Hear No Sievers, See No Sievers: Metrics and the Eddic Commentary Tradition

This article explores a gradual shift towards metrical and linguistic imprecision in the four major commentaries on the Old Norse Eddic poems produced from the 1930s to today. This trend both affects and reflects the state of Eddic scholarship at large. Many scholars today do not avail themselves of metrical and linguistic criteria in the dating and textual restitution of Eddic poems but approach these issues either with agnosticism or by presenting vague arguments in favour of particular dates or variants. During the last three decades, however, testing along many parameters has placed the essentials of Eduard Sievers’ metrical analysis (1893) beyond reasonable doubt, which means that the gradual abandonment of precise description has taken place in a context where the preconditions for such description were improving. These analytical advances and mainstream Eddic scholarship have yet to meet, and the present article aims to facilitate such a convergence. Furthermore, since the Eddic commentaries will remain indispensable heuristic tools for the foreseeable future, the field of Old Norse philology stands to gain from a concise description of the implicit metrical and linguistic assumptions guiding their analysis.


Introduction
This article describes a quiet shift towards metrical and linguistic imprecision, as well as a devaluation of probabilistic reasoning, within Eddic scholarship. In order to limit the scope to representative samples, the article focuses on the four major commentaries on the Eddic poems produced within the last hundred years.
The so-called Elder or Poetic Edda has attracted more scholarly and popular attention than any other ON text or collection of texts. From 1927 to today, four major commentaries have been produced, culminating in the commentary by von See et al. (1997See et al. ( -2019below Kommentar), which runs to approximately 6.700 pages. One profound difference between the first commentary by Sijmons (1927-1931) and the other three, which were produced from the late 1960s onwards, is that Gering and Sijmons are metrically and linguistically precise, whereas this quality is increasingly absent in later commentaries. The Kommentar is most extreme in this regard, which is all the more noteworthy as it is several times longer than the others. Since the Kommentar is an essential reference work for all Eddic scholars, its lack of precision affects the field at large. I therefore direct my main attention at the Kommentar, before exploring the other three by reference to it.
The abandonment of precise metrical and linguistic analysis is probably due in part to a tendency noted by Gustaf Lindblad in 1980, namely that literary historians often dominate the discourse on literary texts but lack the metrical and linguistic skills necessary to evaluate the likelihood of many scenarios (1980: 146). In the first half of the twentieth century, there was less disciplinary fragmentation, and scholars working with medieval literary texts generally possessed a high level of linguistic skill.
The works under study have not signalled their shift towards metrical and linguistic imprecision, and this has hitherto largely escaped comment. Nonetheless, it deserves our serious attention. Metrics and linguistics provide us with frequent and precise criteria, many of which are not the result of the author's active choice, and they are therefore often our best tools for evaluating the poems' dates and earliest retrievable form, which in turn inform most literary and cultural analyses. In addition to the positive effect of enabling probabilistic evaluation, the use of good criteria has the related consequence of imposing limits on scholarly subjectivity. 1 Good criteria for dating and establishing the text are thus of paramount importance, and a fundamental tool for identifying such criteria is a plausible metrical analysis. I turn now to this topic.

The Validity of Sievers' Analysis
The twentieth century saw much debate on Germanic metrics, and especially on whether to accept, adapt or reject the analysis presented in Eduard Sievers' foundational Altgermanische Metrik (Sievers, 1893). From the end of the twentieth century onwards, it has become increasingly clear that Sievers' analysis must be essentially correct. Several observations concern the reality of resolution-the replacement of one long, stressed syllable by a short + one more syllable-which is key to Sievers' four-position system.
First, Sievers' system has been corroborated by the distribution of older linguistic forms, a distribution of which he was not aware and which can only be accounted for by resolution, as predicted by his system (Pascual, 2016a and. A noteworthy feature in this regard is Kaluza's law, which operates in Beowulf and relates 1 The decisive rejection of positivism was presented by Karl Popper. Excellent discussions of how probabilistic evaluation relates to medieval philology are Fulk (1992Fulk ( ): 7-24, (1996 and (2003); Neidorf (2015). to resolution. 2 In instances when a four-position analysis would require resolution, as in Beowulf 2357a frēawine folca, the ending was short on Proto-Germanic (< *-winiz). If resolution has to be suspended, by contrast, as in 2118a ġearo gyrnwraece, the ending was long (< *-wracôz 3 ; here ġearo is resolved). Since the etymological length of the ending is not predicated on the assumption of a four-position system, its correlation with the predictions of that system constitutes a highly effective test of its plausibility. Other noteworthy features that support Sievers' analysis are non-contraction (or hiatus in ON scholarly discourse), such as disyllabic scansion of seon, and non-parasiting, such as wundr for later wundor (Neidorf, 2017: 3). In both instances, the four-position system predicts that the word must contain one syllable less or more than normal, and this prediction conforms to the independent parameter of etymological reconstruction.
Second, verses that do not conform to Sievers' system can nearly always be shown to be due to scribal corruption in cases where we are able to compare more than one manuscript copy of the same poem (Neidorf, 2016 and2017: 126-30;Haukur Þorgeirsson, 2016 and. This parameter is not restricted to resolution but presupposes it among other aspects of the four-position system. Third, R. D. Fulk has shown that resolution was retained also in younger metres, both during the Middle English alliterative revival and in Icelandic late medieval rímur (Fulk, 2002). Since there were no external influences that might lead to the introduction of resolution, its late presence must presumably be due to retained older conventions.
The testing of Sievers' system has mainly been performed in OE, but I would add two parameters from ON. The first of these relates to hiatus versus contracted forms (séa > sjá, saeïng > saeng, etc.). The contraction of hiatus occurred around 1150. The highly regulated dróttkvaett record allows us to see that such forms are never found in later poetry, with the exception of analogical hiatus in endings (e.g., bláum > blám > bláum). The manuscript record begins only around 1150, and except for one passage in the First Grammatical Treatise (c. 1150), there are no indications of scribal recognition of hiatus (Hreinn Benediktsson ed. 1971: 224-26). The form in the First Grammatical Treatise is éarn or íarn for later járn, and it is clear from the author's description that knowledge of this form was dying out at the time of writing.
