West Norse Palatal <i> <scp>r</scp> </i> in Irish Sources

A point of contention in early twentieth-century Norse-Icelandic scholarship was the date of the merger of etymological /r/ and palatal r from */z/ in Old West Norse (henceforth ON). Two main contributions were Bidrag til det norske sprogs historie i Irland by the Norwegian scholar Carl Marstrander and Norsk-Islandske kultur- og sprogforhold i 9. og 10. århundre by the Icelandic scholar Finnur Jónsson.1 In his pivotal study of ON loanwords in Irish, Marstrander investigated the reflexes of the two rhotics in a small subset of the material. It is generally held that Old and Middle Irish had four rhotic phonemes, two of them palatal, and the assumption that ON palatal r would be reflected with a palatal rhotic in Irish before it merged with /r/ is therefore reasonable a priori (the fortis-lenis contrast /R/ vs. /r/ need not concern us here). The bulk of the loanwords was transferred orally from ON to Irish in the ninth and tenth centuries.2The rhotacization of the alveolar sibilant */z/ in Proto-Norse (henceforth PrN) can be seen internally from the effects of final devoicing. Traditionally denoted by r, the resulting phoneme can no longer have been a voiced sibilant when final consonants underwent devoicing, as the otherwise expected change */z/ > /s/ did not occur (ON dagr ‘day’ : Gothic dags).3 On this view, the change */z/ > r took place no later than the fifth century and had already happened by the time of the Blekinge and Eggja inscriptions, preceding the monophthongization of /aix/ > /aːx/ (before final devoicing) and /aix/ (from /aiɣ/ with final devoicing) > /eːx/.4 Final devoicing must predate the syncope of short /a/, as seen from a comparison of ON batt ‘bound’ (first/third singular past indicative active of binda ‘bind’) < *bant < *band with band ‘cord’ < *banda.5 The change in articulation can thus be dated relatively to other sound changes. Finnic and Saamic evidence supports the development outlined here, as /h/ is substituted for */z/ in the earliest PrN loanwords, whereas /r/ is substituted for the later r. Thus, Finnish keihäs ‘speer’ was borrowed from PGmc *gaizaz (ON geirr ‘speer’) before the change */z/ > /h/ in Finnic.6 The later kaira ‘auger’ was borrowed from *gair- after rhotacism (*/z/ > r) in PrN.7Despite the reasonably clear relative chronology and the fact that we can trace the development of the phoneme through external evidence, there is no current consensus on its precise phonetic realization. It may have been pronounced as the apical post-alveolar approximant [ɹ], the voiced postalveolar fricative [ʒ], or the voiced retroflex fricative [ʐ]. The palatalizing or fronting effect r had on preceding vowels, comparable to the workings of i-umlaut, suggests that it had palatal quality.8In West Norse, r merged with /r/ during the Viking Age, where it did not assimilate to other consonants in consonant clusters that arose after syncope (lr, nr, rr, sr > ll, nn, rr, ss). Michail Steblin-Kamenskij argued that the opposition /r/ : r was first neutralized in certain positions (so we get r for /r/ after unstressed /i/ and /r/ for r after dentals).9 Unlike these assimilations and the merger after dentals and labials, the merger in other environments, notably absolute auslaut, cannot be given a date relative to other sound changes in ON. In Marstrander's view, the two phonemes were distinct in West Norse until ca. 1200.10 However, Finnur Jónsson noted that this date is at variance with skaldic evidence (rhyme), which points towards the completion of the merger by the end of the ninth century, and was therefore reluctant to accept Marstrander's principle of regular sound substitution in the ON loanwords in Irish, even though that principle is the obvious precondition for any critical use of the loanword corpus for linguistic or even historical purposes.11 Marstrander, in turn, was skeptical of the evidence of the merger in skaldic rhyme.The present contribution seeks to resolve this conflict by offering an alternative interpretation of the pertinent evidence, which is, in fact, restricted to a couple of proper names compounded from the deuterotheme inherited from Proto-Germanic (henceforth PGmc) *‑gaizaz ‘spear.’ Central to the interpretation offered here is the rebuttal of Adolf Noreen's theory of a shifting accent in PrN dithematic names. I will instead suggest that these names behave according to the regular development seen in common nouns. The central issue is not so much a minor point of ON phonological development as one of method. Recent studies of the linguistic interaction between the Scandinavian world (mainly Norway) and the British Isles and Ireland have moved away from the systematic approach that was fundamental to Marstrander's work.12 Loanwords undergo adaption processes to conform to the target language's phonology and can therefore enrich our knowledge of the structural constraints of that phonology. Integrated loans are often distinguished in the