2.1. The Proterokinetic and Holokinetic Inflections

Proto-Indo-European seems to have had two separate inflections for the word for “fire.” The singular followed a proterokinetic inflection, with an accented full-grade root and zero-grade stem in the strong cases and a zero-grade root and accented full-grade stem in the weak cases. The second inflectional pattern, used to denote the collective, was holokinetic. In the locative form, from which the Gmc. dative case seems to have arisen, one would normally expect the stem to have been accented in either paradigm, with the ending in zero-grade, if present at all (Fortson 2004:108, 111, 165; Schindler 1974:10):

Table 2.

The inflection of “fire” in Proto-Indo-European.

The proterokinetic singular is attested only in Hittite pa ur, pa uenaš, whereas all other Indo-European languages seem to reflect only the holokinetic collective. Ringe (2006:137; 1996:17–18) has sur-mised, based on evidence from Tocharian, that the non-Anatolian reflexes had the nom.-acc. form of the holokinetic inflection restructured to a zero-grade in the root. This originally holokinetic collective was reinterpreted as a singular and had the zero-grade form of the root, ⁺ph₂- leveled throughout the paradigm:Footnote ²

Table 3.

The development of the PIE holokinetic paradigm.

This holokinetic inflection is the most promising candidate as PIE etyma for the Germanic reflexes, as discussed in sections 3 and 4 below. Although it may appear that some Gmc. forms come from ⁺ph₂- én- (see sections 5 and 6), it is the position taken here that they descend from the holokinetic paradigm.

2.2. Laryngeal Metathesis and ⁺ū

A number of etymologies argue that the etyma of Germanic reflexes with long /u/ and /y/ in the root syllable should be traced back to a PIE ū which arose from metathesis of u and h₂ in the sequence ⁺ph₂- u.Footnote ³ Invoking metathesis as an explanation for ⁺ū is problematic. Rasmussen (1989:107, following Hamp), suggests that there might have been a rule in which a laryngeal underwent metathesis with a semivowel (⁺Hu > ⁺uH) in post-consontantal position. Whereas this works for the word for ‘sun’, ⁺ sh₂u-> ⁺suh₂-, when a consonant immediately follows, the same does not hold true for the word for “fire,” where reflexes of the sequence ⁺ ph₂u- have retained a short vowel, perhaps due to the different phonetic traits of /s/ and /p/. In his analysis of the relationship between long and short variants of /u/ in a paradigm, Schrijver (1993:525) is less interested in meta-thesis, as in the search for a factor conditioning the length of the reflex of ⁺Hu. From his examination of Greek reflexes, Schrijver sees stress as the conditioning factor at work. When stressed, the sequence ⁺Hu developed as ū. In pre-tonic position, however, ⁺Hu gives short u.Footnote ⁴ The only reflexes providing direct evidence of laryngeals for “fire” come, of course, from Anatolian, and in those cases the ordering is always HR and not RH.Footnote ⁵ In the following section the four other branches (Greek, Italic, Armenian, and Slavic) that have potential reflexes of PIE ⁺ū are examined.Footnote ⁶ In each case I argue that the vowels in these cognates could also be possible products of independent analogical developments. Alternative explanations for reflexes of PIE ⁺ū among Indo-European cognates in combination with other explanations for sources of long vowels should cast doubt as to whether cognates of the word for “fire” in Germanic possess historically long ⁺ū as a reflex of either PIE ⁺-h₂u- or -uh₂-.

2.3. Greek

Although there are several suggestions as to how PIE ⁺ph₂ ōr(nom.-acc.) and ⁺ph₂ nés (gen.) became Gk. p ū r (nom.-acc.sg.) and pur—s (gen.sg.), one seems to make the most sense.Footnote ⁷ The key to understanding the development into Greek is hinted at in the relationship between the long root vowel in the nominative and the short vowel in the oblique cases. For the most part, Greek has leveled out ablaut grade and length distinctions in its nominal paradigms. N-stem nouns, and a handful of r-stems such as pat ē r, exhibit suffixal full grade in the obliques and lengthened grade in the nominative (Sihler 1995:280). The first change from Proto-Indo-European to Greek involved the leveling of the r-stem throughout the holokinetic paradigm, producing a new genitive singular (with an o-grade ending as found with other athematic declensions in Greek):Footnote ⁸

