Vol.:(0123456789) International Journal of the Classical Tradition (2025) 32(4):383–410 https://doi.org/10.1007/s12138-025-00716-1 ARTICLE From the Aeneid to Accent Theory: The Application of Classical Learning in the Medieval Icelandic Third Grammatical Treatise Sólveig Hilmarsdóttir1   · Haukur Þorgeirsson2  Accepted: 27 July 2025 / Published online: 1 September 2025 © The Author(s) 2025 Abstract This article has two aims. First, we describe a new analysis of the medieval man- uscripts of the Third Grammatical Treatise, a thirteenth-century grammatical-rhe- torical work  by Óláfr Þórðarson focused on phonology and poetic devices, partly undertaken using new discoveries of close Latin models to evaluate Old Norse read- ings. Second, the article focuses on the challenges Óláfr faced in making use of his classical sources and the originality with which he resolved them; previous accounts have often downplayed his originality and inventiveness. We attempt to highlight aspects of Óláfr’s work that appear to result from independent contributions and we further undertake a close reading of Óláfr’s application of classical accent theory to Old Norse. Introduction Medieval Iceland is known for its robust vernacular literary tradition, with roots in oral poetry and storytelling. Nonetheless, it has long been acknowledged that this literary culture owes much to learned writings in Latin. The English philologist Gabriel Turville-Petre famously argued that ‘The learned literature did not teach the Icelanders what to think or what to say, but it taught them how to say it’. He went on * Sólveig Hilmarsdóttir shh35@cam.ac.uk Haukur Þorgeirsson haukur.thorgeirsson@arnastofnun.is 1 Faculty of Classics, University of Cambridge, Cambridge, UK 2 Árni Magnússon Institute for Icelandic Studies, Reykjavík, Iceland http://crossmark.crossref.org/dialog/?doi=10.1007/s12138-025-00716-1&domain=pdf http://orcid.org/0009-0006-3823-0049 http://orcid.org/0000-0002-7872-0601 384 S. Hilmarsdóttir, H. Þorgeirsson to note that the style of the Icelandic sagas showed the benefits of ‘several genera- tions of Icelanders … trained in hagiographic narrative’.1 Spotting similarities between Norse and Latin literature was also a cherished sport for medieval Icelanders. Nowhere is this clearer than in the Third Grammati- cal Treatise, an Old Norse grammatical-rhetorical work written by Óláfr Þórðarson hvítaskáld, ‘the White Poet’, near the middle of the thirteenth century.2 Óláfr explic- itly proclaimed the similarity or even unity of classical and Norse literature. A part of the basis for this idea was the popular medieval doctrine of common descent from Troy. In the Icelandic version of this idea, Óðinn and the other Norse gods, the Æsir, are seen euhemeristically as men of Asia who brought Trojan civilization to the North.3 As Óláfr phrases it: Í þessi bók má gerla skilja at ǫll er ein listin, skáldskapr sá er rómverskir spekingar námu í Athenisborg á Griklandi ok sneru síðan í látínumál, ok sá ljóðaháttr eða skáldskapr er Óðinn ok aðrir Ásíámenn fluttu norðr hingat í norðrhálfu heimsins ok kenndu mǫnnum á sína tungu þess konar list, svá sem þeir hǫfðu skipat ok numit í sjálfu Ásíálandi, þar sem mest var fegrð ok rík- dómr ok fróðleikr veraldarinnar.4 It can be clearly understood from this book [Ars maior by Donatus] that the art of poetry which the Roman scholars learnt in the city of Athens in Greece and then translated into the Latin language is the same as the way of verse-making or poetry which Óðinn and other men of Asia brought here to the northern region of the world, and in their language they taught this art to people as they had arranged and learned it in the very land of Asia, where beauty and wealth and knowledge were the greatest in all the world. The Third Grammatical Treatise is divided into two sections, the first treats phonology and the second explains poetic devices. The first section is, in theory, based on the first two books of Priscian’s Institutiones Grammaticae whilst the sec- ond draws on the third book of Donatus’s Ars maior. However, it is evident that the grammatical works of Priscian and Donatus have been complemented and mediated by other, more recent, Latin grammatical material. In the past few decades, con- siderable scholarly attention has been directed towards identifying those later Latin sources. Most recently, evidence has come to light indicating that Donatus’ work 1  G. Turville-Petre, Origins of Icelandic Literature, Oxford, 1953, p. 142. For a more detailed discus- sion of the influence of hagiographical and ecclesiastical literature on Norse literature, see J. Wellendorf ‘Ecclesiastical Literature and Hagiography’, in The Routledge Research Companion to the Medieval Ice- landic Sagas, ed. Á. Jakobsson and S. Jakobsson, London, 2017, pp. 48–58. 2  M. Males, The Poetic Genesis of Old Icelandic Literature, Berlin, 2020, pp. 178–92; M. Clunies Ross, A History of Old Norse Poetry and Poetics, Cambridge, 2005, pp. 152–3, 186. 3  The theory of the Æsir arriving from Asia to Scandinavia was formulated earlier by Óláfr’s uncle, Snorri Sturluson, in his works Edda and Heimskringla. 4  In this and other quotations of the Third Grammatical Treatise, we follow the text of the A manuscript (AM 748 I b 4to) except where otherwise noted. The text is presented in normalized Old Norse spelling. We reference Finnur Jónsson’s edition for the reader’s convenience in locating the context; Finnur Jóns- son (ed.), Óláfr Þórðarson: Málhljóða- og málskrúðsrit, Copenhagen, 1927, p. 39. 385From the Aeneid to Accent Theory was transmitted to Óláfr through the commentary Ars Brugensis, a work in the insu- lar tradition. This article has two aims. First, we describe a new analysis of the medieval man- uscripts of the Third Grammatical Treatise, partly undertaken using new discoveries of close Latin models to evaluate Old Norse readings. Second, the article focuses on the challenges Óláfr faced in making use of his sources and the originality with which he resolved them; previous accounts have often downplayed his originality and inventiveness. The Latin sources illustrate rhetorical and grammatical concepts with quotations of verses from the Aeneid and other classical works. Óláfr, however, puts forward a series of Norse poetic quotations to exemplify these concepts, often selecting interesting stanzas for comment. Old Norse poetry is a fertile field for the application of stylistic terminology because of its highly developed poetic language with its own vocabulary and syntax. We provide further comparisons between the Old Norse poetic examples and the original Latin examples to throw light on Ólá- fr’s working methods. Another particularly challenging aspect is Óláfr’s applica- tion of classical accent theory to Old Norse. The accents were originally developed to describe the pitch accent of classical Greek, but the system was later applied to Latin as well. Since Latin did not have distinctive pitch patterns in the manner of Greek, the description as applied to Latin is more abstract than concrete. There are good reasons to think, however, that Old Norse did have distinctive pitch patterns so by applying the accent system to Old Norse there was an opportunity to make good use of the theory. Since Old Norse pitch differed from that of Greek, the interpreta- tion of the text is not straightforward, but we argue that the Third Grammatical Trea- tise does contain real information on Old Norse accentuation. We set the scene by first establishing some facts about the Third Grammatical Treatise, its manuscripts and its known sources. The Third Grammatical Treatise The principal works of the Old Icelandic grammatical tradition have no titles in the extant manuscripts. The naming convention used by modern scholarship is to refer to them by their sequence in the fourteenth century Codex Wormianus, a manuscript which preserves four grammatical treatises. The Third Grammatical Treatise is, thus, the third out of the four in that manuscript. Its text is also preserved in three other medieval manuscripts. The manuscripts are as follows: A: AM 748 Ib 4to, Reykjavik. Dated to 1300–1325. W (Codex Wormianus): AM 242 fol., Copenhagen. Dated to 1350–1370. B: AM 757a 4to, Reykjavik. Dated to c. 1400. Contains an abridged text. w: AM 757b 4to, Reykjavik. Dated to c. 1450–1500. Fragmentary copy of W.5 5  For a more detailed discussion of the manuscripts, see T. Wills, ‘The Foundation of Grammar: An edi- tion of the first section of Óláfr Þórðarson’s grammatical treatise’, PhD diss., University of Sydney, 2001, pp. 41–52. The online catalogue of Icelandic manuscripts at handrit.is has detailed listings for each of these manuscripts along with photographs in colour for all but w. 386 S. Hilmarsdóttir, H. Þorgeirsson This number of manuscripts may seem modest, but it is higher than for the other Icelandic grammatical treatises, suggesting that the Third Grammatical Treatise enjoyed some popularity.6 Aspiring scholars presumably found it useful. Further- more, Sturlunga saga indicates that Óláfr Þórðarson took on students and it seems plausible that the Third Grammatical Treatise was part of the material he taught— perhaps it continued to be taught after his death.7 A recent attempt to evaluate the survival rate of medieval documents estimated that seventeen percent of Icelandic medieval manuscripts have survived.8 Therefore, it is likely that the four surviving copies of the Third Grammatical Treatise are only a fraction of the number that once existed. The Third Grammatical Treatise begins with a theoretical discussion of the nature of sound, from which it moves to describe the pronunciation of letters and syllables. Óláfr quotes Priscian on the multiple sounds which can be represented by a single vowel letter and then moves on to analyse the runic alphabet and compare its arrangement with that of the Latin alphabet. This is followed by further discus- sion on syllable theory, with a note that whilst Latin syllables can be formed of up to six letters, Old Norse syllables can have as many as eight or nine. The discussion of phonetics proceeds with material on syllable length, aspiration and accent. The grammatical section of the treatise is then concluded with an overview of the parts of speech. This first section has sometimes been considered more ‘theoretically interesting’, and it certainly contains some quite independent contributions by Óláfr, such as his analysis of the runic alphabet.9 However, the rhetorical part of the treatise is also of great interest. It begins with a reference to Donatus and the claim (quoted above) of affinity between the Old Norse art of poetry and that of the Greek and Latin tra- dition. The author then substantiates this thesis by working through the rhetorical figures described by Donatus in Barbarismus and Solecismus and providing Norse poetic examples for each figure. Many of the vernacular quotations he uses are not known from elsewhere, and some may be his own compositions.10 6  Wills, ‘The Foundation of Grammar’ (n. 5 above), p. 41. 