¹ It may seem rather bold to call this a ‘fact’, but consider the following statistics from Beowulf: only 0.48% of the a verses and 0.76% of the b verses lack alliteration in manuscript form (fifteen lines): 352, 461, 586, 954, 965, 976, 1960, 1981, 2139, 2251, 2298, 2305, 2941, 2972 and 3086. These are non-defective lines that do not contain a pair of alliterating stresses. The following b verses also have no alliteration (the a verses have double alliteration and have been counted): 149b, 307b, 1073b, 1390b, 2341b, 2525b, 2882b, 2929b and 3000b. And only 0.80% of b verses have double alliteration (twenty-six verses): 208b, 367b, 395b, 402b, 457b, 574b, 840b, 1151b, 1227b, 1413b, 1440b, 1518b, 1562b, 1690b, 1704b, 1721b, 1843b, 1884b, 2085b, 2195 b, 2296b, 2352b, 2817b, 2916b and 3032b. The text analysed was that of ASPR. The following lines with indeterminate alliteration because of a defective manuscript were not included in the statistics (twenty-six lines): 20, 62, 389, 390, 1803, 2001, 2002, 2007, 2019, 2020, 2023, 2062, 2186, 2218, 2227, 2228, 2229, 2230, 2275, 2276, 2361, 3151, 3152, 3153,5171 and 3177. The following verses were also excluded from the statistics on the basis of a defective manuscript (they are complete enough for the alliteration of their paired verse to be safely determined): 139b, 159a, 240b, 403b, 723b, 2043b, 2128b, 2146b, 2187a, 2215b, 2216b, 2217b, 2231a, 2232a, 2253a, 2676b, 2792b, 2990a, 3065a, 3150b, 3 168b, 3 172b, 3174b and 3 179a. The statistics are thus based on 3148 a verses and 3 140& verses. The manuscript form was analysed rather than the edited text in the following emended verses, to avoid basing the analysis on the metrical assumptions of the editors: (forty-one a verses) 6, 84, 432, 457, 499, 516, 652, 707, 780, 963, 965, 976, 1068, 1129, 1130, 1329, 1372, 1488, 1513, 1546, 1875, 1889, 1926, 1981, 1983, 2006, 2139, 2305, 2385, 2473, 2488, 2523, 2525, 2589, 2694, 2828, 2941, 3086, 3124, 3136 and 3139, and (forty-five b verses) 84, 149, 307, 312, 332, 431, 461, 530, 586, 648, 722, 764, 765, 811, 949, 954, 1073, 1151, 1165, 1261, 1390, 1404, 1541, 1559, 1960, 2094, 2221, 2226, 2250, 2251, 2296, 2298, 2341, 2525, 2628, 2698, 2814, 2882, 2911, 2916, 2929, 2972, 3000, 3101 and 3102. In some cases the emendations seem quite reasonable and we can be fairly confident that the manuscript does not reflect the poet's intention, but it seemed safer not to eliminate these troublesome verses before the analysis had begun. This analysis, which was performed with the aid of a database program and a personal computer, forms the basis for my A New Theory of Old English Meter (forthcoming, Frankfurt, 1985).Google Scholar

² In these and other examples bold type marks the alliterating sound of the line; the same method is used to highlight self-alliteration. For my abbreviated titles of poems, I follow Mitchell, Bruce, Ball, Christopher and Cameron, Angus, ‘Short Titles of Old English Texts’, ASE 4 (1975), 207–21Google Scholar, and $$$$$'Short Titles of Old English Texts: Addenda and Corrigenda’, ASE 8 (1979), 531–3.Google Scholar

³ See, e.g., Campbell, Alistair, Old English Grammar (Oxford, 1959), p. 34.Google Scholar

⁴ By specifying ‘root morphemes’ I mean to include as self-alliterating compounds only those compounds whose major meaning-bearing elements alliterate, and to eliminate words like sœnœssas (Beo 223a) and leodgebyrgean (Beo 269a).

