Textual criticism and Old Norse philology

ABSTRACT

Old Norse editors typically apply the principles of textual criticism with moderation, mainly correcting obvious errors. The exception is poetry, where the introduction of variant readings and emendations is often necessary. This two-pronged and overall restrictive approach is based on the characteristics of the corpus, but in recent decades, some scholars have called its premises into question, either in theory or in practice. In most examples discussed in this article, arguments against textual criticism are paired with disregard for or misrepresentation of its aims and methods. In other instances, scholars who attempt a reconstructive approach neglect the epistemological preconditions of textual criticism, even when drawing on its assumptions. In order to address these issues, the present article discusses the goals and principles of textual criticism with reference to examples from Old Norse scholarship. Specifically, it explores how scholars relate to scribal errors.

KEYWORDS:

• Textual criticism
• stemmatics
• Old Norse
• Eddic poetry

‘ja, ja, die andacht vor dem schreiber ist heute schon religion!’ (Schröder 1926: 95)

The present article focuses on how Old Norse (ON) philologists of the last three decades approach scribal errors, as well as whether and in what way they employ the basic tools of textual criticism in their evaluation. It focuses on misleading descriptions and implausible claims. This emphasis is not intended to convey the impression that all current ON philological scholarship is epistemologically flawed, but rather, it reflects my belief that evaluation of problems invites awareness. The chronological restriction to the period after c. 1990 should not be taken to imply that earlier scholarship was devoid of problems. It was not, but the scepticism towards reconstruction associated with ‘New’ or ‘Material’ Philology makes the inception of this movement a natural starting point.

The field of Old Norse philology is reminiscent of most medieval philologies in belonging to a conservative tradition in the edition of prose: within reason, editors generally retain the readings of the codex optimus. With poetry, the situation is different. Most scribes, especially after c. 1300, had a limited understanding of the poetry they were copying, some of which was then 400–500 years old. Many poetic passages are therefore nonsensical as they stand in individual manuscripts. With the modern philologist’s knowledge of linguistics, metrics and stylistics, many errors can be corrected with relative ease, whereas others must remain objects of debate. A few manuscripts reflect what must have been a very good scribal understanding of the poetry in the archetype or exemplar – notably R² (Copenhagen, GKS 2367 4to [Codex Regius], c. 1300–50), containing Snorri’s Edda (a poetic and mythographic handbook composed c. 1220–1230) and transcripts of Kringla (the best, but now lost, manuscript of Heimskringla; a history of the Norwegian kings composed c. 1220–1240) – thus providing a solid foundation for textual criticism.

While editions of ON poetry are reminiscent of classical editions in the degree to which they introduce variants and emendations into the text, it should be noted that stemmatic considerations are often of limited importance. In some instances, we possess only one witness of independent, text-critical value, whereas in others, stemmatics can only take us to a relatively late point in the transmission, which was oral at least down to c. 1150. Thus, for instance, stemmatic arguments are rare in the new edition of skaldic poetry (Clunies Ross, Gade, Whaley et al., eds. 2007–). Linguistics, metrics and stylistics are often more central. The traditional primacy of stemmatic over other criteria is thus inverted.¹ Stemmatic observations will enter the following discussion as needed, but more consistent attention will be directed at the basic assumption that it is more likely that scribes are responsible for obvious errors than that the author is and its ramifications.

The validity of this claim can be tested against the manuscript tradition. For instance, below I discuss the ungrammatical foldinnar (no meaning) and the unmetrical tregari (‘sorrow-bringer’). These variants are found in only one manuscript (U = Uppsala, Universitetsbiblioteket, DG 11 4to [Codex Upsaliensis], c. 1300–1325) in one branch in one of the three versions of the poem Vǫluspá, and it is therefore likely that they are innovations by the U scribe. This observation correlates with the fact that foldinnar is not a possible form in the ON language and that the verse tungls tregari has a structure that is otherwise unknown not only in ON, but throughout early Germanic verse (Suzuki 2014: 119). It is thus unlikely that the poet would have produced either form. The independent parameters of grammar and metre thus corroborate observations made on the basis of manuscript evidence. Examples like these can be found in most early texts of any length, and the assumption that errors are more likely to be produced by scribes than by authors is thus not open to reasonable doubt (similarly Fulk 1996: 10).

The term error calls for some elaboration, since it can be used in various ways. Within stemmatics, it is often synonymous to ‘non-original reading’. For the present purposes, this meaning is better conveyed by the terms innovation or secondary reading, covering all identifiable post-authorial readings (see Trovato 2014: 54 for the use of error and innovation; for secondary reading, see Roelli, ed. 2020: 4, 383). By contrast, I here use error in the narrower sense of ‘violation of the morphological, syntactical, phonological or metrical system’. The rules of these systems can be established by statistical means: if a particular form or structure is unique or exceedingly rare, it is likely that it is erroneous, unless there are compelling reasons to think otherwise.² Thus, for instance, in standard English the frequency of the copula is after the pronouns I and you is so low that the speaker will intuitively feel that the phrases I is and you is constitute errors. Of course, there will always be debatable cases, but the principle is clear enough. In the study of literature in dead languages, most of us will not feel the errors as strongly as in our own language, and we must then rely more on statistics than intuition. Noting the errors in a defunct metrical system is even more challenging, but statistics show us that the rules of metrics can be as strict as those of grammar. The definition of error as a violation of a system with clearly identifiable rules therefore entails the inclusion of metrical errors alongside linguistic ones.

Of course, scribes can commit mistakes that do not necessarily break the rules of grammar and therefore do not constitute errors in this narrow sense, for instance through skips of the eye or faulty interpretation of letters. Many non-erroneous mistakes can be useful for textual criticism, but a focus on errors as here defined has the advantage that it facilitates testing and is likely to be acceptable to many, since most will agree that forms such as I is and you is constitute errors in standard English.

In recent decades, some scholars have become wary of the term error.³ As far as I can tell, such scepticism is generally associated with a normative interpretation, errors being ‘bad’. I would therefore point out that within textual criticism, the concept is conditioned by stemmatics and/or statistics and by no other value than credence in a probabilistic approach to the evaluation of competing hypotheses. Normative use of the term introduces irrelevant parameters into the analysis and should be avoided.⁴ As the examples fold í mar/foldinnar and tjúgari/tregari illustrate, scholars need to reckon with errors in order to attribute statistically plausible forms to the author, or even simply for the text to make sense. The concept is thus both valid and analytically indispensable.

What is textual criticism?

The overall methodology of textual criticism is described in books such as Maas (1960), Pasquali (1962), West (1973) and Roelli ed. (2020) and need not detain us here.⁵ Such handbooks mainly take the underlying epistemology for granted, however, and a brief description of the discipline's basic assumptions and preconditions is therefore warranted. First, it should be noted that although textual criticism aims to provide a text as close to that of the author as possible, it does not purport to restore the author’s text. The method is simply designed to identify scribal innovations and to replace these with older readings, and this can only happen when there are strong reasons speaking in favour of one form over another. Edward Schröder formulates this aim – indeed ‘duty’ – succinctly:

‘freilich nicht immer lasst sich die vorlage mit der gleichen bestimmtheit erschlieſsen wie in diesen beiden fällen [where flawed rhymes and manuscript evidence coincide], und es mag immerhin zugegeben werden, dass eine bis in alle einzelheiten gesicherte reconstruction des originals unmöglich ist. das entbindet uns aber nicht von der pflicht, ihr zuzustreben wo irgend wir können.’ (Schröder 1926: 94)

How far this procedure allows the philologist to proceed towards the text of the author varies from case to case, depending on the manuscript situation and which criteria present themselves. Attempts to restore the author’s text beyond what probabilities suggest are not in accordance with the methods of textual criticism and are therefore irrelevant to the evaluation of the method. In any event, excessive Konjekturfreudigkeit was largely abandoned in the early twentieth century and is therefore not a current problem.⁶

Next, a few words on the concept of archetype versus original are expedient. The stemmatic method aims to restore readings from the archetype, which is the common ancestor of all preserved witnesses (Maas 1960: 5–9). In classical philology, the archetype may often be centuries younger than the author’s original. In ON philology, the situation differs in at least two regards. First, in the study of Eddic poetry (anonymous poetry in flexible metre) in particular, we often have only one witness of text-critical value. Reconstruction must then be based on grammar, metrics, language history, plausible meaning, etc.⁷ In such a situation, reconstruction is aimed at either exemplar (e.g. by reversing likely misreadings) or original (e.g. by restoring defunct linguistic forms) rather than the archetype. Second, the distinction between the archetype and the original of prose texts is often of limited practical importance in ON. Thus, for instance, the exemplar of manuscript T (Utrecht, Universiteitsbibliotheek, MS No. 1374 [Codex Trajectinus], c. 1595) of Snorri’s Edda featured an orthography and morphology that indicate that it was written perhaps as early as c. 1250 (Faulkes, ed. 1985: 19–20). This means that the archetype of Snorri’s Edda must date either to Snorri’s lifetime (d. 1241) or very shortly thereafter. In such a situation, the distinction between archetype and original becomes largely theoretical.