7  For Óláfr’s life, see Björn Magnússon Ólsen, Den tredje og fjærde grammatiske afhandling i Snorres Edda, Copenhagen, 1884, pp. xxxii–xxxvii. 8  M. Kestemont et al. ‘Forgotten books: The Application of Unseen Species Models to the Survival of Culture’, Science 375, 2022, pp. 765–9. 9  Clunies Ross, A History of Old Norse Poetry and Poetics (n. 2 above), p. 187; Wills, ‘The Foundation of Grammar’ (n. 5 above), p. 2. 10  Clunies Ross, A History of Old Norse Poetry and Poetics (n. 2 above), pp. 152–3. 387From the Aeneid to Accent Theory Previous Textual Criticism The Third Grammatical Treatise has been published in its entirety in six editions. The first full edition was published by the Danish scholar and philologist Rasmus Rask in 1818, who based his edition only on the Codex Wormianus.11 Sveinbjörn Egilsson prepared the second edition in 1848, identifying A as the best manuscript but using readings from W and B where A is defective.12 The Arnamagnæan edition of 1852 likewise makes some use of all the manuscripts but uses the Codex Wor- mianus text as its base.13 Björn M. Ólsen’s 1884 edition had a more sophisticated approach to textual criticism. Ólsen was the first editor to produce a stemma for the relationship between A, B and W—arguing that A and W share common errors against B.14 Ólsen also identified many correspondences between the treatise and the texts of Donatus, Priscian and other Latin grammatical texts, often relying on Keil’s Grammatici Latini and Thurot’s Extraits.15 The experienced philologist Finnur Jóns- son prepared another edition in 1927. He described A as unquestionably the best manuscript. Whilst declining to comment explicitly on the relationship between the manuscripts, he noted that where B and W agree against A, the majority reading can be assumed to be closer to the original.16 The edition by Krömmelbein in 1998 is based on Ólsen’s work.17 The most recent edition is that of Tarrin Wills in 2001. Although it only covers the first half of the treatise, it is a novel work, based on a fresh examination of the manuscripts. Importantly, Wills corrects some misreadings made by previous scholars. He also presents a new stemma of the manuscripts, in which B is placed on a branch with A rather than W.18 However, our own examina- tion of the text leads us to propose a stemma different both from that of Ólsen and from that of Wills. This is discussed further in section five but first we outline the different content of the manuscripts. The Content of the Manuscripts Although there are four medieval manuscripts which contain some of the text of the Third Grammatical Treatise (A, B, W and w), no manuscript preserves the entire text. The segments contained in the four manuscripts are visualized in Fig. 1. 11  R. Kr. Rask (ed.), Snorra-Edda ásamt Skáldu og þarmeð fylgjandi ritgjörðum, Stockholm, 1818, pp. 297–335. 12  Sveinbjörn Egilsson (ed.), Edda Snorra Sturlusonar, Reykjavik, 1848, pp. 173–200. 13  Edda Snorra Sturlusonar – Edda Snorronis Sturlæi: Tomus secundus, continens tractatus philologicos et additamenta, Copenhagen, 1852, pp. 501–11. 14  Ólsen, Den tredje (n. 7 above), p. lxii. 15  H. Keil. Grammatici Latini, Leipzig, 1855-1880 and C. Thurot, Extraits de divers manuscrits latins pour servir à l᾽histoire des doctrines grammaticales au moyen âge, Paris, 1868. 16  Finnur Jónsson, Málhljóða- og málskrúðsrit (n. 4 above), pp. 18–19. 17  T. Krömmelbein, Óláfr Þórðarson hvítaskáld: Dritte Grammatische Abhandlung, Oslo, 1998; Wills, ‘The Foundation of Grammar’ (n. 5 above). There is also an unpublished partial edition by L. G. Coll- ings, ‘The Málskrúðsfræði and the Latin Tradition in Iceland’, MA thesis, Cornell University, 1967. 18  Wills, ‘The Foundation of Grammar’ (n. 5 above), pp. 52–6. 388 S. Hilmarsdóttir, H. Þorgeirsson Fig. 1   The four manuscripts of the Third Grammatical Treatise and their lacunae 389From the Aeneid to Accent Theory The most complete manuscript is the Codex Wormianus (W) and whilst nothing has been lost from W itself, its exemplar was either missing one leaf or the scribe of W accidentally skipped over a leaf. Comparison with the other sources allows us to estimate that W contains 92% of the text of the treatise. The A manuscript contains ca. 80% of text of the treatise but is missing two leaves. Unfortunately, the second missing leaf of A partly coincides with the part missing from W. No other manuscript covers this section, so a part of the treatise is lost. Since a leaf in A contains approximately 950 words and the text of W covers some 540 words of the second gap in A, we estimate that c. 410 words are lost or 4% of the text. The preserved text as a whole is c. 9070 words. No leaves have been lost from the B manuscript, but its text is abridged, and it only contains 31% of the treatise. The runology sections are omitted, the introduc- tion to Barbarismus is abridged and Solecismus is completely omitted. Further, the B manuscript is quite worn and now difficult to read. To assist with difficult pas- sages, there is some help in an eighteenth-century transcript, AM 744 4to (Reykja- vik), as well as the nineteenth-century diplomatic edition.19 Neither, however, is entirely accurate. The fragment w consists of merely two leaves—the first leaf of the treatise and what must have originally been the fourth leaf. These total some 6% of the text. Nothing else remains of w and we do not know whether it once contained other material as well or only the Third Grammatical Treatise. The manuscripts A, B and W contain many texts besides the Third Grammati- cal Treatise. They all have a version of the early thirteenth-century Prose Edda by Snorri Sturluson. Each manuscript has its own redaction, but commonalities have been noted between the versions in A and B.20 All three manuscripts also contain some poetry. Originally, A included a collection of mythological Eddic poems, and this sec- tion is now preserved separately as AM 748 I a 4to in Copenhagen. The manuscript B also contains poems but of a very different sort, a selection of Christian religious poetry. The compilers of these manuscripts must have seen the Third Grammatical Treatise as relevant for understanding any poetic work, and the treatise itself does quote poetry by both pagan and Christian poets. 19  Edda Snorra Sturlusonar (n. 13 above), pp. 501–11; Wills, ‘The Foundation of Grammar’ (n. 5 above), p. 52. 20  A. Faulkes (ed.), Snorri Sturluson: Edda, Prologue and Gylfaginning, Viking Society for Northern Research, London, 2005, p. xxx. 390 S. Hilmarsdóttir, H. Þorgeirsson The Relationship Between the Manuscripts Previous studies unanimously consider the text of the w fragment to be copied from W and this is our conclusion as well.21 There are 18 instances where w and W share a read- ing against A and B and the single instance where w and A share a reading against B and W consists of the common omission of a preposition, something which can easily happen twice by coincidence. The relationship between A, B and W is a more difficult question but can be approached using the philological method of identifying scribal errors or secondary read- ings. We are well placed to identify secondary readings in the manuscripts since a part of the text is a close translation of known Latin sources. Further, the quotations of Norse poetry in the treatise provide another way to evaluate readings since some of the stanzas which are cited are also preserved in other sources. A third way to identify errors is to look for traditional categories of scribal mistakes such as saut du même au même, where a scribe jumps from one instance of a word to another, or from a word to a similar word, whilst leaving out the intervening text. We have compared the part of the treatise which is preserved in all of A, B and W using the means described above to identify instances where one manuscript has a sec- ondary reading whilst the other two have a more original reading or where two manu- scripts share a secondary reading against a more original reading in the third manuscript. As an example, let us consider the beginning of Barbarismus which defines the subject matter. The Latin text here is well known: Barbarismus est una pars orationis vitiosa in communi sermone. Barbarism is one faulty part of speech in common language. The A and B manuscripts have the same text here, one which is a close transla- tion of the Latin: Barbarismus er kallaðr einn lastafullr hlutr málsgreinar í alþýðligri ræðu. Barbarism is called one faulty part of a sentence in common speech. The W manuscript has a text which differs in two ways from this: Barbarismus er kallaðr einn lastapollr málsgreinar í alþýðumáli. Barbarism is called a puddle of vices of a sentence in the general opinion. We can thus identify two errors in W. The pedantically accurate Latinate phrasing í alþýðligri ræðu (‘in common speech’) found in A and B has been replaced with the simi- lar sounding í alþýðumáli (‘in the general opinion’), which does not mean quite the right thing, as alþýðumál does not elsewhere in Old Norse refer to ‘speech’ as such. Further, the word fullr (‘full’) has been misread as another similarly written word, pollr (‘puddle’) and the definition has become something of a mess. 21  ‘Det indeholder alle de samme fejl som W og er utvivlsomt en afskrift af denne membran’, Ólsen, Den tredje (n. 7 above), p. liv; Wills, ‘The Foundation of Grammar’ (n. 5 above), p. 52; ‘det er en ren afskrift af W’, Finnur Jónsson, Málhljóða- og málskrúðsrit (n. 4 above), p. 18. 