One of the most common pieces of prescriptive grammar that I grew up with was “don’t end a sentence with a preposition”. At least some of the poets composing early Germanic alliterative poetry seem to have had almost the reverse rule: don’t begin a sentence (or rather, a clause) with a preposition. A good illustration of this is provided by the opening line of Beowulf:1

As transmitted, every detail of stress, metre, and syntax conforms to the normal expectations of Old English poetry – but if the second half-line were to be placed first, so that the sentence began simply with the fronted prepositional phrase ˣIn geār-dagum, the result would be a strikingly anomalous configuration.

This restriction, along with a related suite of issues concerning the behaviour of “particles” in Germanic verse forms, has long been noticed, and approached from various angles, with some mixture of phonological, syntactic, metrical, and stylistic constraints generally being used to explain why certain unstressed elements seem to have limitations on where they can occur. The potentially relevant factors are complex, and matters are not made easier by philological problems in the texts, and frequent ambiguities as to how many passages should be parsed syntactically. In this contribution to the problem, I do not intend to settle the matter once and for all, but merely to attempt an integration of metrical, philological, and linguistic approaches that might, despite some outstanding difficulties, be mutually illuminating. Metrically, I turn to the so-called word-foot theory of Russom (1987; 2022), which adopts a more fine-grained approach to “weak” elements in alliterative verse than most models. On the philological side, I will first attempt to reframe the basic data, highlighting the improvements made by, e.g., Whitman (1993: ch. 3) and Momma (1997: ch. 5), to the still-influential discussion of Kuhn (1933). And linguistically, I draw on the more nuanced approaches to clitic attachment that are now available, particularly concerning the possibility of mismatches (even very routine ones) between phonological and syntactic constituency.

In the following sections, I first introduce the workings of Old English metre, including important considerations raised by the word-foot theory of metre (though I attempt, as much as possible, to present my arguments in ways that will also be compatible with other metrical approaches). I then review the phonological frameworks for understanding clitics, emphasizing the possibilities for mismatches between syntactic and phonological constituency that are, though well known to linguists, not yet much discussed with regards to the particles of Germanic poetry. I then outline Kuhn’s second law, the traditional framework for understanding the placement of low-stress elements in Germanic verse. After considering some of the problems involved with the “law” (some of the most salient of which were already noted by Kuhn himself), I turn to Whitman (1993) for a different angle of attack on these problems, examining where different monosyllabic weak elements occur as extrametrical elements (including their place in both verse and clause), when they must be scanned, and where they are usually supported by other material. With these different analytical frameworks in mind, I suggest a synthesis, considering primarily the classes of prepositions, unstressed prefixes, and demonstratives/articles (with some attention also given to the temporal particle þā). I suggest an analysis of prepositions as prosodic-word enclitics and prefixes as affixal proclitics. Demonstratives show an interesting variation, already noted by Kuhn, between at least “classical” Old English verse and Old Saxon, and probably also late Old English. I propose that demonstratives in earlier Old English were prosodic-word clitics, but shifted in the other two corpora to free clitics (a shift probably related to their change of function to articles).

Most Old English poems are composed in one of two (related) metrical modes: a “standard” type and a “hypermetric” type. I will only deal with the standard mode here; on hypermetric verses, see Sievers (1887: 458–475), Simms (2003), and Hartman (2020). The workings of this metre have had to be reconstructed by modern scholarship, and there are a few different approaches and perspectives on just how things work. Mainstream metrical research takes the system of Eduard Sievers (1885a; b; 1887; 1893) as its starting point, though different theorists have since taken his observations in different directions. For my purposes, there are two points of Sieversian metrics that are really important: anacrusis and expanded dips.

Both phenomena can be illustrated by comparing the following two verses from Beowulf:3

Example (2) shows a minimal verse type, in this case a trochaic pattern (though other patterns are possible as well), in which the removal of any syllable would render the verse unmetrical.4 The two stressed syllables, fol- and hyr-, fill metrical lifts or strong positions, while -ces and -de fill dips (weak positions). Example (3) shows extra syllables. The one at the beginning, on-, stands in anacrusis to the verse: it is an additional weak syllable placed immediately before a lift (strong syllable) that would more normally be verse-initial. The second extra syllable is the preposition on, which occurs next to a weak syllable that was already part of the core metrical structure of the verse. In most analyses of Old English verse, -nen and on are regarded as equal constituents in an expanded dip: any dip filled by more than one weak syllable.5 The major division in classic Sieversian metrics is therefore between anacrustic syllables, on the one hand, which are truly extrametrical, and on the other, weak syllables of dips, which all have more or less the same status.6 This distinction will be very important when it comes to assessing the phonological status of the “little words” of Germanic verse.

