous, as it may reflect either Proto-East-Baltic *š or *s (< pie *ḱ or *s). Although this ambiguity can hardly be resolved, it is not an argument per se against the identification of the Latvian suffix with the ḱ-suffix.17 In fact, a suffix *-sa is not paralleled in animal names, and is otherwise unexplained. One could compare the suffix in Lith. vilpišỹs 'wild cat' (but this suffix is even more obscure, see 3.3.), or assume some kind of corruption after lũsa 'lynx' , but this would be ad hoc. Blažek's (1998) attempt to explain the Latvian suffix by starting from a neuter s-stem is semantically unconvincing. Therefore, equating Latv. s with the ḱ-suffix attested elsewhere seems to be the best solution at hand, explaining both the obscure suffixation and allowing us to bring the Baltic forms in line with the other cognates.
A slightly different view of the Latvian form is that it was syncopated from *lapesa, a form more closely resembling that found in Greek (iew: 1179). Although syncope of suffix syllables is indeed a well-attested phenomenon in 16 It has alternatively been suggested that the pie word for 'moon' exhibited an alternation *t *s in the suffix, with nom.sg. *meh₁-n-ōt (> Gothic menoþs 'month') and acc.sg. *meh₁-nes-m (iew: 731; Fraenkel lew: 438-39; Kortlandt 2005: 156). In this case, Lith. meṅuo could simply reflect inherited *meh₁n-ōt with the regular loss of the final dental. However, the Germanic t-stem may instead be of secondary origin (Lubotsky 2019 Latvian (Endzelin 1923: 46-48), the examples generally involve either high vowels (Latv. àuns 'ram'~Lith. ãvinas; Latv. ve̦ cs 'old'~Lith. vẽtušas), or sequences of identical low vowels (Latv. ḕrglis 'eagle'~Lith. erẽlis; Latv. pe̦ lni 'ashes'~Lith. pelenaĩ ). Since the Latvian syncope is itself an irregular change, and the form *lapesa falls into neither of these two categories, it seems better to interpret Latv. lapsa at face value as a reflection of the zero grade.
In summary, the Baltic words for 'fox' (Lith. lãpė, Latv. lapsa) could theoretically continue the same ḱ-stem attested in Greek, Armenian, and Indo-Iranian. It appears that a zero-grade *h₂lop-ḱ-best accounts for the Latvian form. The only remaining question is whether the Baltic words could reflect *ulop-rather than *h₂lop-and thus be cognate instead with Lat. volpēs 'fox' . The s-suffix in Latvian favours the identification with *h₂lop-eḱ-, all the more because pie *ulp-i-already has a reflex in Baltic (Lith. vilpišỹs). If the Albanian evidence for a full-grade *uolp-is accepted, it would constitute another argument against connecting Lith. lãpė and Latv. lapsa with pie *ulp-i-(see Section 3.4).
Regardless of the origin of the element *-erno-, the Celtic forms appear morphologically distinct from the other languages in that they do not contain a velar suffix. Schrijver's reconstruction only allows for a root connection to the remaining forms through a shared root *h₂lop-.18 A theoretical pie *h₂lop-ḱ-s with a zero-grade suffix is unlikely to yield the attested outcome; although a 18 Regular loss of *k in athematic stems is impossible in view of pc *esoxs, gen. *esokos > Gaul. *esox borrowed as Lat. esox 'salmon' .
cluster *-pḱs-is unknown in Celtic, both pie *-ḱs-and *-ps-regularly yield pc *-xs-, surfacing as W -h-~-ch, B -h-~-c'h, OIr. s(s). The segmentation of pc *loɸerno-into *loɸ-(< pie *h₂lop-) and *-ernohinges on the correct identification of *-erno-as a suffix or some other unknown lexical element. However, the origin of this suffix is so uncertain that it remains an open question exactly how the Celtic form fits in with the other Indo-European words for 'fox' .19 If the root connection is correct, then the ḱ-stem inflection found elsewhere (except in Gr. ἀλώπα?) is not shared with Celtic.