391From the Aeneid to Accent Theory Examining the text found in all three manuscripts in this way, we have identified 47 plausible errors. Their distribution is shown in Table 1. To establish a stemmatic relationship between two manuscripts we would need a pattern of shared errors but it is clear that the vast majority of identifiable errors are unique to one of the three manuscripts. The plausible error shared by A and B consists of the common absence of an inflectional m at the end of the word mínum ‘mine’ in a poetic quotation.22 This can very easily have arisen twice. The plausible error shared by A and W is more interesting. In a description of the parts of speech, B is the only manuscript to place the pronoun after the noun, which fits with the text of Donatus and also with the order used shortly before in the Third Grammatical Treatise itself.23 Ólsen saw this as especially compelling evidence for placing A and W together in a stemma.24 It is not, however, decisive. The author of the B text can often be caught intentionally editing the text, and sometimes he genuinely succeeds in improving its flow or sense. In this case, he may well have harmonized the order of the list on his own accord. Ólsen’s other possible examples of common errors in A and W are much less compelling and were not flagged as such in our analysis. In the absence of a clear pattern of common errors, we propose a flat stemma of the manuscripts, as in Figure 2, rather than a binary stemma as proposed by previous scholars. The implication of the flat stemma in Figure 2 is that wherever one manuscript has a unique reading against the two others, the majority reading is the one most likely to reflect the original text. We took this into account in our study of particular readings. We will now move on to discuss the other focus of our article, namely Óláfr’s contributions and how he dealt with pre-existing Latin models. Table 1   Plausible errors in manuscripts A, B and W Manuscript Plausible errors Unique to A 13 Unique to B 9 Unique to W 23 Shared by A and B 1 Shared by A and W 1 Shared by B and W 0 22  Finnur Jónsson, Málhljóða- og málskrúðsrit (n. 4 above), p. 42. 23  Finnur Jónsson, Málhljóða- og málskrúðsrit (n. 4 above), p. 37. 24  Ólsen, Den tredje (n. 7 above), p. lxii. 392 S. Hilmarsdóttir, H. Þorgeirsson Identifying the Latin Sources of the Third Grammatical Treatise After a period of low scholarly interest for several decades, research on the Third Grammatical Treatise picked up again in the 1980s. Significant attention has been devoted to the treatise’s connection to the Latin grammatical tradition and its poten- tial immediate Latin sources.25 Although Priscian and Donatus are the only Latin grammarians mentioned by name in any of the four medieval Icelandic grammatical treatises, they are clearly not the only sources Óláfr uses for the Third Grammatical Treatise, or even his most immediate models. Óláfr seems to access much of their learning through other medieval intermediaries.26 This is clear, for instance in sec- tion 6.4, where Óláfr departs from Priscian’s description of the parts of speech in favour of Donatus’s scheme, but still cites Priscian as his source: En meistari Priscianus telr átta parta málsgreinar þessa. But master Priscian enumerates these eight parts of speech.27 25  An overview of most of the research conducted up until 2005 can be seen in F. Raschellà, ‘Old Icelan- dic Grammatical Literature: The Last Two Decades of Research (1983–2005)’, in Learning and Under- standing in the Old Norse World: Essays in Honour of Margaret Clunies Ross, ed. J. Quinn, K. Heslop and T. Wills, Turnhout, 2007, pp. 341–74. Overviews of more recent progress can be seen in M. Males, ‘Pseudo-Remigius and the Old Icelandic Barbarismus: A Pilot Study’, in Reading Slowly. A Festschrift for Jens E. Braarvig, eds. L. Edzard, J. W. Borgland and U. Hüsken, Wiesbaden, 2018, pp. 321-31, V. Frans, ‘Sub regulis Donati: Ars Brugensis och Den tredje grammatiska avhandlingen’, MA thesis, 2019, University of Oslo and in M. Males ‘The Study of Old Norse Grammatical Literature c. 2000–2024’, forthcoming in Medioevo Europeo. 26  F. Raschellà, ‘The Classical Heritage in Old Icelandic Grammatical Literature’, in L´antichità nella cultura europea del medioevo, ed. R. Brusegan and A. Zironi, Greifswald, 1998, pp. 117–26 (120–1). 27  K. E. Gade, ‘Ælfric in Iceland’, in Learning and Understanding in the Old Norse World (n. 25 above), pp. 321–40 (329). Fig. 2   A proposed stemmatic relationship between the four manuscripts of the Third Gram- matical Treatise Archetype A B W w 393From the Aeneid to Accent Theory However, identifying the intermediary sources has been complicated by the fact that Óláfr may have had access to a wide variety of medieval commentaries and treatises. Many of the Latin texts which had a fundamental position on the conti- nent were well known to medieval Icelandic scholars who frequently studied abroad, particularly in England, Germany or France. Further, book inventories in medieval Iceland (albeit from the fourteenth and the fifteenth century, rather than the thir- teenth) refer to the existence of several works which are relevant for the subjects treated in the Third Grammatical Treatise, such as Priscian’s Institutiones grammati- cae, Isidore’s Etymologiae, Alexander de Villa Dei’s Doctrinale and Ebrard of Bét- hune’s Graecismus.28 Óláfr himself spent time in Scandinavia, at the royal courts of Norway, Sweden and Denmark, which might also have provided him with access to material which did not exist in Iceland.29 Our examination of the Third Grammatical Treatise in light of the Latin gram- matical tradition lends support to recent studies which hold that the Ars Brugen- sis, a work belonging to the insular grammatical tradition, was the immediate model for the second half of the treatise. This knowledge has made it possible to exam- ine Óláfr’s practice in more detail. Yet, some aspects of Óláfr’s scholarly practice remain disputed. In particular, the immediate source(s) or models for the first half of the treatise remain unclear, if indeed any exist. We will discuss some of the sug- gestions that have been made and show why a convincing model is still lacking by closely analysing passages on meaningful and non-meaningful voice from Óláfr’s treatise later in this article. First, however, we briefly outline the identification of the Ars Brugensis as the model for the second half of the treatise as this sheds light on Óláfr’s treatment of his Latin sources and we consider the suggestion of direct influ- ence from the Aeneid. The Second Section of the Third Grammatical Treatise and the Ars Brugensis In the 1990s, Micillo made substantial contributions to research on Latin influences on the Third Grammatical Treatise, especially by identifying numerous parallels between the Third Grammatical Treatise and the works of ninth-century Irish grammarians such as Sedulius Scottus, Murethach and the anonymous work Ars Laureshamensis.30 These works form part of the Carolingian grammatical tradition, which consists of grammatical works written in the ninth century, often associated with the Scotti peregrini. Included in this tradition are the works of Murethach, Clemens Scottus, Sedulius Scottus, and the authors of various anonymous works, such as the Ars Laureshamensis, Ars Brugen- sis and Donatus Ortigraphus.31 These are considered to be based upon a common Irish 28  V. Micillo, ‘The Latin Tradition and Icelandic’, in History of the Language Sciences – Geschichte der Sprachwissenschaften, ed. S. Auroux et al., Berlin, 2000, pp. 617–25 (618). 29  Clunies Ross, A History of Old Norse Poetry and Poetics (n. 2 above), p. 186. 30  See V. Micillo, ‘Classical Tradition and Norse Tradition in the “Third Grammatical Treatise”’, Arkiv för nordisk filologi, 108, 1993, pp. 68–79 and ‘Die grammatische Tradition des insularen Mittelalters in Island. Spuren insularer Einflüsse im Dritten grammatischen Traktat’, in Übersetzung, Adaption und Akkulturation im Insularen Mittelalter, ed. E. Poppe and H. L. C. Tristram, Münster, 1999, pp. 215–29. 31  V. Law, ‘Carolingian Grammarians and Theoretical Innovation’, in Diversions of Galway: Papers on the history of Linguistics from ICHoLS V, ed. A. Ahlqvist, Philadelphia, 1992, pp. 27–38 (32). 394 S. Hilmarsdóttir, H. Þorgeirsson source, now lost, but which would have been a commentary on Donatus.32 Additionally, the vernacular Irish treatise Auraicept na nÉces exhibits significant parallels with these manuscripts.33 Wellendorf has also shown convincingly that several sections of the Third Grammatical Treatise seem directly based on the text and structure of the Doctrinale.34 The independent analysis of Frans comes to the same conclusion.35 Interestingly, Ólsen had already cited in a footnote a passage from the insular tradition in his 1884 edition. He simply quoted a section from Thurot, without any further identi- fication. In Thurot, the passage is found in a section about ninth century grammarians. Thurot earlier identified the quoted manuscript as an anonymous Parisian manuscript, Lat. 7491 A, preserved at the Bibliothèque Nationale.36 A closer look at this manuscript reveals that it comes from a text of Murethach’s commentary on Donatus.37 It was not until in a 2018 pilot-study by Males that several matches were identified with a text that can seemingly only have come from the Ars Brugensis, which is a com- mentary on the third book of Donatus’s Ars maior, attributed to Pseudo-Remigius.38 Males based his study on Micillo’s identification of parallels between Óláfr’s treatise and the commentaries of Murethach and Sedulius. Males argued that a convincing model for the Third Grammatical Treatise should be a single commentary or treatise containing all the relevant matches which had been pointed out between the Barbarismus part of Óláfr’s treatise and the insular commentaries. This can be demonstrated for the Ars Brugensis. In 2019, Frans analysed the correspondences between the Third Grammatical Treatise and the Ars Brugensis more closely and produced an edition of the Ars Brugensis, which had previously been unpublished.39 His close analysis supports the claim that the Ars Brugensis is a likely model, and additionally  that structural 32  V. Law, Grammar and Grammarians in the Early Middle Ages, London, 1997, p. 144; J. E. G. Zetzel, Critics, Compilers, and Commentators, Oxford, 2018, p. 359. T. Mari, ‘Pauca de barbarismo collecta de multis: Observations on a Carolingian grammatical treatise’, Revista Signum, 19.1, 2018, pp. 38–55 (53) has further argued that another ninth-century Irish work, Pauca de barbarismo, probably written by Clemens Scottus, is related to these works and in particular to the Ars Laureshamensis. As Pauca de barbarismo seems to be written chronologically prior to the works which have been mentioned, we had some initial hope that it might bear a stronger resemblance to the Third Grammatical Treatise than the Ars Brugensis does. A brief analysis of the edition by T. Mari, Pauca de barbarismo collecta de multis: Studio ed edizione critica, Pisa, 2017, showed that this is not the case. 33  N. E. Engesland, ‘The intellectual background of the earliest Irish grammar’, Journal of Medieval His- tory, 47.4–5, pp. 472–84 (475–6); E. Poppe, ‘Latin Quotations in Auraicept na nÉces: Microtexts and Their Transmission’, in Ireland and Europe in the Early Middle Ages: Texts and Transmission, ed. Próin- séas Ní Chatháin and M. Richter, Dublin, 2002, pp. 296–312. 