⁵ These classes are based on Campbell, OE Grammar, pp. 30–5; Bliss, A. J., The Metre of ‘Beowulf’’, 2nd ed. (Oxford, 1967), pp. 25–6Google Scholar, and Sievers, Eduard, Altgermanische Metrik (Halle, 1893), pp. 125–6.Google Scholar See also Bliss, , Metre, pp. 55–60Google Scholar, for a discussion of some of the differences in the treatment of the two classes in the poetry.

⁶ These and other verses from Beowulf, unless otherwise specified, are from ‘Beowulf and ‘Judith’, ed. George Philip Krapp and Elliott Van Kirk Dobbie, The Anglo-Saxon Poetic Records 4 (New York, 1953)Google Scholar; this edition will be referred to as ASPR below. Line-end punctuation has usually been deleted.

⁷ It would not be far from the truth to call this the standard view. See Sievers, Altgermanische Metrik, and Cable, Thomas M., The Meter and Melody of ‘Beowulf’, Illinois Studies in Language and Literature 64 (Urbana, Ill. 1974)Google Scholar. I shall not discuss here the treatment of Class One and Class Two words in isochronic theories or how the evidence presented affects such theories, because the use of initial rest-beats greatly complicates the picture. I simply note that John Pope also accepts the traditional view that both classes of words have two stresses (The Rhythm of ‘Beowulf’ (New Haven, Conn., 1942).Google Scholar)

⁸ Some metrists have argued for level stress in such compounds; see, e.g., Bliss, , Metre, pp. 114–15Google Scholar (where he cites Luick), and Kaluza, Max, Englische Metrik in historischer Entwicklung (Berlin, 1909), pp. 90–1.Google Scholar This view is quite attractive, and the evidence presented here tends to support it.

⁹ Bliss, , Metre, pp. 25 and 82–3.Google Scholar

¹⁰ Ibid. pp. 113–17.

¹¹ Campbell, , OE Grammar, p. 35.Google Scholar

¹² The italics in 2106 are in ASPR; the emendation is irrelevant to the present discussion. In Klaeber, F., ed., ‘Beowulf and’ The Fight at Finnsburg', 3rd ed. (Lexington, Mass., 1950)Google Scholar, there is at least one more example (1783a), in which Klaeber prints wiggeweorpad, while ASPR has wigge weorþad, as in the manuscript. Bliss notes the existence of the verses (including the one from Klaeber's edition), but he does not further consider the restriction of such verses to the a verse. Other metrists have also noted self-alliteration in passing, but without making systematic use of it; see Sievers, , Metrik, p. 39Google Scholar, Anm. 1; Kaluza, , Engliscbe Metrik, pp. 8Google Scholar j and 91; and Baum, Paull F., ‘The Meter of the$$$$$ Beowulf, MP 46 (1948-1949), 146.Google Scholar

¹³ Bliss, , Metre, p. 67.Google Scholar

¹⁴ Except for verse 15 38a, which is listed as type E in the ‘Index to the Scansion of Beowulf, the verses listed here are analysed as type d by Bliss, (Metre, pp. 62–5).Google Scholar The count made here confirms his report of sixteen such verses. In his ‘Index’, however, Bliss lists no examples of type d with double alliteration because he considers self-alliteration ornamental, or accidental, and not metrically significant.

¹⁵ There seems to have been some confusion about this morpheme in Old English, as may be seen by comparing its other occurrences in Beowulf: 2636a guðgetawa, 324a gryregeatwum; 674b and 2362a hildegeat-, 2866a eoredgeatwe and 368a wiggetawum. Thus compounds with -geatwe may have been in the process of becoming Class Two words. Bliss, citing Pope, goes a step further and argues that -tawum has a short first vowel; thus Beo 395b becomes type d4C (x x x / x \ with resolution in -tawum) (Metre, p. 65, and ‘Index').

¹⁶ One other peculiar Beowulf verse might seem to belong in this list: 743a ‘synsnjedum swealh; sona hæfde’. In 743a synsnœdum produces triple alliteration and so would seem to provide evidence against the view presented here. That is, if the second alliteration of a Class One word is metrically significant and equivalent to a separate stressed, alliterating word, then -snœdum constitutes a third lift in an a verse - an unacceptable result. A closer look at this word, however, shows it to be a Class Two word beginning with the stressed prefix sin-(or syn-) ‘perpetual, immense’ (see above, p. 77). Self-alliteration in Class Two words will be discussed below.