Even though they are not graphically represented, hiatus forms are often required for fornyrðislag verses to conform to Sievers' system, and the correlation between contracted forms and breaches to Sievers' analysis cannot plausibly be attributed to coincidence. Thus, for instance, in all instances where hiatus forms would be expected in Vǫluspá, Sievers' analysis would allow for their restitution, and in three instances this would be required for the metrical analysis to be plausible: í hǫll Hǫárs (21.5),vǫlu velspáa;(22.3),á Gimléi (64.4). Without hiatus, the first of these reads í hǫll Hárs, which could be scanned as a C-(x / /), but we here encounter the problem that the Vǫluspá poet does not otherwise use this type and is restrictive with catalectic verses generally (Suzuki, 2014: 835;Haukur Þorgeirsson, 2016: 137). 4 The second would read vǫlu velspá, which amounts to a D-(/ / \), conforming to the structure of the verse glaðr Eggþér (41.4). In this instance, the reading vǫlu velspáa is more plausible in light of the poet's restrictive use of catalectic verses, but vǫlu velspá would be possible. The verse á Gimléi would be extremely implausible without hiatus (á Gimlé). Suzuki, 2014 has provided metrical annotation for the entire corpus, and since he ignores hiatus, searches in the electronic version of his lists allows for a precise quantification of structures resulting from contraction. Suzuki takes á Gimlé as type A3 (x / \). A3 with a monosyllabic first dip is extremely rare, however. In fornyrðislag, metrically similar verses are found only in Hyndluljóð 26.3 ok Hjǫrdís, Guðrúnarkviða I ok vinspell and Reginsmál 17.2 á saetrjám. Only the last of these begins with a preposition, like á Gimlé, and is furthermore the only other potential off-verse example of the supposed A3 (or, in an off-verse, rather C-) pattern. In this unique, fully analogous verse, we can also restitute the hiatus form: á saetréum, resulting in a regular type C. 5 Like in the case of Kaluza's law, we see in these hiatus forms that the reconstruction of a historical phase of the language that was unknown to the scribes, and even to Snorri, explains the apparent inconsistencies in a poet who is otherwise conspicuous for his regular metre. 6 This cannot plausibly be explained by coincidence, and it should be noted that these observations are not dependent on Sievers' system, but only on the prosodic structures of the corpus. The crucial point is that Sievers' system allows us to predict that these structures range from mildly to extremely implausible and that this prediction correlates with independent predictions made on the basis of historical linguistics.
The parameter of hiatus does not relate directly to resolution, but rather to the overall plausibility of an analysis that presupposes resolution. With regard to resolution specifically, another set of data from the ON corpus is more important. It 5 While Suzuki analyses verses like the above (one monosyllabic word + a disyllabic compound) as A3, he analyses three monosyllabic words as C-. verses are more numerous, and they display an interesting tendency: in the Eddic fornyrðislag corpus, all verses that do not appear corrupt allow for the reconstruction of an older, disyllabic last word, producing a regular C, as in Hym 9.5 er minn frí > fríi. Thus Vkv 4.6, 20.6; HH II 1.5, 11.2; Grp 22.6, 28.2; Rm 5.7; Sd 16.3; Gðr I 13.4, 19.6; Sg 28.2. HH II 18.8, Hm 25.1, Grt 2.3 and Akv 28.6 are likely corrupt, as they either lack alliteration or give poor or implausible meaning. The only exceptions to the rule that trisyllabic C-allows for the reconstruction of regular C are found in Rþ 43.1 and 46.1 (enn/reið Konr ungr). 6 Snorri's partly failed attempt to replicate such forms in Háttatal is instructive. He correctly produces the forms þjóðár, dreyrfá and járngrá in cases where the metre demands the older forms in -áar, -fáa and -gráa, but reveals himself in the choice of the form vindhlés, in which there was no further syllable in the earlier stage of the language (Faulkes 2007: 7-though note that Faulkes on p. 49 wrongly claims that -hlés would have had two syllables before the loss of hiatus c. 1150). is clear that the highly regulated dróttkvaett metre evolved from fornyrðislag (e.g., Gade, 1995: 7-12, 226-38). In dróttkvaett, a short syllable may carry ictus only if the line is expanded by one syllable (for a more precise description, see Myrvoll 2016: 251-54). The structure of dróttkvaett is so clear that the only reasonable explanation for this correlation is resolution, and there are no other plausible sources for this feature than Germanic, alliterative metre. In order to survive the concerted attempt at standardisation that produced dróttkvaett, resolution must have been an inalienable feature of ON metre.
The reality of resolution is thus no longer open to reasonable doubt, and the distribution of stresses and quantity in the 'strict' Germanic alliterative corpus (notably Beowulf and ON fornyrðislag and dróttkvaett) is such that the four-position system follows as the logical consequence of resolution. 7 Resolution and the fourposition system are the two basic building-blocks of Sievers' analysis, which must thus largely correspond to historical reality. For about a century, however, Sievers' hypothesis awaited testing, and Fulk must take the credit for identifying the parameter of resolution as crucial to this undertaking (Pascual, 2016a). Fulk's concern was not exclusively metrical, however, but he was also attempting to set the dating of OE poetry on a probabilistic footing after a decade of devaluation of measurable parameters (Neidorf, 2014). As we shall see, metrics and dating have been intrinsically connected throughout the period after 1893.

