English iambic pentameter

Here are four lines in iambic pentameter, an accentual syllabic metre. They are all in the same metre, but they are rhythmically different: σ indicates a syllable and σ́ a syllable that is likely to carry stress.

Each line has ten syllables. Even-numbered syllables tend to be more heavily stressed than odd-numbered syllables, but this is a controlled tendency, not a strict requirement. The standard metrical description is that syllables are in subsections called feet; here, each foot contains two syllables of which the second is more likely to be stressed and this is called an iambic foot. There are five iambic feet in the line, which is why it is called iambic pentameter. Foot boundaries are marked with a vertical bar in the line below.

Rhythm can vary within limits. A syllable that carries primary stress in a noun, verb, adjective or adverb is usually in an even-numbered position. Other syllables, including those in grammatical words, can appear either in odd-numbered or in even-numbered positions.

Iambic pentameter presupposes the line as a section of text: it can only hold of a text if the text is in lines. Furthermore, it presupposes only the line; sections of text larger than the line are invisible to the iambic pentameter rules. In Shakespeare's sonnet, lines are grouped into larger sections: three quatrains and a final couplet. But these larger sections are not controlled by the metre and the metre does not depend on them or presuppose them. For example, all the permitted metrical variants refer to the line rather than to any larger metrical section. A common variant is to allow an extra syllable at the end of the line, as shown below, but this is an option within the metrical rules that focuses on the line alone.

In comparison, there is no principle of allowing an extra syllable only at the end of a couplet or quatrain.

Second, though in general the syllable carrying greatest stress in a polysyllable must be in an even-numbered position, the most common rhythmic variation is to allow stress in the first position of the line, which is called a trochaic inversion. This is true of the word ‘dulling’ in the following line, which has stress on its first syllable and is in line-initial position:

There is no equivalent general pattern favouring rhythmic inversions at the beginning of a quatrain or couplet, and this again shows that the metrical rule depends only on the line.

A third relation between metrical rules and the line relates to the metricality of syllables. A sequence of a vowel followed by a vowel can sometimes be treated metrically as two syllables and sometimes as one syllable, and this is a phenomenon that happens only within a line and not, for example, between the end of one line and the beginning of the next. This variable treatment is demonstrated by the word ‘patience’, which is treated as two syllables in the first of the following two eight-syllable lines and as three syllables in the second line (stresses are not shown here).

Tarlinskaja (Reference 211Tarlinskaja1976) noted that the shorter two-syllable form of the word is earlier in the line than the longer three-syllable form, and that this constitutes a general pattern, which might possibly reflect the law of increasing numbers. Again, this is a tendency specific to the line as a poetic section.

The fourth way in which these metres hold of lines, rather than larger sections, involves the possibility of having missing or additional syllables at the ends of lines. Consider for example these lines from a poem by Emily Brontë:

These lines can be analyzed as in dactylic tetrameter, a metre which normatively has twelve syllables to the line, with stress on the first, fourth, seventh and tenth syllables. However, it is possible to add a syllable at the beginning (as in quoted line 5) and one or two unstressed syllables can be omitted at the end, thus varying the number of syllables. These variations are all specific to lines, because they happen at the beginning or end of a line and are not sensitive to position of lines in the stanza, which again shows that metricality is a property of the line and not some higher section. This can be seen by comparing line 5 with line 11, both of which are penultimate in their stanzas, but they have different numbers of syllables. Line 5 adds one syllable at the beginning, while line 11 does not. This is a type of regulated metrical variation that is a property of the line, irrespective of position within any larger section.

An English pattern-poem

Robert Herrick ended his Hesperides (1648) with a pattern poem laid out on the page to produce the image of a pillar:

This is a metrical poem in iambic lines in a mixture of eight-syllable, six-syllable and four-syllable lines, requiring three different metrical rules. The metrical rules presuppose and govern individual lines. The text as a whole is heterometrical (i.e. the lines are in different metres) but the overall heterometricality has no consequence for the metricality of each of the individual metrical lines. The rules for an iambic tetrameter line are the same whether the line is in an isometric text in which all lines are tetrameter or the line is in a heterometric text in which some lines are tetrameter and some in another metre. Further, the heterometricality is governed by something clearly not metrical or even related to language at all, which is the need to create a picture of a pillar on the page. Like many pattern poems, this is a bilaterally symmetrical picture, showing symmetry as a property of visual form. The added form of metre holds of lines while the pattern overall holds of the whole text which is in principle unlimited in linguistic size.

Pashto

The qasida is a genre of poem used widely in the Islamic world, in which long lines are divided into hemistichs. In Pashto (South Central Asia, MacKenzie Reference MacKenzie1958, Reference MacKenzie1996), the qasida is in an accentual-syllabic metre, with twelve syllables in each hemistich and a pattern of stressed syllables within the hemistich. Syllables are in groups of four, and the third syllable in each group must have lexical stress. Here are two hemistichs:

This hemistich is thus like the English iambic pentameter line in that it is the section to which the metrical rules apply. The larger section of the long line is a coherent semantic unit. The metrical rules only require reference to the hemistich which is thus the largest section subject to strict limits, but given the coherence of the long line as a unit, we might speculate that it is also designed to fit into working memory, and note that it is still possibly short enough to do so. (Estimates of this kind are always based on guessing a rough equivalent to about fifteen words of connected English prose as the maximum amount that will fit into short-term memory.)

Havasupai

Hinton (Reference Hinton1990) discusses the metre of a Havasupai (North America) sweathouse origin song, including how it is set to music. The long line is in two hemistichs, and the metrical rules control each hemistich separately, requiring the hemistich to be six syllables long with stresses usually in even-numbered positions, and the fourth syllable in each hemistich has primary stress. The first of the two hemistichs can end on the fifth syllable; thus, although both hemistichs are in the same metre, there is nevertheless some limited variation between hemistichs within a larger section. Because the second hemistich has a more predictable rhythmic form, this means that a whole long line ends more predictably than a hemistich does. This is an example of a cadence, which applies at a higher level of structure than that controlled by the metrical rules, and this suggests that cadence is a different kind of form from the metrical rules. The line as a whole matches a musical section which is divided into three subsections cross-cutting the two hemistichs, and is in 6/8 measure with stressed syllables falling on positions 1 and 4 of the measure. The metrical rules apply to the hemistich, but metrical properties of variation and cadence apply to the long line, and the long line is the section set to music. The hemistich must fit into working memory, but the long line is also short enough to fit.

Cretan

A Cretan mantináda is a pair of long lines (Sykäri Reference Sykaäri2014). Each long line has fifteen syllables, divided by a caesura into eight-syllable and seven-syllable hemistichs, in a type of the Greek dekapentesyllavo. There is obligatory stress on the sixth or eighth syllable and on the fourteenth syllable, which suggests that the metre holds of the fifteen-syllable long line as a whole and can be seen as underlyingly iambic with stress only required in certain positions. The two lines rhyme.

Sykäri says that the first line sets up specific expectations, and the second line expresses the main argument, sometimes humorously juxtaposing the second line against the first. Of particular interest here, she says that ‘improvisers most often say that after figuring out the subject matter of their utterance, they first catch the rhyme words, and then create both verses’. I suggest that the long line can be held as a whole unit in working memory. The whole couplet might be too long, but this would need to be tested. Sykäri's description of composition suggests that the second long line is first composed along with the rhyming word, after which the first long line is composed.

Dyirbal

Dixon and Koch (Reference Dixon and Koch1996) analyse Dyirbal songs (Australia). These are short songs, divided into lines with specific numbers of syllables, some of which are stressed in a particular pattern. As elsewhere in Australian aboriginal song texts, a neologism may be invented in a particular song and not used elsewhere. I briefly discuss four Dyirbal song genres here, all of which control the number of syllables in the line, and two of which control the pattern of stress.

Songs in the gama genre consist of two lines repeated several times, often in alternation. Each line consists of a five-syllable word followed by a two-syllable word followed by a four-syllable word, sometimes with a vocable syllable added at the end. There may be variation in the first part, which can be a two-syllable word plus a three-syllable word, but there is no variation in the last part of the line; this greater fixity might be interpreted as a kind of cadence. The sung text is accompanied by beats on clapsticks, and the first syllable in each of the three parts matches a clapstick beat.