I now turn to an assumption that is implicit within textual criticism, but that needs to be made explicit in response to current philological trends; namely, that the author is more important for the semantic content of texts than scribes are. ON and other medieval philologies have in recent decades seen a growing focus on written transmission, increasingly turning attention away from the authors and towards the scribes. This shift in emphasis has generally not been accompanied by a discussion of the relative importance of original production versus the degree and types of changes in transmission, however. On the contrary, the concept of change has increasingly come to be used without further specification in medievalist scholarship, and this is problematic, since change applies to everything from minute alterations to complete transformations.⁸ The notion therefore has little analytical value unless its individual instantiations are described in some detail. Failing this, it may easily become an instrument of confirmation bias, evoking associations to some other part of the vast scale of change than the one that is likely appropriate to the object under study.

One of the foundational texts of New/Material Philology may serve to illustrate this point. Cerquiglini (1999 [1989]) extolls ‘variance’ as the main characteristic of medieval textual culture, but his examples have little or no semantic consequences. They could have served to illustrate the fact that medieval French, like other medieval vernaculars, had no strict standard orthography and grammar, but little else. Cerquiglini’s interpretation of these well-known facts as signs of a textual culture whose fundamental characteristic was ‘variance’ is obscurantist, obviating testing by means of an abstract concept whose consequences are described as vast and yet are nowhere to be found except in the concept itself. Furthermore, Cerquiglini’s lack of a clear methodology leads him to promote implausible claims. Thus, he contends that the Old French distinction between the nominative and oblique case is a modern invention (Cerquiglini 1999 [1989]: 58). It is true, of course, that scribal errors will accumulate in a period of morphological change, but knowledge of the preceding Latin forms and the resulting French ones, in addition to stemmatic evaluation, aids the scholar in distinguishing authorial from scribal forms. Cerquiglini mentions neither of these analytical tools but instead advocates a mere surface description. This is positivism at its most extreme, and although positivism and relativism may seem like strange bedfellows, they often are in new-philological scholarship (see the following analysis of Quinn). Like Rösli and Gropper below, Cerquiglini sees variants as a reason not to reconstruct, even though they are the main cause for reconstruction. From a scholarly perspective, the fact that a question is analytically challenging is not a valid reason to abstain from analysis.

Rather, a reasonable response to a complex object of study is to delimitate the scope of the parameters involved. The unspecified concept of change is an obstacle in this regard, and it may be reduced by steps. First, a general observation may be useful, namely that all but a fraction of a text’s semantic content and structure is generally traceable to the author. Expressions of scribal interests are often limited to the choice of texts, minor alterations and perhaps some paratextual material (cf. Schröder 1926: 94).

This claim is not universally applicable, however, and we generally need to take a number of additional factors into account in order to evaluate the likely scope and character of change in individual instances. In this article, I mainly discuss the transmission of ON poetry, which put considerable strain on the scribes due to its linguistic and metrical complexity. Changes in poetic transmission will therefore often result in nonsensical or unmetrical readings. This creates optimal preconditions for textual criticism, and I have therefore selected such examples for analysis, but it should be noted that changes in prose are often of a different character.

ON prose texts reflect the language spoken by the scribes, and changes in prose therefore typically result in a comprehensible and coherent text. In ON saga transmission, a common situation is that we have evidence of a slightly longer, thirteenth-century version and a somewhat terser, fourteenth-century one (e.g. Fóstbrœðra saga, Sverris saga and Egils saga; see Males 2020; Þorleifur Hauksson, ed. 2007: xlvii–lii, lxxiv; Bjarni Einarsson, ed. 2001: lix, 160–172). Such versions bear witness to intentional, meaningful changes, but with some noteworthy exceptions, we are essentially dealing with the same text, one version of which uses a more economical mode of expression (a good illustration is found in Sigurður Nordal, ed. 1933: lxxxiii–lxxxv). While such changes are interesting, not least for what they tell us about the evolution of saga style, describing them by means of the unqualified concept of change would do them no justice. Rather, most saga ‘versions’ suggest a remarkable degree of semantic stability despite the seeming flux at the surface level, and only when we take the scope of change versus stability on these different levels into account can we approach a nuanced analysis.

At times, the most consequential changes in transmission stand in no clear correlation with the degree of change within individual texts. Thus, for instance, the compilers of the late fourteenth-century manuscript known as Flateyjarbók, mainly containing sagas of the kings of Norway, could stand comparison with some of the more sensitive saga authors in the way they combine their material for narrative effect. The difference is that while authors produce their own texts, these compilers selected earlier texts which they then conjoined in a sophisticated manner. Authors of sagas about the kings of Norway occupy a middle ground, both compiling and revising older material within the framework of their own narrative. Even if we limit our discussion to a broad outlook on ON historical narrative prose, then, the scope and parameters of change must be specified in order to be analytically useful. Textual criticism has always been committed to such specificity and may serve as a blueprint.

Another central aspect of textual criticism is that it assumes that diachronic and synchronic perspectives are interdependent. Diachronic reconstruction cannot take place without a plausible analysis of scribal behaviour, and conversely, scribal contributions cannot be identified without some concept of either exemplar, archetype or original. These assumptions are not exclusive to textual critics. Thus, for instance, I have encountered no philologist who would deny that the same work can be contained in several witnesses. If we concede this, we also acknowledge that reconstruction is necessary for philology. Since provenance from the same author is the only plausible explanation of multiple copies of the same work, and as a rule, two medieval copies are never completely alike, identifying ‘sameness’ requires the scholar to ignore variation in favour of reconstructing genetic relationships. All scholars who would agree that Egils saga is found in several manuscripts thus admit the necessity of some degree of reconstruction. Without it, there would often be no way of distinguishing author from scribe, since the analysis of both tends to require a comparison of witnesses. This point deserves to be stressed, since most scholarship discussed in this article either disregards or criticises the methods necessary for textual reconstruction. In reality, however, scholars of medieval philology cannot dispense with these without considerable damage to their own analytical capacity.

The trend of embracing change and the material record may initially have been a healthy reaction against, for instance, a tendency within Classics to snub medieval transmission, but today the philological climate has changed so much that it is doubtful how many plausible antagonists are still to be found: one is reminded of how farmers still fear the wolf in areas where it went extinct a hundred years ago. The manuscripts have always been central in vernacular philology, and in Latin, they began to enter centre stage around the turn of the twentieth century. When inaugurating the discipline of medieval Latin philology, Ludwig Traube expressed himself in ways reminiscent of many new-philological publications post-dating him by roughly a century:

In jedes Schriftwerk haben die Schreiber und Leser so viel Lebendiges hineingelegt durch ihre Randbemerkungen, durch sonstige Einträge, durch Verbesserungen, durch die Anordnung des Stoffes, wenn mehr als ein Werk in einem Sammelband seine Stelle findet, – so viel Lebendiges, das der Auferweckung harrt und das dann kündet von allem Möglichen, Persönlichem, Individuellem und Kulturhistorisch-Intressantem […] (Traube 1909, 1: 8)

Through the succession Traube–Lehmann–Bischoff and scholars following their lead, studies of medieval authors and scribes have seen a steady progression and produced impressive results.⁹ New philologists have generally not drawn on these advances, however. Rather, Cerquiglini (1999 [1989]), Nichols (1990) and other seminal new-philological publications describe the scholarly discourse as if Traube, his pupils and the wealth of excellent scholarship influenced by their pioneering achievements had never existed. In fact, however, Edward Schröder drew a sigh of exasperation at the widespread ‘veneration of the scribe’ already in 1926, as illustrated by the opening quotation of this article. The debate on the importance of the scribe is thus a long one, but before the advent of New/Material Philology, it was mainly held in editions and reviews of them (Fulk 1997). The spread of this debate in editions and articles, as well as over several philologies and linguistic areas, is such that many generalisations are bound to be simplistic (Stock 2015). Still, it seems safe to say that there is a general lack of continuity between the ‘Traube school’ and New/Material Philology, and that this disconnect is associated with an important methodological difference between the two.