2
The origin and spread of an Indo-Iranian Wanderwort
On the basis of Uralic *e, Holopainen (2019: 202) asserts that the word was borrowed from pre-pii *h₂reupēćo-(cf. Katz 1987: 259), i.e., before the development of pie *e to pii *a, but after the change *l > *r. However, a reconstruction *h₂reupēćo-is very unlikely, since there is no indication that the root vowel of pii *(H)raupaća-goes back to an e-grade; rather, the evidence from the other branches clearly points to an o-grade. The Uralic vowel *e is therefore problematic, since it matches neither pie *ōnor pii *au particularly well. Holopainen (2019: 202) assumes that Indo-Iranian *-u̯ p-was substituted by Uralic *-p-, since such a cluster may not have been licensed by Uralic phonotactics.21 If correct, one could assume that pii *a was borrowed as Uralic *e, for which there are several parallels (pu *sejti << pii *sHai̯ tu-'bridge' , *rećmä << pii *(H)raćman- 19 One could assume that Italo-Celtic inherited a form without the ḱ-suffix. As this branch is sometimes held to be the first Indo-European branch after Anatolian and Tocharian to split off from the core Indo-European dialect continuum (Ringe et al. 2002: 87), this would imply that the ḱ-suffix was added in late pie after Italo-Celtic split off. 20 Whether one reconstructs final *-i is dependent on whether one believes that consonantfinal word forms existed in Proto-Uralic, which is an issue we will not deal with here. 21 According to Aikio (forthc.: 14-15), clusters of *w + obstruent do not occur in any reconstructed Proto-Uralic vocabulary, nor clusters of *w + the labial *m.
'rope').22 It is still difficult to imagine that *w would be lost without a trace, not even having a rounding effect on the neighbouring vowel, but this scenario appears to be the least problematic one.23 Postulating a borrowing from a later Iranian source (e.g., Oss. ruvas / robas 'fox') would not help to explain the vocalism.
The original PFi. *repäs (gen.sg. *repähen) could be regularly derived from pu *repäć(i). This could also be supported by certain Saami forms, viz., North Saami (Jukkasjärvi) rēbeš, rēhpeh-, Lule Saami riebij, riehpih-, South Saami riepie (< Proto-Saami *reapēš; note the regular sound changes *-š > -j in Lule 22 Note that the idea that pu *e can be a substitution for pii *a goes against the mainstream view on Uralic-Indo-Iranian language contact, whereby pu *e is usually thought to reflect pre-pii or pie *e directly (see Holopainen 2019). However, in pii *sHai̯ tu-'bridge' < *sh₂eitu-the vowel *e would have been coloured to *a by the adjacent laryngeal, probably already in pie times. Moreover, pie *e was phonetically probably a low vowel [ae] (cf. Pronk 2019: 124), and would probably have been closer to Uralic *ä. 23 Another alternative would be to assume that *repäć(i) represents a borrowing from prepii *Hropeć-, i.e., before Brugmann's Law. However, a substitution *o >> pu *e would be unparalleled and difficult to explain phonetically. Similarly, borrowing from *(H)rāpać-, i.e., after Brugmann's Law, but before the analogical replacement of *-ā-by *-au-, requires a substitution *ā >> pu *e, which is equally problematic. 24 We thank Petri Kallio for pointing us towards this reference. 25 Traces of a North Finnic form *repä(h)inen may be seen in the Saami loans Sá. N rieban, Sk. rìèwa̮ n 'fox' (ssa iii s.v. repo). and *-š-> -h-in Lule and neighboring Saami dialects),26 which were probably borrowed from Proto-Finnic, as well as perhaps North Germanic *rebaʀ (see 2.4). It is true that the vocalism of the Saami words would seem to be consistent with pu *repäć(i); nevertheless, Saami *š is not regular from pu *ć, which supports a Finnic origin. Saami *š is sometimes found as a substitution for Finnic *s. In these cases, we can probably envisage a Karelian intermediary, e.g., North Saami šalbmi 'eye of a needle' (<< Kar. šilmä) << PFi. *silmä 'eye' , šaldi 'bridge' (< Kar. šilta) < PFi. *cilta 'bridge' . As final *-s in Karelian proper is typically rendered as -š, the Saami words may have been borrowed from a Karelian noun *reväš, now only preserved in toponyms.
Since there is evidence that pii *ć remained an affricate in Proto-Iranian (Cantera 2017: 492), it cannot be excluded that these Iranian loans were adopted from a more recent source which preserved an affricate pronunciation of PIr. *ć.