In what follows, I make use of a specific theoretical elaboration of Sieversian metrics: the word-foot theory, first proposed by Russom (1987), refined in a number of further articles and two books (Russom 1998; 2017), and presented in a revised form by Russom (2022).7 Russom accepts much of Sievers’ descriptive apparatus in terms of what constitutes a metrically minimal verse, and what patterns of variation are tolerated, but differs in how these are explained. For Russom, a verse consists of a core skeleton of metrical syllables (the members of the “word-feet” that give the theory its name; these are based ultimately on normal word-patterns in the language), some of which may be weak, plus additional extrametrical syllables.8 In word-foot notation, examples (2)–(3) could be represented as follows:

In this discussion, I use S to represent a metrically fully stressed syllable (or resolved equivalent), lower-case s a secondary stress, w a metrically weak syllable, and (x) an extrametrical weak syllable (this is somewhat different from Russom’s notation).9 Word-feet can be divided into classes depending on their phonological weight: those headed by an S are standard, or heavy if they also contain s,10 while low-stress word-feet headed by a w syllable are light. It is important to note that in Russom’s system, unstressed prefixes do not belong to the following word foot: on·gunnen is not a foot of the shape wSw (or xSw), and could only be analysed as a beginning with a distinct extrametrical element, (x)Sw, or potentially as two distinct word-feet, w/Sw.

As can be seen in (4) and (5), there is no fundamental distinction between conventional anacrusis (at the start of a verse before a strong position) and further syllables in an expanded dip: both are simply extrametrical. Still, the traditional “anacrusis” context is much more constrained by word-foot principles (Russom 2022: 57–59). Within a verse, light word-feet are prohibited by general principles of Russom’s system, meaning that a medial preposition like on in (5) cannot be mapped onto a metrical w.11 This metrical rigidity allows for a certain degree of linguistic latitude in terms of what kinds of elements can occur in expanded dips, and the kinds of distinctions I will be examining below do not manifest as strongly in medial position.

At the start of a verse either extrametrical/anacrustic (x) or metrical, light word-foot w are possible. Contrast the anacrustic prefix on- as (x) in (5) with the metrical preposition in as w in Beowulf 1b:

Here the preposition in must be a metrical constituent, since ˣ(x)Ssw = ˣSsw would be an unmetrically short verse. By contrast, (5) would become unmetrical if the initial prefix on- were taken as anything but anacrustic.

One advantage of the word-foot system is that it suggests a phonological interpretation of the difference between initial anacrusis and metrical constituent. In (6), in is, in word-foot terms, a light word-foot, a projection of a word-like unit forming part of the verse’s metrical skeleton.12 Russom does not put the matter in precisely these phonological terms, but it is easiest to understand his word-feet as being based essentially on prosodic words (on which see below). In other words, just based on the few examples considered so far, it would be tempting to speculate that the preference for prepositions to map to word-feet where possible (i.e. in initial position) suggests that they are (often) prosodic words, while the preference for prefixes to be extrametrical in verse reflects their lack of prosodic-word status.13 I will take this as a working hypothesis, naturally not yet fully justified, but something to be tested through application: I attempt to show below that this is indeed a useful and productive approach. I would stress that qualifiers like “preferentially” and “tend” are important. As is the case in many poetic traditions, there is some flexibility in the mapping between linguistic and metrical units, and it is not usually possible to say absolutely categorically that certain types of element are always anacrustic and never light word-feet. The arguments below are based on statistical preferences, not categorical divisions.