Consider now a song in the different marrga genre, which is made from the following four lines performed in the sequence with vocables, 1212121–drr–heey–heey–34343434–drr–heey–heey. This song was accompanied by a rhythm performed by hitting two sticks, accompanied by a lap drum played by the singer's wife, which continues steadily throughout the song.

Each line has eight syllables, and stress is imposed on syllables 1, 4, 6 and 8, freely changing ordinary word stress to fit this pattern. There is an elongated final syllable in performance. Again, we see evidence of cadence because the line ends on a verb with unmarked inflection, and the final part of the line is the most likely to be exactly repeated over different lines; furthermore, the variation in a basically iambic rhythm comes at the beginning of the line, which is a cross-linguistically common place for rhythmic variation in metrical verse. Lines tend to alternate in groups, with the most common pattern alternating the first two lines several times (here, rather stricter than some other examples), then an instrumental interval, then alternating the second two lines several times; thus in performance, higher-level sectioning is expressed by repetition.

The other two song genres in Dyirbal are in syllable-counting metres, with probably no control over where stress falls. Songs in the jangala genre consist of any number of six-syllable lines, usually with a line-final pause in performance. Lines tend to be ordered differently in different performances of the text, except where there is syntactic continuity between the lines. Songs in the burran genre consist of quatrains in a 6–3–6–3 syllable-counting pattern; the three-syllable lines must be single words and the six-syllable lines must contain only words with even numbers of syllables. This might in fact be treated as a stress-based constraint, as words have initial stress; hence, the line has an underlying trochaic rhythm, such that any stressed syllables must fall in odd-numbered positions. Dixon and Koch discuss whether the text should be analysed in single nine syllable lines with a cadence, but note that the singers tend to think of the nine syllables split over two lines.

These Dyirbal song genres are all based on a single layer of poetic section, the line. Metrical rules apply to the line. Cadence holds of the line. The line matches a musical structure, and lines are treated as independent units that can be repeated and reordered in performance. The line is the only section that must, by the principles of the present book, be held in working memory. Note, however, that the text of the whole poem is usually short enough to fit all of it into working memory at the same time.

English ‘stress counting’ metre

Samuel Taylor Coleridge said that his poem ‘Christabel’ was in a metre which involved ‘counting in each line the accents, not the syllables’ (Keach Reference Keach1997: 187). This is a stanza from the poem. Each line can be performed with four stressed syllables, but the number of syllables varies as shown in the numbers on the right.

In iambic pentameter, it is possible to predict whether a syllable will carry stress based on counting syllables (e.g. even-numbered syllables are stressed, counting from the beginning of the line). In this metre, it is not possible to predict which syllables will carry stress. Instead, we can only say that in the quoted lines, there are four stressed syllables per line; stress and counting are thus less tightly connected than in iambic pentameter. It is worth noting that though stressed syllables are counted, there are also some elements of syllable counting; unstressed syllables appear also to be controlled because it is very rare to get three unstressed syllables between the stresses.

Metres in which stressed syllables are counted (and in which there is some additional control over unstressed syllables) have been given very different theoretical accounts. Attridge (Reference 193Attridge1982) suggests that this four-beat line is related to the iambic pentameter line, which though subject to an overall syllable count tends to have four rather than five beats. Hanson and Kiparsky (Reference Hanson and Kiparsky1996) see it as a metre in which prosodic units larger than the syllable are counted. Fabb and Halle (Reference Fabb2008) treat it as a counting metre based on a slightly different mechanism from that found in the strict metres such as iambic pentameter. For the purposes of the present book, the main point is that the metre holds of the line and not of any higher-level section of text. This type of poetry often has rhyme, which tends to be at the end of the line. This means that the rule for locating the rhyming word refers to the same level of section as the rule for metre. The line is always small enough to fit into working memory.

Hayes and MacEachern on English folk song quatrains

Hayes and MacEachern (Reference Hayes and MacEachern1998) look at final rhythmic cadences in English folk songs, songs that are in a metre with four or three stressed syllables per line. They suggest that there are four types of rhythmic cadence in the line with varying strengths of cadentiality. For lines, the most cadential option is for the line to end short on the third stressed syllable, and the least cadential option has the line ending on the fourth stressed syllable. Lines are grouped into couplets, and couplets are grouped into quatrains. For the purpose of their analysis, they suggest that the song ‘hickory, dickory, dock’ can be analysed as such a folk-song quatrain (Opie and Opie Reference Opie and Opie1952: 206 present it on paper as a five-line text).

Lines 1, 2 and 4 end on the most cadential cadences, and Hayes and MacEachern argue that this makes them salient as constutuents. In contrast, line 3 ends on the least cadential cadence, which means that it is not salient as a constituent. This means that the stanza is divided into three salient sections as line + line + couplet.

They explain why the following invented quatrain sounds wrong:

Line 1 ends on the most cadential ending, and so is salient as a section. Line 2 ends on the least cadential ending, but line 3 ends on the most cadential; so lines 2 and 3 form a salient section. Line 4 then forms a section of its own. Here the saliency structure is line + couplet + line. Hayes and MacEachern suggest that this violates a principle of long-last (the law of increasing numbers), which holds in this genre of poetry, and which they formulate as ‘in a sequence of groups of unequal length, the longest member should go last’, where groups are defined by saliency based on cadences.

For our purposes, the particular interest of this account is that the formal property of being a cadence strictly governs the structure of large constituents such as a stanza, and that the formal property of the law of increasing numbers also holds over this large constituent. But the stanza is too large to fit into working memory capacity. This shows that it is possible to have strict formal rules in poetry that do not operate entirely in working memory. Cadence and the law of increasing numbers are in this way different from the metrical rules that count syllables and set rhythms, even though they relate to metricality. I have previously characterized cadence and the law of increasing numbers as ‘formal properties’, which can hold of various different types of poetic (and perhaps also non-linguistic) form, and here we see that they are not constrained to hold over small sections of text that can fit into working memory.

Old English

Old English poetry is usually analysed as being in an accentual metre, and has systematic alliteration. The text is in long lines, and each line is divided into two hemistichs, often called the a-verse and the b-verse, and here referred to as the a-hemistich and the b-hemistich.

Each hemistich must have at least one strongly stressed syllable, accompanied by secondary stresses in various patterns. There is alliteration over the whole long line between one or two strongly stressed syllables in the a-hemistich and one strongly stressed syllable in the b-hemistich. Any vowel can alliterate with any other vowel. There are also patterns in the distribution of secondary stress, syllable quantity and the counting of syllables. Suzuki (Reference Suzuki1996: 348–54) calls this a ‘two-level organization’ and provides evidence that both levels of long line and hemistich are formally relevant. The hemistich is traditionally taken to be the domain of these metrical rules, and hemistichs belong to one of a range of permitted metrical types. West (Reference West1973: 179) suggests that the hemistichs ‘are the original unit and that the long line is in essence a couplet’, evidence for this being that hemistichs can exist on their own or be combined into threes. Formulaic expressions may coincide with the hemistich (Foley Reference Foley and Foley1981: 274), suggesting again that the hemistich is taken as a metrical unit. However, the choice of metrical type for a hemistich depends on the relation between the hemistich and the long line, so the long line is also metrically relevant. This is because the b-hemistich can use a more restricted range of metrical types, and as Suzuki (Reference Suzuki1996: 348) notes, this means that the long line as a whole is more restricted towards the end than towards the beginning, a restriction which does not hold of individual hemistichs. Further, Golston (Reference Golston2009) shows that the two hemistichs within the same long line never have the same rhythmic pattern, whereas adjacent hemistichs can be identical if they are in different long lines, that is, a b-hemistich followed by a different line's a-hemistich. This shows that rhythmic variety is assessed on a line-by-line basis taking the line as a limited domain. Alliteration holds across the long line as a whole but not beyond the long line. The long line boundary tends to coincide with a major syntactic boundary (Suzuki Reference Suzuki1996: 350), suggesting some semantic coherence to the whole long line. Formulaic expressions are used in this poetry and may coincide with the hemistich, suggesting again that it is a compositional unit. Whole long lines should easily fit into working memory. West (Reference West1973: 179) notes that the first hemistich tends to be longer than the second (which is contrary to the law of increasing numbers if the law is based on counting syllables). Cain (Reference Cain2001) describes a macaronic variant of this metre, in a mixture of Latin and English hemistichs; the Latin hemistich, which is usually first in the line, alliterates with the English hemistich, often using Latin words that have initial stress, so as to fit with the general requirements of the metre. It is worth noting that this mixing of languages is constrained by the long line and the hemistich, distinguishing the two hemistichs and their locations within the long line, and it may be that holding the whole long line in working memory enhances the possibility of mixing the two languages within this section.