2.3
Ugric *ra/opać(V) Like Permic *roću, Hungarian ravasz 'cunning' (in Old Hung. 'fox') cannot regularly reflect pu *repäć(i), and rather suggests a form such as *ra/opać(V ). The final consonant could also probably be *s in view of fészek 'nest' (< pu *pesä). A pre-Hungarian *ropas comes very close indeed to the actually attested Oss. ruvas / robas 'fox' . We can therefore be rather confident in attributing the Hungarian word to a later source, probably Alanic. The modern Hung. róka 'fox' (< *rawka) is most likely derived with the diminutive suffix -ka from a secondarily reduced root *raw-(perhaps -asz became analysed as the suffix also found in e.g., kopasz 'bald' , which might be supported by the later specialization of ravasz as an adjective). Such an explanation is more probable than an independent loan from an Iranian *raupaka-(Holopainen 2019: 204) which, to our knowledge, is not attested.

2.4
North Germanic *rebaʀ Old Norse refr (Icel. refur, Far. revur, Nw. rev, Da. raev, Sw. räv) no doubt belongs to the Indo-Iranian Wanderwort cluster as well (Thomsen 1870: 45). The word, which may be mechanically reconstructed as Proto-Germanic *rebaz (or *refaz), has previously been derived from an alleged pie root *ereb(h)-'dark, brown' (cf. iew: 334). However, the reconstruction of this root is fraught with formal and semantic problems. Furthermore, the isolation of the formation to North Germanic rather suggests that the word was adopted in Scandinavia after the Proto-Germanic period. An ultimately Iranian origin seems plausible (cf. Kroonen 2013: 158), but the exact source of the reconstructed Proto-Norse *rebaʀ is unknown. Sarmatian has previously been suggested as the donor language (Brøndal 1928: 10), but in fact the North Germanic form *rebaʀ is formally closest to the cluster of pu *repäć(i) discussed above (cf. Mordvin ŕiveź, Mari rəwəž, as well as North Saami (Jukkasjärvi) rēbeš etc.), and its similarity may be 29 This Permic word is normally considered cognate with Finnish porsas 'piglet' and Erzya Mordvin purcos 'piglet'; however, the regular outcome of Uralic *o(-a) is Permic *o (> Komi u), so the Permic words are better interpreted as independent loans from Iranian (cf. uew no. 1498;Holopainen 2019: 190-93). In fact, even the relationship between the vowels within Permic is irregular. even stronger if we assume that the sibilant *-ʀ was a substitution for the final sibilant in Uralic. Although the exact moment and location of the borrowing process cannot be pinpointed with certainty, we may assume that this originally Iranian word entered North Germanic through Finnic (De Vries 1962: 436;Bergsland 1963: 153-54;1965) or Saami (Keresztes 1997: 146). Since the sibilant of pu *repäć(i) was largely replaced in Finnish and Estonian, borrowing from Saami appears to be the more economic scenario. However, the Saami forms were themselves probably borrowed from Finnic *repäs (see 2.1), and this could also have been the source of the North Germanic word. As a result, neither possibility can be excluded.

2.5
Ibero-Romance raposa, rabosa 'fox' While most Romance languages continue Lat. vulpēs 'fox' , Castilian and Portuguese have an entirely unrelated word: raposa 'fox' . It has been proposed that raposa may originally be a borrowing from Alanic (Brøndal 1928: 10), the unattested language of the Iranian people who established a short-lived kingdom in the South-Central Iberian Peninsula in the 5th c. ad. However, this solution is problematic in view of the absence of any other 'Alanic' elements in the modern Iberian languages.30 Moreover, assuming a foreign origin of the word is unnecessary in light of a plausible inner-Ibero-Romance derivation. The first attestations of Castilian raposa occur next to a variant rabosa, which is still in use dialectally (Corominas 1987: 492). The latter form can easily be a derivation of Portuguese and Castilian rabo 'tail' (< Lat. rāpum 'turnip').31 After the addition of the highly productive adjectivizing suffix -oso/a, the form was lexicalized from 'the one with the tail' to more specifically 'fox' . There are good parallels for such a semantic development, cf. Gallurese codabuffa 'fox' (< coda 'tail' + buffu/a 'funny' , Rubattu 2006: s.v. volpe), Welsh llostog 'tailed; an epithet of the fox' (<< llost 'tail'), Torwali pūš 'fox' (< *pucchin-'tailed' , cdial: 467).