A further complication concerning extrametrical syllables in the word-foot theory comes from runs of verse-initial weak syllables, such as Beowulf 1a:

This could be scanned any of three ways by the word-foot theory: w/(x)Ssw, (x)w/Ssw, or ww/Ssw.14 Under the first option, the pronoun wē would be extrametrical, while under the second the adverbial hwæt would be. Russom (2022: 41, 48) also allows disyllabic light ww feet, and further principles allow a single word-foot to, at least under some circumstances, be instantiated by multiple words. That is, it is not inconceivable, under the general principles of the theory, that hwæt wē could constitute a single word-foot – compare, in different terms, Whitman (1993: 69–83) – equivalent to a single low-stress word such as under in Beowulf 710b:

In (8), there is no doubt that the scansion begins with ww, since a single simplex word cannot under any circumstances contain within it a word-foot boundary. In hwæt wē, scansion as ww is only a possibility – but a possibility that interacts with the question of whether hwæt wē really should be analysed as two “words”, or if hwæt=wē, with an enclitic pronoun, might not be the right prosodic structure to posit here. The possibility of clitic structures being formed within light word-feet will be returned to below.

The term clitic is sometimes used in rather different ways depending on linguistic tradition and area of focus. I am here interested strictly in phonological clitics, which might be provisionally defined, following Anderson (2005: 13), as “prosodically deficient forms” (emphasis original), and which must be adjoined to a neighbouring phonological entity (the host) so that it can stand as part of some larger prosodic unit, namely part of a syllable, foot, prosodic word (PWord, PWd), or prosodic phrase (PPh). Since there are different kinds of prosodic deficiency, the exact relationship between a clitic and its host can be of several types. One influential model posits four types of clitic, reproduced here from Anderson (2005: 46):15

This typology was proposed by Selkirk (1996), though she did not present adjoined prosodic words, as in (9a), as a type of clitic. Anderson (2005: 46) does consider PWord clitics to be a type of clitic, with the prosodic word status of the clitic not being inherent to the element (as it would be for a normal word), but assigned in a process of phonological repair.16 Without being overly committed to this set of options as either exhaustive or universal, this typology does capture some important dimensions along which clitics can vary.

Space precludes an extensive discussion of all of Anderson’s types, but the category of PWord clitic (9a) needs to be considered further, since it plays an important role in the analysis developed below. The really essential point in the present context is that a PWord clitic is in some real sense a clitic (and has a host), but is nonetheless also a prosodic word. This may seem somewhat counter-intuitive, but such entities do seem to exist. An illustration of a probable PWord clitic in early Germanic comes from Gothic verbal “prefix” uz-, found in forms such as us·iddja “went out” and ur·rais “arose”. The underlying form uz- surfaces when a further clitic intervenes between “prefix” and root, as does uh “and” in uz=uh·iddja “and (he) went out”.17 The variant us- is produced by the active process of word-final devoicing in Gothic, which usually applies to “words” (in a phonological sense, therefore, prosodic words), and consequently implies that us- is its own PWord in us·iddja. At the same time, the prefix cannot simply be taken as a fully independent word, since it shows assimilation to the root-initial r- in ur·rais, a process that does not happen between adjacent words within a phrase.18 In other words, it would seem that us-, as a verbal “prefix”, has a status both as a prosodic word, and as a proclitic bound to its host in a relationship closer than that of two adjacent words in a prosodic phrase. In this light, it might be better to model us·iddja and ur·rais, and perhaps PWord clitics in general, not as forming a prosodic phrase (as in Anderson’s representation), but as constituting a larger prosodic word:19

I would stress again that a “clitic type” of this sort is meant only as an abstract generalization of convergent cross-linguistic tendencies rather than a universal linguistic truth. The key point is that there can be clitics that themselves show phonological properties that are associated with “words” (in this example, susceptibility to final devoicing), but which can still be prosodically dependent on an adjacent host element (allowing the assimilation to ur-).

Clitics can further vary in terms of whether they precede their hosts (as proclitics) or follow them (as enclitics). An important point is that this phonological attachment need have nothing to do with syntactic constituency. Anderson (2005: 16–17) gives examples from Kʷakʷʾala (a Wakashan language of the Pacific Northwest) which show determiners that syntactically modify a following noun, but show phonological attachment to a preceding element:

The determiner ida modifies “man”, but, in Anderson’s assessment, attaches phonologically to the preceding verb.

