3.1 Faroese phonology

Before describing the meter, some basic features of Faroese phonology must be noted. Since Faroese orthography is highly etymological, e.g. retaining ð even when silent or a glide in the modern language, an IPA transcription is provided for each line, including indications of primary and secondary stress. Historical long vowels á, í/ý, ó, ú have become the diphthongs /ɔaː ʊiː ɔuː ʉuː/ in modern Faroese, all of which alternate with short counterparts /ɔ ʊI œ y/ when occurring in a closed syllable. Open and closed syllables are defined as in other Scandinavian languages, i.e. a syllable is closed if the nuclear vowel is followed by a long consonant or a cluster. A single word-final consonant does not close its syllable, and hence does not cause a preceding stressed vowel to become short, e.g. lok /loːk/, not */lɔk/. I transcribe vowel length phonemically, e.g. I represent the word at as /εaːt/ even though in rapid speech it tends to be reduced to [at] or [ət]. Unstressed a, i and u are represented as [a I ʊ], somewhat of an idealisation since there is much dialectal variation, e.g. merging /I ʊ/ to [ə] or [ø].Footnote ² I follow Þráinsson et al. (Reference Þráinsson, Petersen, í Lon Jacobsen and Svabo Hansen2012) in transcribing the voiceless unaspirated stops b, d and g as devoiced [b̥ d̥ g̥], and the stops p, t and k with pre-aspiration when appropriate. The transcription is intended to be broad and represents the dialect of an idealised Tórshavn speaker. Perhaps most relevant to meter, the rules for Faroese stress are given in Þráinsson et al. (Reference Þráinsson, Petersen, í Lon Jacobsen and Svabo Hansen2012:28):

1. (2)

   

I assume that the metrical template maps to syllables on a phonological representation of the line. This representation minimally includes main stress, secondary stress and unstressed as discussed in Section 3.4 below. In spoken Faroese, when the alternating secondary stress immediately precedes the article, the article attracts the stress and the weak stress on the inflectional ending is deleted, e.g. fjeppararnar [ˈfjε^hpːa.ɹa.ˌnaɹ], ‘supporters.the.Acc’; the same deletion occurs adjacent to stress in subparts of compounds, e.g. hálvleikur [ˈhɔlv.ˌlaIː.kʊɹ], ‘half.nom (of sport game)’, in which the weak stress assigned by rule (2b) to the final syllable -kur is deleted.Footnote ³ Throughout this paper I use the following notation for the phonological representation of stress: X = primary stress, 2 = secondary stress, x = unstressed.

3.2 Ballad texts

The whole corpus of Faroese ballads, Corpus Carminum Færœnsium (CCF), is collected in Grundtvig et al. (Reference Grundtvig, Bloch, Djurhuus and Matras1941–2003). The University of the Faroe Islands kindly provided me with a digital version of volumes I-V of the corpus; readers may request access to the corpus from the Faculty of Faroese Language and Literature (Føroyamálsdeildin).Footnote ⁴ This excludes the modern ballads known to be composed during or later than the 19th century, two of which I include in my textual sample.Footnote ⁵ The textual sample I used consists of the following ballads (Grundtvig et al. Reference Grundtvig, Bloch, Djurhuus and Matras1941–2003), shown in Table 1:

Table 1. Textual sample of ballads

CCF contains 138,746 lines, 5086 of which are included in my textual sample (hence 3.7%); an additional 807 lines from the modern ballads bring the total in the sample to 5893.Footnote ⁶ I make use of a stress parsing script in order to show, to the extent possible (i.e. in the absence of a complex foot parser for the Faroese meter), that the generalisations made of this small sample also hold of the entire CCF.Footnote ⁷ Moreover, the sample was selected in order to be representative of the range of ballad types: one entire cycle was included, the Sjúrðar kvæði, a multi-part ballad consisting of the three sections indicated in the table by dashes: Regin smiður, Brynhildar táttur and Høgna táttur. As the four-line, 4343 stanza is the most frequent type in CCF, only one two-line ballad, Heljars kvœði, was selected. The two modern ballads Ormurin langi and Virgars kvœði with known authors were also included. The other ballads Runsivals stríðið, Artal kongur í Atlandi and Torsteins kvœði were chosen at random. As can be seen in Table 1, refrains are often of irregular length and a different rhyme scheme than the stanzas; the number of feet can also differ depending on the parse assumed. I have opted for the number of feet proposed for the refrains in the table, since otherwise additional licences are required to a far greater extent even than the most extreme stanza examples, e.g. multiple consecutive unstressed syllables in strong position.