34  J. Wellendorf, ‘Virtues and vices: The Fourth Grammatical Treatise’, in RE:writing: Medial perspec- tives on textual culture in the Icelandic Middle Ages, ed. K. Heslop and J. Glauser, Zurich, 2018, pp. 243–64. 35  Frans, Sub regulis Donati (n. 25 above). 36  Thurot, Extraits (n. 15 above) pp. 3, 6, 68; at the time of Thurot᾽s writing, this was la Bibliothèque impériale de Paris. 37  This is one of the manuscripts on which Ludovicus Holtz based his edition of Murethach, Murethach (Muridac): In Donati artem maiorem, Turnhout, 1977, and he describes it on p. xliii. The manuscript has been digitized by the BnF: https://​archi​veset​manus​crits.​bnf.​fr/​ark:/​12148/​cc667​23x. 38  Males, ‘Pseudo-Remigius and the Old Icelandic Barbarismus’ (n. 25 above). 39  Males, ‘Pseudo-Remigius and the Old Icelandic Barbarismus’ and Frans, ‘Sub regulis Donati’ (n. 25 above). https://archivesetmanuscrits.bnf.fr/ark:/12148/cc66723x 395From the Aeneid to Accent Theory influences from the Doctrinale can be identified. Frans’s analysis of Óláfr’s method of translation also shows Óláfr’s practice essentially seems to have been to trans- late Donatus rather faithfully, and then to select useful additions from the additional comments in Ars Brugensis freely and selectively.40 Before moving onto discussing the first part of the Third Grammatical Treatise, we briefly consider how Óláfr’s treatment of classical poetic and native examples fits into the rest of his practice. The Second Part of the Third Grammatical Treatise and Quotations from the Aeneid? An important facet of the Third Grammatical Treatise are the native skaldic exam- ples which Óláfr uses to exemplify the classical figures he writes about. These native examples replace the poetic examples traditionally used in Latin rhetorical texts, both in Donatus and other medieval texts, which mostly use short quotations from the Aeneid. Previous scholarship has in specific cases noted striking similarities between the Latin examples and the Norse examples which Óláfr uses. Finnur Jónsson noted in particular the close correspondences found in the cases where the Norse exam- ples are not attributable to any known poet. His suggestion was that these might be Óláfr’s own compositions, deliberately composed for the occasion.41 Wills discusses one such instance in the verse used to exemplify acyrology (i.e. ‘improper’ or ‘illog- ical’ use of language). In Donatus’s explanation, a quote is given from Aeneid 4.19, where Dido, speaking to her sister Anna about Aeneas’s departure, says: Hunc ego si potui tantum sperare dolorem (‘If only I could expect/hope for such great sorrow’). The acyrology or ‘impropriety’ is considered to lie in the word sperare (‘to expect’ or ‘to hope for’), which here is applied to a negative context rather than hoping for positive things. Óláfr gives an analogous Old Norse example: ‘Vætti ek harms nema hitta / hǫfuðgulls náim Fullu’ (‘I expect/hope for sorrow unless I can meet Fulla of the headdress [i.e. the woman]’).42 Here, it seems highly likely that Óláfr had the line from the Aeneid in mind when he selected or composed the Norse lines. How- ever, the question is whether this example is typical of Óláfr’s method or not. Wills has argued that Óláfr’s choice of Norse examples is influenced by the Aeneid exam- ples to a much greater degree than scholars had previously thought.43 Wills’s analy- sis was written and published before the suggestion that a version of Ars Brugensis was the model of the Third Grammatical Treatise, and he focused on resemblances to other insular grammatical works, particularly Sedulius. Additionally, he suggests that Óláfr may have had independent knowledge of the Aeneid. 40  Frans, ‘Sub regulis Donati’ (n. 25 above), pp. 32–3, 45–6, 68–9, 87–8. 41  Finnur Jónsson, Málhljóða- og málskrúðsrit (n. 4 above), pp. 14–15. 42  T. Wills, ‘The Third Grammatical Treatise and the Aeneid’, in La letteratura di istruzione nel medio- evo Germanico: Studi in onore di Fabrizio D. Raschellá, ed. M. Caparrini, M. R. Digilio and F. Ferrari, Barcelona and Rome, 2017, pp. 277–302 (290). Note that referents of the Old Norse kenning system are denoted within square brackets. 43  Wills, ‘The Third Grammatical Treatise and the Aeneid’ (n. 45 above), p. 279. 396 S. Hilmarsdóttir, H. Þorgeirsson Frans has already dealt with several of the correspondences suggested by Wills, pointing out that the quotations provided by Donatus himself usually suffice to explain potential similarities, and that the majority of Óláfr’s skaldic examples are not connected to the verses in the source text. We concur and only add one further illustrative example. This is a correspondence in the description of weapons break- ing between Óláfr and Sedulius seen by Wills, at the unattributed St. 76 and at Aen. 11.483–4. These two passages are used to exemplify schesisonomaton, a rhetorical figure in which an idea is repeated using synonymous expressions: Hlíf gnast við hlíf, hjǫrr við mæki, egg lék við egg, þars jǫfurr barðisk. Shield cracked against shield, sword against sword, edge played against edge where the prince fought.44 armipotens princeps belli tritonia virgo frange manu telum Phrygii praedonis Mighty in arms, mistress in war, Tritonian maid, break with your hand the spear of the Phrygian pirate.45 The exact correspondence which Wills cites of weapons breaking is unjustified. Strictly speaking, the Norse example does not refer to weapons breaking but weap- ons clashing together and producing a sound. Further, differently to the ON example, the schesisonomaton found in the Latin example, armipotens princeps belli (‘mighty in arms, mistress in war’), does not come from the part of the verse which describes weaponry breaking or clashing.46 It seems that the desire to find close equivalents to a Latin source text for all parts of Óláfr’s treatise is not fully justified and that it depicts Óláfr as more derivative than a close analysis of his practice suggests. This conclusion bears some significance in the search for the direct source(s) for the first half of the Third Grammatical Treatise which deals with phonology, as will be dis- cussed in the next section. The First Section of the Third Grammatical Treatise Scholars have concluded that Óláfr’s influences might not only come from Latin works but perhaps also from vernacular texts, such as Ælfric’s Excerptiones, the ear- liest substantial grammar of Latin written in a vernacular. Gade has discussed the 44  T. Wills, ed. ‘Anonymous Lausavísur, Stanzas from the Third Grammatical Treatise 25’, in Skaldic Poetry of the Scandinavian Middle Ages: Vol III, ed. K. E. Gade and E. Marold, Turnhout, 2017, p. 543. 45  Wills, ‘The Third Grammatical Treatise and the Aeneid’ (n. 42 above), p. 293. 46  We thank one of the anonymous reviewers for pointing out further support for the lack of correspond- ence between the ON and the Virgilian text in that the Latin schesisonomaton does not hinge on the clashing of weapons unlike the ON example. 397From the Aeneid to Accent Theory similarities between the two works, particularly in their translation of Latin gram- matical terminology where there is considerable overlap. Óláfr would presumably have been familiar with Ælfric’s Excerptiones and its vernacular translations of Latin grammatical terminology, either from his time abroad or in Iceland through a copy or notes.47 Further, a wealth of Latin logical treatises has been suggested as models or influences for the first half of Third Grammatical Treatise. This includes works of Boethius, Pseudo-Probus, Roger Bacon’s Summa grammatica, Petrus Hispanus’s Summulae logicales, Robert Kilwardby’s Priscianus Minor, and Petrus Helias’s Summa super Priscianum.48 Some similarities have also been noted between the Third Grammatical Treatise, Audax’s Excerpta and the Holy Scriptures, as well as between the opening section of the Third Grammatical Treatise and the Arbor Por- phyriana.49 It should additionally be noted that the Old Norse Second Grammatical Treatise also begins with sections on the division of sound and describes various types of sound.50 The doctoral theses of de Pins and Wills provide helpful overviews of parallels between the first section of the Third Grammatical Treatise and Latin scholarly lit- erature by Priscian and medieval authors such as Petrus Hispanus and Roger Bacon. These sources often get quite close to the Icelandic text, but usually exact corre- spondence between the texts is lacking. As an example, we provide an analysis of a section on meaningful and non-meaningful voice with reference to the Latin texts that have been suggested as models. Óláfr’s Old Norse text is as follows: Ritanlig rǫdd er ǫnnur merkilig en ǫnnur ómerkilig. Ómerkilig rǫdd er sú er til engrar merkingar er þrengd, sem ‘buba’, ‘blictrix’. Merkilig rǫdd er ǫnnur af náttúru, ǫnnur af setning eða sjálfvilja. Merkilig rǫdd af náttúru er barna grátr eða sjúkra manna stynr ok annat þvíllíkt. Merkilig rǫdd af setning er sú er framfæriz af sjálfvilja manns sem þetta nafn ‘maðr’ merkir kvikendi skynsam- ligt ok dauðligt. Writable voice is either meaningful or non-meaningful. Non-meaningful voice is that which is not pressed into any meaning, such as ‘buba’, ‘blictrix’. Mean- ingful voice is either natural or arising from design or intention. Natural mean- 47  Gade, ‘Ælfric in Iceland’ (n. 27 above), p. 334. Works by Ælfric existed in Iceland, such as a transla- tion of De falsis diis and a fragmentary translation of Ælfric’s description of the conjugation of the Latin verb amo in AM 921 III 4to. See Sigurður Pétursson, Latína er list mæt: Um latneskar menntir á Íslandi, Reykjavík, 2014, pp. 13–15. 