1. (1) ⁺ ph₂-un-és> ⁺ punés > pur—s

After the creation of the new oblique forms in pur-, the long vowel of the nominative singular could have been restructured according to a model found in the word for ‘wild pig’, as in example 2.Footnote ⁹

1. (2)

In this manner one can account for the length of the root vowel without having to reference laryngeal metathesis or additional and unnecessary sound changes. Rasmussen (1996:77) sees the circumflex accent in p ū r as evidence of an earlier disyllabic form. One should note, though, that words such as h ū s (< PIE ⁺suh_x-s) and m ū s(< PIE ⁺mūs < perh. ⁺muh_xs) possess circumflex accent in historically monosyllabic words. These monosyllables with circumflex accent speak against the necessity of a disyllabic etymon for p ū r, and would suggest rather that an analogical process was at work.

2.4. Oscan and Umbrian

Given the fragmentary remains of Oscan and Umbrian, reconstruction of the history of Umb. pir (< earlier ⁺pūr) is hampered by limited data. Nonetheless, it is still possible to argue that the length of the vowel in the nom.sg. form is the outcome of independent analogical processes similar to those seen in the development of Gk. p ū r.

The attested oblique forms of Umb. pir are acc.sg. purom-e and abl.sg. pure-to, which indicate a long vowel in the nominative and a short vowel in the obliques. Although one expects neuters to have identical forms in the nom. and acc.sg., Umb. acc.sg. purom is a more recent formation derived from the abl.sg. in analogy with masc.acc.sg. in -om (Buck 1904:129). It appears from the abl.sg. pure that the Umbrian favored the holokinetic paradigm, as with the other non-Anatolian Indo-European languages that reflect this word. In its evolution to the attested Umbrian forms, the r-stem of the nominative/accusative was leveled throughout the paradigm.Footnote ¹⁰ One can then view the pre-form of pir, Sabellic ⁺pūr, as the product of analogy with other root nouns where a long vowel in the nom.sg. alternated with a short vowel in the obliques, see Umb. gen.pl. nerum and dat.pl. nerus on the one hand and Osc. níír (< earlier Sabellic ⁺nēr) ‘man’ on the other (Buck 1904:129; cf. Sihler 1995:280).Footnote ¹¹ Although forms for the entire paradigm of either Umb. pir or Osc. níír are not attested, we can conjecture that an analogy would have been possible to produce Sabellic ⁺pūr:

1. (3)

2.5. Classical Armenian

Evidence available from Arm. howr ‘fire, nom.sg’, and hnoc‘ ‘oven’ (< ⁺hun ó c^h a-) (Klingenschmitt 1994:243) is of limited value in determining vowel length, as PIE ⁺ū and ⁺u (along with PIE ⁺ō) merge into Arm. /u/(transliterated as <ow>).Footnote ¹² The variation of howr ‘fire, nom.sg.’ and hroy ‘fire, gen.sg.’ is less indicative of vowel length as it is of earlier Armenian accentuation prior to loss of unaccented syllables. A number of other nominal paradigms exhibit the same u ~ zero alternation in their paradigms (cited here in the nominative and genitive): k‘own ~ k‘n-oy ‘sleep’ (< PIE ⁺swopnos), lowc ~ lc-oy ‘yoke’ (< PIE ⁺yugom), and dowstr ~ dst-er ‘daughter’ (< PIE ⁺d^h u h₂tēr) (Schmitt 1981:38–39). Again it is possible that, as described in sections 2.3 and 2.4 for Greek and Umbrian, Classical Armenian howr ~ hroy is the product of an analogical reformation of the nominative-accusative singular after the model of pre-existing paradigms. In short, one cannot be guaranteed that howr indeed reflects earlier ⁺ū, as it could be the product of analogy or even conceivably derive from the short u of the oblique cases leveled throughout the paradigm.