Metrical Preconditions for Kommentar
In the Kommentar, the editors present no overview of their analytical framework, which they appear to treat as a matter of consensus. Since Klaus von See was the main editor, one may assume that this consensus is to be found in his 1967 Germanische Verskunst, to which the editors occasionally refer. Von See offers an overview of metrical scholarship, including Sievers. Von See does not, however, opt for any individual analysis, but appears to take his starting point in an undefined mixture of 7 Nelson Goering criticises this claim, made in various publications by Fulk, Neidorf and Pascual, stat-ing that "resolution is not dependent on or unique to the four-position theory of metre; it is also a core part of its chief theoretical competitor, the word-foot theory" (2019: 125). Goering is here referring to Geoffrey Russom's work, but Russom does not address the number of positions per verse (Russom 1987: 11-12, 44-46). More importantly, Russom's word-foot theory is largely based on Sievers, and it is thus not so much a competitor as an attempt to reformulate Sievers' system in a way that aims to be purely metrical (although not altogether successfully, cf. Griffith 1997). The crucial point is that resolution allows Sievers to see that the most common number of syllables per verse in Beowulf and fornyrðislag is, in fact, the number of positions found in nearly all verses (with exception made for anacrusis, D* and ON three-position verses). Russom apparently accepts this starting point, as he uses the term "position" in its Sieversian sense, and he would not need to posit resolution if he did not envision a restricted number of positions, just as Sievers did. In a later article, Goering notes that anacrusis and type D* exceed the four positions, and that the four-position system does not account for other, more flexible schemes (Goering 2020: 148-50). This is true, but it is of limited importance for the evaluation of the reality of only four positions in the instances when Sievers posited this. For the pragmatic purposes of the philologist, a system that explains much and allows for accurate predictions is superior to more "linguistic", universal models that offer less aid in the evaluation of emendations and variants. models. Most important in the present context is that he rejects Sievers' analysis of resolution, except in "syllable-counting" skaldic poetry and fornyrðislag influenced by that tradition (1967: 23). In the end, von See's own proposed system consists mainly of the demand for two ictic and an undefined number of unstressed syllables per verse (1967: 18-21).
Von See's observations on resolution and syllable-counting metres are worth quoting in full: Es ist aber fraglich, ob dieses aus der antiken Metrik übernommene Prinzip [Auflösung] der germanischen Verskunst angemessen ist (soweit es sich nicht um die silbenzählende skaldische Dichtung und die unter skaldischem Einfluß stehende eddische Dichtung handelt). (von See, 1967: 23) From the context it is clear that von See sees resolution as more than questionable. In the next sentence he writes that Sievers' scansion of some type C verses with resolution on the first lift, such as en konungr fjǫrvi, "completely hides" (völlig verdeckt) their actual, five-syllable structure. In other words, although this is never clearly stated, von See rejects Sievers' analysis in its entirety, except when applied to skaldic metres, which von See sees as syllable-counting. Von See has here produced an intellectual oxymoron: resolution is what makes Germanic metres position-counting rather than syllable-counting, and von See's restriction of resolution to syllable-counting metres therefore indicates a degree of conceptual confusion that is not easy to disentangle, but whose analytical consequences must in any event be rejected. The idea that skaldic metres are syllablecounting is relatively common (e.g., SkP 1: lxi; Kristján Árnason, 2007: 111-12;Gade, 1995 passim), even among metrists writing in the Sieversian tradition. In order for the term to make sense when applied to ON poetry, von See and others must presumably use "syllable-counting" to mean "isosyllabic" ('containing the same number of syllables in each verse'). Since dróttkvaett treats verses of six to nine syllables as metrically equivalent, however, it is clearly not isosyllabic. How, then, could von See and others arrive at the conclusion that skaldic metres are syllablecounting? I would suggest that this conceptual confusion is ultimately a product of Sievers' own terminological imprecision, which in turn was indebted to Snorri.
In his 1893 Altgermanische Metrik, Sievers generally refers to "positions" (glieder), but in his treatment of dróttkvaett, he abandons this practice in favour of the term "syllable" (silbe). He clarifies that in fact, resolution is relatively common in the first two positions in dróttkvaett, and he never uses the word "syllable-counting" (Sievers, 1893: § 61.2). Still, it may be fair to say that Sievers invited a misunderstanding of his analysis when he stated that "Jedes versglied is der regel nach einsilbig". He goes on to say, however: "Wo im folgenden von einsilbigen formen u.dgl. die rede ist, sind die entsprechenden auflösungen stets mit inbegriffen." (1893: § 61.2). In other words, he changes his terminology to reflect the fact that dróttkvaett exhibits a more regular syllable-count than fornyrðislag, but this is a difference of degree, not of essence. The principles governing the syllable-count remain the same.
In Sievers' articles laying the foundation for his 1893 book, the situation is different. In 1879, he stated that "Auch die sogenannte volkstümliche dichtung der Eddalieder beruht zum grössten teile auf dem princip der silbenzählung" (Sievers, 1879: 298). In 1885, he shifts within a few pages from calling fornyrðislag a viersilbler metre (Sievers, 1885: 218) to discussing Beowulf as gliedrig (1885: 220), in spite of the fact that he strongly emphasises the unity of the two forms. 8 It would thus appear that he had little interest in the terminology and that the silbe/glied distinction reflects little more than various stages in his analytical process. By all appearances, Sievers' 1893 book contains some remnants of this intellectual stratigraphy, since he there restricts the concept of "syllable" to dróttkvaett. Presumably, he saw the syllable as the basic constituent, but one which could be expanded by means of resolution and dip expansion. In the first half of the twentieth century, many scholars would have been familiar with Sievers' articles, and this is likely to have invited a degree of terminological fluidity.
Centuries before Sievers, Snorri composed his Háttatal (c. 1220). Snorri there claims that each dróttkvaett verse consists of six syllables (hverju vísuorði fylgja sex samstǫfur), even though he clearly knew that this was not the case (Faulkes (ed.) 2007: 4). This becomes evident when he presents the first in a series of metrical licences; namely, that a verse may have either fewer or more than six syllables. In his stanza containing "quick syllables" (skjótar samstǫfur), each verse has either nine or seven syllables (Faulkes (ed.) 2007: 7-8).
It seems likely that Snorri's internal contradiction is premised on the grammatical discourse known to him. Models of linguistic and metrical description available in Iceland were prescriptive, presenting rules, errors and licences. They recognised the syllable (and possibly the foot), but they had no room for such a flexible concept as "position". It can in several instances be shown that when grammatical convention was at odds with skaldic practice, Snorri sided with grammatical convention, even though his description became inadequate or confusing as a result. 9 This is probably the case also in his account of the principles governing the syllable count.
Snorri's stanzas and commentary show that he must have been at least intuitively aware of the concept of position, but grammatical convention forced him to account for the syllable count by means of one rule and one licence. In actuality, resolution is governed by rules and is thus not a licence. This is seen, for instance, from the fact that the first ictic position must consist of either a long syllable or one short + one more. 10 By composing the only medieval description of early Germanic metre, Snorri bequeathed his syllable-counting framework to all subsequent generations of Germanic metrists. Sievers was not confused by Snorri's description, but it may have influenced his terminology, leading to confusion among scholars who read Sievers and Snorri alongside each other.