48  Raschellà, ‘The Classical Heritage in Old Icelandic Grammatical Literature’ (n. 25 above), p. 125; Clunies Ross, A History of Old Norse Poetry and Poetics (n. 2 above), p. 189 n. 6; Jón Axel Harðarson, ‘Donat und Priscian in Island: Die Dritte Grammatische Abhandlung der Snorra-Edda’, in Mittelalterphi- lologien heute, Standortbestimmung. Teil 1: Die germanischen Philologien, ed. A. Molinari, Würzburg, 2016, pp. 63–77 (64); Frans, ‘Sub regulis Donati’ (n. 25 above), p. 1. 49  Micillo, ‘Classical Tradition and Norse Tradition in the “Third Grammatical Treatise”’ (n. 30 above), pp. 72–3; Raschellà, ‘Old Icelandic Grammatical Literature: The Last Two Decades of Research (1983– 2005)’ (n. 25 above), p. 355. 50  See e.g. F. D. Raschellà (ed.), The So-Called Second Grammatical Treatise, Florence, 1982, pp. 50-55 for a restored Old Norse text and English translation. 398 S. Hilmarsdóttir, H. Þorgeirsson ingful voice is the crying of children/infants or the groaning of sick people and other things like that. Meaningful voice arising by design is that which is brought forth by a person’s intention, as this name: ‘maðr’ (‘man’), refers to a rational and mortal animal.51 The relevant sections which Wills cites from Priscian are the following: aliae autem sunt, quae, quamvis scribantur, tamen inarticulatae dicuntur, cum nihil significent, ut ‘coax’, ‘cra’. aliae vero sunt inarticulatae et illiteratae, quae nec scribi possunt nec intelligi, ut crepitus, mugitus et similia. There are still others which, even if they are written, are nonetheless said to be inarticulate, since they signify nothing, like ‘coax’, ‘cra’. And others are inar- ticulate and not writable, which can be neither written nor understood, such as creaking, mooing and the like.52 homo est animal rationale, mortale, disciplinae capax. Man is a rational animal, mortal, capable of learning. From Petrus Hispanus he cites a more promising text: Vox litterata est, quae scribi potest, ut ‘homo’; vox non litterata est quae scribi non potest. Vocum litterarum alia est significativa, alia non-significativa. Vox significativa est illa, quae auditu aliquid repraesentant, ut ‘homo’ vel gemitus infirmorum, qui dolorem significat. Vox non-significativa est illa, quae nihil auditui repraesentat, ut ‘bu’, ‘ba’. Vocum significativum alia significat natu- raliter, alia significat ad placitum. Vox significativa naturaliter est illa, quae idem repraesentat apud omnes, ut gemitus infirmorum, latratus canum. Literate voice is that which can be written, like ‘man’; illiterate voice is what cannot be written. Literate voice is either significative or non-significative. Sig- nificative voice is that which represents something, like ‘man’ or the groans of the sick, which signify pain. Non-significative voice represents nothing when heard, like ‘bu’, ‘ba’. Significative voice is either significative naturally or sig- nificative by convention. Naturally significative voice is one that represents the same thing to everyone, like the groans of the sick or the barking of dogs.53 Roger Bacon has another similar text, but not quite matching the Old Norse:54 Vocum alia significativa, alia non-significativa. Non-significativa est per quam nichil auditui representatur, ut ‘bubo’, etc; Vocum significativarum alia signifi- 51  Finnur Jónsson, Málhljóða- og málskrúðsrit (n. 4 above), p. 22. Translation our own but cf. Wills, ‘The Foundation of Grammar’ (n. 5 above), p. 77. 52  English translation by R. Stanton, ‘Sound, voice, and articulation in the Exeter Book riddles’, in Rid- dles at work in the early medieval tradition: Works, ideas, interactions, eds. M. Cavell and J. Neville, Manchester, 2020, pp. 92-106 (95). 53  English translation, lightly adapted, by Wills, ‘The Foundation of Grammar’ (n. 5 above), pp. 169-70. 54  Wills, ‘The Foundation of Grammar’ (n. 5 above), pp. 154-155. 399From the Aeneid to Accent Theory cativa ad placitum, alia naturaliter. Vox significativa naturaliter est que ordinatur ad significandum, ut gemitus infirmorum et omnis vox ferarum vel sonus. Vox significativa ad placitum que ex institucione humana aliquid significat. Voice is either significative or non-significative. Non-significative voice is the one through which nothing is represented when it is heard, like ‘bubo’. Significative voice is either significative by convention or naturally. Naturally significative voice is constituted for signifying, like the groans of the sick and all voices or sounds of wild beasts. Voice which is significative by convention signifies something through human convention. De Pins provides two additional correspondences, firstly one from Petrus Helias:55 [D]icendum est quod vocabula quandoque se ipsa nominant, ut cum dicitur ‘Homo est nomen’. Hic enim non de homine loquimur sed potius de hoc nomine ‘homo’. Et hoc appellabant antiqui ‘materiale impositum’, quod quid sit ut intelligas, materiale impositum est vox repraesentans seipsam, id est, posita ad loquendum de seipsa et dicitur materiale impositum quia nomen, si ita contingit, representat materiam suam, id est, vocem que quasi materia preiacet, ut inde fiat nomen. Ex voce namque fit nomen per impositionem. Quod inde videri potest quia ‘blictrix’ vox est tamen nondum nomen est, sed si alicui rei imponitur nomen erit. And since mention has been made of those things which words serve as names, it should be said that words sometimes serve as the names of themselves, as when someone says ‘man is a name’. For we do not say this about man but about this noun ‘man’. And this the ancients called imposed materially, which you should understand as follows: material imposition consists in an utterance repre- senting itself, that is, imposed to speak about itself. It is called imposed materi- ally because the noun, taken in this instance, represents its own material, that is, the utterance which like a material precedes the noun, which is made out of it. For the noun is made out of the utterance by means of imposition. And this can be seen in the case of blictrix, which is an utterance although it is not yet a noun, but it will become a name if it is imposed on something.56 The second one is from the Abbreviatio Montana which can be found in De Rijk’s edition of early logical treatises from c. 1130 to 1220.57 It does not bear a particu- larly strong resemblance to the Third Grammatical Treatise and as one might add several other sections on significative and non-significative voice from other logical treatises in the same volume (such as from Introductiones Montane Minores,58 from 55  We give the Latin text of L. Reilly (ed.), Petrus Helias: Summa Super Priscianum, Vol. 1, Toronto, 1993, p. 193. 56  Translation by J. Kirk, Medieval Nonsense: Signifying Nothing in Fourteenth-Century England, New York, 2021, p. 190. 57  L. M. De Rijk, Logica Modernorum: A Contribution to the History of Early Terminist Logic, Vol. II, Assen, 1967, p. 78. 58  De Rijk, Logica Modernorum (n. 58 above), p. 11 400 S. Hilmarsdóttir, H. Þorgeirsson Ars Emmerana59 and from the Introductiones Parisienses),60 we let the examples above suffice. What the passages quoted above should show is that whilst the gist of these rhe- torical definitions is largely the same, specific elements from the Icelandic text are missing in all of them. This includes the exact words and syllables which Óláfr gives as examples of non-significative voice, i.e. the combination of buba and blictrix. Priscian’s examples of nonsense words are wholly different, coax and cra. Petrus Hispanus’s text is missing blictrix, Roger Bacon’s as well (and it has bubo instead of buba). Petrus Helias has blictrix but not buba. Abbreviatio Montana, not quoted, has neither. However, this specific combination of the nonsense words buba and blictrix does exist in other logical treatises. Hence, one might think that if Óláfr had a direct source, this source would contain the correct combination of nonsense words. Whilst there are early medieval rhetorical and logical texts which contain this com- bination, such as William of Sherwood’s Introductiones in logicam,61 we have still not found a convincing match as these differ from Óláfr’s treatise in other respects. Kirk has pointed out that blictrix is a garbled form of the stock-phrase blityri. He gives some of the variations found in twelfth- and early thirteenth-century treatises for nonsense words. For instance, Ars burana has blictrix; the Introductiones par- isienses has buba; the Logica ‘Ut dicit’ has buba and plectrix.62 The example buba seems to have remained pertinent in Oxonian circles, with Oxonians such as Bacon using buba as an example, and Walter Burley in the fourteenth century as well, split in two monosyllables, ‘bu’, and ‘ba’.63 There is an additional phrase missing from all the correspondences suggested by Wills and de Pins above. This is Óláfr’s first example of a significative voice com- ing from nature, barna grátr (‘the crying of children/infants’). This appears to be quite an unusual example. As can be seen based on its numerous occurrences above, gemitus infirmorum (‘the groaning of sick men’) was a popular example. Usually, this example stands on its own, or is combined with animal sounds, most commonly with latratus canum (‘the barking of dogs’). However, when searched for, one can find the example of the crying of infants (fletus or ploratus infantium). So far, we have only found this in later treatises, such as in the sixteenth-century work of the Italian priest and philosopher Antonio Rocco, whose In logicam has the following definition: Illa naturaliter significat, quae ex instinctu naturae aliquid praeter se ipsam repraesentat; ut cantus avium, latratus canum, fletus infantium, gemitus infir- morum ... 59  De Rijk, Logica Modernorum (n. 58 above), p. 149. 60  De Rijk, Logica Modernorum (n. 58 above), p. 358. 61  William of Sherwood, Introductiones in logicam, ed. C. H. Lohr, P. Kunze and B. Mussler, New York, 1983, section 1.1.2, p. 223. 62  Kirk, Medieval Nonsense (n. 57 above), p. 55. 63  See the second chapter of Kirk, Medieval Nonsense, (n. 57 above), pp. 52–75, for a good discussion of these circles. 