Klingenschmitt's (1994:234) reconstruction of ⁺pūr ~ ⁺ pun é s from Greek, Italic, Armenian, and Germanic is problematic. Although each daughter family appears to reflect the vowel alternations these etyma would have produced, it is nonetheless a possibility that the reflexes are products of nearly identical, though independent changes. Furthermore, one must still bring ⁺pūr ~ ⁺ pun é s into accord with the data from Anatolian, such that all reflexes can be explained from a common ancestor, PIE ⁺péh₂- and ⁺péh₂- ōr. The change from PIE ⁺péh₂- ōr ~ ⁺ph₂-un- és to ⁺pūr ~ ⁺ pun é s carries with it two difficulties. First, in positing a vowel alternation in ⁺pūr ~ ⁺ pun é s we cannot be certain that the ablaut relationship ⁺e ~ ⁺ē and ⁺o ~ ⁺ō would be applicable to the vocalic variants of semi-vowels. Taking the general rules of Indo-European samprasaraa into account, we might expect the ⁺u of ⁺ pun é s to have been treated as a consonant in secondary derivatives, rather than as a vowel. An analogical reformation of ⁺ pun é s should have produced ⁺p ē -r, or possibly, though less likely ⁺p ē-r (provided also that the heteroclisis is preserved).

The second difficulty lies in the analogical innovation and its chronology relative to the leveling of the consonant stem. In Greek, Italic, and Armenian, the r-stem forms of the paradigm became the norm. However, the Germanic cognates demonstrate that leveling of the para-digm as r-stems or n-stems must have taken place separately in each individual language. For Germanic to have preserved the r-stems and n-stems as morphemes separate from the root and ending, we must suppose the same for its etymon. Producing ⁺pūr from ⁺ pun é s analogically from paradigms with a similar vowel alternation seems all the more unlikely, as it would require a parsing of the oblique into root and stem as pu-n-, and applying ablaut patterns to a non-canonical CV root, rather than interpreting ⁺pu- as the zero-grade of a more regular CVC root shape.

That four branches of Indo-European possess words which look to be derived from a common ancestor ⁺pūr (Gk. p ū r, Umbr. pir, Arm. howr, and ON. f úrr) is not sufficient evidence for the reconstruction of ⁺pūr in non-Anatolian Indo-European. In the cases of Greek and Umbrian, there is sufficient reason to surmise that in each case a new form ⁺pūr had been created based on models present in the languages, which were not present in the proto-language. In the case of Armenian we may imagine the same, though we are hampered by the Armenian vowel mergers on the one hand, and the lack of sufficient examples of root noun paradigms on the other. Finally, as seen below in section 6, there are reasons to view the existence of long u in the Germanic reflexes as products of independent phonological developments and analogical restructuring in late stages of the daughter languages.

2.6. Slavic

Along with Greek, Italic, Germanic, and possibly Armenian, Slavic appears at first glance to preserve reflexes of earlier ⁺pūr in Cz. pýř ‘glowing embers’, Cz. pýřiti and Slov. pýřit’ sa ‘to redden’, and OSorb. pyrić ‘to heat, melt’, among others (Vasmer 1964–1973:419). As seen in sections 2.3 and 2.4 above, however, the long vowel in Slavic could be an independent development due to analogical change. Balto-Slavic is characterized by “neo-lengthened-grade” reflexes of PIE ⁺i and ⁺u (Fortson 2004:367; Carlton 1990:82); for example, Pol rydz ‘lactarius deliciosus’ (a reddish-hued mushroom) < earlier ryd - < PIE ⁺h₁reud^h- ‘red’ (Carlton 1990:91). Examples 4a–c exemplify the Slavic length-ening of PIE ⁺u. One will note that the PIE roots below possess no laryngeals and close in obstruents, which would preclude any possible lengthening of preceding vowels by laryngeals in suffixes or endings:

1. (4)

   1. a.

   2. b.

   3. c.

As the above sections have indicated, there are alternative explanations to reconstruct a long u in the PIE etymon for “fire.” The possibility exists that these similar reflexes are simply the products of coincidental developments in languages which possessed similar lexical items and similar grammatical paradigms.Footnote ¹³ I argue below (section 6) that apparent Gmc. reflexes of ⁺ū could be the result of separate phonological and analogical changes.

4.1. Extension of the e-Grade n-Stem Suffix

Proto-Germanic inherited from PIE the forms for “fire” designated here as stage I (see table 4). Its subsequent development into the attested Germanic word forms was most likely the result of two factors: the salience of the n-stem forms of the heteroclitic inflection as n-stems, on the one hand, and the unusual shape of a zero-grade n-stem in the singular, on the other. Later leveling of the heteroclitic stem in either -r or -n indicates that the stem consonants were still treated as morpho-logical equals within Proto-Germanic. The equivalence of -r and -n as stems also means that the roots of the PGmc. word for “fire” were parsed as ⁺fō- and ⁺ fu-, respectively.