After Sievers, the largely Sieversian 11 metrist Hans Kuhn initially used the term "syllable-counting" loosely, apparently inspired by Sievers' (and perhaps Snorri's) description (e.g., Kuhn, 1929: 66). In his later scholarship, Kuhn's terminology is more precise. Kuhn's terminological imprecision did not affect his own analysis, but it may have contributed to the confusion of later scholars (see the discussion of Gering and Sijmons 1927-1931 below for a comparable example).
Von See represents a second stage in the reception of Sievers' and Snorri's "syllables". The terms "syllable" and "syllable-counting" are now no longer simply shorthand for a more regular syllable-count, but rather, they are used to describe an undefined but somehow essential difference between fornyrðislag and dróttkvaett, which apparently implies that resolution can be found only in the latter. Only at this second stage is the term "syllable-counting" diagnostic of analytical confusion.
In reality, the more regular syllable-count of dróttkvaett is the product of three fundamental innovations. The first of these was the addition of a set cadence of two syllables to the fornyrðislag verse. The second was that the principles guiding the syllable count in stressed positions were extended to cover all positions in which one syllable may be supplanted by more than one: in dróttkvaett, dips, like lifts, can at most contain two syllables, and when they do, the first syllable must be short (Myrvoll 2016: 252-53). 12 A third adaptation was that only the standard, four-position realisation of the fornyrðislag verse was used as a starting point for the dróttkvaett verse, whereas five-or three-position verses were disallowed.
These innovations make the syllable count of dróttkvaett more regular than that of fornyrðislag, but the basic governing principles remain the same. Von See's claim that resolution was an aspect of skaldic poetry, but not of Eddic, is thus untenable. His own analysis of fornyrðislag, based on two stressed and an undefined number of unstressed syllables, would allow for the generation of a vast array of unmetrical structures and would in many instances suggest that prose passages are, in fact, 11 Early on, Kuhn was critical of the fact that Sievers used OE epic and fornyrðislag as the basis for his system, with the result that more flexible forms appear deviant (1934: 21). It is possible that Kuhn later changed this view; in any event, he adhered to a basically Sieversian analysis throughout his career. The validity of Kuhn's critique is doubtful, since Sievers had identified the most salient regularities by means of comparative study. It is therefore likely that these regularities are diagnostic of a shared background, whereas irregularities found in only one tradition have a weaker claim to antiquity. In addition, some of the texts that adhere most closely to Sievers' basic system, such as Beowulf (c. 700) and the Rök stanza (c. 800), belong to the earliest datable items within their respective traditions. (For an overview of recent advances in the dating of Beowulf, see Neidorf 2017: 2-9. The Rök stanza has the structure C-, C, A, D*, A3-(less likely A3), C, A, D. Bo Ralph's suggestion that the first verse contains only two positions is incompatible with the data of the fornyrðislag corpus; Ralph 2007: 150-51;Suzuki 2014: 161-62, 642.) Of course, some scholars prefer synchronic description to analyses that take diachronic developments into account, but since Kuhn did not belong to these, Sievers' diachronic starting point cannot explain Kuhn's scepticism. On the topic of synchronic versus diachronic analysis, it may be noted that metrics are crucial for dating, and for that purpose, a diachronic perspective on metrics is necessary. 12 A further innovation in relation to fornyrðislag is that a dip containing more than one syllable cannot involve the ending of the preceding word. It can contain one disyllabic word with a short first syllable or two fully unstressed monosyllables, e.g. vita, búa, verǫld, spyr ek, þar es (Kuhn 1983: 67). poetic. For editorial purposes, it offers little aid in the scholar's critical evaluation of the text. Von See's metrical analysis of fornyrðislag is thus neither apt nor particularly useful. We turn now to the Kommentar itself.

Metrics, Linguistics and Late Bias in the Kommentar
In the Kommentar, § 7 of the introduction to each poem treats "Strophen-und Versform". As one might expect, the editors recognise the canonical metres: fornyrðislag, málaháttr and ljóðaháttr. They also note if alliteration falls on the second lift of an even verse (e.g., Kommentar 1: 475; 2: 269), as well as the principle that the first noun of a verse should carry alliteration (e.g., Kommentar 1: 74, 475-75). Although pronouns may carry ictus in all Germanic traditions-and indeed must carry ictus if they are not found in the first dip of a clause-the editors list instances of pronouns carrying alliteration (e.g., Kommentar 1: 475; 2: 269). Apparently, the editors do not take note of Kuhn's study of sentence particles (Kuhn, 1933).
Most other comments are of a stylistic rather than a metrical nature, such as if the same stave is retained for more than a couplet or if there are suggestions of internal rhyme. In short, there is little metrical information in the Kommentar, and it is almost never used for dating poems or solving problems. Most notable in this regard is the fact that the editors list verses where alliteration is missing, which is useful for textual criticism and historical linguistics, but as we shall see, they do not avail themselves of their data for these purposes.
The editors' metrical imprecision makes it impossible for them to identify hiatus forms, and the discussion of Vǫluspá is a case in point. In the commentary on this poem, two of the three metrically secured hiatus forms are not even mentioned as such Vǫluspá 21.5),22.3),64.4). Only in the case of Hǫárs do the editors concede that earlier scholarship has reckoned with a disyllabic form "aus metrischen Gründen"-reasons that they do not explain. Instead, they simply state that the genitive Hárs "ist sicherlich zu einer Grundform Hárr zu stellen, die mit dem Adj. hárr ('grau[haarig]') gleichzusetzen ist". In fact, the form Hǫárs in Vǫluspǫ́ is necessary to get the expected number of metrical positions (four instead of three): ok í hǫll Hǫárs. The same is true in some skaldic poems, such as Einarr skálaglamm's Vellekla 10.8 lífkǫld Hǫárs drífu, and Hofgarða-Refr's poem about Gizurr gullbrárskáld 1.2 gall bál Hǫárs stála. In neither case would the form Hárs give the required six metrical positions of dróttkvaett (these verses are printed with the reconstructed form Hǫárs in the main editions; Skj. B 1: 118, 295; SkP 1: 295; 3: 254). 13 The editors of the Kommentar claim that etymological considerations about the origin of the heiti Hǫárr are "irrelevant" (it is probably a compound, perhaps Proto-Nordic *haiha-harjar 'one-eyed warrior'). 14 They also argue that their monosyllabic reading Hárs in Vǫluspá may be found in places in poetry where a disyllabic form is not needed, but no one has questioned the existence of the form Hárr. The two examples given in the Kommentar, Nóregs konungatal 17.4 (late twelfth century) and Ingjaldr Geirmundarson's Atlǫguflokkr 2.6 (c. 1244), are both metrically ambiguous, and both poems were composed after the contraction of hiatus forms and are thus irrelevant to the question (SkP 3: 773; Skj. A 2: 88; B 2: 99).
