¹ Lord, Albert B., The Singer of Tales (1960; repr. New York, 1965), p. 25.Google Scholar

² This study is based on Klaeber's, Fr. text, Beowulf and The Fight at Finnsburg, 3rd ed. with first and second supplements (Boston, 1950).Google Scholar All references to Beowulf are to this edition unless otherwise noted. Other Anglo-Saxon poems are cited according to the texts in The Anglo-Saxon Poetic Records, ed. George Philip Krapp and Elliott Van Kirk Dobbie, 6 vols. (New York and London, 1931–53). Macrons have been added to quotations from ASPR to mark vowel length.

³ For on þœām dage with alliterative stress on þœām, see Christ 1096b and 1371b (ASPR 3, 33 and 41); for þysses liīfes with alliterative stress on þysses, see Genesis 1120b, 1600b and 2452b (ASPR 1, 36, 49 and 73), Gutblac 74b (ASPR 3, 51), Phoenix 151b (ASPR 3, 98), The Gifts of Men 19b (ASPR 3, 137) and Solomon and Saturn 242b (ASPR 6, 40). Cf. the formulaic system … þēos/þās woruld with alliterative stress on þēos/þās, Genesis 1126b (ASPR 1, 36), Christ 1583b (ASPR 3, 47), Guthlac 125b (ASPR 3, 53), The Phoenix 501b (ASPR 3, 108), The Wanderer 58b (ASPR 3, 135) and Deor 31b (ASPR 3, 179).

⁴ The view can be illustrated passim in the papers of Marjorie Daunt, ‘Old English Verse and English Speech Rhythm’, TPS 1946, 56–72Google Scholar, and Baum, Paull F., ‘The Meter of the Beowulf’, MP 46 (1948), 73–91Google Scholar and (1949) 145–62.

⁵ Pace Baum: ‘And we are, accordingly, to expect not a rigid meter but rather a loose and easy manner, a sort of talking style …’ (‘The Meter of the Beowulf’, p. 76).

⁶ It will become apparent, from what follows, that I regard these ‘choices’ as largely illusory. That is, the poetic tradition and the rules of the metrical grammar determine the outcome here as they do elsewhere. But in order to develop the argument 1 speak as though the poet were consciously deciding between alternatives freely available to him.

⁷ In the a verse: 111a, 287a, 413a, 444a, 498a, 573a, 744a, 833a, 932a, 960a, 987a, 1097a, 1129a, 1238a, 1254a, 1389a, 1655a, 1734a, 1756a, 1865a, 2000a, 2188a, 2214a, 2291a, 2443a, 2548a, 2564a, 2624a, 2739a, 3012a, 3031a, 3059a, 3135a and 3168a. In the b verse: 120b, 127b, 130b, 150b, 276b, 410b, 449b, 468b, 602b, 727b, 741b, 876b, 885b, 1072b, 1308b, 1410b, 1792b, 2068b, 2089b, 2120b, 2140b, 2268b, 2413b, 2420b, 2435b, 2578b, 2721b, 2728b, 2821b, 2863b, 2881b, 2908b, 2911b, 2921b, 3138b and 3148b. 357a (MS, eald 7 un hár; Klaeber, eald ond anhār) is not included because un-/an- is the intensive rather than the negative prefix.

⁸ One of the working assumptions of this study is that proper nouns in Beowulf behave differently from common nouns. I have systematically excluded proper nouns from consideration where they directly impinge on the question being investigated. Thus I have nothing to say about the proper noun Unferth or about proper nouns that might be or are compounds. On the other hand I cite as evidence lines which contain proper nouns when they themselves are not at issue.

⁹ Campbell, A., Old English Grammar (1959; corrected repr. Oxford, 1964)Google Scholar asserts: ‘The negative prefix un- was not originally used with finite verbs (though freely added to participles…), and so should always be accented. Occasional unaccented uses, however, occur, e.g. unclœne impure, beside únclœāne’ (§75). With respect to Beowulf 1756a and 2000a cited below, Lehmann, W. P. and Tabusa, Takemitsu, The Alliterations of the Beowulf (Austin, 1958)Google Scholar observe: ’we may account for these two unstressed forms of un- by considering them metrical survivals of past practices, here of proclitic negative prefixes’(p. 8). Bliss, A. J., The Metre of Beowulf (1958; revised, Oxford, 1967)Google Scholar says simply, 'stress on compounds with un- is variable’ (§44). Neither Bliss nor anyone else has attempted to isolate the rules governing this variability.

¹⁰ For the purposes of this paper, an acceptable metrical contour is one of the Sievers types as analysed and catalogued by Pope, John Collins, The Rhythm of Beowulf: an Interpretation of the Normal and Hypermetric Verse-Forms in Old English Poetry (1942; rev., New Haven, 1966)Google Scholar, and by Bliss, Metre. I give the types according to the following modified system, where upraised 1 = a single syllable

¹¹ The metrical type produced by the unstressed form is given first. A₁ or D¹: 444a, 498a, 833a, 960a, 1129a and 3012a; A2 or D*1; 987a, 1097a, 1865a and 3031a; A₃ or C: 111a, 932a, 1655a, 1734a, 2000a, 2291a and 3059a; B1 or E1: 573a; C or A1: 1238a and 1254a; C or D¹: 744a, 1389a and 2548a; and E1 or A1: 287a, 2624a and 3135a.