3.3 Left-headed template

I posit that the Faroese meter is best described by a template of left-headed feet. It allows dactylic substitution, i.e. permitting a ternary grouping of syllables ( X xx ) in place of a binary grouping ( X x ), as well as empty weak positions, i.e. a metrical foot consisting of a single syllable. This places it in the broader family of alternating binary-ternary meters such as dolnik (Tarlinskaja Reference Tarlinskaja1992, Reference Tarlinskaja1997) and the meters of sundry English-language folk ballad and children’s verse types (Hayes & MacEachern Reference Hayes and MacEachern1998), since it is defined primarily by stress peaks and is less strict as to the number of intervening unstressed syllables.

The basic structure of a four-line stanza is shown in Figure 1 below, where ‘|’ indicates foot boundaries and parentheses indicate optionality:

Figure 1. Basic metrical template of 4343 stanza.

This template holds for all four-line ballad verses in CCF; additionally, in medieval cycle ballads, empty weak positions are permitted line-medially (see Section 3.5).

The need for a left-headed parse becomes clear upon even a surface-level examination of the ballad texts. Aside from the trochaic pattern of Faroese stress assignment (Árnason Reference Árnason2011: Chapter 13.2.1), a number of pieces of evidence point to such a conclusion: first of all, right-headed parses would require more structure-altering licences. An iambic parse of the following line requires both initial trochaic inversion and an extrametrical syllable at the end of the line (3); contrast a trochaic parse (4), which only requires permitting more than one unstressed syllable between stress peaks.

1. (3)

   

1. (4)

   

This problem is even clearer with lines consisting entirely of trochees, which are best assumed to be what they appear (6), rather than positing a right-headed parse with mismatches, e.g. (5).

1. (5)

   

1. (6)

   

Although anacrusis (an initial extrametrical syllable) is optional, an anacrustic line is better analysed as a left-headed line with a preceding weak position, rather than a right-headed line without anacrusis: otherwise, lines with initial trochees would require parses like (5). Indeed, lines with multiple initial unambiguously left-headed feet are common. Table 2 below shows the frequencies of lines beginning with three consecutive trochees or dactyls by ballad in the text sample. I exclude sequences of two or more initial unstressed syllables, which do not favour left- or right-headed parses by stress alone; I allow for a single initial extrametrical syllable. I have included one instance of the refrain associated with each ballad or section.

Table 2. Frequencies of lines with three initial left-headed feet by ballad

These results show that for all ballads in the sample other than Regin smiður, a majority of lines begin with three or more initial left-headed feet.Footnote ⁸ Those which do not, usually lines beginning with a string of monosyllabic function words, may also be parsed with left-headed feet, but their stress pattern does not force such a parse. Moreover, this generalisation holds of the entire CCF: of the 138,746 lines, 42,008 begin with a sequence of three consecutive dactyls or trochees (30.3%), which would be an incredibly large proportion of primary stresses in supposedly weak positions. Therefore, this is best parsed as a trochaic meter with dactylic substitution.

A core claim of this paper is that the adoption of this trochaic-dactylic metrical template enables us to make empirically sound descriptive generalisations about the meter itself. In the following section, these generalisations are summarised. Data from the textual sample will demonstrate not only that the template is a prerequisite to achieve empirical coverage, but also to rule out unattested line types.

3.4 Summary of descriptive generalisations

Any empirically adequate theory of the basic Faroese meter will have to account for at least the following:

1. 1. Left-headed feet

2. 2. Strong position admits:

   1. a. any one stressed syllable, including secondary stress

   2. b. monosyllabic words not bearing lexical stress

   3. c. at end of line only: unstressed inflectional endings of polysyllabic words

3. 3. Weak position admits:

   1. a. no main word stress

   2. b. up to three unstressed syllables, including whole trochaic function words

   3. c. at end of line only: empty

4. 4. Up to two extrametrical syllables permitted before line only

Having discussed left-headed feet in Section 3.3, I present evidence from the ballad corpus for generalisations 2–4 in Sections 3.4.1–3.4.3 below. In Section 3.5, further licences permitted in cycle ballads are discussed; again, I argue that the metrical template provides a means of capturing the systematic differences between the stricter and looser Faroese meter.

3.4.1 Strong position

Stressed syllables. In strong metrical position, (a) monosyllabic words bearing stress, (b) main word stress in polysyllabic words and (c) secondary stress must all be permitted. The relevant syllables in examples (7–8) from the text sample are given in bold.Footnote ⁹

1. (7)

   

1. (8)

   

Only one stressed syllable may correspond to a strong position; four-foot lines such as the constructed examples (9–10) with more than one stress in strong position are unattested in CCF.