target language only by their origin, as they are subject to the same linguistic changes that affect the inherited lexicon. The phonetic substitutions inform us about the acoustic relation between the phones of the two languages.13 For comparative linguistics, however, it is equally essential that the loanwords naturally escape subsequent linguistic change in the giving language and can thus be used as a source to the phonological makeup of that language at the time of borrowing. Loanwords, accordingly, supplement the comparative method.In the following, I aim to show that Finnur Jónsson's critique of Marstrander's take on the chronology of linguistic change was justified but that his reproval of the central tenets of the study was not. Finnur Jónsson based his conclusions on unetymological (obligatory) rhyme in aðalhendingar.14 The skaldic evidence suggests a terminus ad quem at ca. 900.15 We have five examples of rhyme between r and /r/ in Haustlǫng, a poem by the Norwegian poet Þjóðólfr ór Hvini (fl. late ninth–early tenth century), which demonstrate the merger of the two phonemes. Finnur Jónsson rejected Marstrander's principles as if they depended on his theories, rather than the other way around, but in fact, the principles in question simply amount to a systematization of the evidence. There is little reason to assume that skaldic poetry permitted rhyme between r and /r/ before their merger, however, as Marstrander did.I am not aware of any unequivocal example of the substitution in Irish of a non-palatal /r/ for ON r. There is, therefore, no direct Irish evidence of the merger of the phonemes in ON. However, we know that Irish substituted palatal /rj/ for ON non-palatal /r/ in the vicinity of front vowels. Examples of this are: accaire /ˈakərjə/ ‘anchor’ from ON akkeri /ˈakkeri/; sceir [ˈʃkjerj] ‘skerry’ from ON sker /ˈsker/; and stiúir /ˈʃtjuːrj/ ‘rudder’ from ON stýri /ˈstyːri/ (these ON words are all neuters). The same substitution is seen in Scottish Gaelic placenames of ON origin. Two examples from Leurbost, Isle of Lewis, are [ˈkðeːnəval] (name of a hill) from ON Grǿnafjall /ˈgrøːnaˌfjall/ and [ˌtɔb ˈLjeːðəvaj] (name of a bay) from ON (hóp) Leiruvág /ˈleiruˌvaːɣ/ or /ˈleiruˌvɔːɣ/.16 These examples show that the substitution of a palatal Irish rhotic for ON r in the vicinity of front vowels could have been triggered by the phonetic environment.Within the subsystem of ON loanwords, the quality of the consonants is phonologically redundant, as it is largely predictable from the vocalic environment.17 Therefore, evaluating Marstrander's hypothesis is straightforward: the only diagnostic items are Irish words of ON origin where /rj/ is substituted for an ON palatal r that is neither preceded nor followed by a front vowel or diphthong. In ON (as in Old Irish), we find a strongly reduced inventory of vowel phonemes in unstressed syllables. ON had nine vowel phonemes in stressed position against three phonemes in unstressed position, which in normalized orthography are denoted as i, u, and a. Hreinn Benediktsson has argued that the phonetic realization of the high unstressed vowels was stable in Old Icelandic, as [ɪ] and [ʊ], despite their shifting orthographic representation in the manuscripts.18 According to this view, the orthographical representation changed when the short front unrounded vowels in stressed position were lowered so that unstressed [ɪ] could be identified with stressed /i/ rather than /e/ and, somewhat later, [ʊ] with stressed /u/ rather than /o/. Therefore, the scribal conventions are evidence of changes to the vowel system in stressed syllables and not to the unstressed system. The corollary is that the unstressed vowel denoted in ON manuscripts by <e> or <i> triggered the substitution of palatal Irish /rj/ regardless of the etymological origin of the ON rhotic. Any front vowel in the range close [i] to near open [æ] (with or without rounding) seems to have triggered the substitution of a palatal rhotic. This observation does away with the bulk of Marstrander's examples. Another limitation of the evidence is the fact that the frequent nominative singular ending -r (> -r) is not reflected in the Irish material, as the ON oblique cases formed the basis for the Irish forms.Both the strongest piece of evidence and the key to the resolution of the conflict between Marstrander and Finnur Jónsson belong to a series of dithematic ON names in which PGmc *gaizaz is realized as either of the deuterothemes *‑geirr (< *gairar with the changes *ai > *æi and *rr > *rr after syncope) or *-arr (> *-arr) (weakened as part of compound).19 We thus have Þórarr (< *-ārr < *-gārr < *-gairar) and Þórgeirr, Hróarr (from *Hrōþwārr < *Hrōþu[ʒ]ārr, cf. OE Hrōðgār) and Hróðgeirr and so on.20 The internal reconstruction of the development of the deuterotheme *gaizaz towards ON (> *‑arr) is supported by Finnic evidence. The attested Gmc languages have a term for ‘helicoidal hand drill’ that goes back to a PGmc compound of *nabō ‘nave’ and *gaizaz ‘speer.’21 That compound was borrowed into Finnish as napakaira ‘auger,’ presumably directly from PrN *naᵬaʒair(a).22 The regular ON reflex of the compound is nafarr ‘auger,’ with reduction of the deuterotheme exactly parallel to that seen in Þórarr. An earl named Þórarr appears in an entry in the Annals of Ulster for the year 847 (848):23Bellum re nOlcobur, ri Muman, ⁊ re Lorggan m. Cellaig co Laighniu for gennti ecc Sciaith Nechtai[n] in quo ceciderunt Tomrair erell, tanise righ Laithlinne, ⁊ da cet dec imbi.24[Ólchobor, king of Munster, and Lorcán son of Cellach with the Leinstermen, won a battle against the genti at Sciath Nechtain, in which fell the jarl Tomrair, tanist of the king of Lochlainn, and twelve hundred about him.]25Þórarr (*Þṍrarr), one of the earliest attested ON names in the Irish annalistic record, is here rendered as Tomrair in the nominative singular.26 The most reasonable explanation for the palatal /rj/ at the end of that form is that it was substituted for ON palatal r, as Marstrander suggested.27 Prior to the publication of Bidrag, the name Tomrair was thought to reflect ON Þórir.28 This equation is ruled out on formal grounds but has nevertheless been rehearsed in more recent literature.29 If Marstrander's analysis is accepted, Irish Tomrair from ON *Þṍrarr is among the latest unambiguous examples of a retained (audible) distinction between /r/ and r in ON that can be dated with confidence. Marstrander's suggestion is compatible with internal ON runic and skaldic evidence of the merger, which it corroborates by furnishing an additional terminus post quem.30Marstrander established the upper chronological limit on two names that occur in later entries in the Irish Annals, namely Torgair and Suartgair. The first of these would give a terminus post quem in the 1170s, which is close to three hundred years later than we would expect from the internal ON evidence, the latter a terminus in the early eleventh century.31 Marstrander took these items as evidence for a retained distinction on the assumption that ON had a third variant of the deuterotheme, namely *‑gar‑, shortened from *‑gār‑.32 However, this deuterotheme is unattested in ON; Marstrander's comparison with Sigarr is flawed, because ‑g‑ [ɣ] here is part of the prototheme.33Sigarr occurs in several Eddic poems, among them Guðrúnarkviða II (16.7: Sigarr ok Siggeirr) and in skaldic poetry with reference to a legendary sea-king (in kennings).34Furthermore, the assumption of a variant **‑gar‑ is incompatible with the sound changes in early ON, evidenced by nafarr ‘auger’ and Þórarr. It is also unnecessary to account for the Irish forms on which it builds. Instead, the Irish forms likely reflect ON -geirr, the deuterotheme of a new compound based on the same two simplexes, after the semantics of Þórarr had become obscured, and the palatal /rj/ is triggered by the front vowel segments in the ON diphthong (/æi/), rather than the palatal quality of the ON consonant.35 Marstrander took his cue from a notion that Adolf Noreen advocated in his grammar of ON and elsewhere, namely that compounds such as *Þunra-gaizaz “schon in der Vikingerzeit schwankende Betonung hatten, so dass bald das erste, bald das letzte Glied haupttoning sein konnte” (had shifting accentuation already in the Viking Age, so that sometimes the first, sometimes the last member could carry the main stress).36 This assumption is unnecessary if we assume that compounds such as Þórarr and Þórgeirr were made at different stages. Instead of accepting the unlikely development of PGmc *Þunra-gaizaz to both ON Þórarr and Þórgeirr at the cost of the fixed initial stress that triggered the reduction of the deuterotheme in the first place, we can assume that Þórarr was the regular outcome and that a new name Þórgeirr, compounded of the regular reflexes of the simplices, was coined in early ON. The latter compound must postdate the shortening of *[a:] > [a] and the loss of *[ʒ] in the second element of compounds.37 This explanation accounts for other ON dithematic names of this type as well, such as Hróðgeirr.38Marstrander's other examples of Irish /rj/ from ON palatal r are not diagnostic and have no bearing on the question.39 By way of example, Irish beóir from ON bjór(r) is prima facie a good candidate. The problem of the lack of r‑mutation in the ON form, however, remains to be explained and suggests that the commonly accepted reconstruction *beuza cannot be correct.40 Even if an etymology with */z/ is accepted, this evidence does not carry much weight, as the word might have undergone paradigmatic leveling in Irish before its first attestation, analogous to the development of the -ā-stem túath ‘tribe’ in Middle Irish.41 Accordingly, the Irish evidence is fully in tune with the internal ON evidence, corroborating the notion of a retained distinction between ON /r/ and r in the first half of the ninth century. No evidence has yet been presented from the Gaelic languages which could be used to shift this date forward.