401From the Aeneid to Accent Theory It is naturally significative, the voice which by natural instinct represents something besides itself; such as the singing of birds, the barking of dogs, the crying of infants, the groaning of sick men ... Another sixteenth-century scholar, Franciscus Titelmans has a long definition of vox naturaliter significativa in his Compendium Naturalis Philosophiae, where he gives as examples the human sounds ‘ploratus infantium, gemitus infirmorum, sus- pirium dolentium’ (‘the crying of infants, the groaning of sick men, the gasps of men in pain’) and the animal sounds ‘latratus canum, mugitus boum, balatus ovium’ (‘the barking of dogs, the mooing of cows, the bleating of sheep’). It seems the example of children’s weeping was floating around in the medieval tradition. It is quite difficult to know what to make of the existence of bits and pieces of Latin examples matching specifics from the first half of the First Grammatical Trea- tise. We might conclude that a close exemplar, or perhaps a text with specific margi- nalia, has yet to be identified, much like in the case of the second half of the treatise and the close correspondence with Ars Brugensis. Alternatively, the relatively little luck scholars have had with the first half of the treatise so far should perhaps be taken as an indication that there is no such exemplar for this part. As already main- tained above, Óláfr’s practice may not have been as derivative as has sometimes been suggested. Perhaps the case for the first section as a whole is that Óláfr simply worked from multiple sources and combined various examples from these. Here, the recognition of structural influence from Doctrinale in addition to the use of Ars Brugensis is relevant, as this is already evidence of Óláfr’s independent combination of different sources. Perhaps his method in the first half was a more pronounced ver- sion of such practice. Before concluding, we move on to discuss Óláfr’s treatment of accent theory, a subject where it was particularly challenging for the Norse author to apply the Latin sources to his native language. Accent Theory On three occasions, the Third Grammatical Treatise offers an introduction to accent theory, each time giving Norse examples for the three types of accents defined in the classical tradition. These instances contain different and somewhat contradictory information, and we may ask how this state of affairs arose. Finnur Jónsson hypoth- esized that one of the clauses was an addition by a later author.64 This is possible and, in fact, the text itself seems to suggest that the application of accent theory to Old Norse was a topic of debate amongst the scholars of the day. An early copyist might certainly have inserted his own take on the subject. An alternative solution is that the text of the Third Grammatical Treatise was a work in progress, never properly completed by the author. Óláfr may have made more than one attempt to fit classical accent theory to the Norse language, leading to 64  Finnur Jónsson, Málhljóða- og málskrúðsrit (n. 4 above), pp. 16–17. Positing clumsy additions by later copyists was something of a go-to explanation for Finnur Jónsson when he encountered imperfec- tions in the preserved texts, cf. Ármann Jakobsson. ‘Den kluntede afskriver: Finnur Jónsson og Morkin- skinna’, Opuscula 11, 2003, pp. 289–306. 402 S. Hilmarsdóttir, H. Þorgeirsson both repetition and inconsistency. Furthermore, since we have reason to believe that Óláfr was a teacher, it seems likely that the Third Grammatical Treatise was part of the material taught at his school. The text we have might have originated as a compi- lation by his students, sometimes reflecting different stages of their master’s thought. Only one of the accent discussions in the treatise closely reflects a known Latin source, namely Priscian’s Institutiones Grammaticae. The Norse text gives the fol- lowing information: Priscianus segir at hverr raddarstafr hafi tíu hljóð eða fleiri, svá sem a ef þat er skammt hefir fjórar hljóðsgreinir; hvassa hljóðsgrein fyrir útan áblásning h sem hér: ari, þunga hljóðsgrein enn fyrir útan h sem hér: <...>, hvassa hljóðs- grein með áblásning h sem hér: hafi, ok þunga hljóðsgrein sem hér: hafandi. Langt a hefir sex hljóð; ef þat hefir áblásning h þá berr þat annathvárt hvassa hljóðsgrein eða þunga eða umbeygiliga sem hér: <...> Slíkt hit sama ef þat hefir eigi áblásning hljóðar þat þrjár leiðir sem þessi nǫfn: ári, áranna, ára.65 Priscian says that each vowel has ten or more sounds, such as a—if it is short it has four divisions of sound; a sharp accent without aspiration h, like here: ari [‘eagle’], a grave accent without h like here: <...>, an acute accent with aspiration h like here: hafi [‘have’], and a grave accent like here: hafandi [‘hav- ing’]. Long a has six sounds; if it has aspiration then it can either have an acute accent or a grave or a circumflex like here: <...> The same if it does not have aspiration, it sounds in three ways as in these names; ári [‘year’, sg. dat.], áranna [‘of the years’], ára [‘of years’].66 The Norse text cites Priscian, and this clause is indeed a translation of a section from the Institutiones.67 The examples in the Latin text are ‘hábeo, hàbemus’, ‘ábeo, àbimus’, ‘hámis, hàmorum, hâmus’, and ‘árae, àrarum, âra’. Clearly, the translator has attempted to find Norse examples that correspond phonetically as closely as pos- sible to these Latin words. It seems that no Norse equivalents to ‘ábeo, àbimus’ or to ‘hámis, hàmorum, hâmus’ came to mind so these fields were left blank. This pas- sage is preserved in all three textually important manuscripts so clearly the omission of the planned examples goes back to the archetype and, presumably, the original. This is one indication that the treatise was never properly completed. 65  Finnur Jónsson, Málhljóða- og málskrúðsrit (n. 4 above), pp. 25–26. 66  The Norse words which are meant to illustrate the different accents have no accent marks here in any of the manuscripts. Or, at least, none of the words are marked with an acute or grave accent. The word ári has an acute accent in A and that is the word meant to illustrate that accent. However, A sporadically uses an acute accent to mark long vowels (as normalized Old Norse orthography does) and that may be all we are seeing here. The same applies to áranna in B. The writing is now rather faint, but the first vowel appears to have an acute accent to indicate that the vowel is long. 67  quippe cum singulae vocales denos inveniantur sonos habentes vel plures, ut puta a litera brevis quat- tuor habet soni differentias, cum habet aspirationem et acuitur vel gravatur, et rursus cum sine aspiratione acuitur vel gravatur, ut ‘hábeo hàbemus’, ‘ábeo àbimus’. longa vero eadem sex modis sonat: cum habet aspirationem et acuitur vel gravatur vel circumflectitur et rursus cum sine aspiratione acuitur vel grava- tur vel circumflectitur, ut ‘hámis hàmorum hâmus’, ‘árae àrarum âra’. Priscian, Institutionum Grammati- carum Libri I–XII, ed. Martin Hertz, vol. 2 of Grammatici Latini, ed. Heinrich Keil, (n. 15 above), p. 7. 403From the Aeneid to Accent Theory The passage suggests an author in the first stages of applying accent theory to Old Norse. The text cites Priscian and attempts to imitate his examples as closely as possible—suggesting little or no attempt at independent analysis. However, the other two passages introducing accents are not translations of any known Latin work. The first is as follows: Hvǫss hljóðsgrein er sú er skjótliga er fram færð með upphǫldnu hljóði sem þessi samstafa: hvat. Þung hljóðsgrein er sú er af lítillátu hljóði hefz ok dregz niðr í enn lægra hljóð sem hin fyrsta samstafa í þessu nafni: háreysti. Umbeygilig hljóðsgrein er sú er hefz af lítillátu hljóði ok þenz upp sem hvǫss hljóðsgrein en fellr niðr at lyktum sem þung hljóðsgrein, svá sem þetta nafn: hraustr.68 An acute accent is one which is delivered quickly with a raised sound like this syllable: hvat [‘what’]. A grave accent is one which begins with a modest sound and is pulled down into an even lower sound, like the first syllable in this name: háreysti [‘commotion’]. A circumflex accent is one which begins with a modest sound and is raised like an acute accent but falls down in the end like a grave accent, like this name: hraustr [‘healthy’]. The definition of the accents here does not closely reflect the classical descrip- tions, but it is evidently influenced by the orthographic shape of the diacritics—the acute is a rising sound, the grave is a falling sound and the circumflex is a combina- tion of the two. Classical Latin theory entails that a monosyllable has a circumflex accent if its vowel is long but an acute accent if its vowel is short.69 The Norse examples here follow this idea, with hvat juxtaposed with hraustr. The choice of examples, how- ever, is interesting. The word hvat is a pronoun with a short vowel followed by a sin- gle consonant. The word hraustr is an adjective with a long diphthong followed by three consonants. The intention seems to be to show one syllable that is as short as possible and one that is as long as possible. In Old Norse poetry, the length of sylla- bles is an important factor and it interacts in an interesting way with word class. The length of an adjective like hraustr is regulated more tightly than that of a pronoun like hvat—a phenomenon known as Craigie’s law.70 The Third Grammatical Treatise does discuss quantity, both in its theoretical sec- tion and in examples from Norse poetry. Classical learning on the length of seg- ments in Latin was an abstract theory in the Middle Ages since medieval Latin was pronounced quite differently. Quantity theory could, however, be applied to Old Norse without major problems since both Old Norse and classical Latin had distinc- tive vowel length and quantity-based metres. Thus, knowledge which had become theoretical book-learning was applied as originally intended to describe audible 68  Finnur Jónsson, Málhljóða- og málskrúðsrit (n. 4 above), p. 24. 69  P. Probert, Latin Grammarians on the Latin Accent, Oxford, 2019, pp. 8, 187–8. 70  N. Goering, Prosody in Medieval English and Norse, Oxford, 2023, pp. 201–214; W. Craigie, ‘On some Points in Skaldic Metre’, Arkiv för nordisk filologi 16, 1900, pp. 341–84. 