1. (9)

   

1. (10)

   

Monosyllabic function words, which can be stressed or unstressed, can occupy strong metrical positions. A reviewer notes that it could be the case that such words could be parsed as word or phrasal clitics, or as prosodic words in their own right (Selkirk Reference Selkirk, Morgan and Demuth1996, Ito & Mester Reference Ito, Mester and Parker2008); given the lack of preceding work on this specific aspect of Faroese prosody, we cannot assume definitively that these are unstressed. There seems to be no restriction on which strong position in the line these syllables may occupy, as illustrated in examples (11–12) from the textual sample.

1. (11)

   

1. (12)

   

Monosyllabic function words in strong position are frequent in CCF: 6,092 lines begin with a sequence of three monosyllabic words, which includes parses of either an initial extrametrical syllable followed by a trochaic foot ( x | x x ), or a trochaic foot followed by the strong position of the following foot ( x x | x ).

Definite article and compounds. Two subtypes of secondary stress appear to be more restricted in distribution than second elements of compounds: the suffixed definite article, and the rhythmic stress described by rule (2b). The definite article is attested in non-final feet (14), and not only in cycle ballads with the looser meter:

1. (13)

   

1. (14)

   

However, occurrences of the definite article in strong position are found disproportionately at line-ends. Table 3 shows some relevant statistics, in which ‘parsed’ indicates a parse that must hold lest there be an unexpected number of feet in the line.

Table 3. Frequencies of lines with kongurin and -ríki compound

A number of striking results emerge here, some of which appear to be categorical restrictions. First, all 150 lines with the definite article of kongurin line-finally must be parsed with the article in a strong position, lest the line have one foot less than expected in each case. Of the remaining 612 with kongurin elsewhere in the line, only 49 (8%) could be parsed kongur|in ( s w | s ) and not result in an unexpected number of feet. This contrasts with second elements of compounds, which may be bisyllabic and therefore have an unstressed syllable in the final weak position. For example, of the 503 line-final occurrences of the second element –ríki ‘realm, kingdom’ in CCF, 372 (74%) consist of a four-syllable word like himmiríki ‘heaven’ and must be parsed as a trochee, whereas the other 131 (26%) have three syllables like Nøríki ‘Norway’ and Svøríki ‘Sweden’, with the final unstressed –i in strong position. All bar two of the 223 non-line-final occurrences of these words in CCF must be parsed as a dactylic foot, the exception being two instances of the same line, for Nøríki rœður ‘rules Norway’, in which Nø|ríki ( s Ø | s w ) is the most natural parse. This indicates that there is indeed a line-final strong position which permits a wider range of syllable types than line-initially or medially.

The contrast between the four-syllable compounds and suffixed article may simply reflect that the latter is never trochaic or dactylic; however, it also reflects a tendency to avoid weak cadences when they would violate saliency or parallelism, conflicting constraints that favour the 4343 structure (Kiparsky Reference Kiparsky, Dresher and Friedberg2006). For example, the following parse of (13) violates parallelism by creating a three-foot line where four feet are expected:Footnote ¹⁰

1. (15)

   

This same observation holds for the three-syllable compounds with ríki: a line-final parse of Nøríki ( s w ) results in one foot too few in all 131 examples of this type in CCF, while Nø|ríki ( s Ø | s w ) is possible but would introduce systematic misalignment with performance rhythm (see Section 4). Hence, although the meter allows both the definite article and second parts of compounds in strong position, these types of secondary stress are distributed differently in frequency. I propose that this occurs as a result of both a purely metrical pressure (preserve parallelism), and word-phonology (compounds can end in trochees, the suffixed article does not). A result as striking as this, i.e. that 100% of line-final articles in kongurin and 100% of three-syllable -ríki compounds maintain the 4343 stanza structure if and only if their final syllable occurs in strong position, ought not be ignored. If metrical constraints exist at all, we would expect them to produce systematic correspondences precisely as they do in the Faroese data. The null hypothesis, that there is no metrical template to which the phonological representation is mapped, should not yield such patterns.