Golston and Riad (Reference Golston2003) offer a radically different analysis. They argue that the metre is quantitative, and holds of the long line (not the hemistich); the long line consists of eight phonological feet, each of which consists of either one or two morae, thus giving a long line which varies between eight and sixteen morae, which they say is true of 99 per cent of the lines in Beowulf (scanning all syllables and ignoring none). This analysis still retains the long line as the unit of structure that must fit into working memory.

Other metres that count stressed syllables

There are other metres that appear to count stressed syllables, but only weakly if at all count unstressed syllables. Akkadian poetry is described by Speiser as in lines with normally a unit of two distinct halves with two beats in each half (Pritchard Reference Pritchard1955: 60). Hittite is described by Güterbock (Reference Güterbock1951: 142) as having a normal line with usually four stresses and about twelve to seventeen syllables; Beckman (Reference Beckman and Foley2009: 256) says that in the Hittite Clothes of Nesa each line consists of four stresses, in two cola. Emovon (Reference 198Emovon, Abalogu, Ashiwaju and Amadi-Tshiwala1981: 267) argues that Edo poems (Nigeria) have two rhythmical segments in each line. Bailey (Reference Bailey2002) argues that some Bedouin Arabic metres count stressed syllables (Fabb and Halle Reference Fabb2008: 251–3).

Mediaeval Welsh

The bards of mediaeval Wales composed poetry in many metres that count syllables in each line, and also prescribe rules of rhyme and other types of sound patterning. The Welsh metres strictly count syllables but do not control rhythm. The named metres describe whole stanzas, which can consist of lines of different lengths. One such metre is englyn unodl union, which is a four-line stanza of 10+6+7+7 syllables, and with rhyme at the ends of the last three lines and on the seventh, eighth or ninth syllable of the first line. In this text, the rhyme on -i is on the seventh syllable of the first line and the last syllables of the other lines:

I suggest that we see the line as the metrical unit, the unit that must be of a certain length and to which the generative metrical rules apply. The line is also the section relative to which the rhyme is located. For example, it is on the seventh, eighth or ninth syllable of the first line, and this location is never dependent on any characteristic of any other line. Thus, the line is the distinctive poetic section to which the added forms apply. Each line is clearly short enough to fit into working memory.

Each named Welsh metre is the name of a pattern for a whole stanza or poem, and not the name of the pattern found in a single line, but the whole stanza is too large to fit into working memory. I have suggested that we treat metre as a collection of modules. The metre englyn unodl union consists of a module which is the generative metrical rules that sets the number of syllables, and a module that locates the rhyme; both of these modules relate to a unit of structure, the line, small enough to fit into working memory. There are two separate modules for the two patterns: one controlling the pattern of which lengths of line combine in which order, and one controlling the pattern of which lines rhyme with which other lines. Specific line lengths are combined in a heterometrical pattern into the larger stanza. There appears to be no principle that governs how a specific metrical line will fit into a larger pattern; instead, it is just stipulated. The rules for patterns appear to be arbitrary or governed by external factors such as being a pattern poem: the englyn stanzas are arguably visual patterns, drawing angels’ wings on the page. This all supports a modular approach, in which one part of the form is not caused by another. The patterns are distributed over the whole stanza, which is too large to fit into working memory, and so this element of the forms cannot be calculated in working memory. The patterns are made from component parts which can be reordered and reused. The englyn is an angel's wing with a shaft (the 10+6 pair) and two wings (the two 7 syllable lines); and the parts can be reordered (7+7+10+6 is englyn unodl crwca). Another englyn has 10+7+7, with the ten-syllable line having the same internal rhyme structure.

The rhyme pattern has a mid-line word rhyming with a following line-final word. In contrast, in South-East Asian hook rhyme traditions, the line-final word comes first, and the mid-line word comes second, and when I come to these traditions I will suggest that a preceding rhyme is easier to find if it is line-final, but this is not true of this Welsh metre. However, the mid-line rhyme in the first line must be followed by a pause, and perhaps this is a device to allow an initial rhyme to be found when it is in the middle of the line.

Consider now another mediaeval Welsh metre, of the awdl type.

The metre is cyhydedd hir, which Williams (Reference Williams1953: 239) describes as ‘a line of nineteen syllables which, for convenience of writing or printing, may be divided into ten and nine, or into sections of five, five, five and four, or five, five and nine syllables. The first three sections rhyme together and the fourth carries the main rhyme’. In principle, given how short the stanza is, we might indeed take the whole stanza to be the basic poetic section and it might just fit as a whole into working memory. However, I suggest that instead we treat it as a 5+5+9 syllable text, as laid out on the page by Williams, with a line-internal rhyme in the last line. A reason not mentioned by Williams for treating it as in three lines is that each line has the sound-sequence parallelism called cynghanedd, where a sequence of consonants in the first half is repeated in the second. This is true within each of the lines only if they are laid out as 5+5+9: line 1 (n-d-r repeated), line 2 (g-n repeated) and line 3 (n-r-dd repeated). Since cynghanedd holds within a single line, whereas rhyme can be line-internal in this tradition, this is the lineation that optimally allows for the various forms which are dependent on the line.

Javanese

Javanese tembang macapat or matjapat is a genre of poem and song, discussed by Kartomi (Reference Kartomi1973), Hatch (Reference Hatch1976) and Brinner (Reference Brinner2008: 76–87). The macapat metres have also been borrowed into Madurese, Balinese, Sundanese and other languages. Different metres are considered appropriate for different emotional affects or types of poetry, and a poem may mix sections, each of which is in a different metre. A metre in this tradition is a set of rules that govern a heterometrical section, which Kartomi calls a ‘verse’, such as the following ten-line verse which is in the dandanggula metre:

The metre controls the number of syllables in each line as shown by the number to the right of each line, and the final vowel in each line as shown by the vowel on the right of each line; so, for example, the first line must have ten syllables and its final vowel is [i]. This text shows a certain amount of variation from the rules, because two of the lines have one syllable more or less than required (bracketed vowels are not included in the count). A different metre pangkur must have seven lines of 8, 11, 8, 7, 12, 8, and 8 syllables each, the ending vowel sounds being [å, i, u, å, u, å, i]. Another metre midjil, must have six lines of 10, 6, 10, 10, 6 and 6 syllables each, ending in [i, o, é, i, i, u]. Some metres are more homogenous, such as the popular kinanti metre, which has six lines of 8 syllables, with alternate lines ending in [i].

Lines with the same number of syllables and the same ending are found as components of different metres, but the tradition as a whole seems not to use a system for combination. Kartomi (Reference Kartomi1973: 43) says that native speakers have told her that macapat means ‘reading in fours’, and that ‘when people read or sing aloud old stories in matjapat metres (kidung), as a general rule they should try to breathe or make breaks only after each fourth syllable of a line. This popular explanation fits in somewhat with the generally accepted pedotan (caesura) rules, which favour divisions of four syllables for all line lengths’. The pedotan rule is for performance, regulating pauses in singing, and is allowed to shift pauses so that words are not split in two. However, it is difficult to make much of this in terms of the various actual line lengths other than to note that it might resemble traditions in which morae are grouped in fours, as in Tashlhiyt Berber and Indian metres.

The line is the unit to which the metre refers. In other ways, lines show the following structural characteristics (Kartomi Reference Kartomi1973: 52–5, 94, 143), none of which are strict added forms, but all of which show that the line is a unit of structure. There may be assonance, alliteration or rhyme within a single line. A line-final word may be repeated as the initial word of the next line. Lines can be arranged to form an acrostic such as the poet's name. In singing, the musical sequences separated by rests usually coincide with poetic lines, and in performance, the last syllable of the line can be extended in various ways, including in an undulating melody called eluk.