Association of the oblique cases with the regular n-stem inflection opened the door to an analogical reformation of ⁺ funiz and ⁺ funi. These analogical reformations replaced the unusual zero-grade n-stem of ⁺fun- with the more typical suffix of n-stem singulars, ⁺-in-.Footnote ¹⁶ One can make the following analogy with the PGmc. neuter noun ⁺ hert-in- ‘heart’:

1. (5)

As this analogical process would have only affected n-stem forms, the nom.-acc. ⁺fōr would have remained unchanged. At the end of the process, Proto-Germanic would have possessed two distinct inflections for “fire,” the older paradigm (stage I) with zero-grade n-stems, and the newly formed paradigm (stage II) with Gmc. reflexes of e-grade n-stems:

Table 5.

The stem variants of “fire” in Proto-Germanic, stage II.

As I argue below, the newly created form ⁺fuïn- is the source for the NWGmc. words for “fire” in -r. Alternate explanations for ⁺fuïn- are problematic. Rasmussen (1989:77, n10) argues for a derivation of PGmc. ⁺fu n- from a Lindeman's Law variant of the oblique proterokinetic inflection in PIE, that is, ⁺ph₂u én-.Footnote ¹⁷ The operation of Lindeman's Law with bilabial glides possesses few unambiguous case examples, and what little data there are would suggest ⁺ fwin- as the expected Gmc. outcome of PIE ⁺ph₂u én- (Ringe 2006:121–122, 277).Footnote ¹⁸ Voyles (1992:238; also section 6.3 below) looks to a pre-Germanic form of ⁺pú-er- as a possible source of Gmc. ⁺ fuïr-, from which he derives ON fýr. However, ⁺p ú-er- is an unlikely source, because it is evident that the leveling of -r in the paradigm takes place after the Proto-Germanic period. Beekes (1996:6) also attempts to derive ⁺fuïn- from the proterokinetic inflection, though with a metathesized laryngeal and analogically reintroduced n-stem suffix, ⁺puHen-. In a sense, Voyles and Beekes approach what the correct processes probably were; however, their relative chronologies for when these changes occurred are off. Ringe (2006:277) proposes that ⁺fuïn- derives from ⁺funin- via dissimilatory deletion of the first -n-, a change that might better be explained in another manner, discussed below in 4.2.

4.2. Blending of the Two Paradigms

In addition to accounting for the origin of forms based in ⁺fuïn-, substitution of PGmc. ⁺-in for ⁺-n in the oblique cases also motivates the creation of the pre-forms of Go. funins (gen.), funin (dat.) as well as ON funi, funa ‘flame’.Footnote ¹⁹ Coexistence of competing oblique bases ⁺fun- and ⁺fuïn- creates an environment in which a form such as ⁺funin- could easily have arisen as a result of a speaker's uncertainty regarding which of the two forms should be used. One could characterize this pheno-menon as hypercorrection where the speaker, in doubt regarding the form of the oblique ending, uses both.Footnote ²⁰ Hock (1986:169) notes in reference to analogical change that “traces consist in the coexistence of the old form with the innovated form […]. Doublets of this sort are very common as transitional stages in all analogical change. Occasionally, however, they may persist for a longer period. In such cases, they are often subject to the analogical process of blending […].”

Such a process is observable also in two Latin reflexes of IE heteroclites: iecur ~ iocineris and iter ~ itineris. In order for itineris to have come about, a third, unattested form ⁺iteris must be assumed. For a time two paradigms coexisted, older ⁺iter ~ ⁺itinis and newer ⁺ iter ~ iteris. Latin speakers formed a new genitive by attaching the newer -eris onto the preexisting itin- stem (Sihler 2000:85). It is reasonable to treat PGmc. ⁺funin- as the product of a similar process (see table 6).

Table 6.

The stem variants of “fire” in Proto-Germanic, stages I–III.

The interpretation above has two benefits:

• • It accounts for the two stems ⁺fu-n- and ⁺fu-in-.