Weak secondary stress and unstressed inflections. Finally, there is evidence that both weak secondary stress assigned by (2b) and unstressed inflectional endings can occur in line-final strong position in the stricter meter:

1. (16)

   

1. (17)

   

Of the 56 occurrences of bardagar ‘battle’ in CCF, only two are not line-final: two instances of the same line contain bardagar | hildust | á [ˈb̥aɹ.d̥eː.ˌaɹ ˈhIl.d̥ʊst ɔaː] ‘battle continued on’, which cannot have the weak secondary stress in strong position without adding a foot or resulting in a dactylic final foot that does not rhyme. The parse in (17), as opposed to the most obvious alternative Virgar | leyp úr | gull|stóli, is supported by a performance by the Sumba dance society, in which the unstressed final –i of gullstóli receives a strong dance step and strong melodic beat.Footnote ¹¹ This lends further support to the purely metrical preference for a line-final strong position, even when the occupying syllable is unstressed, a generalisation difficult even to formulate without the template. Thus we have evidence that any syllable type is permitted in strong position, but following a kind of markedness hierarchy, presented in (18) from most to least marked:

1. (18)

   $$\scale70%{\left\{ \matrix{ \,\,\,\,\,\,\,{\rm{Unstressed}} \hfill \cr {\rm{inflectional}}\,{\rm{ending,}} \hfill \cr \,\,\,\,\,\,\,{\rm{Rule}}\,(2{\rm{b)}} \hfill \cr} \right\} \gg \,{\rm{Definite}}\,{\rm{article}}\, \gg \,\left\{ \matrix{ \,\,\,\,\,\,\,{\rm{Stress}}\,{\rm{in}}\,{\rm{second}} \hfill \cr {\rm{part\,of\,compound,}}\,\,\,\, \hfill \cr \,\,\,\,\,\,\,\,\,\,\,{\rm{Main}}\,{\rm{stress}} \hfill \cr} \right\}$$

To summarise the empirical evidence for this hierarchy, although it is impossible to count every instance of the relevant syllable types due to the absence of a Faroese foot parser, some high-frequency examples are instructive: unstressed inflectional endings (Nørík-i and Svørík-i) and weak rhythmic secondary stress (bardag-ar) only occur in strong position line-finally in the stricter meter; the definite article in kongur-in tends to occur in strong position at line-ends but is attested line-medially; and both stress in second parts of compounds (himmi-ríki) and main word stress are attested in any strong position. Further work is required to establish that these trends hold more broadly across the ballad vocabulary, but it is clear that for these frequent tokens at least, the generalisation in (18) holds, and can only be understood in terms of the metrical template.

3.5 ‘Loose’ ballad meter

There are two striking exceptions in CCF to the generalisations laid out in Section 3.4. It turns out that some ballads permit (i) empty weak positions in feet other than the last in the line, and (ii) unstressed inflectional endings in strong positions other than line-finally. The ballads in which we find unequivocal examples are of medieval origin, the majority of which constitute multi-part cycles. 155 lines in CCF consist of only three syllables; of these, 138 come from long medieval cycles, including the Sjúrðarkvæði or Sigurd cycle, Karlamagnusar kvæði or Charlemagne material, or similar cycles whose medieval source is not disputed; the remaining 17 are found in non-cycle ballads of medieval origin.

In order to account for lines consisting only of three or four syllables, we have two options: posit empty positions (36) or empty feet (37).Footnote ¹⁴

1. (36)

   

1. (37)

   

A clear reason to avoid option (37) is the problem of overgeneration, since it is unclear how extremely short unattested lines, e.g. a line of a single syllable, could be ruled out. It would be more plausible to allow empty weak positions, since such a licence exists in English folk verse, and has already been suggested for meters which allow multiple unstressed syllables between strong positions, e.g. Hopkins’ sprung rhythm (Kiparsky Reference Kiparsky, Kiparsky and Youmans1989). There is evidence that in some ballads, any foot can be monosyllabic (38), and any strong position can be occupied by an unstressed syllable (39).

1. (38)

   

1. (39)

   

Such stanza lines appear to be limited to three ballads in the text sample: Regin smiður, Brynhildar táttur and Runsivals stríð, all of which are sections from heroic cycles. However, this looser meter is not limited only to multipart cycles in the entire corpus, as indicated by e.g. Hermundur illi. This typology of meter represented by the stricter and looser meter is readily captured by the templatic approach: in the former, certain licences are permitted only in the final foot, whilst in the latter they can also occur in line-medial feet. It remains to be seen whether these same facts can be accounted for without assuming the template model.