¹⁷ Bliss, , Metre, p. 12.Google Scholar

¹⁸ Ibid. p. 67. Bliss's figures are thirteen out of 174 verses, or 7.5%. For the verse not listed by me, see above, n. 12: my figures are twelve out of 173 verses, or 6.9%.

¹⁹ Bliss, , Metre, p. 64.Google Scholar

²⁰ Ibid. p. 66.

²¹ Ibid. pp. 114 and 115. He argues that secondary and tertiary stress are metrically distinct and that the secondary stresses may actually be primary. Then he comments on evidence from Latin borrowings and remarks that the only relevant one in Beowulf is giganta, whose three occurrences he claims are inconclusive as to secondary or tertiary stress. However, the evidence seems sufficient in the present context, because two of the three Beowulf verses are bverses. Giganta is, then, almost certainly a Class Two word. And this conclusion supports Bliss's already strong arguments.

²² See Ross, A. S. C., ‘Philological Probability Problems’, Jnl of the R. Statistical Soc. B 12 (1950), 19–59.Google Scholar

²³ The figures include the verses from Beowulf that have already been cited. These examples were found by reading the headwords in A Concordance to ‘The Anglo-Saxon Poetic Records’, ed. B. Bessinger, Jr, programmed by Philip H. Smith, Jr (Ithaca, NY, 1978)Google Scholar; some rechecking was done by using the ‘Index of Compounds’, compiled by Michael W. Twomey, pp. xvii-xxxvii. This book is cited below as CASPR.

²⁴ The other examples of triple alliteration in a verses that are caused by Class One words are Fort 71, GenB 444,Dan 539,And 1275,Mald 96,Ruin 20,ChristC 1006 and MSol 446.

²⁵ Campbell, , OE Grammar, p. 155.Google Scholar

²⁶ Kaluza, , Englische Metrik, pp. 85 and 91Google Scholar. (Ex 482a is printed as three words in ASPR.)

²⁷ See Dobbie's notes to these lines (ASPR 4); Professor A. J. Bliss (personal communication) called this collateral problem to my attention.

²⁸ Some differences between poets seem clear: in Class One words Beowulf has only one anomalous alliteration in the a verse, while The Metres of Boethius has five anomalous and Andreas five anomalous and one triple.

²⁹ For simplicity, I generally refer to ‘second’ elements of Class Two words, even though some examples involve other elements, e.g., unscomiende (double alliteration on un- and -ende?) or unscylidigne (un- and -ig-?).

³⁰ Sievers gives a few examples of this kind of syllabification: fund-di-an, fun-di-en-de and þo-li-an (p. 125). Wiesław Awedyk helpfully surveys a large number of theories of the syllable in The Syllable Theory and Old English Phonology (Wrocław, 1975)Google Scholar and argues for a view similar to that of Charles Hockett, F., A Manual of Phonology, International Jnl of American Linguistics 21 (Baltimore, Mol, 1955)Google Scholar. See also well, Robert P. Stock, ‘The Phonology of Old English’, Stud. in Ling. 13 (1958), 13–24Google Scholar, and other references in Awedyk. The majority opinion, with which I agree, is on the side of phonetic syllabification. Note that generative phonology, which does not recognize a significant phonemic level, sidesteps this problem. See Chomsky, Noæm and Halle, Morris, The Sound Pattern of English (New York, 1968)Google Scholar, for the classical exposition of this view, esp. pp. 7–14 and 66–9.

³¹ For the method of discovery, see above, n. 23; a few dubious or problematic examples have been omitted, as have some emended verses.

³² It would be more accurate, perhaps, to call the alliteration in these verses ‘non-anomalous’, or ‘unobjectionable’, because I know of no metrist who would consider them to have double alliteration; the same is true of the ‘normal’ verses in the next list.

³³ See the references above, n. 30.