• • It provides for the extension of the e-grade n-stem endings to only the oblique forms, while retaining the original ablaut in the nom.-acc. form, without having to recast the word as an n-stem.

We should take heed to note, however, that the innovation of ⁺funin- involves the double use of an n-stem, and not the reanalysis of the word's root as ⁺fun -, to which an n-stem ending is added.Footnote ²¹ The reasoning behind this is that reanalysis of ⁺fun- with -n as part of the root rather than of the stem would suggest that ⁺fun- was no longer perceived as possessing an n-stem. If speakers no longer perceived it as an n-stem, reanalysis as a root-noun would seem more logical, and not the addition of another n-stem. Reanalysis of the root would require the paradoxical state of simultaneously belonging to an n-stem inflection and not belonging to an n-stem inflection.

4.3. Leveling of Stem throughout Paradigms

As mentioned above, this heteroclitic noun remained saliently hetero-clitic at the earlier stages of Proto-Germanic. Only later did speakers level both stem consonants such that the heteroclisis disappeared. This is also indicated by the ordering of the steps. For example, reflexes based on ⁺fuïr- (OHG fuir, OE, OFris, OS, and OHG fiur, ON f úrr, fýr, and fýri) must have had their -r replace -n after the spread of -in to forms in ⁺fun-. Otherwise we might expect forms such as ⁺funir-, which is not evidenced. Nonetheless, a few matters remain unclear. First, it is unclear whether these levelings took place toward the end of Proto-Germanic, or whether the leveling took place independently among the daughter languages. One should note, though, that in order for the stems to be leveled throughout the paradigms, two things are required:

• • There must have been two separate paradigms, with homophonous forms in the nom.-acc. singular, that is, ⁺ fōr ~ funiniz and ⁺ fōr ~ ⁺fuïniz, as illustrated in stage III.

• • The stem consonants must have been salient morphological units with respect to their roots at the time of leveling.

The first point is suggested by the ways various dialect groupings of Germanic preserved the paradigms: East Germanic preserved only n-stem varieties, West Germanic preserved r-stem varieties, and North Germanic preserved both, though with sociolectally restricted usage. Had we been faced with a situation in which there was only one nom.-acc. form, but multiple obliques (rather than two homophonous nom.-acc. forms) at the point during which ⁺ fōr became ⁺ fōn, then there would have been no way for the -r to spread to ⁺fuïriz. The converse, however, is still possible. Forms like ⁺fuïriz could have taken on -r prior to the change from ⁺ fōr to ⁺ fōn. It would be helpful to have a surviving reflex of ⁺ fōr; however, none exists. Reflexes bearing -r in Northwest Germanic represent the only traces that it had existed.Footnote ²²

In order for leveling to have taken place at all (that is, the change from -r to -n and from -n to -r), a notion of morphological equivalency must have been in play. Although the changes between stage I and stage II, where the oblique stem -n was replaced by its more common form -in, suggest a connection to n-stem nouns, the word(s) for “fire” remained heteroclitic. Only later is there an indication that forms such as funin- were treated as n-stem nouns, as indicated by the ON funi (as discussed in 5.3. below).Footnote ²³

Table 7.

The stem variants of “fire” in Proto-Germanic, stages III & IV.

One may speculate that the motivation behind the direction of leveling in each paradigm was driven, in part, by the double-marked stems of ⁺funin-. As these stems have predominantly more “n” than ⁺ fōr ~ ⁺fuïn-, it is natural to see the -n leveled into the nom.-acc., producing ⁺ fōn. Conversely, in the context of two competing heteroclitic paradigms, the domination of one paradigm by the n-stem blocks, in a sense, the leveling of the n-stem throughout the competing paradigm of ⁺ fōr ~ ⁺fuïn-. Therefore, when leveling does take place, it is the r-stem which prevails. The leveling of the stem consonants is a prerequisite for the reformation of these words as members of other inflectional categories in all Germanic languages except Gothic.

5.1. Gothic

Besides Toch. A puwar and Toch. B pwar, Gothic is the only other historically attested Indo-European language to preserve a clear reflex of the vocalism in PIE ⁺ph₂- r. Provided that the leveling of the n-stem in the paradigm is not an independent innovation in Gothic, which the data from ON would suggest it is not, the Gothic forms are more or less the same as their PGmc. predecessors, with necessary phonological changes deleting the final unstressed vowels in inflections:

Table 8.