The editors' decision to quote only non-diagnostic examples and bypass diagnostic ones, like the ones found in Einarr skálaglamm and Hofgarða-Refr, must probably be intentional, since the heading in Finnur Jónsson, 1931 will have led them to all occurrences. The evidence for hiatus in Hǫárs is found in dróttkvaett stanzas, which von See sees as syllable-counting, and his flexible metrical analysis of fornyrðislag therefore cannot explain why the editors have excluded diagnostic examples only. Furthermore, there is no ambiguity in Vǫluspá: the disyllabic form is expected according to the structure of fornyrðislag generally, but also according to the metrical patterns that may be observed in Vǫluspá itself (see above and Haukur Þorgeirsson, 2016: 130).
As this example illustrates, metrical and linguistic imprecision are intrinsically linked, and this has chronological consequences. The fact that contraction is never metrically required, whereas hiatus is metrically secured in three instances, strongly suggests that Vǫluspá was composed before c. 1150. By contrast, the editors present a vague date before c. 1225 (Kommentar 1: 73). This they do only with reference to Snorri's Edda (c. 1225) and textual influences, most of them postulated on vague grounds.
While the example of hiatus illustrates the methodological problems with the Kommentar's disregard for metrical detail, its chronological consequences are somewhat limited, since a date around 1150 would still make Vǫluspá a relatively young Eddic poem. Alliteration in vr-as opposed to later r-, by contrast, can be dated to before c. 1000 in the skaldic record, where the dates are reliable and where three staves per couplet is an absolute rule throughout the tradition. This criterion has withstood thorough testing and may now be considered one of the most reliable dating criteria (Haukur Þorgeirsson, 2017). The editors' treatment of the occurrence in Hávamál 32.3 en at virði vrekask is revealing.
The editors note that "Formen dieses Verbs mit v-Anlaut sind zwar im Altwestnordischen nicht bezeugt" (Kommentar 1: 574). "Nicht bezeugt" must here mean "not attested in writing", since the form is metrically secured here and in Vafþrúðnismál 53.3 þess mun Víðarr vreka. As one might expect, the scribes were unable to produce this or other vr-forms, which had been defunct for over 250 years at the time when the relevant MSS were written (Haukur Þorgeirsson, 2017: 49-53). In this instance, the editors do not accept even the handful of metrical restrictions described by von See, 1967 as sufficient grounds for attestation, since von See sees internal alliteration as the defining feature of the ljóðaháttr full verse (i.e., the third and sixth verse of the ljóðaháttr stanza; 1967: 52). Furthermore, they do not mention dróttkvaett occurrences of vr-, which are secured by absolute rules, as described by von See and all other introductions to the metre (1967: 42). The Kommentar thus represents an even more liberal perception of ON metre than von See, 1967, apparently in order to accommodate the MS readings.
Furthermore, the editors do not evaluate the likelihood of a missing stave based on the evidence of the poem, even though they collect and discuss this evidence. In fact, only one full verse (Hávamál 80.3) is missing internal alliteration, and in that instance, the poet compensates with alliteration to preceding verses. This irregularity appears to be motivated by the traditional collocation reginkunnar rúnar ('runes derived from the gods'), found also on the Noleby stone (sixth century) and the Sparlösa stone (c. 800; Kommentar 1: 696). Hávamál 80.1-3 read: Þat er þá reynt. er þú at rúnum spyrr. inum reginkunnum.
Apparently, alliteration in full verses is an absolute rule to this poet, and even when he breaks the normal pattern in order to accommodate a traditional phrase, he retains alliteration according to the normal fornyrðislag pattern. In other than full verses, a stave is missing only in the first couplet of stanza 36, and the same couplet is then repeated in the following stanza: bú er betra | þótt lítit sé ('[your own] dwelling is better [than no dwelling], even if it is small'). The content suggests that the couplet may be proverbial and that both stanzas serve to elaborate on the proverb, which is certainly not the case with the verse where vr-would be expected: at virði vrekask ('[…] to argue over a meal'). Whatever the reason for the missing stave in stanza 36, this flaw is clearly exceptional in all kinds of verses in Hávamál. 15 The situation in Vafþrúðnismál, where the other likely vr-form of this verb is found, is comparable. No full verses are missing internal alliteration, unless one reads reka rather than vreka. In other verses, the editors count three missing staves, but one of these is due to an inversion of the principles of textual criticism (Kommentar 1: 989). Stanza 32.4-5 reads hvé sá bǫrn gat | inn aldni (R)/baldni (A) jǫtunn ('how that old/defiant giant conceived children'). The editors note that the word baldinn is otherwise found only once in the poetic record, whereas aldinn jǫtunn is found in four other instances in Eddic poetry, and they therefore opt for the latter reading (Kommentar 1: 1083-84). They do not seem to realise that they are here describing the reason why a scribe would change baldni to aldni. Obliviousness to such a basic principle as lectio difficilior potior is disconcerting, and when combined with alliteration, the matter should have been decisively settled in favour of baldni.
The two remaining instances of a missing stave are 5.4-5 at hǫllu hann kom | ok átti Íms faðir and 38.4-5 hvaðan Njǫrðr um kom | með ása sonum. Neither in Hávamál nor in Vafþrúðnismál is alliteration lacking in any full verse, not even in the abnormal inum reginkunnum. The assumption that the verses containing the verb reka are the result of flawed composition thus demands the improbability that both poets would disregard the rule of alliteration in full verses only when using this verb, and that the fact that reka clearly had vr-before c. 1000 (Swedish vräka, English wreck) is coincidental. On closer inspection, then, the editors' assumption of a missing stave turns out to be implausible, and this is clear from the collection of data that they themselves present.