In an extensive review of diachronic and experimental evidence, Lahiri & Plank (2010) make the same case for a variety of Germanic languages, suggesting that both determiners and prepositions are routinely phonological enclitics, despite their syntactic constiuency with what follows. The evidence surveyed includes diachronic reductions,20 metalinguistic intuitions from various commentators (Lahiri & Plank 2010: 372–375), synchronic clitic formation in various Germanic languages (Lahiri & Plank 2010: 385–390), and timing evidence from modern experiments (Lahiri & Plank 2010: 392–393). This last set of evidence comes from Dutch, where Wheeldon & Lahiri (1997) had found that, when prompted to respond immediately, participants took longer to begin the response ik drink het water “I drink the water” than either ik drink water “I drink water” or (crucially) ik drink vers water “I drink fresh water”, a finding they plausibly attribute to (ik drink=het) (water) having a longer initial prosodic word than (ik drink) (vers) (water). While a single experiment of this sort is not of course decisive, the results are impressively consistent with the wide range of evidence marshalled for Germanic in the article as a whole, and with the more global survey of Himmelmann (2014). It seems to me that there are at least reasonable grounds for adopting a working hypothesis that for Old English and Old Saxon, syntactically preposed modifiers – most relevantly for my purposes, prepositions – could be phonological enclitics. This idea was already anticipated by Whitman (1993: 82–83), as discussed below.

With regards to example (1), Beowulf line 1, I noted the strong tendency to avoid clause-initial prepositions in Old English poetry. This constraint, along with a number of other features of word order, has traditionally been seen through through the lens of the famous satzspitzengesetz formulated by Kuhn (1933: 43):

This particular framing turns on the distinction made by Kuhn (1933: 4–5) between satzpartikeln and satzteilpartikeln, the former being syntactically independent (such as substantive pronouns, various adverbial particles, most finite verbs, and conjunctions), the latter being syntactically bound to an immediately following constituent (prefixes, prepositions, and many demonstrative, possessive, and genitive pronouns). Since the preposition in is a satzteilpartikel, Kuhn’s rule would exclude it from the clause-initial position unless supported by a satzpartikel.

Both of Kuhn’s laws have been much discussed,22 and the literature on them is too vast to review properly here. With regards to the second law, the essential point is that even Kuhn (1933: 43–49) find a noteworthy number of irregularities and exceptions. These are elaborated on by, especially, Momma (1997: 65–75), who observes not only that different Old English poems show varying numbers of breaches of the law, but that these breaches are not all of the same type. For instance, Genesis A – probably one of the very earliest surviving Old English poems, likely from the late seventh or early eighth centuries (Fulk 1992: 61, 348–351)23 – shows a number of verses with clause-initial demonstrative pronouns, as in 1172:

This begins a new clause (punctuated by editors as the start of a new sentence), but its upbeat contains only the demonstrative se “the/that”, which Kuhn’s second law predicts should not happen (Momma (1997: 70 and note 42) finds 28 comparable examples in this poem). On the other hand, Exodus – another early poem, though probably somewhat younger than Genesis A (Fulk 1992: 348–349) – shows a number of clauses beginning with prepositional phrases, the point with which I opened this article. Momma (1997: 71) quotes the following example:

She lists four to five “other examples of this construction” in Exodus in a footnote, but none of these other verses involve a bare preposition. More representative is 397:

That is, these examples all involve a preposition followed by a demonstrative. This type is certainly just as much a violation of Kuhn’s second law as (14), but the difference is potentially prosodically interesting.24

If we accept conclusion that, at least in Old English, Kuhn’s second law should be “disqualified” (Momma 1997: 73–75), and that Kuhn’s generalizations (careful and intelligent as they were) conflate and obscure various matters,25 then a slightly different set of questions arises. Kuhn’s law has its uses as a focus for attention, but it may leave other important considerations out of focus. Kuhn’s major contribution is in drawing attention to the distinction between clause-initial and clause-medial position, and in distinguishing between different kinds of particles – though as his own discussion suggests, in practice an even more fine-grained distinction is often called for, rather than simply lumping prepositions, prefixes, and demonstratives all together as satzteilpartikeln. Perhaps the most important element missing, however, is the question of syllabicity, since both Kuhn and some of his critics, such as Momma, tend to lump cases such as (14), beginning with in, and (15) with tō þām, together, or to only think about the word classes represented in these upbeats. But it is potentially just as important, or more so, that in is monosyllabic and tō þām disyllabic.