The development from PGmc. stage IV to Gothic.

A more problematic cognate of Go. fon is Shetl. f⊘n, a taboo replacement for “fire.”Footnote ²⁴ Shetl. f⊘n is first attested in Edmondston's (1866:36) thesaurus of Shetlandic and Orkney Norn, cited as fūn with the note that the vowel's pronunciation is a long /y/, as found in French. Jakobsen (1928:205) cites it later under the form f⊘n. There are a couple of ways of explaining a form such as f⊘n ~ fūn. The pre-Norn form of both could be either ON ⁺ f —n or ⁺fún.Footnote ²⁵ If we consider ⁺fún as the preform, then we are forced to provide an explanation of the n-stem and the ū (which would have to have come from ON fúrr). This course of explanation is fraught with ad hoc changes. One would have to ask whether ⁺fún is the product of carrying the vowel of fúrr over to funi and turning it from an n-stem to an a-stem noun, or whether it is the product of carrying the -n- of funi over to fúrr despite the fact that neither word exhibits heteroclisis, among other problems? Following Edmondston's (1866:36) interpretation of fūn as a variant of Shetl. fona (< ON funi), one might also see it as the product of open-syllable lengthening fol-lowed by apocope. This is not without problems, as one must contend with the survival of fona without lengthening and apocope. If, however, we look to hypothetical ON ⁺f—n as the pre-form of Shetl. f⊘n ~ fūn, then the possibility exists that PGmc. ⁺fōn, as found in Go. fon, survived in North Germanic in the remote islands to the north of Scotland, preserved only as a taboo variant in the speech habits of fishermen when away at sea. Caution is warranted here, though, as the amount and quality of data available on Shetlandic Norn is lacking.

5.2. Old Norse

Apart from Shetl. f⊘n (which suggests that PGmc. ⁺fōn was not eliminated from North Germanic by the developments portrayed in table 9) the usual n-stem formation found in Old Norse was recast as an n-stem noun, albeit with a gender change from neuter to masculine and standardized PNorse ⁺-an- < PGmc. ⁺-on-.Footnote ²⁶ This need not have occurred as a result of analogy with ON eldr (Pokorny 1959:828), but rather because the overwhelming majority of n-stems are masculine. This occurred when subjected to the analogical reformation in 6. There are no traces of the n-stem inflection in West Germanic, which only provides evidence of this change as occurring in North Germanic, though it could have conceivably happened at any point from the late PGmc. period on.

1. (6)

Table 9.

The development from PGmc. stage IV to Old Norse.

ON funi ‘fire, flame’ continues to exist in Modern Icelandic, despite eldur remaining as the default word for “fire.” The n-stem forms are also attested as late as the 18th to 19th centuries in Norn and Norn remnants in the Shetland Islands. In Norn one finds reflexes of ON funi as Shetl. fona (with variants fenna, finna, fenni, and finni).

6.1. Old High German, Old Low Franconian, and Old English

The Old High German corpus presents us with the best direct evidence of former ⁺fuïr. This variant is present in OHG texts from the eighth century to the mid-ninth century, after which point one finds only OHG fiur (Braune 1987:52), for example, in the Abrogans Footnote ²⁸, as well as the Regula Benedictorum Footnote ²⁹, Tatian Footnote ³⁰, the Weißenburger Katechismus, and the Muspilli Footnote ³¹. Of these texts, however, Muspilli 59a is of particular interest, as the manuscript, Cod.monac.lat. 14098, preserves the spelling uugⁱ r, which likely indicates a disyllabic pronunciation.Footnote ³²

Old Low Franconian also preserves a reflex of unmetathesized ⁺fuïr. The OLF Psalms of the Diez Manuscript preserve five examples of fuir: Psalms 57.9, 65.10, 65.12, 67.3, and 73.7, in all four grammatical cases.Footnote ³³ I would take the MS spelling at face value, representing [fuïr] or something similar, rather than viewing the sequence <ui> as a grapheme for a long high front rounded vowel. Fortunately, problems in the interpretation of the graphemic data, that is, whether fuir represents [fuïr] or [fy:r], are not an issue here, as the likeliest source for [fy:r] would have been [fuïr].Footnote ³⁴