Scholars writing in Traube’s tradition typically retain the methods of textual criticism and use these as needed, even when focusing on scribes. This combination goes a long way towards explaining how they have been able to produce so many new and plausible answers. By contrast, new/material philologists often either disregard the stages preceding the extant manuscripts or treat these with agnosticism. As noted above, such an approach leaves the philologist analytically hampered, and the problems emerging from an elimination of diachronic perspectives have become increasingly apparent in recent decades. Thus, for instance, present scholarly interest in scribes is often paired with an exaggeration of their capacity and agency, bringing them close to the role of the author. In the case of Vǫluspá (an Eddic, prophetic poem, likely composed c. 950–1000), such analyses have been shown to be incompatible with grammar and metrics (see the discussion of Quinn below), and in the case of Old English poetry, they have not held up to probabilistic evaluation of the data (Orton 2000: 28 et passim; Neidorf 2017: 103–32). Syntactic obscurities tend to co-occur with problems of meaning, alliteration and metre, a fact that rules out coincidence as a plausible explanation and suggests that the term corruption is often appropriate in such cases (Fulk 1996: 10). A survey of skaldic poetry (attributed poetry in strict metre) in manuscripts suggests that many scribes were content with comprehension of the overall subject matter of single stanzas, whereas incomprehensible elements were retained as stylistically suggestive metrical fillers (Males 2010).

In short, the evidence conforms with what one might expect: if a modern person without palaeographic, linguistic and metrical training were set to transcribe poetry written by hand in the nineteenth century, it would be reasonable to expect some rate of error. The copyist ought neither to merit praise nor blame for behaving as one might expect. Such irrelevant parameters can only cloud our analysis, of this hypothetical scribe as well as of medieval ones. Exceptionally faithful or careless or inventive scribes may be interesting but should not be allowed to colour our perception of normal scribal activity.¹⁰ In addition, a conflation of the roles of scribe and author renders obvious causal and chronological sequences illogical. Our analytical self-interest should thus encourage us to a balanced approach in the evaluation of the relative importance of primary causation and secondary contributions.

From time to time during the last two millennia, philologists have needed to defend their focus on the author’s text against their critics, such as was the case with, for instance, Jerome and Erasmus. Today, we have again reached such a juncture. I would therefore point out that denying the importance of the author is often incompatible with the exigencies of textual interpretation. If scholars do not take the author’s historical circumstances into account, they will typically be unable to explain central characteristics of a text. Thus, for instance: why would a poet compose a eulogy to a king and ask him for a reward if that king was not alive to remunerate him? The ‘death of the author’ is thus often at least partially nonsensical to philologists.

For a more detailed example, we may contemplate the staggering improbabilities required in order to study, for instance, the poems of Catullus as products of the fourteenth century, when nearly all of them are first attested. Attributing Catullus’ fierce individualism to the scribes would be nonsensical, and his literary influences and references to current events can only plausibly be explained as the products of an urbane author of the Late Roman Republic. On a more general note, scribal contributions to a given text cannot logically come anywhere near those of the author. If they did, we would no longer recognise the text as a witness to the same work, and the scribe would then be the author of the new work. As long as we are discussing witnesses to the same work, then, their semantic content must primarily have been dictated by the author. For these reasons, a focus on the author is a viable and often indispensable approach within philology.

It is true, of course, that manuscripts copied by someone other than the author are most often our only witnesses to the works of the authors and that scribal alterations necessarily affect our access to the authors' works. Restricting the study to the text of the witnesses might thus appear to be the ‘safest’ option, but we need to remind ourselves that there is no epistemological ‘safe space’ for scholars. Thus, also in less extreme cases than the transmission of Catullus’ poetry, a restriction to synchronic perspectives may entail considerable improbabilities. Thus, for instance, individual manuscripts would suggest that skalds often produced incomprehensible gibberish. By selecting variants from several manuscripts, however, it is generally possible to construe meaningful stanzas, and the probability that coincidence would allow the scholar to achieve this result is vanishingly small. The only plausible explanation is that the skalds did in fact compose comprehensible poetry, and that linguistic incoherence is the product of scribal transmission. A concrete example from Eddic poetry may further illustrate the point that a plausible analysis may require reconstruction.

The poems Hávamál and Vafþrúðnismál (both Eddic poems, likely composed in the tenth century) contain one instance each of the verb reka (here ‘avenge; fight’) in a full ljóðaháttr verse, and such verses must contain internal alliteration. In order to achieve alliteration, the older form vreka must be substituted (Hávamál 32.3 en at virði vrekask; Vafþrúðnismál 53.3 þess mun Viðarr vreka).¹¹ Forms with vr- had been defunct for at least 250 years by the time of the scribes, and it is therefore not surprising that there are no traces of them in the two manuscripts of text-critical value (R: Copenhagen, GKS 2365 4to [Codex Regius of Eddic poetry], c. 1270 and A: Reykjavík, AM 748 Ia 4to, c. 1300–1325). Closer inspection of the poems reveals that they contain no full verses without alliteration except for the two in question according to their manuscript reading (reka/rekask). The only partial exception is Hávamál 80.3, where the full verse retains the alliteration of preceding verses rather than displaying internal alliteration (von See et al., eds. 1997–2019, 1: 475). The stanza is abnormal also in exhibiting only full verses in the second half:

Þat er þá reynt

er þú at rúnum spyrr

inum reginkunnum,

þeim er gørðu ginnregin

ok fáði fimbulþulr;

þá hefr hann bazt ef hann þegir.

It is evident then, when you ask about runes, the runes derived from the gods, those that the great gods made and the mighty speaker (Óðinn) painted; then he is best off if he stays silent.

The metrical irregularities of this stanza appear to be motivated by the presentation, for the first time in the poem, of the runes and their divine origin. The collocation reginkunnar rúnar (‘runes derived from the gods’) was apparently traditional, since it is found both on the Noleby stone (sixth century) and the Sparlösa stone (c. 800), and the inclusion of this idiom coincides with the abnormal alliteration pattern (von See et al., eds. 1997–2019, 1: 696). The peculiar distribution of alliteration and types of verses in this stanza is thus clearly intentional, and the third verse does not, in fact, lack alliteration, even if one would normally have expected internal alliteration. By contrast, the two verses where the verb (v)reka(sk) occurs do not alliterate with preceding verses.

Defending the reading reka as authorial therefore demands that the scholar believe that both poets would break the rule of alliteration in full verses only when using the verb reka, and that the fact that this verb clearly had vr- before c. 1000 (Swedish vräka, English wreck) is coincidental. Belief in such unlikely coincidences is incompatible with a probabilistic approach. As this example illustrates, reconstruction is often necessary in order to postulate probable scenarios. The text of the manuscripts must serve as our starting point for investigating competing hypotheses, but in the investigation itself, the text is only one of an unpredictable number of factors that must most plausibly and economically be explained by the preferred hypothesis. In the example above, the hypothesis that both poets used the form vreka, which was subsequently replaced by reka, explains both why the poets appear to break the rule of alliteration in full verses only once each and why the scribes failed to recognise the alliteration. In fact, the poets adhered to the rule, and the scribes simply used the language they knew.

Probabilistic considerations such as these inform textual criticism and explain why its methods are often necessary for postulating plausible scenarios and testing competing hypotheses. I turn now to a more focused analysis of how ON scholars relate to textual criticism. To ensure that the analysis is representative of the state of the art, I have excluded non-specialists and early-career scholars and included only those of whom the highest level of ON philological expertise may be expected. Furthermore, I discuss only publications that either deal explicitly with textual criticism or that explore questions where its methods are indispensable, such as dating. In the study of practical applications, I have restricted the study to scholars who have retained their methodological approach through several publications, so as not to include occasional blunders.

Misrepresentations of textual criticism

This section analyses the presentation of textual criticism in two publications of 2010 and 2021. Their authors are key members of the ON academic community, and both publications are found in edited volumes focusing on the transmission of ON literature. Neither publication corroborates its description of textual criticism with empirical evidence. This suggests that the authors think of their views as mainstream, a perception that makes these two publications interesting for the evaluation of the current ON discourse on textual criticism.