An important step in asking precisely these questions was made by Whitman (1993: 69–83), though his discussion appears to have made little impact on later investigators. Like Kuhn, Whitman is interested in clause openings, but his generalization regarding them is rather different. He notes that some types of verses which should conform well enough to Kuhn’s second law are nonetheless very rare (as rare as the openings disallowed by Kuhn’s law). One type is shown by Beowulf 706b, 967b (Whitman 1993: 71):

This verse, which recurs twice as a formula in Beowulf,26 is both times a single-verse subordinate clause. It adheres to Kuhn’s second law, since the sentence particle þā occurs in the upbeat, meaning that there must be some other explanation for its rarity.27 As Whitman notes, the constraint is not syntactic, since verses opening with siððan “after, since”, a syntactically similar particle, does occur regularly in clause-initial position:

Whitman (1993: 72) counts some 35 examples like (17) in that poem alone. His conclusion is as follows:

This “suspicion” must surely be more or less correct in some form. A monosyllabic particle like þā can and does frequently occur clause-initially in verses like the following:

Whitman (1993: 70) claims that this type occurs sixty times in Beowulf. It may be worth noting that in (18), wæs is plausibly itself analysed as a clitic (Y. Suzuki 2008: 122–134, 182–186). Above, I noted that sequences like hwæt wē in (7) at least had the possibility of being taken as hwæt=wē and scanning as a single disyllabic word-foot, ww. I would take Whitman’s assessment of þā as implying just such a process. Where þā on its own is prosodically unsuited to be a proclitic clause opener (either because it is not a PWord clitic, or because it is dispreferred as a proclitic), if it is promoted to be the host for a clitic like wæs (or other elements, such as c(w)ōm “came”, a demonstrative, or a pronoun), the enlarged þā=wæs becomes eligible to be a metrical, clause-initial word-foot. This enlargement matters either for turning þā into a prosodic word by promotion, or for allowing þā to occur as something other than a proclitic. As it will be useful below to suggest a strong dispreference for proclisis for other elements, and since þā rarely occurs as a clearly anacrustic element, I suggest that the second option is correct: that verses like (16) are rare because þā is strongly dispreferred as a proclitic.

In þā=wæs, the prosodically deficient þā would be contextually promoted to the status of PWord (arguably no longer remaining a clitic at all; cf. note 19 above), with wæs attached to it as an enclitic. The resulting phonological and metrical structure might provisionally be represented, in a somewhat simplified manner, as follows (WF here stands for word-foot):28

This kind of synthesis, adapting Whitman’s broad approach to fit within the frameworks of clitic phonology and word-foot metrics, is what I will attempt to apply in the remainder of this discussion to three common classes of “little word” in Germanic verse.

The number of positions þā can appear in is limited by the fact that outside of clause-initial position, it is always metrically stressed. It is easier to develop the prosodic profile of light elements by turning to prepositions (Whitman 1993: 81–83). Returning to example (1), it is clear that there is at least some parallelism with þā: it is extremely hard to find a clause in Old English verse whose upbeat consists only of a preposition. Exceptions like (14) can be found, but they are very rare, just as is the type with þā represented in (16). On the other hand, such verses (i.e. like in geār-dagum) are very common indeed in clause-medial position. Whitman (1993: 82) generalizes that “each [verse of this sort] can be read without pause in the rhythm of the preceding half-line”. This can be understood as a suggestion that such prepositions are enclitic to elements in the preceding verses, though they are at the same time able to stand as light word-feet in their own verses. I suggest that prepositions are PWord clitics as defined above: prosodic words that retain their clitic status (and are not promoted out of clitichood). The following is an attempt to represent this suggestion schematically, using Beowulf line 1 (the dotted line shows the clitic attachment across the verse):

If prepositions and þā are taken as prosodically more or less parallel – even if this parallelism is limited by the aforementioned inability of unstressed þā to occur clause-medially – then the following hierarchy of preferences might be suggested for both þā and prepositions:

When verse-initial, both þā and prepositions normally map to light word-feet in the metrical system, which I take to be an indicator that these are specifically PWord clitics.30 As has long been noted in the metrical literature, there are only a handful of examples where these elements occur in anacrusis, a context which is (by my working hypothesis) prototypically not associated with prosodic words (Cable 1971). Two of these occur in Beowulf,31 six or so in Genesis A:32

All of these examples are clause-medial. A clause-initial example with an adverb derivationally related to a preposition is found in Beowulf 2093a:

The overall picture is that prepositions, like þā, strongly prefer to map onto light word-feet when this is possible, i.e. in verse-initial position. These few exceptions are just that: instances where PWord enclitics have, exceptionally, been treated as extrametrical rather than as word-feet. It may be worth noting that the repetitions found in (21b) and (22) are typical for the examples in Beowulf, and in Genesis A 2660a, all of which show echoing prepositions before the second lift of the verse.33