If rabosa is indeed the original variant, being derived from rabo, then the -p-in raposa requires an explanation. Corominas (1987: 492) ascribes this to influence from Asturian rapiega 'fox' or rapiña 'robbery ' (cf. Grimm 1834: xxv). 30 Since Alanic is scarcely attested, it is difficult to evaluate whether it is possible to formally derive raposa from that language. Judging from the form of the word in the only surviving modern descendant of Alanic, Oss. ruvas / robas 'fox' , the vocalism does not match. 31 The meaning 'turnip' in Castilian and Portuguese has been taken over by nabo < Lat. nāpus 'turnip' , leaving rāpum to designate 'tail' . While a shift from 'turnip' to 'tail' may seem counter-intuitive, parallels include Castilian nabo 'turnip; base of a tail' , German Rübe 'turnip; fleshy part of a horse's tail' (Corominas 1987: 489 Although it has been suggested that this rapiega was derived from rabo 'tail' as well (Cano González 2009: 194), in which case it, too, would have an unexplained p, it is in fact more likely that it was directly derived from the verb rapar 'to shave, snatch, rob' .32 That the suffix -iega could be added to verbal roots is possibly demonstrated by parallel formations such as labriega/o 'farmhand' << labrar 'to work' (Corominas 1987: 350;Rainer 1993: 533).

2.6
Summary It appears that at least three different borrowing events resulted in the various Uralic 'fox'-words. The earliest is pu *repäć(i) as attested in the westernmost branches of Uralic, which probably originates in pii, even if the vowel substitutions are not entirely clear. A later, perhaps Proto-Iranian, borrowing yielded Permic *roću. A third borrowing from a presumably Alanic source yielded Hung. ravasz (>> róka). The Finnic successor of the originally Indo-Iranian word pu *repäć(i) was subsequently borrowed into Saami and North Germanic.
It is not immediately obvious why a word for 'fox' would be so frequently borrowed as seems to have been the case here. We may only speculate that trade in the fur of the (Arctic) fox may have been a factor in the transmission of the word. The importance of fox hunting to the ancient Iranians is supported by finds of fox remains in Scytho-Sarmatian settlements and burial sites (Smirnov 1966: 128 apud Abaev iesoj ii: 434).
According to De Vaan (2000: 289), pii *raupi-'fox' and PIr. *rupi-'marten' (see Section 1.3.3) must also ultimately derive from pie *ulp-i-, as he finds a semantic shift from 'wildcat' >> 'fox' in Latin unlikely. Instead, De Vaan argues that *rupi-and *raupi-derive from *u̯ r̥ pi-with metathesis in the initial syllable. He assumes that pie *ulp-i-originally meant 'marten' , with a semantic shift 32 This verb, too, could be considered a possible source of secondary -p-in raposa. Whether it was rapiega, rapar, or rapiña that ultimately influenced rabosa cannot be established with certainty.
from 'marten' >> 'fox' in Latin and 'marten' >> 'wildcat' in Indo-Iranian and Baltic. While this is conceivable, it is difficult to say with any certainty precisely which animals would have been considered (dis)similar by the speakers. Furthermore, the irregular metathesis assumed for *u̯ r̥ pi-is a complicating factor. Since *raupi-is pii, the supposed metathesis must have been early. At the same time, MiP gurbag preserves the non-metathesized form. In De Vaan's scenario, this implies that original pii *u̯ r̥ pi-'marten' split lexically into 'marten' and 'wildcat' , of which only the former underwent metathesis. However, this leaves the shared semantics of the Persian and Lithuanian words unexplained.33

3.3
Lithuanian vilpišỹs 'wildcat' A cognate of the Latin and Persian words is Lith. vilpišỹs 'wildcat' . However, the formation is somewhat obscure. A similar suffix is found in only a handful of words, largely of obscure origin: cf. Lith. takišỹs, Latv. tacis, Pr. takes 'fishing weir' (with no plausible cognates), Lith. lašišà 'salmon' (a word with a North European distribution matching that of the fish itself), and Lith. lop(i)šỹs 'cradle' (obscure, but remarkably similar to pu *le̮ pći 'cradle'). Besides this, a productive suffix -ìšius is found mainly in an agentive function, e.g., ėd-ìšius 'one who eats a lot' (<< eṡti 'to eat, guzzle'), sald-ìšius 'someone with a sweet tooth' (<< saldùs 'sweet'), merg-ìšius 'womanizer' (<< mergà 'girl'). Even though we do find a variant vilpìšius, it is difficult to explain the motivation for appending such a suffix to the word for 'wildcat' .34 It is possible that vagìšius 'thief' 33 Unless, of course, it is assumed that Indo-Iranian and Baltic underwent a shared lexical split and semantic development 'marten' >> 'wildcat'; this is very speculative, however. 34 A possible, although speculative, explanation could be that the inherited *vilpi-gained a suffix *-š-under the influence of the Proto-East-Baltic stem *lap-š-'fox' . The derivative could have played a role, considering the various potential semantic parallels for 'fox'-words noted in Sections 1.3 and 2.5. In any case, the unclear suffix does not give us cause to doubt the etymological connection with the Latin word.