404 S. Hilmarsdóttir, H. Þorgeirsson features of a living language. We will argue that the same applies to the Old Norse application of classical accent theory.71 The evidence showing that Old Norse was a pitch-accent language is extensive. Most modern Swedish and Norwegian dialects have a pitch contour distinction that etymologically must be traced to processes active during the Old Norse or Proto- Norse periods.72 The Danish stød, a form of glottalization or creaky voice, is a related phenomenon with a common origin. To be sure, some varieties of the Nordic languages have lost these distinctions and notably modern Icelandic does not have them. There is, however, substantial evidence that Icelandic had a pitch distinction at an earlier stage. In the 18th century, Icelandic scholars described the dialect of the eastmost part of Iceland as having ‘tonal’ pronunciation similar to Swedish and Norwegian. This dialect feature was stigmatized and regarded as archaic by contem- porary scholars.73 Another source of evidence is medieval Icelandic poetry which observes a distinction between two classes of disyllables. These two classes accu- rately mirror the distinction made in the mainland Scandinavian languages between words with two forms of pitch contours.74 Overall, we have good reasons to believe that when Óláfr Þórðarson sat down to engage with Graeco-Latin accent theory, he had the chance to apply it to a living language with pitch distinctions. The final passage in the Third Grammatical Treatise to define accents is similar to the previous one but seems a little more refined. Sú samstafa hefir hvassa hljóðsgrein er hefz af lítlu hljóði ok endiz í hvassara hljóð sem þessar samstǫfur, var, þar, ok er hon svá merkt ´. Þunga hljóðs- grein hefir sú samstafa, er hefz af lítlu hljóði ok endiz í lægra hljóð sem þessar samstǫfur enar fyrri, hára, sára, ok er sjá hljóðs-grein svá nóteruð `. Umbey- giliga hljóðs-grein hefir sú samstafa, er hefz af lítlu hljóði ok dregz upp í hvas- sara hljóð en niðr at lykðum í lægra sem þessar samstǫfur, árs, sárs, ok er sú hljóðs-grein svá nóteruð ^.75 A syllable with an acute accent begins with a little sound and ends in a sharper sound like these syllables: var [‘was’], þar [‘there’], and it is marked like this: ´. A syllable with a grave accent begins in a little sound and ends in a lower sound like the first syllables here: hára [‘of hairs’], sára [‘of wounds’], and this accent is noted like this: `. A syllable with a circumflex accent begins with a little sound and is drawn up into a sharper sound but in the end, it is drawn 71  For Greek and Latin vowel length as an abstract rather than audible feature, see Probert, Latin Gram- marians on the Latin Accent, (n. 70 above) pp. 171 ff. 72  T. Riad, ‘The origin of Scandinavian tone accents’, Diachronica 15, 1998, pp. 63–98. 73  Kjartan G. Ottósson, ‘Indicier på tonaccentsdistinktion i äldre isländska’, Íslenskt mál 8, 1986, pp. 183–90 (87–8). Earlier commentators were unaware of what may be the clearest 18th century source: Páll Vídalín, Skýríngar yfir Fornyrði Lögbókar, Reykjavik, 1854, p. 126. 74  This was first pointed out by Jón Helgason, ‘Ortografien i AM 350 fol.’, Meddelelser fra Norsk foren- ing for sprogvidenskab 1, 1926, p. 55. The evidence was further analysed by Kristján Árnason and Haukur Þorgeirsson, ‘Tonality in earlier Icelandic’, in Nordic Prosody, Proceedings of the XIIth Confer- ence, Trondheim 2016, eds. Jardar Eggesbø, Jacques Koreman, and Wim A. van Dommelen, Frankfurt am Main, 2017, pp. 51–62. 75  Finnur Jónsson, Málhljóða- og málskrúðsrit (n. 4 above), pp. 35–6. 405From the Aeneid to Accent Theory into a lower sound, like these syllables: árs [‘of a year’], sárs [‘of a wound’], and this accent is noted like this: ^. Like before, the distinction between the acute and the circumflex accent is exem- plified with monosyllabic words. Here we have var (‘was’) and þar (‘there’) which have a short vowel followed by a single consonant and belong to parts of speech which are usually lightly stressed. On the other end we have árs (‘of a year’) and sárs (‘of a wound’) which are nouns with a long vowel followed by a consonant cluster.76 This follows the pattern established in the previous clause but appears to be an attempt at producing a cleaner set of examples where each word has a short or a long a followed by r. The development of the example for the grave accent is also interesting. Instead of a trisyllabic word on the Latin model like háreysti (‘commotion’), here we get hára (‘of hairs’) and sára (‘of wounds’). The author may have come to the (probably correct) conclusion that there was no crucial dif- ference between the accent patterns of disyllabic and trisyllabic words in Old Norse, in which case it seemed simpler to use disyllabic examples. The passage concludes by sketching a three-dimensional model of syllable structure and then adds a caveat. Hljóðsgrein merkir hæð samstafligrar raddar, en andi digrleik hennar, sem skilja má, at hvern raddarstaf verðr at nefna meir með sundrloknum munni, ef áblásning fylgir. Tíðar tilfelli merkir lengð samstafligrar raddar, sem fyrr var sagt. En með því at þess konar greinir heyra lítt nórænuskáldskap at flestra manna ætlan, þá tala ek þar um ekki fleira at sinni.77 Accent refers to the height of the syllabic voice and aspiration refers to its thickness, understandably, since each vowel must be pronounced with a more closed mouth if it has aspiration. A distinction in time refers to the length of the syllabic voice, as said earlier. But since distinctions of this sort are, in the opinion of most, not important for Norse poetry, I will not discuss this more for now. The caveat that this knowledge may not be relevant for the analysis of Norse poetry can hardly refer to aspiration, which the treatise discusses in a previous chap- ter as important for alliteration. Nor is it likely to refer to syllabic length, which is so obviously fundamental to Norse poetry. It presumably does refer to accent theory, the main subject of the section. The reference to ‘the opinion of most’ suggests that this was a topic of discussion in scholarly circles of the thirteenth century and, at the stage when this section was written, it was not clear to Óláfr and his colleagues that accent theory had any practical application for the analysis of Old Norse poetry. However, in the second part of the Third Grammatical Treatise, it turns out that Óláfr has, after all, found analytical uses for accent theory. As the examples of vari- ous poetic or rhetorical devices are illustrated, accent theory comes up three times. This is all the more surprising since the Latin sources do not have any corresponding poetic examples. Since the Latin accent as described by the grammarians is perfectly 76  Or ‘overlong’ or ‘hypercharacterized’ syllables, see Kristján Árnason, Quantity in Historical Phonol- ogy, Cambridge, 1980, pp. 114–16. 77  Finnur Jónsson, Málhljóða- og málskrúðsrit (n. 4 above), p. 36. 406 S. Hilmarsdóttir, H. Þorgeirsson predictable based on stress and vowel length, it was not straightforward for the clas- sical theorists to think of interesting examples where one might refer to accent to elucidate something about a line from, say, the Aeneid. Donatus mentions the possibility that accent could be added, subtracted, changed or moved but conspicuously declines to give examples of this.78 The Ars Brugen- sis, Óláfr’s source, gamely attempts to rise to the challenge and come up with some examples. First, the text points out that the proper pronunciation of malesanus only has an accent on ‘sa’ and one would commit a barbarism by addition of an accent if one were to place an accent also on ‘le’. As another example, it would be a bar- barism to move the accent in dominus from the first syllable to the second syllable. Neither of these examples mentions the type of accent involved, a nod to the reality that the salient distinction in Latin is between two types of syllables—stressed and unstressed. The author of the text appears to have some difficulty in imagining the subtraction or change of an accent: Detractionem et inmutationem in una parte intellegere possumus, quia, quando aliqua pars circumflexum habet accentum, sicut ‘metha’ et ‘murus’, detrahendo illum non possumus aliter ut alium non addamus, id est acutum.79 Subtraction and change must be understood as occurring together since, when a circumflex accent is involved, as in metha and murus, we cannot subtract this without adding another, i.e. an acute. This example appears to be rather theoretical—there was probably no real way to mispronounce murus so that it has the wrong accent. Notably, none of the examples here show a minimal pair where a word means one thing when it has one accent pattern and another thing when the accent is different. Nor are there any citations of poetry. Thus, none of the examples in Ars Brugensis appears to be a close model for the very interesting Norse examples that surface in Óláfr’s text. To illustrate barbarism by the change of an accent, the Third Grammatical Trea- tise cites two lines of poetry which contain a series of elaborate word plays. When solved, the sense is that the poet has taken a liking to a married woman, the wife of a man named Máni (‘moon’). The sentence involving accent is hefir fljóð ef vill, (‘has a lady if one wants’), which is first substituted by the synonymous konu má ná, (‘a woman can be obtained’). To complete the pun, this latter version is then altered to konu Mána, (‘the wife of Máni’). Accent comes into play as the treatise explains the difference between má ná and Mána. Hefir fljóð ef vill—konu má ná. Hér er máli skipt. Konu Mána. Hér er seni skipt ok hljóðsgreinum. Mâ nâ. Hér er hvár tveggi hljóðsgrein umbeygileg. Máná – hér er hvöss hljóðsgrein yfir báðum samstöfum ok er hér bæði skipt 78  ‘Toni quoque similiter per has quattuor commutantur. nam et ipsi adiciuntur, detrahuntur, immutantur, transmutantur, quorum exempla ultro se offerunt, si quis inquisierit.’ Frans, ‘Sub regulis Donati’ (n. 25 above), Bilaga [Appendix] I, p. 15. 79  Frans, ‘Sub regulis Donati’ (n. 25 above), Bilaga [Appendix] I, p. 15. 