Avestan

Avestan is the language of the Gāthās, which were composed in South Central Asia before 1000 BC and attributed to the prophet Zarathuštra. According to Peabody (Reference Peabody1975), Avestan metres count syllables and do not control rhythm. The text is divided into lines that are grouped into pairs or triplets, which are in turn grouped into pairs or triplets, so that the text is divided into sections on three layers. Here are two triplets in the eight-syllable metre:

The eight-syllable line is presumably the poetic section to which the metrical rules apply, and it is certainly small enough to fit as a whole into working memory. However, this is problematized by the second stanza above, which demonstrates a systematic type of variation. In this stanza, the second line has nine syllables and the third line seven; the overall number of syllables remains at twenty-four, but it is as though the second line has borrowed a syllable from the third. Does the metre actually control the whole triplet, making it twenty-four syllables with a flexible caesura after the eighth and sixteenth syllables? This would create a very long section, probably too large to fit into working memory. An alternative analysis is that the lines of the second stanza are indeed eight syllables long but that the word xra-tǝ̄uš is split across lines.

Malay

The Malay syair is a poem in four-line stanzas, all lines having the same rhyme, with the first part of the line being loosely five syllables with variation, followed by a caesura which is then followed by a strictly five-syllable sequence (Braginsky Reference Braginsky1991: 140).

The strict control over the end of the line is an example of a line-final cadence. The whole line is the unit that must fit into working memory, because this is the unit over which the metre holds, and because rhyme is located relative to the line.

Quechua

Bolivian speakers of Quechua and Spanish have a type of sung duel exemplified by the coplas de Todos Santos, which is performed in Mizique (Solomon Reference Solomon1994: 388). The song unit, or copla, is a quatrain divided into two couplets.

The first couplet is sung twice and then the second couplet is sung twice. Each couplet is internally semantically coherent: for example, the first couplet might be a comment and the second an insult. The metre holds of the eleven-syllable line, which has six syllables of improvised Quechua text, followed by a fixed five-syllable Spanish phrase, which is a type of cadence. Our hypothesis states that the line must fit as a whole unit into working memory. Though the metrical rules apply to the line, other properties hold of higher sections; the couplet is semantically coherent, and the couplet is a unit when set to music, as evidenced by the repetition. We have seen elsewhere that semantic coherence and musical setting may relate to a higher-level section than the section controlled by the metrical rules. Though the couplet need not be fitted into working memory, it is short enough that it could.

Other syllable-counting metres

The syllable-counting metres of French, and those of Latvian dainas are the subject of chapters in Fabb and Halle (Reference Fabb2008), and Jan Kochanowski's Polish adaptation of French syllable-counting metres is discussed in Fabb and Halle (Reference Fabb and Halle2006a). Zaborski (Reference Zaborski1996:257) analyses an Afar ritual text, which he says is in lines, most of which are eight or nine syllables long. Chadwick and Chadwick (Reference 196Chadwick and Kershaw Chadwick1940: 544) say that Oromo texts (a.k.a. Galla, Ethiopia) are usually in eight-syllable lines with rhyme. Grijns (Reference Grijns, Ras, Robson, Roolvink, Teeuw, Herbert and Milner1989: 133, 140) describes a Bare'e poem (Indonesia) as in stanzas of four eight-syllable lines; the kajori genre has AAAA or AABB rhyme, while in bolingoni the second pair of lines parallels the first with variation in word order or the use of synonyms. Erdener (Reference Erdener1995: 83, 174) describes Turkish rhyming songs with 7, 8, 11 and 15 syllable metres, usually isometrical but with some mixing possible. Norris (Reference Norris1968) describes Ḥassānīyametres from Western Sahara, some of which just count syllables, ignoring the weight distinction between syllables, and others which count syllables but require a specific syllable to be superheavy (Fabb and Halle Reference Fabb2008: 253–5).

Minangkabau

Phillips (Reference Phillips1981: 105) describes the Sijobang, a sung narrative poem of the Minangkabau (West Sumatra). The poem is in lines corresponding to syntactic units, which ‘if spoken, could be followed by a mid-sentence or end-sentence pause’, of which about 80 per cent are eight or nine syllables long, suggesting that it is in a syllable-counting metre. In one performed version, 37 per cent of lines are in parallel pairs, such that the second line has a similar meaning to the first line, either with the same syntactic structure as in the first example included here or with a different structure as in the second (both from Phillips Reference Phillips1981: 114–5):

This is an example of a poem with both metre and parallelism. The metre controls the line. Each line is a parallel member. In this poem, both metre and parallelism presuppose the same section, the line, which can clearly fit as a whole into working memory.

Sulawesi

George (Reference George1993: 702) describes a highland Sulawesi (Indonesia) headhunting song, which is in triplets of eight-syllable lines, sung to a set tune, and with some ‘semantic and rhetorical integrity’ to the triplet:

The leader sings the first line, and the leader and chorus together sing the second and third, and the practice is typically competitive with different groups singing against each other. The metre applies to the line and the parallel member is the line, so it is the line that must fit into working memory.

Toda

Toda (India) songs have been described by Emeneau (Reference Emeneau1966). Toda songs are in sections called song-units, which are three or sometimes four syllables long and consist of one, two or three words each. Words are altered to fit the formal requirements. Words are contained within song units and are not split across their boundaries, and song-units are typically fixed formulae that are reused in other songs; a song unit fills up a melodic line and matches a musical rhythmic unit. Groups of song-units may combine to form a sentence, which is often formulaic. The following song is in six units forming two parallel members:

The section to which the metrical rules apply is the three-syllable unit. In this song, a triplet of units is the parallel member, and so we say that the triplet which is here represented as a single line is held as a whole unit in working memory. The metre controls the song unit while the parallel member is the line, so the line must fit into working memory.

In Toda, parallelism involves paired words, and in some cases the meaning is provided by one word, with the other word not providing any meaning. However, for this meaning to arise, both words must be sung, as in the following pair of song-units. The literal translation here does not express the actual meaning of the pair: the pair can be used either to refer just to tigers, or just to refer to Kurumbas, and never refers to both.

Greek dactylic hexameter

Dactylic hexameter is the Greek metre used by Homer; we saw the first two lines of the Odyssey in Chapter 1, and here are the first two lines of the Iliad, transliterated from Greek script:

Each line is a sequence of words whose syllables are classified as light or heavy. The line is made of six feet, and each foot is a dactyl with a heavy–light–light pattern, symbolized – ⏑ ⏑, or a spondee with a heavy–heavy pattern, symbolized – –. The last two feet must be a dactyl followed by a spondee. The last syllable can be light instead of heavy, following the general convention of brevis in longo found in many quantitative metres. These constraints on foot combinations mean that the line is between thirteen and seventeen syllables long and specifies a rhythm but allows for variation. This can be seen in the fact that though the two lines above both end on a dactyl-spondee sequence, which is a fixed rhythmic ending or cadence, the lines otherwise differ in the patterns of their second, third and fourth feet, giving the lines different quantitative rhythms.

The line is conceived of as underlyingly a sequence of six dactyls, hence its name dactylic hexameter. A dactyl can be replaced by a spondee. This ‘spondaic substitution’ is not equally common throughout the line; Miller (Reference Miller1977: 22) cites statistics on spondees in the first five feet of the line in the first twenty books of the Iliad, showing that 30 per cent of spondees are in the first foot, 31 per cent in the second and 23 per cent in the fourth. The third foot has 12 per cent and the fifth foot 5 per cent. This is an asymmetry that applies to the line as a whole poetic section. Further, there is a highly predictable end to the line: 95 per cent of the lines have a dactyl followed by a spondee. This is one of the most commonly cited examples of a cadence, which is a predictable ending to the line. Allen (Reference Allen1973: 106), discussing cadences in general, notes that there are many Greek examples of final strictness including, for example, ‘the anapaestic systems of tragedy, where spondaic variants are generally common, but where the last full foot of the final line is almost invariably a pure anapaest’, a sequence of two light and one heavy syllables. The approach to metre I take in this book is modular: that is, I distinguish between metrical rules and tendencies such as asymmetries and cadences, which in principle can be separated off from the metrical rules. In this metre, all the forms converge on the same line, including the metrical rules, which must presuppose the line. The cadence and other asymmetries also coincide on the line but in principle could have held of larger sections.

There is an obligatory caesura that requires a word boundary to fall in the third foot, or sometimes the fourth, though never exactly in the middle of the line. The caesura is marked by the symbol ‘:’ in the quoted text. West (Reference West1982: 35) suggests that this caesura breaks the line into two sequences and that these sequences and not the line are the sections controlled by the metre. The sequences are called cola (singular colon) and are in the pattern – ⏑ ⏑ – ⏑ ⏑ – with potential replacement of a pair of light syllables by one heavy syllable. The two sequences are separated by a small sequence of two light syllables or one heavy syllable, and there is a final light/heavy syllable at the end. This alternative pattern is shown for the first two lines of the Iliad below:

West notes that ‘many of the repeated phrases of epic are designed to fill one or other colon’. This is an example of how some metrical traditions have been subjected to very different analyses.