3.4
Albanian dhelpër / dhelpën 'fox' A further potential reflex of pie *ulp-i-is Alb. dhelpër / dhelpën 'fox' . The nsuffix is recent as shown by the dial. (North Gheg) variant dhelpë. The Albanian word can hardly continue a zero-grade in the root: if pie *ulkʷo-'wolf' gave Alb. ujk (dial. ulk), one would expect *ulp-i-to yield a similar outcome (**ylp, vel sim.). If the Albanian word does belong here, then, it must reflect a different ablaut grade. Since pie *e regularly gave Alb. -je-(e.g., vjehërr 'mother-in-law' < *sueḱruh₂-), the vowel in dhelpë is more likely to represent an umlauted ograde (*uolp-i-> PAlb. *välp(i)-). This preform could of course be identical to that of Lat. volpēs. The only issue is the irregular anlaut dh-for expected *v-. Yet for this an almost exact parallel seems to be found in Calabrian Albanian dhespër 'evening' (Mann 1948: 90) << Italian vespero. As the substitution dh-for v-is unparalleled in Romance borrowings, we must be dealing with a dissimilation *v-p > *ð-p. Indeed, additional support for such a development can be found in the borrowing of the South Slavic word for 'vampire' , Bulg. vampir, SCr. vàmpīr, as Alb. dhampir (see Topalli 2003 for a discussion). A similar dissimilation can therefore be assumed for 'fox' (cf. Stier 1862: 144), which may be supported by a secondary association with dhelë 'cunning; caress; flattery' (cf. Demiraj 1997: 156;Orel 1998: 81). Note that as an adjective, dhelpër means 'cunning, treacherous' , and such polysemy is by no means unusual; compare Gr. ἀλώπηξ 'fox; sly or cunning person' . An association with dhelë is in any case assured by the dial. (North Gheg) variant ledhpë 'fox' , which seems to have been created to ledhë 'a caress' on the analogy of dhelpë~dhelë.
Given that dhespër 'evening' and dhampir 'vampire' are both loans, one may wonder whether dhelpë could also be a loan. A Latin origin cannot be maintained, as Alb. *s regularly surfaces as sh in Latin loans (Matzinger 2006: 84). Furthermore, Latin already had -u-in 'fox' by the start of the Classical period, which could not have yielded Alb. -e-. Since Lat. u developed again into *ọ in most Romance dialects (cf. Italian volpe), one might be tempted to consider a later Romance source. However, this is chronologically difficult, as later Romance loans were not affected by the Albanian umlaut. A direct substitution *vilpiš-ja-, which underlies the attested Lithuanian form, would then have been formed on the basis of this extended root.

4
Reconstructing a pie athematic ḱ-stem As Greek, Armenian, and Indo-Iranian all provide unambiguous evidence for a ḱ-suffix, and the same suffix could neatly account for Latv. lapsa, we consider it optimal to explain all of these forms from a single Proto-Indo-European ḱstem. Since these words are so widespread across the branches of ie, yet within each branch so morphologically isolated, they can hardly be seen as independent post-pie innovations. Besides these forms, there is also marginal evidence for a suffixless form *h₂lop-which could have been the basis of Gr. ἀλώπα 'fox' (in Alcaeus) and Celtic *loɸerno-'fox' . A suffixless form could also provide a simpler explanation for Lith. lãpė, although as argued in 1.4, the Lithuanian form could have been extracted from the original ḱ-stem, which is probably preserved in Latvian. It is therefore theoretically possible to reconstruct a shorter form *h₂lop-beside the ḱ-stem *h₂lop-eḱ-for pie, but it should be stressed that the evidence for the former is relatively isolated, while the most reliable evidence points to a ḱ-stem. The remaining question is how the paradigm of *h₂lop-eḱ-should be reconstructed for pie.

4.1
The morphology of pie *h₂lop-eḱ-The forms we have analysed show evidence for both root and suffixal ablaut. In the root syllable, Gr. ἀλώπηξ and probably Arm. ałowēs show a lengthened ō-grade, while Lith. lãpė and Latv. lapsa imply a short *o. In the suffix syllable, the aforementioned Greek and Armenian forms, as well as Indo-Iranian *(H)raupaća-and probably Lith. lãpė, support the reconstruction of a lengthened ē-grade. A short e-vowel in the suffix syllable is continued by the Greek oblique stem ἀλώπεκ-and possibly by some Iranian reflexes of *(H)raupaća-. Finally, in our opinion, Latv. lapsa provides fairly compelling evidence in favour of a zero-grade allomorph *-ḱ-.