4.1 The ring dance

The Faroese ballad tradition is performed in a unique combination of text, music and dance. The performers sing the text in unison while simultaneously dancing in a chain formation. One individual, the skipari ‘skipper’, is in charge of leading the song, which requires remembering the correct lyrics of the stanzas in order. There is usually a particular melody associated with the given ballad, and the stanza melody often differs from that of the refrain, which can have more rhythmic irregularities than the stanza. The singing is unaccompanied by instruments, and there are no ‘parts’ or harmonisation, i.e. only the melodic line itself is sung. The dance consists of a simple repeated L(eft)-L(eft)-R(ight) pattern of strong or ‘leading’ steps ad infinitum in a closed circle, with a following movement of the other foot, which is often inaudible. The strong steps provide a rhythmic backbone to the performance, and reflect the musical structure, in that the strong step in most cases marks the beginning of a musical measure (for further information on the relation between the dance and melody, see Luihn Reference Luihn1979 and Clausen Reference Clausen and Clausen2003).

Importantly, this repeated LLR pattern of strong steps does not restart at the beginning of a line, stanza or refrain: rather, the pattern continues throughout the performance, regardless of where lines or stanzas begin or end. Therefore whether a given line begins with a L or R strong step (or a weak following step) depends entirely on how the steps on the immediately preceding line happened to end. As a reviewer notes, a consequence of this will be that even if one strong step occurs per metrical foot, the fact that many lines have four feet means that a stanza line may occur with an LRLL, RLLR or LLRL pattern, depending on where the steps fell on the previous line.

The key insight brought by the performance data is that, barring mismatches as discussed in Section 4.2, if there is a strong dance step, there will be a strong metrical position. There is not a single instance of a strong dance step falling on a weak metrical position in the Faroese-language data collected.Footnote ¹⁵ Hence this may be brought to bear on determining the best of multiple possible parses for metrically ambiguous lines in the Faroese-language ballads. The recorded selection of ballad excerpts are listed below; the recordings consisted of audio, video of the entire dance, and video focused on the feet of the dancers, in order to be certain of when the leading steps occurred. Complete audio and video files of the recordings, as well as dance step annotations on the texts, are accessible permanently from the Stanford Digital Repository.Footnote ¹⁶

1. 1. Regin smiður, medieval heroic ballad from Sjúrður (Sigurd) cycle: 131 verse, 4-line ballad, 4-line refrain, vv. 1–25.

2. 2. Torkils døtur, medieval legendary ballad: 50 verse, 2-line ballad, 4-line refrain with refrain fragment in first verse line, vv. 2–17.

3. 3. Ormurin langi, modern ballad on heroic theme: 85 verse, 4-line ballad, 4-line refrain, vv. 1–13.

4. 4. Flóvin Bænadiktsson, medieval ballad of chivalry: 25 verse, 4-line ballad, 2-line refrain, vv. 1–16.

5. 5. Ólavur Riddararós, medieval supernatural ballad: 38 verse, 2-line ballad, 3-line refrain with first line between verse lines, whole ballad.

6. 6. Grindavísan, modern vísa in gøtudanskt (historical variety of Danish spoken on the Faroes): 58 verse, 4-line ballad, 2-line refrain, vv. 1–10.

In Section 4.2, the relation from meter to rhythm in the Faroese performance data is discussed; in Section 4.3, evidence from the performance is brought to bear on the metrical template hypothesis.

4.2 Meter and performance

The locus classicus on meter and musical rhythm is Lehrdahl & Jackendoff (Reference Lehrdahl and Jackendoff1983), which presents metrical structures as inferred by listeners to music, in the form of a grid representing alternating beats. Their model imposes the grid on non-vocal music also, and does not necessarily entail identity of musical and textual meter. I assume the ‘meter’ of Lehrdahl & Jackendoff (Reference Lehrdahl and Jackendoff1983) to be in fact a representation of musical structure only, and the meter of the text to be distinct, for reasons delineated below. This view finds support in the ethnomusicological literature, for example Temperley (Reference Temperley2000:85–86), who notes that there are mismatches between clapping and strong metrical beats in African children’s songs, suggesting that the relation between these levels of rhythm is not always a one-to-one mapping.

The key generalisation about the relation between meter and performance in the Faroese data is given in (40).

1. (40)

   

This principle is not bidirectional: crucially, not all strong metrical positions line up with strong dance steps, as shown below in a refrain from Regin smiður (41).

In the examples below, I notate dance steps as follows: the uppercase L(eft) and R(ight) represent strong/leading steps, and the lowercase l and r indicate weak/following steps. Here the colon indicates a splitting of the metrical foot into two equal bars of music. On the fourth line I have indicated musical bars by double pipe and strong beats by asterisks. The same symbols are used in the annotations of the recordings available at the Stanford Digital Repository. The strong positions that do not line up with strong dance steps in (41) are bolded.