³⁴ ‘Prosodic Features in Proto-Germanic’, Toward a Grammar of Proto-Germanic, ed. Frans Van Coestsem and Herbert L. Kufher (Tübingen, 1972), pp. 112–13.Google Scholar

³⁵ See Ibid. pp. 102–6 and 111–13, on juncture and stress in Proto-Germanic. Bennett argues for the same stress on Class One and Class Two words in Proto-Germanic, but he cautions that many of these stresses are ‘not applicable to the historic Germanic languages …’ (p. III).

³⁶ See Bliss, , Metre, pp. 113–17Google Scholar, for an explanation of this evidence that does not involve a half-stress on the second elements of Class Two words. Bliss's argument was proposed on independent grounds but is confirmed by the conclusions reached above.

³⁷ Bliss, , Metre, p. 122Google Scholar, says there are twenty-six such verses in Beowulf.

³⁸ In his discussion of Bliss's light verses and the two-stress principle, Cable presents the following argument for the two-stress assumption that seems not to rest on alliteration: ‘If a principle is established in the great majority of lines, it is reasonable to assume that ambiguous verses are constructed upon that same principle instead of upon a completely different one. If the principle requires a minimum of two stresses to the verse, it is reasonable again to assume that the poet and his audience felt two stresses in those occasional verses where the stress is less obvious to the modern reader’ (Meter and Melody, p. 24). However, it is only the existence of double alliteration in a verses (and in some of Bliss's light a verses) that allows Cable to claim that the principle is established in the great majority of verses. I believe the argument presented here is stronger than the one Cable makes and that it is as strong an argument as can be made; it is not a straw man.

³⁹ Bliss, whose figures are based on rather different assumptions, counts 508 one-stress a verses (just under 16% of all a verses) in Beowulf. Poems such as Cuthlac A and Andreas have much higher percentages of A3 verses than Beowulf, so that the difficulty with treating type A3 as exceptional is much greater than one would think if only Beowulf were considered.

⁴⁰ ‘The Prefix un- and the Metrical Grammar of Beowulf‘, ASE 10 (1982), 39–52.Google Scholar

⁴¹ Ibid. p. 43, n. 16.

⁴² Ibid. pp. 43–4.

⁴³ Ibid. p. 44, n. 17.

⁴⁴ It will be noted that this is essentially the traditional view of alliteration; see, e.g., Sievers, , Metrik, pp. 36–7Google Scholar. My own research has uncovered no reason to doubt this universally accepted definition of what constitutes alliteration, though my views on the placement of aliteration and on the relevance of the alliteration of specific words, syllables, stress levels and so forth are quite different from the traditional view.

⁴⁵ Both ‘stress’ and ‘heavier’ are somewhat vague, but a full explanation of the rule is beyond the scope of the present article. Let me simply state that in my theory only primary and secondary stress are metrically significant and that a difference of one stress level is, consequently, a fair definition of ‘heavier’.

⁴⁶ See Kuhn, Hans, ‘Zur Wortsellung und -betonung im Altgermanischen’, BCDSL. 57 (1953), 1–109.Google Scholar For a new and ambitious revision of Kuhn's work, see Kendall, Calvin B., ‘The Metrical Grammar of Beowulf. Displacement’, Speculum 58 (1983), 1–30.CrossRefGoogle Scholar (The conclusions of Kendall's interesting article are largely vitiated, in my view, by too much reliance on stress and too great a willingness to disregard the evidence of alliteration.)

⁴⁷ For the full theory, see ch. 5 of my A New Theory of Old English Meter (forthcoming, Frankfurt, 1985).Google Scholar

⁴⁸ For a survey of bases for metre other than rhythmic, see Versification: Major Language Types, ed. W. K. Wimsatt (New York, 1972)Google Scholar; John Lotz, ‘Elements of Versification’ (Ibid. 1–21) is especially interesting.

⁴⁹ I am grateful to Professor A. J. Bliss, whose perceptive, insightful and probing comments on an earlier draft of this article provided numerous local improvements and corrections, and also forced me to rethink and restate my conclusions and their metrical consequences.