Lukas Rösli and Stefanie Gropper are editors of the volume In Search of the Culprit. Aspects of Medieval Authorship (2021b). Throughout the book, the term error is used only twice in relation to medieval textual transmission. Both occurrences are found in a short passage in the editors’ introduction:

Some variants might be errors or mistakes, although we should make such claims only in restricted cases and should be wary of viewing them as indications of deterioration: in some cases, errors can result in productive change to a text. (Rösli & Gropper 2021a: 12)

For lack of definition of ‘error’, ‘mistake’ and ‘restricted’, the precise import of this passage is elusive. It seems clear, however, that the editors are contrasting the concept of ‘deterioration’ with the one conveyed by the term ‘productive’. Context shows that they attribute the idea of ‘deterioration’ to textual criticism, and this suggests that they view its evaluations as based on normative values, rather than on probabilities. As discussed above, however, textual criticism does not focus on good versus bad readings, but on the plausibility that some are older than others. Other statements convey equally misleading notions of textual criticism, a discipline based on the premise that:

[…] it is often easier and more comfortable to neglect the debate on transmission and instead to return to viewing ‘the text’, or even ‘the work’, as the product of one (probably male) educated being at a specific time in a specific surrounding. (Rösli & Gropper 2021a: 11)

Textual criticism to a large extent constitutes a debate on transmission, and the claim that it would neglect it is therefore nonsensical. The parameters of ease and comfort are not very relevant to scholarship in general, but they ought certainly not to be attributed to scholars who draw on the widest possible range of evidence in order to evaluate probabilities. With regard to gender, modern philologists have no power to alter medieval social norms and should not be expected to do so. This description is thus riddled with claims that are either nonsensical or irrelevant.

In In Search of the Culprit, only one of twelve contributions treats authors in the conventional sense of the word (Sigurður Ingibergur Björnsson, Steingrímur Páll Kárason and Jón Karl Helgason 2021). Rather, the book’s title in conjunction with its contents serves to convey the impression that authorship is mainly a question of transmission and reception. As discussed above, such a position does not hold up to scrutiny, and the editors present no empirical support for it. Rather, like Cerquiglini (1999 [1989]), they base their generalisations on abstract terms such as mouvance and variance. Overall, the editors are indebted to post-modern discourse, as evident from the fact that they open their introduction with a reference to Roland Barthes’ La mort de l’auteur. Throughout their introduction, the editors’ discourse is relativistic, the author being constructed mainly through reception: authorship is ‘pluralistic’, ‘undetermined’, ‘discursively and intertextually produced’, ‘distributed’, and so on (Rösli & Gropper 2021a: 14–15). Indeed, the author as an individual originator of text is said to be ‘mythical’ (Rösli & Gropper 2021a: 10). The concepts of stemma and error are both treated with suspicion, apparently due to the editors’ emphasis on medieval mouvance (Rösli & Gropper 2021a: 11–12).

The editors treat textual criticism as a thing of the past: ‘[…] it was for a long time the aim of philologists and the editions they produced to present a text as close as possible to the lost original […]’ (Rösli & Gropper 2021a: 11). This is misleading. If anything, digitisation and other developments have for some time been making textual criticism into a dynamic field of scholarship. The Münster School of New Testament stemmatics is now famous, and the Handbook of Stemmatology (Roelli, ed., 2020) bears witness to the methodological sophistication and wide-ranging application of present-day textual criticism. Stemmatics and other aspects of textual criticism have also begun to see a revival in ON, through the exploration of new methodologies and refined testing of an increasingly greater number of criteria.¹² Far from being obsolete, textual criticism is gradually proving itself more and more useful in providing a solid textual basis for scholars of a large array of disciplines.

The editors refer to the fact that we generally do not possess the autograph of the author as a factor that speaks against reconstruction, even though the rarity of autographs is the reason that the methods of textual criticism exist. They then go on to describe the changes in transmission that have always been the basis for textual criticism as if they had been recently discovered: ‘[…] only in the late 20^th century have we become more aware that the transmission of these texts is best understood […] as an ongoing process of adapting and reproducing texts […]’ (Rösli & Gropper 2021a: 11). The logic by which the preconditions for textual criticism have made its methods obsolete is nowhere presented. The editors embrace the study of change in transmission, but they do not clarify how this is to be done, if the method for identifying such changes has been abandoned.

The editors treat their views as a matter of consensus, and their introduction therefore suggests that the presentation of misleading claims about textual criticism, without adducing empirical support, is relatively unmarked in the present scholarly discourse. This is puzzling, and it seems likely that the debate has evolved gradually to a point where such a renunciation of probabilistic standards and intersubjective scrutiny might appear acceptable. I turn next to a publication that predates In Search of the Culprit by eleven years and is suggestive of the discourse leading up to it.

As professor of ON philology at the Armamagnæan Institute in Copenhagen, which has prime responsibility for ON manuscripts in Denmark, Matthew James Driscoll may be taken to represent views current among experienced researchers in the field. In a chapter on philology from 2010, Driscoll describes textual criticism thus:

Although the stemmatic method is all very neat and its logic nothing short of majestic, it has a number of shortcomings, the most significant being that it hardly ever works with real textual traditions, since it assumes, among other things, that no two scribes will ever independently make the same mistake, which they frequently do, that they will always work from a single exemplar, which they frequently do not, and that most scribes will tend to reproduce their exemplars exactly, which they almost never do, at least in the case of vernacular literature. (Driscoll 2010: 89)

Driscoll seems to suggest that the neatness and majesty of the stemmatic method is intrinsically connected to its detachment from the reality of the manuscripts. To some extent, this is true. Like most scholarship, textual criticism aims to explain that which is not self-evident, and it therefore cannot be restricted to a mere surface description. Driscoll then proceeds to describe a method that does not exist, however. Scribal propensity to make the same mistakes is the reason for the division into trivial or polygenetic and significant or monogenetic innovations.¹³ This distinction is one of the most fundamental preconditions for stemmatics and textual criticism generally. With regard to scribes working from more than one exemplar, this is a universally recognised problem, giving rise to the phenomenon known as contamination (West 1973: 12–13, 35–37). It is true that this creates difficulties for the reconstruction of the archetype, but a glance at the dotted lines between the branches of the stemma in countless editions of classical and medieval texts is enough to demonstrate that this problem is generally taken into account by editors. Finally, the claim that textual criticism assumes that most scribes tend to reproduce their exemplars exactly is a strawman argument. If that were the case, there would be no need for a stemmatic method nor any other aspect of textual criticism, and it is not reasonable to assume that adherents of the method believe it to be superfluous. All three claims about textual criticism are thus misleading.

Driscoll goes on to state:

And, indeed, criticism of the method has chiefly come from medievalists working in vernacular traditions, most notably the French scholar Joseph Bédier (1864–1938), who rejected the claims of stemmatic analysis to scientific objectivity and advocated an editorial policy which involved choosing a single ‘best text’ and reproducing it conservatively, that is, with as little emendation as possible (only in cases of obvious scribal error). […] Although initially criticised by many, Bédier’s ‘best-text’ method has the advantage of reducing damage to the text through subjective editorial emendation (by editors, who, Bédier alleged, tended to see themselves as collaborators with the author), and presenting the reader with, if not the text, then at least a text which had actually existed. (Driscoll 2010: 89–90)

One notes here that Driscoll does not mention Bédier’s key claim, namely that the preponderance of two-branched stemmas in editions is due to the fact that editors wish to be able to choose freely among the variants in order to shape the text according to their wishes (Bédier 1928). Paul Maas and Sebastiano Timpanaro have pointed out, however, that while the observation is correct, the conclusion does not follow. If there are three witnesses to a text, logic dictates that roughly 60–70% of the stemmatic possibilities will be two-branched, and the preponderance of two-branched stemmas therefore has no bearing on the validity of the method (Maas 1960: 29–30; corrected values in Timpanaro 2005 [1963]: 163). Odd Einar Haugen has shown that two-branched stemmas dominate also in the Arnamagnæan editions – the most ‘scientific’ of ON editions – even though their editors have no reason to favour them.¹⁴ Other philologists have refuted numerous unfounded claims in Bédier’s argumentation, as well as demonstrated its incompatibility with empirical observation (Montanari 2003: 358–404; Trovato 2014: 77–108). Driscoll mentions no specifics and has no reference to Bédier’s critics, and the matter thus appears to boil down to a difference of opinion rather than of probabilistic validity.

In the quotation above, Driscoll uses both ‘objective’ and ‘subjective’ to describe textual criticism. His point seems to be that while claiming to be objective, textual criticism is, in fact, subjective. Neither characteristic is relevant, however, since textual criticism is probabilistic: it imposes restrictions on subjectivity by testing hypotheses against known scribal patterns of behaviour, historical linguistics, palaeography, metrics, etc.¹⁵ As one gathers from the phrase ‘a text which had actually existed’, Driscoll’s desired alternative is exact reproduction of the text. Towards the end of the chapter, it becomes clear that his main concern is that all manuscript readings should be included in editions, and he sees digital reproduction as the solution (Driscoll 2010: 102–04). Today, his wishes have in one sense come true, since the manuscript images are now available online.