The Greek heterometrical strophe

The following Greek text is the first section of Pindar's eleventh Pythian ode, with the sequence of heavy and light syllables shown.

The stanza is a heterometric section of text, with its lines (here charcterized by West as ‘periods’) in different metrical patterns. It is followed by another stanza, which consists of a five-line sequence with the same patterns of heavy and light syllables, a large-scale copying called responsion; the first stanza is called the strophe, the second the antistrophe. This repetition shows how deliberate and controlled the pattern is, even though it varies, apparently at random. Here again, we have metrical poetry in which the line is controlled by the metrical rules, but a larger section is controlled for a heterometric pattern, showing that metrical rules and metrical pattern are distinct kinds of form.

Various generative approaches have sought to find some regularity in Greek lyric metres, including Golston and Riad (Reference Golston2005) and Fabb and Halle (Reference Fabb2008). In West's (Reference West1982: 62) analysis, the text is analysed into five sections called periods, which in this text coincide with the lines, though periods can sometimes be much longer, each period being made from smaller metrical components such as cola. Thus, the first line is made from four parts: (i) an anaclastic telesillean colon (x ⎽ x ⎽ ⏑ ⏑ ⎽) where x stands for either a light or a heavy syllable, (ii) two light syllables or a resolved heavy syllable, (iii) a telesillean colon (x ⏑ ⏑ ⎽ ⏑ ⎽) and (iv) a syncopated iambic metron (⎽ ⎽). West notes that this may seem to be an ad-hoc analysis of what is a unique metrical scheme in each poem, but suggests that the component parts are reused elsewhere. In a new analysis, Willett (Reference Willett2002) criticizes the tradition of dividing the text into periods that can, in other texts, extend over long sections of text, because periods are too long to fit into working memory; he makes the assumption (which I have subsequently also made) that the section presupposed by the metre must be small enough to fit into working memory.

Classical Arabic

The classical Arabic metrical system called ˁarūḍ was established by al-Khalīl in the eighth century AD, and the metres have been widely adapted in Islamic literatures. The standard western account of Arabic metres is Weil (Reference Weil and Gibb1960), and contemporary theoretical discussions include Golston and Riad (Reference Golston and Riad1997) and Fabb and Halle (Reference Fabb2008). Bayt is the Arabic term for what might elsewhere be called a long line or possibly couplet. The bayt is divided into two hemistichs, as illustrated by the following bayt with one hemistich written above the other:

This bayt is in the ṭawīl metre, where each hemistich is 14 syllables long, in a sequence which can be represented as follows, where ⏑ must be a light syllable, ⎽ must be a heavy syllable, and x can be either light or, more usually, heavy:

It is in the x position that the two hemistichs vary above: they are rhythmically slightly different but in exactly the same metre. The sequence repeats a ⏑ ⎽ x pattern followed by a ⏑ ⎽ x x pattern, and then the pair is repeated. Thus, there is evidence for internal constituency on two levels within the hemistich: a larger seven-syllable sequence that repeats, and smaller three- and four-syllable sequence. This is a multilayered periodicity. Many metres allow variant endings to the hemistich, and the second hemistich is allowed a choice from more variations than the first, which means that the bayt-final hemistich is less restricted in principle, unlike some other traditions. Any variation (called ˁIlal ‘disease, defect’) is established at the beginning and must be kept in all hemistichs of the poem (Weil Reference Weil and Gibb1960: 671–2).

The metrical rules hold of the hemistich, not the bayt. But though rhyme is occasionally also located at hemistich end, it is more usually located at the end of the bayt. Thus, one added form presupposes the hemistich and another the bayt, and I suggest that it is the bayt that is held as a whole section in working memory. The bayt as a whole is also a coherent syntactic unit, which as here often correlates with the unit that is held as a whole section in working memory.

Moroccan Arabic

Elmedlaoui (Reference Elmedlaoui, Bendjaballah, Faust, Lahrouchi and Lampitelli2014) describes the metres of malħun songs in Moroccan Arabic. These are quantitative metres with line-final rhyme, where the line need not be a complete syntactic unit. One or two lines (rarely three) form a group that is a coherent and complete syntactic unit. A sequence of two three-line groups is given below.

Lines may be of any length but manifest a controlled pattern of heavy and light syllables, in groups that are ordered in a specific way. Syllables are grouped into two types of sequence, called ‘dipodies’ by Elmedlaoui: either a light dipody of four light syllables LLLL or a heavy dipody of a light and two heavy syllables LHH. These dipodies are combined in sequence, subject to the constraint that a light dipody cannot follow another light dipody, and the sequence is then repeated. Thus, the poem cited above follows the sequence shown below. The first line, (a), is a full manifestation of the sequence, and the second line, (b), repeats the beginning of the sequence, which is then picked up by the third line, (c).

In this illustration of three-line patterns, dipodies are separated by a dash, and square brackets enclose heavy or light positions that are not filled by textual syllables, though the singer may extend a syllable melismatically over the gap. The second line ends in the middle of a dipody and the third line begins in the middle of a dipody.

Elmedlaoui describes the metre as resembling a wheel that keeps turning, consisting of an LLLL dipody followed by an LHH dipody followed by an LLLL dipody and so on, but allowing an extra LHH at any point. Where there are breaks in the metrical sequence between lines, as there are in all the lines above, there is a pause in performance, which corresponds to the missing syllables. Elmedlaoui argues that the metre of this verse does not control the length of the line but only the pattern of heavy and light syllables. Thus, all the songs are controlled by a single metre, which has many different length variations. This proposal makes malħun a radically different kind of metre from others discussed in this chapter. One possible analysis is to take the text which is covered by the whole metrical sequence as a poetic section that fits as a whole into working memory; thus, in the quoted text, line (a) is a unit fitting into working memory and lines (b) + (c) together constitute another unit which fits into working memory. Mohamed Elmedlaoui says (personal communication) that in attempting to recall lines with specific metrical patterns, he finds that the sequence of words making up a line is best remembered if it forms a metrical pattern constituting the beginning of a sequence, such as line (a) above.

Japanese

Japanese metres are based on the counting of morae. This means that lines in the same metre can have the same number of morae but vary in the number of syllables. For example, aki no kaze (‘autumn wind’) is a five mora line consisting of five light syllables, and kyō nite mo (‘I am in Kyoto’) is a five mora line consisting of one heavy syllable and three light syllables (both by Bashō; Carter Reference Carter1991: 358). Many Japanese metres are based on lines of either five morae or seven morae. Haiku have 5+7+5 morae. The whole poem can have no more than seventeen words, and so in principle it might be possible to fit the whole poem into working memory; indeed, the overall length might be limited specifically for this purpose. However, the metrical rules refer to the five-mora line or seven-mora line and so by our hypothesis, only the individual line must be held as whole unit in working memory.

A looser counting of morae characterizes higo biwa oral narratives, performed in post-war Japan almost exclusively by blind performers (de Ferranti Reference De Ferranti1994):

The nagashi metre seen here organizes the text into long lines normally of twelve morae, divided by a caesura into two hemistichs of seven and five morae, in which the seventh mora is word-final, and usually each hemistich is a coherent phrase. The second and third lines of the above song fit this pattern. The first line does not, but still ends on a five-mora sequence, which thus functions as a kind of cadence. The metrical rules apply to the long line, which by hypothesis is the section held as a whole unit in working memory.

It is sometimes implied that Japanese mora-counting poetry is dependent on matching morae to a timing template. Gilbert and Yoneoka (Reference Gilbert and Yoneoka2000) discuss a performance style for haiku in which there are silent beats, which match the performance to three timed sections, each of which can be measured as eight timing units long; the timing units are filled by morae or by timed silences that can be located at the ends or even middle of the lines. They suggest that this twenty-four unit template explains why extra morae can be added to the 5+7+5 pattern by some poets. However, the nagashi text shows that mora counting can also be completely independent of timing, showing that the counting of morae is a matter of how the text is organized at the linguistic level, not how it is performed.