1. (41)

   

What example (41) shows is that for refrains at least, a dance step does not necessarily correspond to a separate foot, but a single foot may receive more than one dance step. In this refrain, each dance step whether strong or weak corresponds to a musical stress. Rather than following the dance steps and musical stresses exactly for my metrical parse, which would yield a structure of 6646, I have assumed that dance steps may occur on weak positions (as is clear from other performance data, e.g. Flóvin Bænadiktsson in which all weak positions receive a weak dance step). Thus it seems that stipulating separate but linked metrical and musical grids leads to a more consistent overall characterisation of the system. Mismatches operate on a regulative principle of binarity (42):

1. (42)

   

In (41) the music is effectively ‘double-speed’ relative to the meter: two strong musical beats per metrical foot if there is a mismatch. We see from line 2 here that the music-to-text correspondence may change even within a line, since the ratio for the first two feet is 1:1. Nonetheless, this observation on ratio falls out simply from the fact that the music cannot be dissociated from the dance, which since humans are bipedal creatures is going to operate on a principle of two. It is clear then that the position that the meter and musical rhythm are one and the same is not tenable for Faroese, although there is a general mapping principle—and one robust enough that a strong dance step is a very reliable predictor of a strong metrical position. The data therefore support a position which construes the metrical and musical structure as generally congruent but not identical (cf. Hayes & MacEachern Reference Hayes and MacEachern1998); the lack of a one-to-one mapping, however, does not render performance data irrelevant to meter (pace Fabb & Halle Reference Fabb and Halle2008). These theoretical issues are discussed further in Section 5 below.

Generalisation (40) is robust enough to enable us to distinguish between potential metrical analyses. For instance, a reviewer raises the question of whether the apparently tripodal stanza lines in the 4343 pattern could be catalectic, i.e. have a fourth foot that lacks prosodic exponence but may show up in the music or choreography. This is the analysis of 4343 stanza types in English folk verse proposed by Hayes & MacEachern (Reference Hayes and MacEachern1998), for example. The dance patterns in the data collected show that this is not the right analysis for the Faroese 4343 stanzas, however: across the board, a three-foot line in such stanzas maps either to three strong steps, e.g. Lr|Lr|Rl, or a combination of three strong or weak steps, e.g. L|r|L or r|R|l, depending on whether weak steps are also mapped to strong positions. Catalexis would predict instead four strong steps or another combination of four steps, i.e. something like Lr|Lr|Rl|(Lr) or L|r|L|(r), in which the foot in parentheses is only expressed in the music and dance but has no prosodic expression. This is not what the recordings suggest, however; to illustrate, a representation of the performance of fourth stanza of Ormurin langi is given in (43):

1. (43)

   

As seen in the second line of (43), blá does not receive two dance steps, and there is no dance step between the end of this line and forgyltan at the beginning of the next. If there were a catalectic fourth foot, we would expect the strong L step at the beginning of the third line instead to fall after blá and prior to the stressed syllable of forgyltan, which is not the case. For the 4343 ballads, there were no examples of a line with three prosodic feet followed by a prosodically empty foot that received a dance step or musical measure.

In the following section, examples from the recordings are presented in support of the metrical template.

4.3 Support for template hypothesis

Left-headed feet. The proposal that feet are left-headed in this meter is supported by the fact that a trochaic performance is given to lines with multiple unstressed syllables (44–46).

1. (44)

   

In (44) a strong dance step falls on á, suggesting that it occurs in strong position, and hence that the foot is trochaic. Similar observations hold of (45–46):

1. (45)

   

1. (46)

   

Note that (46) contains an example of ‘filler words’ sung in performance before the beginning of the line, indicated by brackets in the parse as not belonging to the metrical grid. This implies a preference for a ternary rhythm in performance, since these anaptyctic vowels are inserted only between the final strong position of a line or stanza and the first strong of the following line or stanza.Footnote ¹⁷

Dactylic substitution. Examples of more than one unstressed syllable in weak position abound in the recordings, three of which are given below (47–49).

1. (47)

   

1. (48)

   

1. (49)

   

Empty weak positions. There are multiple examples of strong dance steps occurring on line-final syllables (50–51) and in lines without enough syllables to construct more than two feet if empty weak positions are not permitted (52). Note that when each metrical foot receives two dance steps and the weak/following step is not notated on the final weak position of the line, as in (51), the weak step in question occurred on an initial extrametrical position of the following line.