407From the Aeneid to Accent Theory hljóðsgreinum ok aftekning hljóðsgreinar þvíat hvöss hljóðsgrein er gör af umbeygiligri ok tekin svá af þung hljóðsgrein.80 Hefir fljóð ef vill—konu má ná. Here there is a change of expression. Konu Mána. Here there is a change of sense and accents. Mâ nâ. Here both the accents are circumflex. Máná. Here there is an acute accent over both syllables and there has been both a change of accents and the removal of accents since an acute accent is made out of a circumflex accent by the removal of a grave accent. The idea that a change of accents occurs together with the removal of an accent may be inspired by the previously quoted text in the Ars Brugensis. But the Third Grammatical Treatise has a different idea of how this happens since it conceptual- izes the circumflex accent as being a combination of acute and grave. The monosyllabic words má (‘can’) and ná (‘obtain’) should indeed have a cir- cumflex accent according to the Latin theory since they have long vowels. But it is quite interesting to see the word Mána (‘of Máni’) described as having an acute accent over both syllables. The first vowel is long whilst the second is short so within the framework of the Latin theory we would expect it to be described as hav- ing a circumflex accent on the first syllable and no accent on the second one. No disyllabic word in Latin has an accent on both syllables. With the above in mind, how did Óláfr end up with the form Máná with a double acute accent? Since this has no apparent basis in classical theory, it seems natural to ask whether this was Óláfr’s attempt to describe the actual pronunciation of the Norse word. In his description, an acute accent indicates a rising pitch. Is it, there- fore, possible that his pronunciation of words like Mána included two rises in pitch? The answer is that this is very possible and to see why we must briefly describe the facts of Scandinavian pitch patterns. Modern Norwegian and Swedish make a distinction between two accent patterns in polysyllabic words. The realization of this distinction differs between the dialects but, for the majority of speakers, words with one of the patterns (‘accent one’) have one pitch peak and words with the other pattern (‘accent two’) have two pitch peaks. For Swedish, the distinction is often illustrated with the minimal pair anden (‘the duck’), which has accent one, and anden (‘the spirit’), which has accent two. These words have the same phonemes in the same order, and both have stress on the first syllable but their pitch contours make them distinct. An excellent popular introduction to Swedish pitch accent uses an interesting notation for explaining accent two—an arrow pointing up and to the right above each vowel symbol.81 This notation looks quite similar to the double acute accent over Mána in the Third Grammatical Treatise and may have been as intuitive in the thirteenth century as it is now. Forms like Mána, i.e. Old Norse disyllables with no epenthetic vowel or syllabic article, have accent two in Swedish (cf. Swedish måne 80  Finnur Jónsson, Málhljóða- og málskrúðsrit (n. 4 above), p. 45. 81  Academia Cervena, 2014, ‘Understanding the Swedish Pitch Accent’, (last accessed 3 May 2025). https://www.youtube.com/watch?v=lXp7_Sjgm34&t=236s https://www.youtube.com/watch?v=lXp7_Sjgm34&t=236s 408 S. Hilmarsdóttir, H. Þorgeirsson ‘moon’) so if Old Icelandic had similar pitch patterns to modern standard Swedish, Óláfr’s notation would be perfectly reasonable.82 The Third Grammatical treatise offers another striking case where accent theory is used to explain a play on words. In that case, the skaldic quotation is as follows: Um viðrlagning hljóðsgreinar verðr barbarismus sem Skraut-Oddr kvað: Ef væri Bil báru brunnins logs, sús unnum — opt gerik orðaskipti einrœnliga — á bœnum. Hér er bǽnúm sett fyrir bæ̂num.83 A barbarism by the addition of an accent happens as Splendour-Oddr com- posed: If the Bil of the burnt flame of the wave [woman]84 that we love were—I often exchange words in an eccentric way—‘á bœnum’. Here ‘bǽnúm’ is placed instead of ‘bæ̂num’. The representation of the accent marks in the three manuscripts is inconsistent but we follow the reconstruction of Myrvoll and Skomedal.85 As they point out, the manuscripts indicate that the second instance of bænum in Óláfr’s commentary was meant to have a circumflex accent, bæ̂num. The only way in which the first bænum could have ‘the addition of an accent’ compared to that, is if it had an accent on both syllables—i.e. bǽnúm.86 These acute accents have mostly disappeared in the transmission which is unsurprising since the scribes were used to acute accents as optional symbols to mark long vowels or the letter i. The commentary on this stanza is laconic in the Third Grammatical Treatise but clearly the pun involves the substitution of á bǽnúm (‘at prayer’) for á bæ̂num (‘on the farm’). The poet expresses a wish that his beloved is piously praying whilst really implying that he wished she were at the homestead, and perhaps available for a romantic meeting. The forms bǽnúm (‘prayers’, dat. pl., accent 2, with two pitch 82  Myrvoll and Skomedal note the possibility that Óláfr is describing a pitch contour with two peaks like in East-Norwegian and Central Swedish but downplay it in their conclusion in favour of a single-peak interpretation on the West-Norwegian model. See K. J. Myrvoll and T. Skomedal, ‘Tonelagsskilnad i islendsk i Tridje grammatiske avhandling’, Maal og minne 102, 2010, pp. 68–97 (88, 94). 83  Finnur Jónsson, Málhljóða- og málskrúðsrit (n. 4 above), p. 46. 84  The bracket denotes the referent of the kenning, as in n. 42 above. 85  Myrvoll and Skomedal, ‘Tonelagsskilnad i islendsk’ (n. 83 above), p. 92. 86  Jón Axel Harðarson attempts to analyse this example from a theory-internal perspective, suggesting that bænum ‘on the farm’ could be written with a circumflex because monosyllables with long vowels are supposed to have a circumflex accent. The form bænum is derived from the monosyllabic dative bæ with the addition of the suffixed article and in the oldest Icelandic manuscripts there are occasional exam- ples of the article being written as a separate word. Jón A. Harðarson, ‘Donat und Priscian in Island’ (n. 49 above), pp. 70–1. This is an interesting idea, but it has some weaknesses. The examples cited of the suffixed article written separately are from manuscripts significantly older than the Third Grammati- cal Treatise and the preserved manuscripts all write bænum as one word. Secondly, this line of analysis would leave us with no explanation for the form Máná. 409From the Aeneid to Accent Theory peaks) and bæ̂num (‘farm’, dat. sg. definite form, accent 1, with one pitch peak) are a minimal pair distinguished only by their pitch contours—something which Latin did not have but which Greek did. Here the classical analysis of pitch accent has been put to good use to describe a situation similar to the one it was originally designed for. Conclusion The Third Grammatical Treatise by Óláfr Þórðarson represents an astute and ener- getic effort to build a bridge between Latin learning and Old Norse poetics. The theory used to justify this enterprise was the idea that the Norse people shared an Asian origin with the Greeks and the Romans. By producing Norse poetic exam- ples to illustrate classical rhetorical devices, Óláfr not only meant to introduce use- ful analytical concepts to his readers but also to prove that Norse culture had a valid claim to the fruits of classical learning. One of the goals of this contribution is to place study of the Third Grammati- cal Treatise on a more solid philological footing with a new analysis of the medi- eval manuscripts and their relationship, partly by using the new discoveries of Latin models to evaluate the Old Norse readings. We detect no pattern of common errors in the manuscripts A, B, and W—when the text is preserved in all three manuscripts, the majority reading is most likely to represent the archetype. Another goal has been to contribute to the study of the Latin sources of the treatise and Óláfr’s treatment of these. We have sought to highlight the aspects of Óláfr’s work where he appears to provide independent contributions and/or rework- ings of other scholarly material available to him at the time. It seems that one should resist the temptation to search for an exact Latin model for all parts of Óláfr’s trea- tise. This may be particularly relevant for the current state of knowledge about the sources of the first half of the treatise. Though future discoveries may change the present picture, our analysis of the section concerning meaningful and non-mean- ingful voice seems to indicate that Óláfr may have worked from and combined mul- tiple sources, similarly to how we find additional influences from Doctrinale on his use of the Ars Brugensis in the latter half of the treatise. As part of this goal, we have undertaken a close reading of the accent theory presented in the treatise. This may have been the aspect of the source material that caused Óláfr the greatest difficulty. There are traces of multiple attempts to come to terms with the phenomenon and even an expression of doubt that accent theory could be usefully applied to Old Norse poetry. But when examples are needed, the treatise rises to the task. The contrast between the forms bǽnúm (‘prayers’) and bæ̂num (‘the farm’) is especially interesting in using classical accent learning to describe a minimal pair distinguished by different pitch contours. For the gram- marians of the Middle Ages, accent theory had long been an abstract field of study, but we argue that here it was again applied to describe audible features of a living language. 410 S. Hilmarsdóttir, H. Þorgeirsson Acknowledgements  We are grateful to the anonymous reviewers of this article for valuable feedback and helpful and constructive comments. We also especially thank Hið íslenzka fornritafélag for making this research possible. We would further like to express our many thanks to Jonas Wellendorf, James Clack- son, Victor Frans, Mikael Males, Philomen Probert, Tarrin Wills, Tommaso Mari, Cyril de Pins, Valeria Micillo and Jón Axel Harðarson. 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 permis- sion directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/ licenses/by/4.0/. Publisher’s Note  Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. http://creativecommons.org/licenses/by/4.0/ http://creativecommons.org/licenses/by/4.0/ From the Aeneid to Accent Theory: The Application of Classical Learning in the Medieval Icelandic Third Grammatical Treatise Abstract Introduction The Third Grammatical Treatise Previous Textual Criticism The Content of the Manuscripts The Relationship Between the Manuscripts Identifying the Latin Sources of the Third Grammatical Treatise The Second Section of the Third Grammatical Treatise and the Ars Brugensis The Second Part of the Third Grammatical Treatise and Quotations from the Aeneid? The First Section of the Third Grammatical Treatise Accent Theory Conclusion Acknowledgements