Ponapean

Fischer (Reference Fischer1959) describes Ponapean metrical songs and poems from the Eastern Caroline Islands, which have the text mostly organized into couplets of seven or five morae, and usually in a pair of seven morae lines as in the following text.

The words are sometimes altered to fit the metre, for example, by adding the vowel [i] between two consonants, as parenthesised in the text above. The text is recited in couplets, with the first line ending on rising intonation and a pause, and the second on falling intonation and a longer pause. Fisher's translation shows that in the quoted text, the couplet is the smallest semantically coherent unit. The odd-numbered morae are more likely to be stressed, thus producing a trochaic rhythm. In these texts, we can say that the metre controls the individual line. It may be that the line pair is actually held as a whole unit in working memory, as this is a semantically coherent unit.

Fischer says that in some chants with accompanying stick rhythms, the pause after a five-mora line is systematically longer than that after a seven-mora line, and that there are pauses equivalent to three morae and one mora respectively. This is similar to the claim made in the previous section about the performance of Japanese haiku. Note, however, that when the text is sung, there is no systematic relation between the length of notes and the one- or two-mora lengths of syllables, and this suggests (as for Japanese) that the counting of morae is fundamentally a matter of the composition of the text, not how it is performed.

Nanti

Michael (Reference Michael2009) describes karintaa chants in Nanti (an Arawak language of Peruvian Amazonia). The chants that draw on fixed refrains form an inventory of over a hundred, some refrains consisting of vocables. The chants also have improvised sections, divided into lines in couplets. The two lines in a couplet have a modified phonology to make them the same length in morae, and Michael particularly emphasises that the words are modified to fit the moraic requirements of the line. Each line is usually seven morae long, sometimes eleven morae; in some cases the number of morae in the line is copied from the number of morae in the refrain. More skilled chanters will add a sound-sequence parallelism in which each line is a parallel member, such that the two lines in the couplet have the same sequence of heavy and light syllables. By my hypothesis, only the individual line need fit as a whole unit into working memory, but in fact the whole couplet is probably short enough to fit.

Hindi

A common Hindi metre is dohā, in which morae are organized into groups. A rhyming couplet is given in this section, divided into two long lines, each long line in two hemistichs called pāda divided by ‘||’. Each hemistich is usually a clause, and the long line is a complete thought. Each line has twenty-four morae (mora = mātrā in Indian theory), organized into different-sized groups of 6+4+3 and 6+4+1, each group called a mātrik gaṇa, divided by ‘|’. The parenthesized vowels are counted as morae but not pronounced, which incidentally again shows that the counting of morae is a characteristic of the composed text, not of the performance.

Heavy syllables cannot be split across mora-group boundaries. The four six-mora groups are each realized as a different pattern of three, four or five syllables, and a six-syllable group would also be possible, if all six are light. Though the number of morae is fixed, the number and pattern of syllables varies within limits set by the mora-group boundaries.

There are some additional stipulations. The line cannot begin with a light–heavy–light sequence, and the first pāda in a line cannot end on a heavy–light sequence. The line must end on a heavy–light sequence, which is the strictest stipulation and hence an example of a line-final cadence. The metrical rule holds of the two-pāda line, and the line is also the section to which the rule locating rhyme refers, which both suggest that specifically the line is held as a whole sequence in working memory.

There are many other Hindi metres. For example, in the classical Hindi language Braj Bhāṣā (Snell Reference Snell1991), some varṇik metres organise the line by repeating sequences of syllables: one repeats a pattern of heavy–light–heavy, with each heavy syllable capable of substitution for two light syllables. Others count morae into the lines: arill has a form with rhyming couplets, each line being twenty-one morae long divided into pādas as 11+10 or 12+9, with no further internal organization into groups. The lines are sufficiently small to fit as whole units into working memory.

Punjabi

The Punjabi dohṟe metre counts morae in groups of four. Each group begins with a stressed syllable, which can in principle be light or heavy:

The long line is divided into two hemistichs, each of which has seven groups of four morae to produce twenty-eight morae total. Each hemistich is divided into two parts, with a caesura rule forcing a word boundary after the fourth group. The long line is probably short enough to fit as a whole into working memory.

Classical Sanskrit

Classical Sanskrit has metres that fix the pattern of heavy and light syllables. Here, for example is the first of four lines in a stanza in the mālinī metre (from the Rāghavapāṇḍavīya); all four lines in the stanza must have exactly the same sequence of heavy and light syllables. The colon punctuation indicates a caesura:

The line is not a distinct syntactic or semantic unit in this tradition; only the four-line stanza is. The syntactically or semantically coherent stanza as a unit is too large to fit as a whole into working memory. This is consistent with the fact that there is no general requirement that syntactic or semantic processing be confined only to material in working memory.

Below, I schematically show three of the Classical Sanskrit metres, each controlling the weight pattern of syllables within a line. The pattern is fixed, and unlike most metres, the overall sequence is often not made up of smaller repeating patterns (it is aperiodic):

Fabb and Halle (Reference Fabb2008) divide the line at the caesurae into two or three distinct metrical sections, each of which is separately controlled by its own metre (this pursues a suggestion made to Fabb by John Smith). Sometimes, as in the line quoted earlier, a single long word takes up a single metrical unit. These metrical sub-sections are recycled into other named metres; thus the last part of sragdharā is the same as the last part of mandākrāntā, and the penultimate sections of all three are similar, differing in length. In this account, the metrical rules apply to a short section; a metre like mandākrāntā is a combination of three different metrical sections into a larger unit. We could, in principle, allow just the sub-part of the line to be held in working memory, and then say that the three sub-parts of the line are in a heterometrical sequence that is then repeated throughout the stanza. However, as elsewhere, it is likely that the longer unit, the long line, is sufficiently small to fit into working memory even though it need not do so for metrical purposes.

Deo (Reference Deo2007) takes a different approach, in which the line as a whole is presupposed by the metrical rules. She analyses these as metres which count morae in groups of four or five, as in some other Sanskrit metres such as āryā and Hindi dohā or Punjabi dohṟe, which uncontroversially operate by grouping morae. Thus, she analyses mandākrāntā into moraic groups with the structure 4+4+3+4+5+5+2, as shown here:

Though organized into moraic groups, these Classical Sanskrit metres allow no variation within the groups, which makes them different from Hindi dohā or Punjabi dohṟe. In this metre, the first group must be two heavy syllables and cannot be substituted by four lights or a heavy and two lights. This rigidity of pattern is explained by Deo's second proposal: what are standardly described as different Classical Sanskrit metres are actually the same metre in different variants. She notes that in many of the world's metrical traditions, the metre does not fully specify the rhythmic form of the line, but instead allows for variation. She suggests that the same is true of many of the apparently diverse metres of Classical Sanskrit, which can be reduced to a much smaller number of underlying metres. In this account, a stanza will set a particular variant on the underlying metrical pattern in its first line, and then will keep to the same variant in the next three lines; thus, rather than all the lines being in a specific metre of their own, the four lines are identical variants of an underlying metre. For example, mandākrāntā would be a metre whose underlying structure is the moraic grouping, and the actual syllabic weight sequence would be one possible variant on this, which is fixed for all four lines in the stanza, thus producing the effect of a distinct metre called mandākrāntā. If Deo is right, then we might confidently identify the whole line as the unit to which the metrical rules apply, and the line is sufficiently short to be contained as a whole unit in working memory. In addition, there is a strict rhythmic parallelism, with each line a parallel member, in a pattern that extends over four lines. Note that only the parallel member need be kept as a whole unit in working memory, not all members of the pattern.

Somali

Various Somali metres are based on a five-mora group, which Orwin and Riiraash (Reference Orwin and Riiraash1997) call the waax, in an analysis based on discoveries by Maxamed Xaashi Dhamac known as ‘Gaarriye’ and Cabdullaahi Diiriye Guuleed. The five-mora waax must end on a light syllable, in the pattern ⏕ ⏕ ⏑, with four possible expansions: ⎽ ⎽ ⏑ , ⎽ ⏑ ⏑ ⏑ , ⏑ ⏑ ⎽ ⏑ and ⏑ ⏑ ⏑ ⏑ ⏑ . The gabay metre combines four such groups into a twenty-mora line, with an optional extra short vowel syllable at the beginning. The line is divided into two hemistichs at the caesura after the twelfth mora:

The second hemistich must contain at least two long vowels, a restriction in the latter part of the line that is therefore a type of cadence. There is also alliteration throughout the gabay poem: every hemistich in the poem must contain one instance of a particular word-initial consonant, which in this poem is [g]. The two types of added form depend on different layers of section: the metre depends on the line and the alliteration rule depends on the hemistich. This is another split-level arrangement in which different added forms depend on two levels of section. The larger of the two sections, here the line, must be held as a whole unit in working memory.