1. (50)

   

1. (51)

   

1. (52)

   

Again we see that instances of lines with non-line-final monosyllabic feet come from the long heroic ballad cycles, in this case Sjúrðarkvœði. The same phenomenon occurs in refrains, which seem always to permit more metrical licences than stanzas (53).

1. (53)

   

Stressed syllable types in strong position. The recordings also provide support for permitting (a) monosyllabic words bearing stress, (b) main word stress in polysyllabic words, and (c) secondary stress in strong position (54–56).Footnote ¹⁸

1. (54)

   

1. (55)

   

1. (56)

   

I also found evidence for the suffixed definite article in strong position:

1. (57)

   

Unstressed syllables in strong position. As proposed in Section 3.4.1, there is evidence for (a) monosyllabic function words and (b) unstressed inflectional endings in strong position (58–61).

1. (58)

   

1. (59)

   

1. (60)

   

1. (61)

   

The only ballad in the recorded sample which allows an unstressed inflectional ending in non-line-final strong position is Regin smiður, i.e. part of the Sigurd cycle; this is consistent with the observation that the long medieval ballads have a looser meter.

Up to three unstressed syllables in weak position. Several examples of trochaic function words along with a third syllable in weak position are found in the performance, two of which are given below. The lack of a strong dance step lends further credence to these syllables occupying an interval between strong positions.Footnote ¹⁹

1. (62)

   

1. (63)

   

4.4 Summary of performance data

In this section, it has been argued that a general principle of mapping from metrical structure to rhythm in performance is borne out by recordings of dance performances of the kvæði. It was proposed that this principle is violated by mismatches between meter and musical rhythm which can and do occur, but that when they occur the mapping is subject to a principle of binarity: performance restrictions will trump meter, but in a regulated way. Given the robustness of the prediction that a strong dance step will mark a strong metrical position, this can be employed in support of claims made about the meter (as in Section 3.4 above). In the following and final Section 5, I will discuss alternative approaches which either do away with the template altogether, or collapse metrical structure into music or purely phonological constraints. I discuss their implications in reference to the Faroese tradition, and conclude that the approach I have advocated here constitutes a better overall analysis of the data.

5.1 No template, disconnected performance?

The model put forward here contrasts with one recent proposal, Fabb & Halle (Reference Fabb and Halle2008), which although it shares some assumptions inherent to other generative approaches, is unique in that its basic claim is that all meters are generated via bottom-up parsing. It assumes a combination of iterative rules and constraints, but does away with the metrical template, and represents a substantial rethinking of the correspondence between meter and phonology. I contend that due to these problematic starting assumptions, Fabb & Halle (Reference Fabb and Halle2008) will by necessity both overgenerate and miss generalisations, in contrast to the approach I have outlined above.

Although Fabb & Halle (Reference Fabb and Halle2008) do not provide an analysis of Faroese meter, it can be demonstrated that their rejection of the template will lead to greater difficulty explaining the descriptive generalisations laid out in Sections 3–4. One of their central claims is that ‘the metrical grid is generated from each individual line, not preconstructed’, and that ‘the computation that generates the grid follows an ordered application of a licensed set of rules’ (Fabb & Halle Reference Fabb and Halle2008:11). The metrical structure is generated by iterative rules which insert parentheses on gridlines; this functions essentially as a grouping mechanism for items that correspond to syllables or higher nodes in hierarchical structure. This hierarchical structure is not templatic but built up bottom-to-top from the base gridline where syllables are projected. Example (64) illustrates how a well-formed grid is generated from a line of verse.

1. (64)

   

Syllables project as asterisks to gridline 0, after which an iterative rule groups asterisks by insertion of a parenthesis at a given interval. The heads of each grouping project to gridline 1, and again an iterative rule applies; this continues until there remains only a single ungrouped asterisk on the highest gridline, the main criterion for well-formedness, though further filters and constraints may apply.

On this theory, since the rules for generating the grid make no reference to metrical positions other than syllable groupings, the regularity of correspondence between strong dance steps and strong metrical positions is not accounted for. Headedness of syllable groupings might be a good candidate for an equivalent to strong positions, but this will vary greatly with respect to stress, e.g. the least stressed syllable in a line may end up as the head of a grouping. It is possible to stipulate non-iterative rules which privilege stressed syllables, as Fabb & Halle (Reference Fabb and Halle2008:67–69) do for looser meters: they allow for non-projecting syllables and insertion of parentheses to the left or right of ‘maxima’, which mitigates the headedness problem. However, despite undergoing these readjustment processes, maxima may or may not end up corresponding to strong beats in performance. In addition, other empirical generalisations lose the straightforward explanations they have on the template-based approach. For example, the typological differences between the stricter and looser meter, such as allowing unstressed syllables in strong position line-finally versus elsewhere, raises the question of how the rules yield regularities in an apparently arbitrary fashion. Moreover, it is difficult to rule out unattested line types, such as those with lexical stresses in weak position, if weak positions do not exist. These issues arise not from a specific analysis of Faroese, but as a direct result of rejecting the metrical template and positing that meter and performance are disconnected.