The remaining challenge, then, is the same as it ever was; namely, that of interpreting the text. This activity consists of a progressive testing of hypotheses, and in any but the shortest fragments, it involves some degree of invasiveness. This is borne out by Driscoll’s own edition of Ágrip, where he sometimes corrects the text even where it is grammatical (e.g. Driscoll, ed. 2008 [1995]: 6 e, 10 d). Ágrip is found in a single, early manuscript, however, and it therefore requires minimal contributions by the editor.

In many instances, the material record both allows for and requires more active testing of hypotheses, for instance due to the number of witnesses or the likelihood that the text under study is considerably older than these. This leads us to our next point. How well equipped are ON scholars to evaluate the probabilities involved in textual criticism? It is conceivable that the descriptions of the method so far discussed serve the interests of polemic more than those of evaluation. If so, they may not tell us very much about how ON scholars approach textual criticism in their daily work. I turn now to practical applications.

Textual criticism: examples

In the following, I discuss a selection of scholarly analyses of scribal forms and how these relate to the assumptions of textual criticism. Since the edition of poetry involves more parameters and often older texts than prose, the examples pertain to poetry.

I begin with an article on the editing of Vǫluspá by Judy Quinn (2001). Quinn makes similar points in two later publications (Quinn 2016a and 2016b), but her 2001 publication has the most informative discussion and I have therefore selected it for analysis, adding comments on the other two as needed.¹⁶ Together, Quinn’s three articles belong to the most influential new-philological studies of Eddic poetry, and they are thus of some importance for the evaluation of the overall debate.

Like Driscoll, Quinn refers to textual criticism as ‘subjective’ (Quinn 2001: 72; similarly Quinn 2016b: 53). Throughout her article, Quinn gives no indication that textual criticism is based on probabilities. Instead, the only criteria for selection that she attributes to textual critics are their own ‘cultural and ideological pre-occupations’ (Quinn 2001: 72). Her critique of fanciful reconstructions of Vǫluspá is valid, but this is another strawman, since the method cannot be evaluated by means of examples of scholars who do not adhere to it (Quinn 2001: 75–77). Overall, her description of textual criticism is parodic:

Because there is often no scientific way of proving the superiority of one editor’s choice over another’s, the debate between editors and commentators within this tradition is characterised by the rhetoric of derision. (Quinn 2001: 73)

Quinn ought perhaps to have been wary of attributing ‘the rhetoric of derision’ to others, since by identifying only cultural and ideological preoccupations as guidelines of textual criticism and by discussing examples that are irrelevant to its evaluation, she invites a similar analysis of her own text. More importantly, it here becomes clear that her earlier use of the term ‘subjective’ has a ‘scientific way of proving’ as its logical counterpart. Apparently, scholars can either provide proof or be relegated to rhetorical posturing. Quinn thus appears to subscribe to a positivist belief in certain knowledge, although this view of scientific knowledge production has been thoroughly discredited since the appearance of Karl Popper’s Logik der Forschung in 1934. There can be no scientific ‘proofs’, but at best increased probability through evaluation. Subjectivity is indeed an epistemological problem, and the scholarly solution is to impose probabilistic restrictions on the evaluation of competing hypotheses or, in Popper’s terminology, to attempt to falsify them. It seems likely that Quinn’s positivist view accounts for her own emphasis on the readings of individual manuscripts to the exclusion of other parameters, since these readings may be considered ‘proven’ in the sense that they are often more or less self-evident to a scholar used to reading ON manuscripts. Problems emerge, however, when Quinn attempts to project the ‘truth’ of the manuscripts onto their hypothetical prehistory.

After her discussion of textual criticism, Quinn turns to her own evaluation of a number of textual variants in Vǫluspá, and she focuses in particular on those found in U. Regarding these she writes:

When the texts of quoted verses differ significantly between manuscripts of Gylfaginning we must wonder at the state of the original text or of the draft that scribes were copying from, as well as about their own acquaintance with the poem from oral tradition. (Quinn 2001: 83–84)

It is true U has independent text-critical value and that its variants may therefore be relevant for the reconstruction of the oral original of the poem, as well as for the archetype of Snorri’s Edda. In order to impose restrictions on scholarly subjectivity, however, textual criticism works under the assumption that it is unlikely that two or more transmitters have independently made the same change to the text, unless that change is typologically common. Quinn goes on to discuss two variants that are unique to U, and thus to one out of four manuscripts of Gylfaginning (the first part of Snorri’s Edda) with independent text-critical value. The three other manuscripts belong to two branches (R²T versus W: Copenhagen, AM 242 fol. [Codex Wormianus], c. 1340–70), and it is thus unlikely that shared readings in them are the result of scribal change (Haukur Þorgeirsson 2017a). In addition, in the two instances in question, their readings are also shared by the other two versions of Vǫluspá (R and H: Reykjavík, AM 544 4to [Hauksbók], c. 1350), and this makes it highly unlikely that U’s readings are older than those of R²TW. The first possibility that Quinn suggests – that the written archetype of Gylfaginning had U’s reading – is thus inherently improbable. Her other option – that U’s variants are due to independent knowledge of oral tradition – must be evaluated through recourse to other than stemmatic criteria. I turn now to Quinn’s discussion of the variants.

The first variant is sigrfoldinnar, which elsewhere reads sígr fold í mar ‘the earth sinks into the sea’. Quinn translates sigrfoldinnar as ‘triumph of the earth’, and notes that ‘this description reinforces the tragedy of ragnarǫk’ (Quinn 2001: 87). Sigrfoldinnar is an impossible form with no meaning, however. Apparently, Quinn assumes that it may stand for sigr foldarinnar, which would indeed mean ‘triumph of the earth’, but that is not the form found in the manuscript. Since Quinn here defends the manuscript reading, one must assume that she postulates a different grammar than that described in standard handbooks. ON morphology is well known, however, at least to such an extent as to rule out this assumption as extremely implausible (Noreen 1923: §§ 386, 391). Had Quinn taken the common phenomenon of minim confusion (confusion of letters consisting only of minims, such as i, n, m, u) into account, however, the visual cause of the variant would have presented itself. This explanation is so obvious and the alternative morphology so implausible that Quinn’s interpretation must be taken as an extreme stance against reconstruction, so much so that probabilities become irrelevant.

The rules of metrics may be as absolute as those of morphology and are therefore useful to philological evaluation. Quinn’s second variant illustrates this point. As noted above, U’s tungls tregari for tungls tjúgari in other manuscripts displays a structure that is unknown in all early Germanic traditions. Quinn interprets U’s tregari within a larger mythological framework, thus attributing considerable intentionality to the originator of the reading, and she takes no note of its metrical characteristics (Quinn 2001: 84–85). Statistically, this word is equally implausible as foldinnar, however, and since tjúgari is a hapax with no obvious underlying verb, it is likely that tregari represents an attempt to produce a comprehensible form, based on the verb trega (‘to bring sorrow’) and/or the noun tregi (‘sorrow’). This is thus most likely an instance of the common phenomenon of trivialisation (Maas 1960: 11). Leonard Neidorf’s extensive study of the transmission of Beowulf shows that trivialisation often leads to nonsensical readings, which suggests that it is in most instances devoid of co-authorial intent (Neidorf 2017: 62–68, 103–32). Since the likely reason for the variant is apparent, namely the scribe’s wish to achieve comprehensibility, it is inadvisable to posit other reasons that lack empirical support. In addition, the fact that tregari is unmetrical indicates that the scribe did not aim to contribute to the poem in a sensitive way, but only to make it easier to comprehend.

Quinn’s suggestion that these variants may be more original than those found in other manuscripts is thus not only implausible from a stemmatic perspective, but also based on grammar and metrics. Quinn takes a similar position on metrics in a later publication (Quinn 2016a: 65–66). She there argues against the emendation fé > fekk in Vǫluspá 29.3 on the basis that ‘across the Eddic corpus, a significant number of lines do not fit with the versification “rules” derived by Eduard Sievers for early Germanic poetry’. She does not specify which rules, nor the fact that the metrical rules of fornyrðislag are more rigid and better understood than those of other Eddic metres, nor that we are here dealing with a break of not one, but two strong conventions: lack of ictus on a stress-word and lack of alliteration on the first noun. Most importantly, Quinn ignores the crucial question whether the Vǫluspá poet broke these conventions. Only if he can be shown to have done so is her critique valid, but by all appearances, he did not (Haukur Þorgeirsson 2020: 51–53).