The CoC-eC-type inferred from the above examples neatly accounts for the o-grade of the root and the ē/e/Ø-ablaut of the suffix found in the pie word for 'fox' . Still, it fails to explain the lengthened ō-grade in the root as implied by the Greek evidence.
One hypothesis could be that lengthened *ō in *h₂lop-eḱ-was analogically introduced from a root noun *h₂lōp-, where lengthened grade would be expected. The pie lengthened grade (in athematic nouns) has been argued to have essentially two regular origins: 1) Szemerényi's Law38 and 2) monosyllabic lengthening39 (Streitberg 1894;Wackernagel 1896: 66-68). In the case of *h₂lōp-, the latter would apply. However, the evidence for a monosyllabic root noun is virtually non-existent, since Gr. ἀλώπα 'fox' is of debatable etymological value (and in any case, is not a root noun) and the original stem class of the derivationally unclear Celtic *loɸerno-'fox' is obscure. In other words, only the lengthened grade itself could be seen as evidence for a root noun *h₂lōp-. To use this hypothetical root noun as an explanation for the same lengthened grade would amount to circular reasoning.
On the other hand, we note that pie *h₂lóp-ḱ-s is crucially different from these forms in that it was phonetically monosyllabic and therefore liable to lengthening. The phonetic motivation behind the presence of long *ō in *h₂lop-ékversus its absence in other CoC-eC-stems can thus be found in the consonantal as opposed to vocalic realizations of the corresponding suffixes.
Regarding the subsequent development in the different daughter languages, we observe that Greek and Armenian levelled the lengthened ō-grade of the nominative *h₂lṓp-ēḱ-s. In Indo-Iranian, the ḱ-stem was thematicized based on the nominative stem *h₂lṓp-ēḱ-or perhaps the accusative stem *h₂lop-éḱ-in part of Iranian. In Baltic, it appears that short *o was levelled throughout the paradigm, but the ablaut in the suffix was retained. While we accept that there may be other ways to account for the data, we consider this the option that best accounts for all the relevant evidence, while also being in line with what we otherwise know about pie nominal ablaut.

4.2
The pie ḱ-suffix While a suffix *ḱ is not particularly common, there is solid evidence for an adjectival ḱ-suffix, perhaps originally limited to the positive degree of the word for 'young' , cf. OIr. óac, W ieuanc 'young' (< *h₂iuHn-ḱo-, cf. Skt. yuvaśá-) beside comp. OIr. óa, mw ieu (< *h₂iéuH-ios-, cf. Skt. yávīyas-), and Lat. iūnior < *h₂iuH-n-ios-.41 In pie, positive forms with and without the suffix may have 40 The lengthened ē-grade in the nom.sg. may originate in a subset of stems due to some phonological development (cf. fn. 38) but eventually became morphologically conditioned. Here, the crucial development is the extension of the full grade suffix to the nominative. 41 Perhaps a ḱ-suffix was also present in the word for 'old' , in view of Lat. senex 'old man' (gen. senis), comp. senior and Gothic sineigs 'old' , which uniquely loses its suffix in the superlative sinista. These examples are ambiguous, however. For the Gothic form it must be assumed that the ḱ-suffix was replaced by the productive suffix *-īga-. For Lat. senex, Martinet (1955) assumed hardening of *h₂ > *k before s in the nominative. An anonymous reviewer suggests to us that an original, problematic nominative *sēns (> */sēs/) was replaced by the nominative of a derived noun senex.
our discussion of the ie evidence. A remaining problem is that the Uralic vocalism is hard to derive from any possible (pre-Proto-)Indo-Iranian source form. The least problematic option seems to be pii *(H)raupać(a)-, which when borrowed lost its *u due to Uralic phonotactics. Finally, we have explored the morphology of the ἀλώπηξ family, arguing that it preserves an athematic paradigm that was rare, but not unparalleled, in pie. Moreover, we have argued that the suffix *-ḱ(o)-, present in some animal names and adjectives, might be related to the suffix in the ἀλώπηξ family. On the one hand, the Indo-European words for 'fox' have frequently been subject to secondary developments, lexical innovation, and borrowing; on the other hand, they have preserved traces of archaic morphological patterns.