The editors also mention the possibility of archaization with vr-, which would produce verses that conform to von See's and others' metrical analysis, but they are not open to the possibility that these or any other verses demanding alliteration in vr-are old (Kommentar 1: 575, 1171). Their rejection of the possibility that such forms are old is corroborated by their discussion of the age of the poems, where this criterion is not mentioned (Kommentar 1: 496,999). This leads us to a paradox: How could the poets have archaised if the metrical system was as flexible as the editors assume? Such a system could presumably only reflect synchronic phonology. And if we are indeed dealing with archaisation, how is it that neither the R nor the A scribe shows any sign of understanding it?
The editors do not address these problems, and their approach must therefore be reconstructed from their practice. The three possible explanations of the verses containing (v)reka(sk) are that they are the result of flawed, archaising or early composition. Only the last of these explains all observable data, and yet the editors opt for the other two. Such a choice cannot be based primarily on an evaluation of probabilities, and some bias must therefore have overruled this parameter. Since the Kommentar dates most poems only with reference to Snorri c. 1225 or Codex Regius c. 1270, depending on whether they are attested in Snorri's Edda or not, it would appear that written transmission is the decisive factor for dating, to the exclusion of other observations. This emphasis on written transmission speaks in favour of the scribal reka(sk).
The editors' preference for scribal evidence is apparently strong enough to overturn the basic metrical analysis, even according to von See's liberal interpretation and the data on alliteration in Hávamál and Vafþrúðnismál that the editors present. This observation cannot explain, however, why they also find the form vrekask plausible, provided that it is young. The common denominator to both assumptions is dating, and more precisely that of excluding interpretations that might suggest that the poem is old, whether one adheres to the manuscript reading or not. We saw a similar tendency in the editors' selection of occurrences for the evaluation of hiatus, where diagnostic forms were left out. With regard to (v)reka(sk), the editors do not address the improbabilities involved in their two prosed interpretations: in the case of reka(sk), that both poets would break their own rule only when using this verb, and in the case of a young vreka (sk), that an old form should in fact be young and that both scribes should nonetheless fail to comprehend it.
The editors thus depart from probabilistic method in two ways. First, they take manuscript evidence as their sole guide to authorial readings, to the exclusion of metrical evidence (reka(sk); hiatus). In so doing, they take the self-evident surface level as diagnostic of that which is not self-evident; namely, the poet's text and the date of the poems. Scientific method, by contrast, aims to explain that which is not self-evident by recourse to as many tests as needed until the desired degree of probability is achieved. 16 In the interpretation, editing and dating of Eddic poetry, this generally translates to as many tests as possible, since relevant criteria are few. The editors' restriction to surface phenomena creates problems not only for reconstructing a plausible poetic practice, where the poets adhere to observable rules, as well as for dating, but also for the overall metrical analysis. Scribal mistakes and incomprehension are so common that no metrical system, not even that of the Kommentar, is easily reconciled with a mere surface description. The Kommentar's treatment of vr-reflects this tension. 17 Second, the editors allow their bias for late dates to interfere with their evaluation. Thus, when discussing a form that is demonstrably old, since in skaldic poetry it only occurs before c. 1000, they assume that it is young, and they present only ambiguous examples of hiatus, even though many unambiguous forms are attested. These two departures from scientific method are interrelated, since a preference for surface phenomena is also a preference for phenomena from the time of the scribes.
The editors' renunciation of scientific method and bias for late dating is also found at many other points in the Kommentar. Verifiable linguistic and metrical evidence is generally disregarded, and the value of absolutely datable skaldic sources is downplayed or ignored in favour of the-at best-only relatively datable Eddic sources. With regard to Hávamál, the editors seem to lean towards a date after 1150 because of its supposed reliance on the Disticha Catonis, even though it is a wellknown fact that gnomic literature world-wide has many features in common, and similarities therefore cannot be taken as diagnostic unless they are strongly atypical, which the ones in question are not. 18 Similarly, the editors date the poem Lokasenna to the twelfth century because it, like some Classical texts, features drinking and verbal exchange, as well as one uninvited and one late-coming guest 384). They do not consider the possibility that such features may occur spontaneously in descriptions of feasts.
In contrast to the strained attempt to find parallels in Latin literature, the unique correspondence between Hávamál 29 "tunga […] opt sér ógott um gelr" (lit. 'the [glib] tongue often chants un-good for itself') and Lokasenna 31 "tunga […] ógott um gala" is taken as plausibly coincidental . The substantivised ógott is found only in these two places, and similarly, only in these two instances is gala 'chant' used of ordinary speech (Finnur Jónsson, 1931: s.vv.). The alliterative scheme shows that we are not dealing with a formula that might aid composition: Hávamál's opt: ógott is expected, whereas Lokasenna's ógott: gala is a relatively rare and marked choice. 19 Direct influence is therefore likely, and metrical precision would have allowed the editors to draw this conclusion. In addition, it is untenable to give diagnostic precedence to vague, near-universal similarities to culturally distant texts over specific and unique similarities to texts in the same tradition.
As in the case of Hávamál, the linguistic criterion of alliteration in vr-for later ris disregarded in the dating of Lokasenna (Kommentar 2: 384). Lokasenna 15.4-5 requires alliteration in vr-according to the metrical analysis of von See and all others (Kommentar 2: 411;Haukur Þorgeirsson, 2016: 35). Lokasenna 18.6 and 27.6 also most likely require alliteration in vr- (Haukur Þorgeirsson, 2016: 36). In these instances, the Kommentar fluctuates between agnosticism (416, 440-441) and dismissal (376: "in 18 6 and 27 6 ergäbe die Annahme solcher Formen einen überzähligen Stab"). The dismissal presupposes that all initial consonants must belong to the alliterative scheme, irrespective of prosodic stress and ictus, and it is thus based on an imprecise and implausible analysis (cf. Haukur Þorgeirsson, 2016: 36). This problem notwithstanding, the fact that von See's analysis requires alliteration in vr-in stanza 15.4-5 and that this feature is not mentioned in the discussion of the date of the poem means that the most unambiguous and precise dating criterion in the poem has been excluded. Instead, the poem is dated with reference to vague and distant analogues, as well as to the fact that Snorri quoted it. While the second observation is relevant, it is also self-evident and in no need of evaluation. In sum, apart from observations of self-evident fact, the Kommentar's evaluation of the date and intertextual connections of Hávamál and Lokasenna is based on a full inversion of normal scientific method, rejecting specific criteria in favour of vague, non-falsifiable ones and thereby eliminating probabilistic restrictions on scholarly subjectivity. By means of this approach, the Kommentar dates both poems as late as MS transmission and the chronology of prose works allow.