If þā and prepositions can be taken as PWord enclitics because they normally map to light word-feet (where this is possible), rarely stand in anacrusis, and rarely occur in contexts where they must be proclitic, then the prosodic status of unstressed “prefixes” must be analysed differently. Unstressed prefixes are common in Old English: typical examples include ā·gifan “give, grant” and ge·wyrc(e)an “make”. There are good reasons to regard such prefixes as standing outside the prosodic word of the root they are attached to. Minkova (2008: 28–29) cites four features that are suggestive of a significant prosodic boundary between prefix and root:

Not all of these points are equally strong – we have no direct evidence for the lack of voicing in the Old Enlgish period itself, and the later voiceless outcomes could, conceivably but perhaps not plausibly, all be analogical34 – but they all point the same way, and may receive some support from scribal habits.35

Metrically, too, the word-foot theory requires unstressed prefixes to be outside the prosodic word. Words like folc “people”, hyrde “shepherd”, byrnende “burning”, and ende-lāf “final remnant” all project word-foot contours used to form verse-shapes (S, Sw, Ssw, and Sws, respectively). By contrast, as noted above, there are no word-feet of the shapes ˣwS or ˣwSw that might be expected if ge·trum “troop, company” or ge·wyrcean were normal prosodic words. This decision to treat prefixes as prosodically distinct “words” is an essential component part of the theory, which will not generate the correct outputs of verse types if “rising” word-feet are allowed,36 an analytical necessity which aligns well with the phonological considerations raised by Minkova.

If prefixes are not simply a part of the same prosodic word as their host, what is their status? They are presumably clitics, but they would seem to be a different type of clitic than prepositions. This can be seen clearly by contrasting on as a preposition and a prefix in a sample of Genesis A (lines 1001–1500); the preposition occurs in this passage a total of 77 times, and the prefix 20 times. In this section, the majority of both varieties of on appear in medial dips, which are, as noted in §2, metrically uninformative.

Only in verse-initial position will the metre potentially treat different kinds of light syllable differently.37 For prepositions, the passage in question contains 29 examples serving as a light word-foot, as in:

By contrast, of the 20 instances of on as a prefix in this same sample, just one serves as a light word-foot:38

That is, about 38% (29 out of 77) of prepositional ons clearly serve as light word feet, while only 5% (1 out of 20) of prefixes do. Under the terms used here for attempting to integrate metre and phonology, the implication is that while prepositions are PWord enclitics (as argued above), prefixes lack any status as prosodic words. Examples like (24) are probably best regarded as exceptional mismatches of linguistic and metrical structure, allowed here and there, but not particularly optimal.39

This suggestion is reinforced by the relative readiness of prefixes, unlike prepositions, to stand in anacrusis (Cable 1971). In Beowulf, there are perhaps 52 examples of prefixal anacrusis,40 compared to the three examples of prepositional anacrusis given in note 31. The difference is not so stark in Genesis A, but clear nonetheless, with some 22 examples of prefixal anacrusis in the entire poem against the six or so prepositions given in note 32 above.41 It would seem most natural to consider prefixes as Affixal clitics.

Whether prefixes are pro- or enclitic is not entirely clear, though on the whole I would lean towards proclisis. A weak argument for enclisis might come from verses such as Beowulf 1753b:

Since eft ge- forms a word-foot, an analysis as eft=ge could perhaps be entertained. That said, this point can be turned on its head, since verses like the following are considerably more common than ones like (25):

This preference might suggest that cases like (25) are considered less optimal in terms of language-metre mapping, supporting a proclitic analysis. This may be reinforced by the simple observation that prefixes are inseperable from their roots in Old English: ge=drēoseð, blocking any further insertion after the prefix.42 Verbal prefixes are also much more common in clause-initial position than prepositions,43 which is easier to understand if prefixes are already proclitic. At the least, it seems clear that proclisis is not as disfavoured as it is for prepositions.

Overall, it is clear that prefixes differ in their prosodic status from prepositions, and are treated quite differently in metrical terms. To reconcile the observed facts with the theoretical frameworks adopted here, the simplest suggestion is that, where prepositions are PWord clitics, prefixes are Affixal clitics. Furthermore, where prepositions seem to be rather strongly preferentially enclitic, I would tentatively suggest that prefixes were (as an innovation relative to older Germanic),44 preferentially proclitic. The relatively uncommon use of prefixes as light word-feet is – as has often been suggested by people working in various metrical frameworks – best regarded as a point of latitude in language-metre mapping.