The choice of going against so strong probabilistic parameters is highly marked, and most notably so in the first two examples. This may perhaps find its explanation in aspects of Quinn’s perception of textual criticism which become clear in one of her later publications (Quinn 2016b). She there criticises stemmatic reconstruction of ‘oral-derived poetry’, focusing on Ursula Dronke’s hypothesis on the stemma of Vǫluspá (more on Dronke’s stemma below). Stemmatics have never been prominent in the study of ON poetry, however, since much of it is oral-derived and most scholars wish to reach beyond its written transmission. Dronke’s study is a rare and somewhat eccentric exception. In her scepticism towards stemmatics in the study of likely oral variants, Quinn is in agreement with most scholars, but, as noted above, she differs from them in not adhering to stemmatic principles even when her focus on internally related manuscripts would have made these useful.

Quinn is also at odds with other philologists when she refers to Dronke’s conjectures against all manuscript evidence as instances of what Maas calls divinatio, without further qualification (Quinn 2016b: 65, 68; similarly Quinn 2001: 74). In fact, Maas specifies crucial parameters for divinatio: trivialisation, typology of corruption, transmission history, language history, palaeography, orthography, etc. (Maas 1960: 11). This is also where grammar and metre belong in Maas’ scheme, although he appears to have thought these too obvious to mention. By contrast, as Quinn notes, Dronke conjectures on purely aesthetic grounds (Quinn 2016b: 66). Maas’ parameters are open to testing and intersubjective scrutiny while Dronke’s are not, and yet Quinn treats the two as epistemically equal. The fact that grammar and metre belong to the parameters of divinatio and that Quinn views divinatio as inherently subjective may explain why she sees grammar and metre as irrelevant to the evaluation of variants.

To sum up: By extending the concept of subjectivity to cover many parameters open to testing and intersubjective scrutiny, Quinn paints a misleading picture of textual criticism, and by projecting likely secondary readings onto the prehistory of the text without further testing, she inverts its principles. In so doing, she abandons the probabilistic restrictions on scholarly subjectivity inherent to the methods of textual criticism.

Another article on Vǫluspá – this one by Karl G. Johansson – presents an attempt to use the stemmatic method to argue for a written stemma, rather than an oral–written one. This is an interesting case, since Johansson here and elsewhere disregards dating criteria relating to language history and metrics (see below). Unlike Quinn, who rejects both stemmatic and other parameters for textual reconstruction, Johansson’s article thus allows us to address the state of textual criticism in ON philology from the angle of a scholar who accepts a stemmatic model while rejecting other parameters.

Following the analysis by Ursula Dronke against scholarly consensus, Johansson assumes that the two main versions of Vǫluspá – R and H – represent branches going back to a written archetype rather than two separate recordings from oral tradition (Dronke 1997: 62–92; Johansson 2000). Unlike Dronke, however, Johansson assumes that there is contamination from H to W, W being one of four witnesses to Gylfaginning.¹⁷ Johansson does not clarify whether he accepts Dronke’s assumption that quotations of Vǫluspá in Gylfaginning also go back to the same archetype as R and H, but the overall impression he conveys is that he wishes to add only contamination from H to W to Dronke’s analysis (Johansson 2000: 78). It therefore seems likely that Johansson, like Dronke, assumes that all witnesses to Vǫluspá go back to the same written archetype.

With regard to the connection between R and H, Johansson’s only diagnostic criterion is the shared error þrjár for þrír in stanza 17, in all likelihood due to influence from a similar verse in stanza 8 (Johansson 2000: 69–70). In the manuscripts, Vǫluspá 8.6 and 17.1 both read ‘unz þrjár kómu’, but 17.1 must read ‘unz þrír kómu’ in order to correlate with the following masculine plural subject. Since the scribes would have been used to repeated verses from refrains, and since þrjár for þrír would have been identical except for one letter, the error is probably best seen as trivial, that is, both the R and H scribe may have independently been influenced by stanza 8.¹⁸ Furthermore, only part of the error is shared, since H echoes the entire couplet ‘unz þrjár kómu | þursa brúðir’ from stanza 8 (although stanza 8 has the synonymous and metrically equivalent ‘þursa meyjar’), whereas R adopts only the first verse.¹⁹ Taking such a single, partly shared and probably trivial error as grounds to assume a stemmatic connection is not easily reconciled with the assumptions of textual criticism, even though Johansson’s analysis presupposes a text-critical approach.

Concerning the connection between H and W, a single variant again carries the argument, namely Bláins (dwarf name) in H and W for blám/bláms in other manuscripts (Johansson 2000: 79). Jón Helgason, Jakob Benediktsson and Stefán Karlsson believed that the W scribe is the scribe of Vǫluspá in H, and Johansson takes this assumption as his starting point for arguing that Bláins is diagnostic of contamination rather than a shared archetype.²⁰ I see no reason to question the identification of W’s hand in H, but beyond that, Johansson’s analysis presents two main problems.

First, like other stemmatic affiliations, contamination is shown through shared innovations, whereas forms from the archetype may turn up in any manuscript of text-critical value. In order to argue that Bláins in W is due to contamination from H, Johansson would thus have had to demonstrate that Bláins is an innovation, but he does not address this question.²¹ An evaluation of the variants suggests that the opposite is true. Bláins is the lectio difficilior and has analogues in other dwarf names (e.g. Þráinn). In addition, blám (‘blue’, dat. pl.) and bláms (no meaning) are easily explained as trivialisation and corruption of Bláins through minim confusion (in > m), and this assumption is corroborated by the fact that they make poor sense in the context and that one of them has no meaning at all. Minim confusion is one of the most trivial errors, especially in cases like this, where it transforms an obscure name into a known word or at least a more common sequence of letters (Mårtensson 2013: 152–63 and forthcoming; Neidorf 2017: 73–101). This supports that Bláins is the oldest form and that it is irrelevant to the evaluation of contamination.

Second, the hypothesis of contamination is difficult to reconcile with the orthography of the two attestations: ‘blaens’ (W; AM 242 fol. 7r 20–21) versus ‘blains’ (H; AM 544 20r 17).²² Like other fourteenth-century scribes, the W scribe normally uses i for the upper front vowel in unstressed positions, and it is unlikely that he would deviate from his own practice if he consulted a manuscript that adhered to it, or even if he just remembered having written the text there.²³

The spelling with i became the norm in the course of the thirteenth century, and ‘blaens’ thus presumably goes back to an early point in transmission (Hreinn Benediktsson 1965: 72). It seems likely that the W scribe abandoned his own orthographic norm and followed that of his exemplar because he was uncertain of the phonological makeup of this mythological name. The archaic nature of ‘blaens’ is supported by the fact that the variants blám and bláms suggest minim confusion from an underlying -in-, meaning that the spelling Bláins came to dominate in a period antedating our preserved manuscripts. Johansson has thus overlooked two crucial factors: first, that a shared form from the archetype is not indicative of contamination, and second, that the orthography of W speaks against contamination. Johansson discusses two more variants (byleiz versus byleistz; af versus at) but notes that these are ‘not substantial’ (Johansson 2000: 79–80).

On closer inspection, then, Johansson’s stemma is not compatible with text-critical evaluation. Johansson accepts the assumption that preserved witnesses of the same text are somehow related, but he disregards the importance of innovation, trivialisation, orthography/palaeography, significant versus trivial variants and number of shared variants. When stripped of its tools for evaluation, stemmatics no longer has the power to aid us in the testing of our hypotheses, and it is thus no longer a method. The use of a stemmatic framework without its concomitant method is a marked choice, and it may therefore be useful to explore whether Johansson’s elimination of testing may be connected to some preferred conclusion that is not easily compatible with methodical evaluation. I would suggest that this is indeed the case, and that a bias against early dating and oral tradition has considerable explanatory power.