¹² A1 or D¹: 120b, 127b, 727b, 741b, 885b, 2068b, 2120b, 2413b, 2578b, 2821b, 2881b, 2921band 3148b; A2 or D^*1: 2863b; A2 or D2²: 1792b, 2420b, 2721b and 2728b; B1 or E1: 130b, 150b, 276b, 410b, 602b, 876b, 1072b, 1410b, 2140b and 2268b; and C or D¹: 449b, 468b, 1308b, 2089b, 2908band 3138b.

¹³ I agree with Pope's reservations about the possible instances; see Rhythm, p. 372 (F₃), and Preface, p. xxxi.

¹⁴ The uninflected formative element -ing can hardly take secondary stress in this position. Otherwise one might argue for the possibility of type E1 here and in 2188a. See Huguenin, Julian, Secondary Stress in Anglo-Saxon (Baltimore, 1901). pp. 3–11.Google Scholar

¹⁵ Bliss, Metre, §84, discusses the sequence where (×) stands for a word boundary. I cannot accept his conclusion that the ‘ caesura’ (word boundary) suffices to make this ‘the simplest and most fundamental variety’ of type E. For a critique of Bliss's view, see Cable, Thomas, The Meter and Melody of Beowulf, Illinois Stud. in Lang, and Lit. 64 (Urbana, 1974), 45–64.Google Scholar

¹⁶ Contrary to some current views (e.g. Cable, Meter and Melody, p. 67, ‘stress, and not alliteration, is the basic element of Old English meter’), I regard alliteration as an integral part of metre. Alliteration in one half-line is not determined by the alliteration of the other half-line, but by the rules of the metrical grammar. Once we know, for example, that the initial ictus in 1756a falls on -múrn-, a fact that we derive from the rules of the metrical grammar, we know that the half-line alliterates on m. Comparison with 1756b merely confirms that fact. The group of lines with alternative scansions is interesting precisely because the stress rules of the metrical grammar allow two realizations and we cannot know which the Beowulf poet chose until we compare the corresponding half-line or apply other rules. The alliteration and scansion were still determined internally by his choice.

¹⁷ Words of high alliterative rank are, by my definition, initially stressed compounds with two fully meaningful elements; nominals (nouns, descriptive adjectives and most other adjectives, infinitives and participles); a few other classes and specially marked words (e.g., forms beginning with œāg-, the pronoun self); and other parts of speech raised to high alliterative rank by transformation rules. See Sievers, Eduard, Altgermanische Metrik (Halle, 1893), pp. 22–46;Google ScholarHeusler, Andreas, Deutsche Versgcschichte mit Einscbluss des altenglischen und altnordiscben Stabreimverses I, Paul's Grundriss der germanischen Philologie 8 (Berlin and Leipzig, 1925), 105–13;Google Scholar and Slay, D., ‘Some Aspects of the Technique of Composition of Old English Verse’, TPS 1953, 1.Google Scholar

¹⁸ Above, pp. 42–3.

¹⁹ The seven verses are 307a, 402a, 560a, 702a (where, however, double alliteration depends on an emendation of the Thorkelin transcript), 905a, 1501a and 3031a. The alliterative stress on the finite verb may be ornamental. If so, these would be ‘light verses’. See below, n. 24. The stress in 3031a falls on un- in any case.

²⁰ 477a, 548a, 633a, 667a, 1077a, 2469a, 2487a, 2508a, 2566a, 2748a and 2792a (Pope's types E3, 6, 12 and 15; Bliss's type 2E2a).

²¹ See below, p. 51, for its determination.

²² See Pope, Rhythm, line index (his types D₁ and D₂), and Bliss, Metre, index to the scansion (his type 1D1).

²³ Elsewhere in Old English poetry unlyýtel appears unambiguously both with the prefix stressed (Genesis 1614a, 2407b and 2552b: ASPR 1, 50, 72 and 76) and unstressed (Riddle 40 75a: ASPR 3, 202).

²⁴ 21a, 31a, 54a, 160a, 163a, 165a, 288a, 322a, 398a, 436a, 449a, 487a, 551a, 554a, 592a, 598a, 692a, 732a, 742a, 816a, 868a, 936a, 1109a, 1409a, 1641a, 1845a, 1847a, 1895a, 1897a, 1919a, 1927a, 1948a, 1954a, 2025a, 2042a, 2090a, 2112a, 2118a, 2226a, 2239a, 2263a, 2266a, 2271a, 2273a, 2315a, 2368a, 2408a, 2414a, 2476a, 2517a, 2557a, 2563a, 2582a, 2642a, 2827a, 2902a, 2915a and 2950a. Included in this list are certain verses which in my view are not genuine type D¹ s, but light verses of type C. Since the rescansion automatically places alliterative stress on the first element of the compound, it does not affect the present argument except by reducing somewhat the size of the list.