I now, and finally, consider the demonstratives. These are interesting not only for their exceptional behaviour in Old English poetry, as noted already by Kuhn (1933: 44–45), but also for the way their metrical status contrasts in Old Saxon alliterative verse. In Old English, demonstratives account for a large chunk of the violations of Kuhn’s second law, as discussed above in relation to (13):

This certainly begins a new clause, with the demonstrative alone filling the light word-foot of the verse, which scans w/Sws. The relative frequency with which demonstratives can stand in comparable positions suggests two things. First, that in contrast to prepositions, demonstratives are more freely accepted as proclitics, since enclisis is not possible in verses like (13). I would not claim that demonstratives are preferred as proclitics, but it seems that they at least lack the bias against proclisis observed for the prepositions. Why demonstratives and prepositions should vary on this point is not obvious – naturally so, since it is currently not clear why prepositions should be so apparently strongly enclitic, or why þā should be more frequently tolerated as a proclitic in Genesis A than in Beowulf.

The second thing suggested by verses like (13) is that demonstratives are, like prepositions, prosodic words in their own right (whatever else they may also be), and preferentially map to (light) word-feet in the metre – by the working assumptions adopted here, I would understand them to be PWord clitics, mapping to w rather than (x) or S. The resulting expectation that they should be avoided in anacrusis, where they would have to be extrametrical, holds true in practice. Genesis A shows a single example of a demonstrative in anacrusis:45

The most distinctive feature of demonstratives comes not from their behaviour within Old English, but in comparison to Old Saxon. The sizeable corpus of alliterative metre represented by the Heliand and the Vatican Genesis, both dating from the ninth century, is clearly closely related to the Old English poetic corpus, and arguably should be treated as a part of it as a matter of course (though this is not the current habit of Old English scholars).46 By and large, the light elements of Old Saxon behave much the same as in Old English: thuo (þā) is almost never a lone light word-foot at the start of a verse (cf. (16)),47 and prepositions strongly prefer to occur verse-medially.48 The following are typical for Old Saxon; compare (18) and (1) for Old English:

Prefixes too behave much the same as in Old English: they rarely occur as metrical word-feet,49 and when verse-initial are usually in anacrusis.

The mere fact that prefixes can stand in anacrusis is not particularly meaningful in Old Saxon, where such extrametrical additions are both much more common and more tolerant of a wide range of fillers (in terms of element class and number of elements) than in Old English (Hofmann 1991a: 66–68; S. Suzuki 2004: 160–175). Prepositions and even verbs can stand in anacrusis (in the following example, the alliteration is on the vowels):50

The difference between prepositions and prefixes lies not in whether they ever occur in anacrusis, but rather in the near-invisibility of prefixes. Prepositions very often stand as light word-feet, and are only more occasionally anacrustic, while prefixes show the reverse behaviour: routinely occurring in anacrusis and only less commonly standing as light word-feet.51

If demonstratives were broadly comparable in Old English and Old Saxon, then they should behave much like prepositions (but perhaps with a greater freedom to occur clause-initially), but in practice they resemble prefixes much more closely. The first 300 lines of the Heliand contain some 16 examples of a lone demonstrative in anacrusis (cf. the 19 instances of prefixal anacrusis listed in note 51):52

At the same time, demonstratives (in contrast to prepositions) rarely serve as metrical word-feet.53 The first 300 lines of the Heliand contain no examples at all, though Hofmann (1991a: 140) does cite one instance from much later in the poem (biti is resolved):

For comparison, lines 1001–1300 of the Old English Genesis A contain 11 examples of word-foot demonstratives:54

That demonstratives have become more acceptable in anacrusis has often been observed (Hofmann 1991a: 66–67; Russom 1998: 156–157), but Hofmann (1991a: 140) maintains that they are still unlike prefixes in more often being metrical constituents (light word-feet, in the terms used here). This does not seem to be correct, and the standard explanation for demonstrative behaviour in Old Saxon probably needs to be somewhat emended. The greater latitude towards unstressed elements in general, and demonstratives in particular, is often attributed to their greater syntactic necessity, and the decreased possibility of deleting such elements (Russom 1998: 156–157), but this explanation can only account for the increased presence of demonstratives in anacrusis, not their decreased presence as word-feet. Instead, though I would continue to see the change as driven by the shift of demonstratives to definite articles, I would suggest that this has led not merely to a syntactic change (more weak syllables that need to be metrically accommodated), but to a specifically prosodic one. Specifically, demonstratives – or perhaps better now, articles (Kuhn 1933: 45) – have stopped being PWord clitics, and have become Free (or, less plausibly, Affixal) Clitics, with no status as a prosodic word at all.