Tashlhiyt Berber

Dell and Elmedlaoui (Reference Dell and Elmedlaoui2008) describe what they call the ‘straight metres’ of songs in Tashlhiyt Berber (Morocco), building on discoveries by Jouad (Reference Jouad1995). Here are three two-line stanzas of a song, each stanza consisting of a text line followed by a refrain line:

Each stanza-initial line is a sequence of four-mora groups, in the following fixed pattern of heavy and light syllables:

All lines in the straight metres have fixed patterns of this kind. The line is always divided into four-mora groups, in one of three possible syllable-weight patterns: (i) light–light–light–light, (ii) light–heavy–light, or (iii) light–light–heavy. Every second group must be light–light–heavy. These are metrical texts that are set to music. The second mora in each group must match a musical beat, and so the second mora in each group must be either a light syllable or the beginning of a heavy syllable, where the attack can fall; this corresponds to the restriction against a heavy–heavy or heavy–light–light group. Dell and Elmedlaoui (Reference Dell and Elmedlaoui2008: 29), however, argue that the text is metrical independently of its being set to music. The metrical rules apply to the whole line, which is sufficiently short to fit as a whole into working memory.

Jouad (Reference Jouad1995) discovered that each line in these metres has a ‘flatted syllable’ (syllabe bémolisée) in a specific numbered position that remains the same throughout the poem. A flatted syllable begins with a voiced obstruent; no syllable that follows it in the line may begin with a voiced obstruent, but the voiced obstruents [ʕ] and [ʁ] are ignored in checking this. In these lines, the fifth syllable is the flatted syllable. Other sound-level rules which apply to the line are that syllables without an onset are allowed only at the beginning of the line, and the final consonant in the line is discounted in assessing the metre, which here is [n] or [k], with the final vowel also ignored. These precise controls over phonetic form, though unusual, might be treated as types of systematic added form, and it is worth noting that they apply to the line as the relevant layer of section, as do the metrical rules.

In a small minority of songs, a word may be split across line boundaries, which is remarkable. Even more remarkable is the fact that a syllable can also be split across line boundaries, so that the onset is line-final and the nucleus line-initial. In performance, there may be a pause in vocalization and a few instrumental notes midway through the word or syllable, near the end of the line (Dell and Elmedlaoui Reference Dell and Elmedlaoui2008: 120). As they note, this shows that the metrical rules treat the line not as a sequence of syllables but as a sequence of mora-bearing units, and since syllable onsets play no role in bearing morae they can be separated off from the rest of the syllable (Dell and Elmedlaoui Reference Dell and Elmedlaoui2008: 127–31).

Serbo-Croatian

The Serbo-Croatian epic decasyllable is a line of ten syllables forming a syntactic unit (Jakobson Reference Jakobson1966b: 418). There is a tendency to stress odd-numbered syllables in a trochaic rhythm. If the seventh and eighth syllables are heavy, they may not carry stress. If the ninth syllable is light it may not carry stress, and any stressed syllable must here be a heavy syllable.

This is a cadence controlling weight and stress. West (Reference West1973: 173) comments that with the exception of Russian, the Slavic metres regulate the quantity of syllables only in the last four syllables of the line, as here. There are bridge and caesura rules which have the effect that the fourth syllable must be the last syllable in a polysyllabic word, and the line may not end on a monosyllable. Other rules govern the placement of words, including a rule that at least one word must begin with an odd-numbered syllable. All the metrical rules hold of the line as a unit, which is clearly small enough to fit into working memory as a whole.

Finnish and the Kalevala metre

The Kalevala metre is the metre of much older Finnish and Karelian poetry, some of which was anthologized and edited by Elias Lönnrot into the Kalevala (1835). There are eight metrical positions in the line, each of which is filled by a syllable except that two or three syllables can fill the second position. This is a line-initial looseness found in many metres, and note that it is specific to the line, not to some larger unit. Each word in the language begins with a stressed syllable and its placement is controlled by the metre. As a general principle, which is ignored at the beginning of the line but enforced increasingly strictly as the line progresses, a heavy stressed syllable must be odd numbered, while a light stressed syllable must be even numbered (Kiparsky Reference Kiparsky and Gribble1968: 167).

The metre controls the number of syllables and the location of stressed syllables depending on their weight, and the metre depends on the line, which is the unit that must be held as a whole sequence in working memory. Another distinct kind of form that holds over the line is ‘winnowing’, in which the words of a line are arranged in order of increasing length and the line is forbidden from ending on a monosyllable (Leino Reference Leino1986: 133). This is a version of the law of increasing numbers. Asymmetric variations in length hold also in non-poetic material such as Finnish names, where Kalle Kustaa with increasing length is a common name but Kustaa Kalle with decreasing length is almost unknown. This suggests that the asymmetric variation in length, though it holds of the line just as the metre holds of the line, is actually a distinct kind of form from the metre. The line is also characterized by alliteration, which usually holds between two words in the same line, and by parallelism as I discuss in Chapter 6. Various different kinds of form converge on the line here.

Chinese regulated verse: metre and rhyme

Chinese regulated verse, or lüshi, is the form of a poem normally in eight five-syllable lines or eight seven-syllable lines, divided into four couplets (Cai Reference Cai2008). I focus on the five-syllable regulated verse here. Words are monosyllabic, so five syllables are five words and five written characters, which gives the poem a regular visual shape. Syllables are divided into two tonal classes: (i) level tone L includes syllables with flat tone, such as mā, and syllables with rising tone, such as má; and (ii) oblique tone O includes syllables with falling-rising tone mă and with short falling tone mà. Each line must contain one level pair and one oblique pair plus one other syllable. This produces four possible tonal sequences OOLLO (seen below), LLOOL (seen below), LLLOO, and OOOLL.

Within the couplet, the second line replaces each L with an O and each O with an L, as seen above. The rules are followed with various degrees of strictness; though there are no violations in the first two lines of the poem, in the remaining six lines three of the four violations are in the first syllable of the line, a characteristic place for metrical looseness cross-linguistically. There is also a metrical principle that affects the metrical pattern of the poem as a whole, and is manifested as a relation of partial sound-sequence parallelism between the last line of a couplet and the following first line of a couplet, which must begin with the same pair of tones and end with a different tonal sequence. Thus the next line of the poem above should have an LLLOO sequence.

Rhyme is couplet-final and holds across the whole poem, which means that the last word in each couplet must rhyme with all other couplet-final words. The rhyme must be on a level tone syllable, which means that the final syllable in each couplet must therefore have level tone. In principle, it is also possible to have rhyme between the first two lines of the poem, but given the requirement that rhymes be on level-toned syllables, this forces an unusual couplet combination such as LLOOL/OOOLL. This means that the rule requiring rhyming syllables to be level overrides the rule requiring the inversion of tonal patterns between adjacent lines in a couplet. The requirement that the last syllable in a couplet must have level tone creates a cadence for the couplet as a whole.

The metrical rules can be stated as holding of the individual line, but I suggest we take the obligatory alternation in tone as showing that the metrical rules actually hold of the couplet. The rhyme is couplet-final. These two added forms coincide to suggest that the couplet is the unit held in working memory as a whole. We return to look at parallelism in this type of poetry in Chapter 6.

Thai

The following Thai stanza is in khlooŋ sii suphâap (Cooke Reference Cooke1980), one of a set of metres that control the number of syllables, rhyme and tone over a whole stanza.

The annotations specify the added forms. For example, the first line has a five-syllable sequence ending on a caesura followed by a two-syllable sequence, and this can optionally be followed by a two-syllable sequence, though here it is not. The number 1 indicates that the syllable must be spelled with the mai ek tone mark or ends in a short vowel or a plosive consonant; the number 2 indicates that the syllable must be spelled with the mai tho tone mark. Otherwise, the syllable must be spelled with no tone mark. The tonal rules, and to some extent the rhyming rules, refer to spelling, not to pronunciation; this is subject to some adjustment in practice. The letters A and B indicate which syllables must rhyme, in complex patterns. The last syllable of the fourth line marked C rhymes with the first one or two syllables of the first line of the next stanza. Though not annotated here, and not regulated, alliteration within the line is desirable, particularly between the last syllable of the first hemistich and the first syllable of the second hemistich. Line boundaries tend to coincide with major syntactic breaks.