In cases where verbal correspondences with skaldic poetry are so close as to strongly suggest quotation, the skaldic poems are often seen as doners, even though it is in skaldic poetry that we otherwise find intentional quotations. 20 The direction of borrowing is left open even in such a clear case as the shared couplet of Vǫluspá 57.1-2 and Þorfinnsdrápa 24.1-2, where both metrical criteria and Þorfinnsdrápa's concomitant, unambiguous quotation from another older poem indicate that Vǫluspá is the giver and that at least Vǫluspá 57 was therefore most likely composed before the 1060s. 21 The possibility of oral transmission of Hávamál is ruled out as a matter of principle, because of its length, rather than being investigated (Kommentar 1: 496: "Die Länge der Háv. schließt sicherlich aus, daß dieses Lied mündlich tradiert gewesen ist"). The fact that this poem is short in comparison to the poems of some oral traditions is not mentioned. A twelfth-century or later date is suggested for Vafþrúðnismál, because a number of words in it are otherwise found only in texts from that period (Kommentar 1: 999). On closer inspection, the reader realises that the editors are referring to one occurrence each of the construction hverfa + accusative and the adjective baldinn ('defiant') (Kommentar 1: 991-92, 1062(Kommentar 1: 991-92, , 1083. There is no clear post quem in the language for hverfa + accusative or baldinn, and there is thus no reason to connect these occurrences to a date. Furthermore, as we have seen, the editors see the reading baldni as the result of scribal corruption, and it ought therefore to have been rejected for the purposes of dating. Other occurrences to which the editors refer are found in prose: a corpus that did not begin to take shape until the twelfth century. By this method, all poetry and all runic inscriptions containing words that are otherwise only found in prose must be dated to the twelfth century or later. This is not a viable assumption, since the reason why some words are attested only once in poetry or in an inscription and more often in manuscript prose is generally not the date of the poem or the inscription, but the vastly greater size of the prose corpus.
As seen from these observations, the editors display a strong bias for late dates. I will not speculate on their reasons, and scholarly bias need not be a problem, provided that it is regulated by means of probabilistic evaluation. The key point is rather that the inversion of normal scientific principles serves to enhance, rather than to check, the editors' pre-existent bias. This raises the question of how the editors could view this extreme renunciation of the scientific project as reasonable. Must there not have been some previous scholarly developments that made such an approach appear more justifiable than one might otherwise have expected? After all, if a team of Classicists decided to write a commentary on Homer-which would have to run to around 50.000 pages in order to correspond to the scope of the Kommentar-and abandoned all diachronic analysis, both with regard to textual criticism and language history, as well as all metrical restrictions except, perhaps, that each verse must contain five-six long syllables, one of which is the penultimate, this would appear bizarre in the extreme. 22 For any number of reasons, this is unlikely ever to happen, but many of the same reasons would a priori exclude the likelihood that the Eddic poems should be the object of such a commentary. This raises the question of whether there may have been some preconditions in the scholarly tradition that served to undermine the viability of probabilistic evaluation in relation to metrics, linguistics and dating in the eyes of the editors. In the following presentation of some possible preconditions, I allow myself a degree of speculation, since the correctness or otherwise of the analysis does not alter the fact that a non-probabilistic approach calls the entire rationale of philological investigation into question.

Possible Preconditions
As noted above, Sievers' lax use of the term "syllable" may have contributed to von See's confusion, and possibly that of others. The result of this confusion was an illusion of a poetic tradition that featured two sets of guiding principles whose mutually distinctive characteristics are not susceptible to rational description. Such a model does not command credence, and Sievers' analysis of ON metrics may therefore have appeared less reliable than it is. In addition, as we have seen, the overall correctness of Sievers' analysis has only recently been placed beyond reasonable doubt. Before that, a degree of scepticism was rational.
Another factor, too, is attributable to Sievers himself and to the transmission of Sieversian lore. As an old man, Sievers promoted a universal metrical theory of euphony -Schallanalyse -that has eluded scholars and deservedly gained little to no acceptance. 23 It was in this period that Sigurður Nordal, one of the most prominent ON philologists of his generation, tried to absorb Sievers' insights and famously stated the following: Ég kynntist Sievers og vinnubrögðum hans talsvert í Leipzig haustið 1921. Þótti mér mikið til mannsins koma, en um niðurstöður hans virtist mér naerri því jafntorvelt að daema fyrir aðra menn og um tíðindi þau, er ófreskir menn segja frá öðrum heimi. (Sigurður Nordal (ed.) 1933: ix) I was thoroughly acquainted with Sievers and his methods in Leipzig in the autumn of 1921. I thought highly of the man, but I found it nearly as difficult to convey his conclusions to others as I would the tidings that visionaries bring from another world.
Although Sigurður did not say so, he was referring to Sievers' Schallanalyse, not to his four-position system of an earlier period. Sigurður's statement is found in the introduction to the first edition to come out in the series Íslenzk fornrit (Egils saga), which would have a tremendous impact on the ON field at large. The statement would come to be read by many ON scholars, and unless they chose to investigate the matter, they would only know that one of the most respected scholars in the field thought of Sievers as an impressive madman. This was for a time true of myself, as well as of Haukur Þorgeirsson, who has previously commented on the unfortunate impact of Sigurður's statement, and probably of many scholars before us (Haukur Þorgeirsson, 2016: 118). This is a problem specific to ON philology, where the introductions in Íslenzk fornrit are a staple, which is not the case in other Germanic philologies.