5.2 Metrical and musical rhythm identical?

A second alternative proposal concerns the relation between metrical structure and musical rhythm: one could assume that the metrical and musical grids are not distinguishable abstractions, i.e. that strong beats in performance and strong metrical positions are one and the same. Although it is not the main concern of their paper, Hayes & MacEachern (Reference Hayes and MacEachern1998:475) make this assumption, adopting the rhythmic grid representation of Lehrdahl & Jackendoff (Reference Lehrdahl and Jackendoff1983), an assumption also made in Hayes & Kaun (Reference Hayes and Kaun1996). While there is not space to discuss these approaches in detail here, it is noteworthy that the Faroese tradition speaks especially to the issue of rhythm in meter and music, and calls this position into question. Examples like (41) in Section 4.2 seem to falsify the hypothesis that metrical strong positions and musical strong beats are identical: while there is a general alignment of strong beats, strong positions and stressed syllables, examples abound in the ballad recordings of refrains with strong metrical positions that do not align with strong dance steps and/or strong musical beats. For example, in Ólavur Riddararós alone, every instance of a refrain contains 5 correspondences of weak/following dance steps and weak musical downbeats with strong metrical positions, which amounts to 180 such misalignments in the recorded sample; an example is shown in (65) with misalignments bolded.

1. (65)

   

It is significant that these examples all come from refrains rather than the regular stanzas, suggesting that more textsetting variation is permitted in the former, but it nonetheless remains clear that the metrical and musical grids cannot be identical in these cases.

5.3 Meter reducible to phonology?

If it were possible to account for the descriptive generalisations laid out in this paper without meter-specific apparatus, and instead rely on more general phonological constraints, we would arguably have a more parsimonious theoretical solution. Tomas Riad (p.c.) commented that the Faroese data do not provide evidence one way or the other with respect to the traditional templatic approach versus a theory in which metrical prosody obeys a phonological constraint more strictly than in the prosody of normal language (Riad Reference Riad2017). For example, the generalisation that a main word stress never occurs in weak position could be an instantiation of adherence to a privileged constraint such as NoCLAsʜ prohibiting adjacent prominences.

However, there are a number of considerations that render the templatic approach still the better option. Firstly, if we allow performance to have some say in our evaluation of metrical hypotheses, in the Faroese tradition, a strong dance step is predictive of a strong metrical position, not a word-stress—i.e. the head of the metrical prosodic constituent. Hence, NoCLAsʜ penalises adjacent stresses within the prosodic word (i.e. verse foot), but does not capture the tracking of the strong step to the strong position regardless of whether the syllable in that position is stressed or not. For instance, lines like (66) contain only one lexical stress and therefore no adjacent stresses within prosodic words, and therefore would not incur violations of NoCLAsʜ, but the strong dance step in the line predictably occurs on a strong metrical position:

1. (66)

   

Therefore, we can call the binary feet ‘unmarked prosodic constituents’ of a metrical tree, but we still need to explain the lining up of strong beats to the head of these constituents. This could not be achieved with language-internal phonology alone, since the alignment is between an extra-linguistic phenomenon and the metrical structure. Moreover, as Blumenfeld (Reference Blumenfeld2015:84) notes, in many works of verse the template is in fact the unifying feature; this is certainly true of the longer cycle ballads, where line length can vary greatly, as can the number of licences used in a given line. Blumenfeld (Reference Blumenfeld2015: 84–5) provides further argumentation and a more detailed discussion of the templatic versus holistic approaches, and reaches the same conclusion as this paper regarding the template’s usefulness. It seems to me that accounts like Riad (Reference Riad2017) in fact are not dispensing with the template anyway, since the metrical prosodic structure plays essentially the same role. It is doubtless a sound methodological principle to avoid reification of a theoretical construct, but given the remarkable consistency of the Faroese data with the template hypothesis, it seems to be as necessary to metrics as onsets or codas are to phonology of the syllable.