OE f r is also a reflex of ⁺fuïr.Footnote ³⁵ This form alongside the ON forms fýr, fúrr, and fýri(r) have been the main reason for the positing the existence of historical ū in this word's etyma.Footnote ³⁶ Although i-umlaut is certainly at work here producing the front rounded vowel, I think it likeliest that the length of this vowel (and that of ON fýr below) is the product of coalescing vowels. As Winter (1965:192–193) points out, the vowel sequence -ui- is unusual in Germanic and was likely restructured to [iw] (see section 6.2). Despite the rarity of the -ui- sequence in Germanic, the development of -ui- > in Old English is paralleled in the loanword from Old Irish druí ‘druid, sorcerer’ > OE dr ‘sorcerer’. One may assume a near identical, though independent, sequence of events for the change from ⁺fuïr to ON fýr, discussed below in section 6.3.

6.2. Old Frisian, Old High German, Old Saxon, and Anglian OE

Whereas Old English and Old Norse dealt with the unusual CV-VC sequence present in ⁺fuïr by the coalescing of syllables, the remaining WGmc. forms “solve” the unusual syllable sequence by metathesis of the two elements, as first pointed out by Winter (1965:192–93).Footnote ³⁷ Here one finds OFris., OHG, and OS fiur. Old FrisianFootnote ³⁸ and Old SaxonFootnote ³⁹ exhibit only the metathesized forms. As mentioned above in section 6.1. OHG fiur tends to dominate and is the sole attested form by the end of the ninth century.

Old English seems to have used f r almost exclusively. There is, however, one possible remnant of a metathesized form more similar to its continental cognates, OS, OFris., and OHG fiur. Verse 45b of the Old English Riming Poem cited without emendation from the ASPR edition of the Exeter Book (Krapp & Dobbie 1936:168) in the context of ll.44b–47a reads:

The lack of rhyme between half lines causes line 45 to stand out in the Riming Poem.Footnote ⁴² In addition to the absence of end-rhyme, 45b is also metrically deficient. At the very least the verse lacks an unstressed syl-lable after deop and/or feor. A hypermetric off-verse would not be out of the question, as 45a is a relatively rare instance of weak hypermetric on-verse and the verb-initial syntax of 45b is in line with typical hypermetric constructions.

Given verse 45b's metrical and morphosyntactic deficiencies, let us assume that dyre of 45a represents something close to the original rhyme-pair of line 45.Footnote ⁴³ The MS spelling of this word, a reflex of Proto-Germanic ⁺diur-ja-/⁺diuz-ja ‘dear’, would be typical of late West-Saxon (Campbell 1959:126–128).Footnote ⁴⁴ This spelling, though, reflects perhaps the last stage in the poem's transmission, as clues in the poem's rhymes point to an original composition in an Anglian dialect (Fulk 1992:365).Footnote ⁴⁵ The Northumbrian cognate of lW-S d re would have been dīore, and a Mercian form could have been either dēore or dīore (Brunner 1965:113). Taking the first half of the rhyme pair as either dīore or dēore we may posit a possible match with fīore or fēore, as the case may be.Footnote ⁴⁶ End rhyme and sense could be introduced if we suppose the original composition of verse 45b to have been along the lines of the following:

Why OE fīor would have surfaced only in this text is tied, in part, to the same reason that Old Norse retained a multitude of cognates (see in particular section 6.3). In the OE Riming Poem the formal poetic requirements (in this case, end rhyme) held a dominant position over lexis. The poet employed here a perhaps less common, though under-standable, variant of more usual f r in order to meet the requirements of both semantics and rhyme.