In addition to these preconditions relating to Sievers and his work, the posthumous Fidjestøl, 1999 has been important to many ON scholars' perception of our ability to date the Eddic poems. Perhaps in reaction to the confirmation bias of some earlier Eddic scholars, Fidjestøl consistently interpreted his results in a negative light, noting that the various criteria may not be trustworthy. In fact, some of Fidjestøl's results were promising but needed further testing to exclude coincidence or archaisation as potential explanations of the distribution of criteria (testing and corrections are performed in Haukur Þorgeirsson, 2012 andMales, 2020: 39-94). Fidjestøl's doubts appear to have affected Eddic scholarship more than his results. Von See, 1967show that Fidjestøl, 1999 appeared too late to account for the methods and biases of the Kommentar, but it may have served to confirm the editors' belief in their approach.
While these analyses must remain speculative, I think that it may be useful to point to specific factors for why ON scholars may have been hesitant to adopt Sievers' analysis. Their hesitation has resulted in the proliferation of a view of Eddic metrics that is half wrong. Many scholarly publications present Eddic metre as extremely flexible and simple, whereas in reality, it is extremely flexible and complex. The Kommentar's simplification of Eddic metre cannot explain the editors' bias for late dates, but it eliminates many potential restrictions on this bias. We now turn to the three Eddic commentaries that were produced before or alongside the Kommentar. In order to evaluate their methods as compared to the Kommentar, I use the same examples of hiatus and vr-as discussed above.

Other Eddic Commentaries
The first major commentary on the Eddic poems was Sijmons 1927-1931. These scholars clearly accept resolution and the four-position analysis, and Gering employed Sievers' system throughout his career (e.g., Gering, 1902Gering, , 1924Gering, and 1926 With regard to the verse vǫlu velspáa (MSS velspa) the editors write that the -a must be supplied in order to achieve a Viersilber Sijmons 1927-1931, 1: 28). Their use of Silbe (syllable) rather than Glied (position) harks back to Sievers' pre-1893 usage and is suggestive of a discourse that may later have contributed to the confusion of scholars such as von See, but in spite of this terminological imprecision, the underlying analysis is apt: on linguistic grounds, the editors reject attempts to create four positions by adding a -v-in vǫlvu, which goes to show that they see vǫlu as resolved. This is of course also evident from the fact that vǫlu velspáa consists of five, not four, syllables. We here see that metrical and linguistic precision go hand in hand, just as they are both absent in the Kommentar. In the commentary on the verse mistilteinn (type A-), the editors explicitly refer to Sievers, and it is clear that their analysis overall is based on his system Sijmons 1927-1931, 1: 44).
In Dronke 1969-2013 metrical analysis has been abandoned, but it is not clear what has replaced it. For instance, Dronke refers to the type D-verse Geyr Garmr mjǫk as an "error", even though D-occurs 30 times in the corpus and other catalectic types are found in up to 388 verses (Dronke 1997: 65;Suzuki, 2014: 75, 79, 82, 105, 121;Suzuki's numbers for some catalectic types are inflated, but their existence is not in doubt). Dronke's statement can therefore probably only be explained by ignorance.
Like the Kommentar, Dronke comments only on Hǫárs of the three metrically secured hiatus forms in Vǫluspá. Unlike the Kommentar, Dronke states that "a disyllabic form would be expected" (130) and indeed "required" (131), but even though she introduces other, more drastic emendations into the text, she retains the scribal Hárs. Dronke does not resolve this contradiction. Overall, Dronke's lack of precision is comparable to that of the Kommentar, but she appears to harbour fewer consistent biases.
The least methodologically transparent commentary is Jónas Kristjánsson and Vesteinn Ólason 2014. This work differs from the others in that it is aimed at an Icelandic educated readership, and the editors can therefore presuppose a high degree of linguistic competence in their readers. In this regard, Jónas Kristjánsson and Vesteinn Ólason 2014 is more similar to Sijmons 1927-1931 than to the other commentaries, since Gering and Sijmons could expect their readers to possess a degree of philological expertise that is today somewhat rare outside Classics. Jónas's and Vésteinn's philological sense is more intuitive and less probabilistic than that of Gering and Sijmons, however. Hiatus forms in Vǫluspá are neither mentioned nor restored (2014, 1: 128-31, 296, 307), and although the editors restore vr-in Hávamál 32, this does not affect their twelfth-century dating of the relevant part of the poem (2014, 1: 166-67, 328).
Jónas and Vésteinn seem to take acceptability to the Icelandic educated ear as their main point of departure. This is an invaluable and often underestimated asset, but it does not relate to probability in a principled way. In their capacity as native speakers of Icelandic as well as philologists, the editors are probabilistically inclined to correctness in their assumptions, but this is a matter of competence, not of method. By the same token, they are inclined to favour continuities and ignore discontinuities, which is an obvious problem in the study of early texts.

Conclusions
For reasons that are not entirely clear, but at least some of which relate to scholarly reception and transmission of Sievers' system, the Eddic commentary tradition has gradually dispensed with metrical and linguistic precision, and obliviousness to the scientific principle of falsifiability is today acceptable within Eddic scholarship. This epistemological shift has gone largely unnoticed, and drawing attention to it would itself be sufficient motivation for the present article: the Kommentar is in need of a commentary in order for scholars to use it with benefit. For the future progress of the field, however, more than this is required. While this article has focused on a negative development, it has also discussed the considerable advances that have taken place in recent metrical scholarship. The work of scholars such as R. D. Fulk, Rafael Pascual, Leonard Neidorf, Haukur Þorgeirsson and Klaus Johan Myrvoll now facilitates an unprecedented degree of precision and evaluation. Scholars are thus not relegated to the sad prospect of reverting to the stage of Eddic scholarship in the 1930s but may proceed to producing more and more reliable knowledge.
Non-probabilistic approaches not only in ON, but also in OE, have been instrumental in provoking a probabilistic reaction. 24 This was needed, since even though scholars in the first half of the twentieth century often had an overall probabilistic approach-for instance, they would normally treat early linguistic forms as indications of early composition-few displayed epistemological rigour. Recent developments have forced the hand of probabilistic scholars, several of whom now present and evaluate their arguments with unprecedented levels of precision and testing. Gradually, the evolution described in this article is becoming an epistemic success story. Especially in ON, this development is still in its infancy, and it promises to yield advances that will leave the scholarship of the 1930s far behind.
Funding Open access funding provided by University of Oslo (incl Oslo University Hospital).

Conflict of interests
The author has no competing interest in relation to the present work.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/.