Much the same trend may be observable in late Old English verse. In The Battle of Maldon (hereafter just Maldon), a poem composed in or (probably not long) after 991 and consisting of 325 surviving lines, there are three very well-known examples of demonstrative anacrusis (Russom 2017: 95–96; Griffith 2024: 64–65, 161):55

This is certainly a much lower rate of demonstrative anacrusis than is found in the Heliand (found in less than 1% of lines, whereas the first 300 lines of the Heliand show a rate of about 5.3%). What is more striking, however, is that, just as in Old Saxon, demonstratives almost never stand as the sole light word-feet of verses in Maldon. The sole example is 121a:

The contrast between Old Saxon and Maldon, on the one hand, and on the other Genesis A, and its 11 examples per 300 lines (note 54) is striking. I would suggest that prosodically, demonstratives (or articles) had already lost their PWord clitic status by the time of Maldon, but the metrical or stylistic mode of the poem had not moved as far towards embracing anacrusis as had Old Saxon poetry a century and a half earlier.

Reconstructing the prosody of languages now only known through textual corpora is never an entirely obvious or straightforward task. In the case of the early Germanic languages, poetry is one of the most valuable sources of information: that the metre involves the prosodic regulation of at least some aspects of the languages is obvious. Actually exploiting this source requires adopting several working hypotheses at once, looking for points of productive congruence or alignment, and seeing how useful the whole exercise is in the end at explaining the data (and even a failed exercise may have some use in drawing attention to new aspects of the data). The main working hypotheses I have adopted here are:

The data in question comes from verse-initial low-stress elements in alliterative poetry, with the main dimensions being whether they are extrametrical, metrically necessary, or in “expanded dips” (in which case the metricality of any given constituent is less clear-cut).

The extent to which this attempt at theoretical integration has been successful depends on how useful it is in understanding the data. The main results are summarized in Table 1. This lays out the behaviour of prepositions (Prep), prefixes (Pfx), and demonstratives/articles (Dem) in classical Old English poetry of roughly the eighth century (represented here by Genesis A (GenA), agreeing in most respects with Beowulf), the Old Saxon Heliand (Hel) of the ninth century, and, back in English, Maldon (Mald) of the late tenth.56 Each word class in each corpus is labelled for whether it should be considered a PWord clitic (PWd) or not (as assessed by its preference for serving as light word-feet versus standing in anacrusis) and whether it is preferentially pro- or enclitic (Pro or En). Demonstratives are here labelled as Pro/En, since they seem to show no particular bias against being proclitic (unlike prepositions, they can occur in positions where there is no host for them to be enclitic to), but this does not mean they are necessarily preferentially proclitic – and in later Germanic, they are often enclitic.

This kind of synthesis certainly has its limits. The reasons why demonstratives/articles should apparently be more flexible in their direction of attachment is not obvious (a comparison with Norse, where a different demonstrative stem developed a very strong preference for enclisis, might be instructive), and there is a further trajectory apparent in the shift of prefixes from separable, often enclitic elements in Gothic and (presumably) Proto-Germanic to closely bound proclitics (a status that is then stable across the corpora examined here). Even so, it seems to me that drawing a distinction in clitic “strength” (modelled here in PWord/non-PWord status, though this could potentially be formulated in other terms) is clearly useful in explaining not only the synchronic distinctions between, e.g., prepositions and prefixes in Genesis A, but also in accounting for the diachronic history of demonstratives. The well-known syntactic development of these into articles can be paralleled by a simple categorical shift in status to the weaker (non-PWord) type of clitic. The idea that prepositions are often (even strongly preferentially) enclitic in early Germanic also emerges as a strong conclusion, anticipated by Whitman (1993). These are encouraging results, and suggest that the integration of metrical theories, phonological frameworks, and philological data can, despite the inherent uncertainties and difficulties, be productive.