6.3. 6.3. ON Forms in -r

Old Norse preserves the greatest variety of reflexes of “fire.” In addition to funi, funa examined above in section 5.2., Old Norse preserves three reflexes of PGmc. ⁺fuïr-: fúrr, fýr, and fýri(r). Of these three, fúrr and fýri(r) are likely innovations within North Germanic alone. Unlike funi, these three words had a nearly exclusive poetic usage, which explains in part their variety and survival. The switch to ON eldr as the default lexical item for “fire” enabled the earlier synonyms to take on a secondary role as poetic vocabulary.Footnote ⁴⁸ In West Germanic, where reflexes of PGmc. ⁺fuïr- remained the default lexical item for “fire,” this secondary role as poetic vocabulary was less likely, hence the presence of only two variants, whose differences are purely phonological in nature. Given the demands of skaldic poetry, it would have been advan-tageous to possess synonyms with varying syllable numbers, weights, and vocalic qualities for internal rhymes. In eddic poetry the only variants exhibited are funi and fúrr. The ja-stem variant, fýri(r), is found three times in the corpus, and only in the works ascribed to one skáld, Einarr Gilsson. Each cognate is examined and discussed in turn below.

ViÐb—tarþulr pp.2 (Þl IV pp.2) is home to three of the synonyms, fýr, funi, and fúrr, in lines 3, 7, and 8, respectively. This stanza, found among a listing of 60 heiti for “fire”, contains the only attested use of ON fýr and is extant in only two manuscripts of the Prose Edda (AM 748 and AM 757):

• Læygr vafrlogi

• linn r ok farr i

• bran dr fýr tsi

• brími ner till.

• viti eimyria

• velann fasti

• fvni hyrr fæÐir

• fvr rok eisa.Footnote ⁴⁹

Other than the presence of a long, rounded front vowel, there is no reason to suspect that f ýr reflects fýrr, a masc. i-stem variant of earlier ⁺fūr-.Footnote ⁵⁰ It is apparent from preceding brandr and following t si that these words appear in the nominative form. One would expect the scribe to have written f ýrr, had it been masculine.Footnote ⁵¹ A more tenable solution lies in Egilsson's (1913–1916:163) note: “eller fýr ntr?” Whether PGmc. fuïr- would have become ON f ýr, just as it had become OE f r, is unknown, as no additional examples or counterexamples are available to determine the development of PGmc. -uï-. It seems likely, though, that ON fýr reflects earlier ⁺fuïr- rather than metathesized ⁺fiur-.Footnote ⁵² Barring analogical influence from words such as d ýr ‘animal’ (PGmc. ⁺diuz-a-), one would expect a metathesized variant ⁺fiur- to have become ON ⁺ fj —r, cf. ON stj—r, þj—r.Footnote ⁵³

Rather than viewing fýr as an i-stem derivative of fúrr, we should entertain the converse; namely, that fúrr is derived from fýr. Voyles (1992:238) suggested that the [y] of f ýr was reinterpreted as a morpho-logically conditioned umlaut, and that masculine fúrr is the product of R ückumlaut.Footnote ⁵⁴ This would be possible in analogy with the dative singular of root nouns:

1. (8)

Subsequently fúrr became an a-stem noun.Footnote ⁵⁵ ON fúrr continued to survive in the taboo vocabulary of Shetlandic fisherman as Shetl. furin (= fūrin), foorin with the earlier suffixed definite article fused to the noun, that is, from ON fúrrinn (nom.sg.) or fúrin (acc.sg.) (Jakobsen 1928:204).

Just as fýri (dat.sg.) made it possible to recast earlier fýr as fúrr, so too was it possible to create a new variant, modeled after ja-stems:

1. (9)

Here one should note that all three occurrences of ON fýri(r) are not only restricted to poetic usage, but to the poetry of one fourteenth century individual, Einarr Gilsson.Footnote ⁵⁶ The three attested instances share other common points as well: they are all in the genitive (leaving us unable to tell whether the noun is neuter or masculine); they all occur in kennings for “man;” they are all found as the final cadence of an off-verse; and they all participate in a Ðalhending rhyme. No other variant of “fire” could have stood in the same position. Funa lacks the requisite syllable weight and rhyme, and neither f ýrs nor fúrs would provide a trochee, though they provide aÐalhending and skothending, respectively. One might question whether this form would have been part of the common poetic stock, or whether it was perhaps the novel coinage of this one clever skáld who required a trochee with a trimoraic first syllable to serve in the dr—ttkvætt and hrúnhent cadences while still satisfying the requirements for aÐalhending in the off-verse. As umlaut in Old Norse during this period was morphologically conditioned (see, for example, Iverson & Salmons 2004:105–107), others would have been capable of deciphering just such a poetic neologism. The relationship between f úrr and fýri (r) would have been morphologically transparent.