²⁵ Only compounds consisting of un- + stem beginning with vowel or diphthong (there is none in Beowulf) could not be correctly stressed at sight, although compounds in second position theoretically remain ambiguous. That is, if góld unrìme is properly so scanned, then 960a, for example, could be scanned ^*éafoð uncɫuþes, ignoring the vocalic alliteration of un-, without violating proposition (1). But, since double alliteration in these cases always fits an acceptable metrical contour and single alliteration sometimes does not, the theoretical possibility can be ignored.

²⁶ Cf. Campbell, Grammar, §§87 and 88.

²⁷ See above, pp. 43–4.

²⁸ A verb which has been displaced from the first dip of the verse clause is assimilated to the stressed elements of the clause and does have high alliterative rank. The verbs in 741b, 2578b and 2865b have not been displaced. See Bliss, Metre, §16.

²⁹ Proof of the statements made in this paragraph must await another occasion.

³⁰ See above, n. 17.

³¹ Bliss, Metre, §44, arrives at the same scansion of 2863b on different grounds. See also Stanley, E. G., ‘Verbal Stress in Old English Verse’, Anglia 93 (1975), 307–34Google Scholar, esp. 309–10. Bliss stresses the un- in 741b and 2578b. Pope, Rhythm, stresses the un- in all three cases.

³² In fact, Pope expresses a strong reservation about this scansion: ‘the stressing of un- [in 2921b] may be questioned, because there is evidence on both sides. To read this verse as type A would therefore be legitimate, and the rhythm would be smoother’ (Rhythm, p. 77).

³³ Metre, §61.

³⁴ The alternative, type D2² with a three-syllable drop, is ‘barely possible’, according to Pope, Rhythm, p. 331. Bliss, Metre §86, objects that ‘ungemete(s), though unusually long, is an adverb of degree, and must be proclitic o n the adjective it qualifies’. He relegates the four half-lines to the group of unclassifiable ‘remainders’. Although Bliss's objection is weighty, Pope's scansion has the merit of keeping the verses, whose authenticity is vouched for by the repetition of the pattern, within the compass of the normal metrical contours.

³⁵ See above, p. 46.

³⁶ When the stress on the second element was weakened as a result of the fusion or near-fusion of the two elements into a single form, the restriction seems to have been relaxed. Thus, not only compound proper nouns but also forms like œāgbwylc (621b, 984b and 2887b), missēra (153b, 1498b, 1769b and 2620b) and ōretta (2538b) appear in the second (N) position of the b verse, as do compounds composed of stem + suffix like innanweard (991b and 1976b).

³⁷ With the addition of brūnfāgne helm (2615a), we have every non-alliterating compound in Beowulf in either half-line. Line 2615a violates the fundamental alliterative principle, according to which brūn-, not helm, should alliterate; it is clearly anomalous. Max Reiger proposed reading byrnan bringde (2615 b), which solves the problem very neatly (‘Die alt- und angelsächsische Verskunst’, Zeitschrift für deutsche Philologie 7 (1876), 21).

³⁸ E.g. Dobbie (ASPR 4, 66); cf. Bliss, Metre, §63.

³⁹ Metre, §§61, 62, 77, 78.

⁴⁰ Ibid. §32.

⁴¹ Ibid. index to the scansion. Evidence of phonological weakening would strengthen Bliss's case, which apparently rests entirely on metrical criteria. Some evidence can be found. For example, phonological weakening accompanies the stressed prefix af-: af-ēst >œfst. On the other hand, I think it possible that the true explanation of certain forms in these lists may prove to be that the second element is more heavily stressed than the first.

⁴² See above, p. 46.

⁴³ Ibid.

⁴⁴ Bliss, Metre, §32, gives the prefix un- as an example of a stressed prefix which causes stress reduction in the following stem. Thus he would exclude all verses with un- from the category of fully stressed, two-element compounds. There seem to me to be two objections to his solution. (1) The prefix un-, unlike some of those that occur in the list of verses above, p. 50, has been an active and productive feature of the language in every period. One would expect to find abundant evidence of phonological weakening. So far as I know, there is none. (2) Two unrelated explanations are required to account for the evidence of the text – stress reduction and the alternation of stressed and unstressed un-. But alternation alone suffices to account systematically for the distribution of the forms when the alliterative patterns are taken into account.

⁴⁵ To recapitulate, I have argued for five instances of the unstressed form in the a verse out of thirty-four: 498a, 833a, 1756a, 2000a and 3012a; and for eighteen out of thirty-six in the b verse 120b, 127b, 727b, 741b, 885b, 1792b, 2068b, 2120b, 2413b, 2420b, 2578b, 2721b, 2728b, 2821b, 2863b, 2881b, 2921b and 3148b.

⁴⁶ ‘Accent Marks in the Vercelli Book’, NM 72 (1971), 703.Google Scholar

⁴⁷ Bliss, Metre, §§33 and 61–5, implicitly arrives at a similar conclusion by somewhat different means.

⁴⁸ I am most grateful to Professor Pope for a meticulous critique of a first version of this article. I am grateful also to Professor Greenfield for helpful comments.