While disregard for text-critical evaluation would be compatible with any conclusion, Johansson consistently uses it to argue against separate recordings from oral tradition and for late contamination. He presents no evaluation of the relative strength of his own criteria versus those that suggest three separate recordings or, in the case of H and W, a shared archetype.²⁴ He does not explain this uneven treatment of indications, but his emphasis on the period of the manuscripts is consistent with other statements in the article, as well as those of two later publications. Thus, Johansson notes that other scholars argue that the major differences between the versions are due to oral transmission. He states that the differences may equally well be explained as the result of written redaction, but he does not clarify how (Johansson 2000: 76). With regard to shared similarities, he mentions three possible explanations: derivation from the same written archetype, contamination and reoralisation (Johansson 2000: 71). He does not mention that they may be due to the poem’s oral background. Absence of testing, or even of mention of an obvious logical possibility, is thus consistently associated with statements in favour of written transmission (possibly through the intermediary of reoralisation).

This approach may be compared to that of Johansson’s later article on Vǫluspá, as well as one on Sonatorrek (a skaldic lament, c. 961), where he bypasses oral transmission without any attempt at evaluation, and linguistic and metrical arguments are wholly absent. Instead, Johansson dates the poems based on vague textual analogues and the chronology of preserved prose texts and manuscripts (Johansson 2013 [Vǫluspá]; Johansson 2020 [Sonatorrek]). In both articles, he argues for very late dates, and the possibility of testing his hypothesis against linguistic and metrical criteria is not even mentioned.

The correlation of absence of testing and late dating is thus perfect. Of course, correlation need not imply causation, but it often does. In this instance, testing would have suggested that Johansson’s hypothesis is flawed (features like hiatus and expletive of point to early dates for both poems, and there are no indications of a late date), and absence of testing is a marked choice, since it goes against widely accepted scholarly ideals. These factors support a that the correlation of absence of testing and late dating is due to a preference for the latter. I return to the topic of bias below.

I now turn to another group of scholars and a different aspect of textual criticism; namely, the principle of lectio difficilior potior (‘the more difficult reading is preferable’) (Maas 1960: 11). The assumption is that scribes are more likely to simplify things than to produce additional complexity. They may transform an unknown word into a known one, or they may change it into something they believe is more natural or correct. Unlike previous examples, which are not found in editions and therefore need not affect future scholarship very much, inversions of the principle of lectio difficilior potior have made their way into standard editions of Eddic poetry.²⁵ Thus, the editors of the Kommentar zu den Liedern der Edda (1997–2019) choose the variant aldni over baldni in Vafþrúðismál 32.4–5: hvé sá bǫrn gat | inn aldni (R)/baldni (A) jǫtunn (‘how that old/defiant giant conceived children’).²⁶ They 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 (von See et al., eds. 1997–2019, 1: 989 -90). They do not seem to realise that they are here describing the reason why a scribe would change baldni to aldni. The key point here is that alliteration shows that baldni is the word that fits the metre, while its rarity explains why a scribe would change it.

This example highlights an important feature of textual criticism: apart from the stemmatic or phylogenetic model, which is shared with disciplines such as historical linguistics and evolutionary biology, it is a method, not a theory. Based on the experiences of many philologists, some patterns of scribal behaviour have been identified, but the standard approach remains probabilistic and devoid of dogma. In this instance, personal experience is relevant, since one may assume that all individual philologists know that they themselves may sometimes be prone to adapt quotations to their own idiom. It is therefore reasonable to assume that medieval copyists occasionally made similar adaptations. This assumption would speak in favour of the variant baldni, which may then be tested against the independent parameter of alliteration. In the fornyrðislag corpus, Seiichi Suzuki counts twenty verses missing alliteration in a total of 5.998 verses. Unfortunately, Suzuki gives no number for ljóðaháttr, the metre of Vafþrúðnismál, but the fornyrðislag corpus gives a general indication of the frequency of missing alliteration. Six of the twenty examples are obviously missing their sibling verse, since the stanza or half-stanza is an anacoluthon, and one is damaged (Suzuki 2014: 335–36). With a percentage of roughly 0.3%, lack of alliteration is clearly unmetrical, even without consideration of missing or damaged verses. The abnormal surroundings of verses missing alliteration corroborates this assumption, since there is no obvious explanation for this correlation except corruption. These observations speak strongly in favour of the variant baldni. The Kommentar’s reading of Vafþrúðismál 32.4–5 thus amounts to a renunciation of probabilistic reasoning generally, rather than of textual criticism specifically.

At other points, the editors’ disregard for probabilistic restrictions correlates with a bias for late dating. Numerous examples are analysed elsewhere, and one may suffice here (Males 2022: 9–15). I discussed earlier the considerable improbabilities involved in positing that the form reka, rather than the older form vreka, is original in Hávamál and Vafþrúðnismál. The editors are nonetheless open to this possibility, but their main efforts are spent in arguing that with or without v-, the form is young (von See et al., eds. 1997–2019, 1: 574–575, 1170–1171). In other words, the editors accept two improbabilities; namely, that the poets would break a rule that they otherwise never break or that an early form is in fact late. The only option that they do not accept is the one that explains all observable data; namely, that the form is early.²⁷

The Kommentar shares the combination of disregard for formal criteria and a bias for late dates with Johansson. This approach has become common in scholarship on ON poetry from the 1960s onwards (Males 2020: 215 n. 82, 252 n. 216). This leads us to the problem of bias. Since scholars are necessarily biased to some degree – if for no other reason, then at least due to previous experiences that may or may not be applicable to the question at hand – the task of scholarly knowledge production relies on probabilistic restrictions. Scholarly methods, such as those of textual criticism, provide us with the blueprints. Disregard for such methods therefore means that scholars cannot perform their task of providing the answers that are most likely to be correct. Although this article focuses on the limited field of ON philology, the larger issue at stake is nothing less than the most obvious social justification of academic institutions: to conserve, produce, improve and provide knowledge. I therefore believe that the implications of the topics treated in this article are of greater and more general importance than might first meet the eye.

Favourable factors

The picture that emerges from the preceding analyses suggests that during recent decades, it has become acceptable to attribute a variety of undesirable qualities to textual criticism, including some that are diametrically opposed to it, without any attempt at documentation. Such a discourse is not conducive to serious debate, and it is likely to elicit a degree of animosity in scholars who find their aims or methods described in this manner. It may therefore be useful to consider whether philologists ought to extend the ideals of probability and intersubjective scrutiny not only to their sources, but also to their descriptions of fellow scholars.

On the side of applied philology, we have seen disregard for grammar, metre, language history, innovation, trivialisation, orthography, palaeography, significant versus trivial variants and number of shared variants, in cases where these parameters would have been crucial for the evaluation of competing hypotheses. By eliminating the most obvious means of testing, publications like these pose a challenge to the epistemological credibility of the field, especially when produced by senior scholars. For philology, which traditionally boasts a ‘scientific’ reputation, this is an unfortunate development, and it invites us to explore ways to recalibrate the scholarly discourse. This is desirable not only for scholars who focus on the original production of texts, but also for those who take an interest in transmission and reception, as well as literary interpretation. New Criticism never made its entry into ON studies, and medieval philologists generally remain committed to some degree of historical contingency. This being the case, ON scholars in general stand to benefit from a methodological basis for distinguishing author from scribe and errors or mistakes from intended readings. In the following, I point to some factors that may contribute to a balanced future debate.

Earlier generations of philologists tended to take basic philological training and a probabilistic outlook for granted. They did not think that they needed to point out, for instance, that early linguistic forms are likely signs of high age. As a result, good overviews of the epistemological underpinnings of medieval philology have been lacking for a long time. Only when faced with unprecedented rates of non-probabilistic arguments did some scholars begin to write such overviews. Thus, for instance, R. D. Fulk emphasises that in philology, the linguistic nature of the sources usually makes the reliability of different alternatives quantifiable, whether the scholar chooses to quantify them or not, and that this provides a sound probabilistic basis for a competition between hypotheses (Fulk 2003: 16). Editorial choices and the dating of texts should therefore be informed by likelihood, and the main obstacle to a sound progression of philological research is the tendency to treat competing hypotheses as equal options from which the scholar may choose at will. In order to contribute to knowledge production, the scholar must aim to make the best-informed choice possible, to the exclusion of other explanations (Fulk 1992: 7–24; 1996: 1–24; similarly Neidorf 2015).

R. D. Fulk and Leonard Neidorf have written useful overviews of this kind, focusing on the closely related discipline of Old English philology (Fulk 1996; Fulk 2003; Neidorf 2015). In ON, Haukur Þorgeirsson has revisited central topics of both stemmatic reconstruction and emendation, and his articles serve as accessible introductions (Haukur Þorgeirsson 2017a; 2020). For philology more generally, Roelli, ed. (2020) is a landmark, since it does not require the same degree of expertise in classical philology as Maas (1960) or West (1973). It is thus now easier to acquire a sound methodological basis also for philologists with limited background in Classics. In this article, I have focused on problems engendered by disregard for text-critical method and the need for a probabilistic common ground in order to maintain a debate where subjectivity is kept in a responsible relation to observation. I also hope to have shown that there is nothing mysterious about the methods of textual criticism.

This point may be elaborated further. First, even the most abstract component of textual criticism, that of reconstructed manuscript stemmas, is readily understandable by reference to known analogues from biology: it is the features that we share with the greater apes, and not those that both of us share with fish, that show that we and the greater apes are more closely related than any of us are to fish. The same principle of shared innovations allows scholars to posit manuscript stemmas. Second, the distinction between trivial and significant innovations is easy to grasp without recourse to analogy. Logic dictates that it is vastly more probable that two scribes might independently change an ‘and’ to an ‘or’ than that they would both spontaneously change a string of words in the same way. Third, with regard to the principle of lectio difficilior potior, every literate person knows that they sometimes simplify things by mistake, or even just for economy. It is inherently plausible that medieval scribes occasionally behaved in the same way. Fourth, the concept of minim confusion is eminently understandable, for instance by considering the word minim itself. Medieval scribes did not dot the is,²⁸ and connecting strokes were often faint and narrow, so that this word would consist of ten vertical strokes of the same length, allowing for any number of interpretations.

As these examples show, the fundamentals of textual criticism are easily understood, and their validity is susceptible to testing in various ways. Nonetheless, many scholars of ON philology reject these assumptions in part or in full, either tacitly or by reference to yet other notions that are not held by textual critics. This is a challenging discourse to navigate even to the experienced scholar, and I have therefore opted to analyse a selection of misrepresentations of text-critical method, as well as some examples of the consequences of disregarding its principles. My reason for doing so is simple: an article of this kind would have been useful for me in my formative years as a student of ON philology. In the event that this may be true also of others, I hope that the present article may promote an awareness of the variety of epistemological underpinnings of ON philologists, as well as providing some input for methodological reflection.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Notes on contributors

Mikael Males

Mikael Males is professor of Old Norse philology at the University of Oslo. He specialises in Old Norse poetry and grammatical literature, as well as the interface with the Latin and Irish traditions.

Notes

¹ Cf. Maas (1960: 11): ‘Die Begründung der durch divinatio (oder selectio) vorausgesetzten Fehler spielt in der Textkritik eine beträchtliche Rolle, aber immer eine sekundäre’.

² Errors are included under what Maas calls ‘anomalies’, whereas hapax legomena, for instance, correspond to his ‘singularities’. The latter are not in themselves suspicious (Maas 1960: 10–11).

³ This is evident in the analysis of Quinn (2001; 2016a), Rösli and Gropper (2021a) and von See et al., eds. 1997–2019 below, and the tendency is common in publications associated with New/Material Philology.

⁴ An example of normative use is discussed below (Rösli & Gropper 2021a: 12; similarly Quinn 2016b: 72).

⁵ Both Pasquali and West conceived of their contributions as complementary to Maas, not as replacing him. With Roelli (ed.), the digital dimension has been added to the methodological discussion. These four books thus give a good overview of the advances within stemmatics during the last century.

⁶ See Fulk (1997: 43–44) for Old English; Quinn (2001: 75–78) for Old Norse (the rich editorial tradition of Vǫluspá being an excellent test case).

⁷ Maas (1960: 5).

⁸ Alternatively, the mere possibility of change is used to discourage reconstruction and dating. Cf. the following statement: ‘One may gain more by asking in what ways the poems may or may not be considered old, and from whose perspective they were regarded as old, than by searching for the lost moment when a poet performed something which eventually became the extant text.’ (Thorvaldsen 2016: 90).

⁹ A good overview of the debate on scribal versus authorial focus from Traube to the twenty-first century is Stock (2015).

¹⁰ Fisher (2012) is a case in point. Fisher studies exceptional scribes who combined the roles of scribe and author, but in his introduction, he extends this mixture of agencies to medieval scribes in general.

¹¹ For an evaluation of occurrences of alliteration in vr- throughout the corpus, see Haukur Þorgeirsson (2017b).

¹² For stemmatics, see e.g. Hall and Parsons (2013); Haukur Þorgeirsson (2017a). Other aspects have mostly been explored for the purpose of dating poetry, e.g. Haukur Þorgeirsson (2017b; 2020); Myrvoll (2020).

¹³ The traditional terms are trivial and significant, but Trovato prefers polygenetic and monogenetic (Trovato 2014: 54–56).

¹⁴ Interestingly, Haugen finds that ON editions have roughly the same percentage of bifurcation (83%) as Old French ones (82,5%), which is somewhat higher than what one might expect on statistical grounds (roughly 60–70%) (Haugen 2015: 607). The reason is presumably what Haugen calls ‘the force of dichotomy’, meaning the editors’ aim to find variants that distinguish one manuscript from another (Haugen 2015: 608). The dynamics of the procedure may thus lead to a slight exaggeration of two-branched nodes, but this is not a considerable obstacle to the method. Rather, it is a reminder that the stemma is a working tool which should not be taken as an exact description of reality.

¹⁵ In fact, the probabilistic method of science could be reconciled with Karl Popper’s use of the term objective to mean ‘intersubjective’, but Driscoll describes an absence of possibilities for intersubjective scrutiny (Popper 1972). He cannot, therefore, be using the term in Popper’s sense.

¹⁶ Aspects of the following discussion of Quinn have previously been presented by Haukur Þorgeirsson (2020: 49–50).

¹⁷ W was likely compiled at Þingeyrar. See Jakob Benediktsson (ed. 1980: 10–12); Stefán Karlsson (1982: 320–22); Guðbjörg Kristjánsdóttir (1983): Johansson (1997: 10–16).

¹⁸ Johansson writes that he ‘cannot accept’ that two scribes would independently make ‘exactly the same mistake’ (‘Det skall genast sägas att jag inte kan acceptera att två skrivare, oberoende av varandra eller en förlaga, gör exakt samma fel i de två versioner av samma dikt som av slumpen har bevarats till vår tid’; Johansson 2000: 70). He does not explain why he cannot accept such mistakes, which are common and constitute so-called trivial errors, and he does not at this point mention the fact that only part of the innovation is shared.

¹⁹ Johansson writes that there is ‘no hint’ (ingen antydan) that stanza 8 has influenced the R scribe (Johansson 2000: 69). On the contrary, the fact that the scribe changes the first verse to become identical to 8.6, even though it then becomes ungrammatical, is a strong indication to this effect.

²⁰ Johansson (2000: 77). For references on W’s hand in H, see Johansson (1997: 162).

²¹ Johansson states that the form Bláins suggests a ‘a common point of origin’ (gemensam utgångspunkt; Johansson 2000: 79), but this is true both of readings from the archetype and significant errors and is thus irrelevant to his hypothesis.

²² Similarly W’s ‘þroenn’ for H’s ‘þrainn’ (AM 242 fol. 7r 24; AM 544 20r 21).

²³ On 6v–7r, I count 110 spellings with i versus nine with e. Three of the nine are found in mythological names (‘fenres’, ‘blaens’, ‘þroenn’), where scribes would often be prone to follow their exemplar.

²⁴ Johansson does not describe the arguments in favour of separate recordings in any detail (Johansson 2000: 64, 69).

²⁵ This statement can be qualified. Heimir Pálsson (ed.) and Faulkes (trans. 2012: 82) emend sigrfoldinnar to sigrfold<ar>innar, but theirs is an edition of U only. Still, the emendation is based on an inflated view of scribal intentionality and competence, presupposing that the scribe had a meaningful form in mind, even though he did not produce such a form and minim confusion is the obvious explanation. If the editors wanted to correct, they should rather have changed sigrfoldinnar to sígr fold í mar, which not only has the support of the other manuscripts, but also requires a less invasive emendation (nn > m). On inflated views of the agency of copyists, see Neidorf (2017: 103–32).

²⁶ Since R and A are the only two manuscripts of independent text-critical value, the choice between variants must be based on non-stemmatic criteria. On the archetype of R and A, see Lindblad (1980). The choice between two stemmatically equivalent variants is called selectio in Maas’ terminology (Maas 1960: 11).

²⁷ The editors’ rejection of this possibility may be corroborated by consulting their discussion of the dates of the poems, where the vr-forms are not mentioned, although these are the strongest formal dating criteria in both poems (von See et al., eds. 1997–2019, 1: 496, 999).

²⁸ Diagonal strokes over the is are sometimes found, however, presumably to avoid minim confusion.