This metre shows a modular dissociation between those forms that are specific to the hemistich or line, and those forms that hold as a pattern over a larger part of the text. Syllable count is controlled by the metrical rules for the line, because the line always begins with a five-syllable sequence, followed by a two or four syllable sequence with some line-final optionality. On the other hand, the metre of each line depends on where the line is in the heterometrical stanza, which is too large to fit into working memory, and so we must conclude that when heterometricality is stipulated in this way, it is not stipulated by rules that operate solely in working memory.

The placement of syllables with specific rhymes and tones is restricted in placement in each line, though again the placement within a line depends on the larger pattern of where the line is. We can thus distinguish between line-internal rules that may operate only within working memory and rules that operate over larger sections of text and do not depend on the whole text being held in working memory.

Australian aboriginal songs

Most traditional Australian (aboriginal) poetry is sung, and there has been extensive research into the forms of songs and their relation to musical form, some of which I note here. Ford (Reference Ford2005: 28–9) discusses Marri Ngarr songs where the requirement to set text to metrical music controls textual possibilities, forcing the omission of certain words in song-versions of sentences that would be required in ordinary speech, and reducing the syllable count of words, and forcing the use of vocables. Garde (Reference 200Garde2005: 77) discusses the Kun-borrk genre of songs performed in western Arnhem Land, and suggests that fitting the text to metrical music governs the choice of textual elements. Kartomi (Reference Kartomi1984) notes that the stressing of syllables is changed in song in the Pitjantjatjara, Bundjalung and Kayadild areas. Barwick, Birch and Evans (Reference Barwick, Birch and Evans2007) discuss Iwaidja songs in the jurtbirrk genre, in which the text seems to be composed to match a musical metre. For example, in one song, the musical units are each made of three clap-stick beats, and each word matches one unit, thus giving the song a certain number of words. There is some control over the number of syllables per word and hence alteration of words to fit into the song. They comment that ‘[a]s has been widely reported for other repertories of Australian song, there is a strong overall tendency for words to be set using short durations for syllables at the beginning of a word and longer durations for word-final syllables’ (Barwick et al. Reference Barwick, Birch and Evans2007: 18). Note that this is an asymmetry relating to the word and its setting to music, an illustration of the fact that asymmetries can arise at any level of structure.

Turpin (Reference Turpin2007a, Reference Turpin2007b) describes a repertoire of Kaytetye songs, called akwelye, where songs consist of two lines that have a set repetition pattern, such as AABB. The musical structure of the song is organized into rhythmic cells, each the length of two crotchet beats but with varying rhythms, and two to four rhythmic cells make up a rhythmic line. The text is composed to match the rhythmic structure: the number of phonological words in a song matches the number of rhythmic cells, so that, for example, four phonological words match four rhythmic cells, which is the maximum line length. A textual line matches a rhythmic line, and each phonological word in the line matches a rhythmic cell, and each syllable within the phonological word matches a note (a quaver/eighth-note). The text is subject to an interesting constraint within the rhythmic line: cells are ordered in decreasing number of notes per cell, and in combination with the rules matching phonological words to cells, this means that words mapping to shorter cells will tend to follow words mapping to longer cells.

Many song texts in Australian languages are quite short (often two lines), and the song itself involves much repetition of this short textual material. The whole song text could probably be fitted into working memory.

Children's songs

Cross-linguistically, children's verse may have a form determined by its being recited by matching to an external template, either a regular beat or a rhythmic tune. Both the musicologist Constantin Brăiloiu in 1954 (Brăiloiu Reference Brăiloiu, Brăiloiu and Lloyd1984) and the anthropologist Robbins Burling (Burling Reference Burling1966) have independently argued that there are cross-cultural similarities in how this matching happens. Brăiloiu proposed that there was a rhythme enfantin, or child's rhythm, such that ‘children's rhythms are spread over a considerable surface of the earth . . . identical throughout Europe . . . and more or less identical among the Kabylians, the Tuareg, [the people] of Senegal, of Dahomey [Benin] and of the Sudan, and the inhabitants of Formosa [Taiwan]’ (Brăiloiu Reference Brăiloiu, Brăiloiu and Lloyd1984: 207). Burling argued that metrical poetry, or something resembling metrical poetry, is found as a potential universal in children's chants and songs, even in languages that do not otherwise have metrical poetry in the adult literature (Burling Reference Burling1966: 1434). Burling took his examples of children's poetry from Chinese, Bengkulu (Sumatra), Cairo Arabic, Yoruba, and Serrano (California); he acknowledges the sparseness of the data as the basis for a claimed universal.

Brăiloiu and Burling both describe a performance practice involving a rhythmic template or tune rather than a metrical restriction on texts. The tune is the claimed universal, and because the text is matched to the tune in roughly similar ways, the texts are thus also rhythmically somewhat alike. For Brăiloiu, the basic rhythm is a sequence of eight beats grouped into pairs, each beginning with a stressed downbeat. The two beats making up the pair can be of different lengths. Importantly, this is a sequence based on musical timing and is not a sequence based on syllables: that is, it is a tune, not a textual constraint in itself. The text is matched to the tune. Though often one syllable matches one beat, it is also possible to have silent beats, so that fewer than eight syllables fill the line, or to have more than one syllable fill a single beat, so that more than eight syllables fill the line. Burling (Reference Burling1966) suggests that children's nursery rhymes are structured so that there are four strong beats to the line, evenly spaced in time, where strong beats normally match stressed syllables with up to three unstressed syllables between, but the fourth beat can be unmatched to a syllable. Again, it is clear that the text does not have to be metrical but must be matched to the tune and thereby acquires its regularity. Thus, Burling says, ‘the trick of simple poetry is to fit the words to the pattern, adjusting them in such a way that their stresses will somehow fit the rhythm of beats that our ear demands’ (Burling Reference Burling1966: 1424). Though Brăiloiu and Burling differ in approach, both basically claim a universal pattern in children's poetry that involves rules for matching a text to a tune where there are normatively four strong beats to the line. Similar local claims are also made elsewhere without universal implications. Thus, Moyle (Reference Moyle1987: 235) notes that children's game songs from Tonga tend to be in eight-beat lines. The text-to-tune matching normally has long vowels twice or three times the length of normal short vowels, and short vowels being allowed to vary greatly in length: ‘the constraints of the pulse evidently override those of spoken linguistic usage’, again showing that this is a regularity in the tune, to which the text is matched.

Brăiloiu's and Burling's proposed universals in children's poetry are sparsely evidenced in their own papers. They have been contested, for example, by Blacking (Reference Blacking1995: 160), who cites Brăiloiu but says that ‘it may not be possible to speak generally of rhythms peculiar to children’. In the Venda children's songs Blacking analyses, while four-strong-beat lines are predominant, they are not the only rhythmic pattern, and also a wider range of text-to-tune matchings is found than predicted by Brăiloiu. For example, Venda children's songs make extensive use of triplets, which Brăiloiu disallows as a text-to-tune matching in children's poetry. Minks (Reference Minks2002: 391) discusses Brăiloiu's influence and cites work critical of his methodology which problematizes the purported universals. She points out, for example, that adult literatures often have the same rhythmic patterns, and suggests that children may also have other different kinds of rhythm in their songs: thus there is nothing special, formally, about children's poetry. However, even if the Burling-Brăiloiu approach is right, it is not clear what these findings about children's songs tell us directly either about lineation or about metre. Instead, they may tell us about performance practices and rhythms, and perhaps universals in text-to-tune matching.

We are left with two unanswered questions. First, does children's poetry show general characteristics across languages, which might be associated with cognitive similarities across children? Second, are these general characteristics interestingly different from those of adult literatures, again showing some relation to the cognitive differences between children and adults? Children have reduced working memory performance in comparison with adults in complex ways and for complex and not fully understood reasons (Conlin et al. Reference Conlin, Gathercole and Adams2005). Does children's poetry have lines that are shorter or formally simpler, or have added form that requires less processing effort? We might wonder if the tunes and text-to-tune matching principles proposed by Brăiloiu and Burling are designed to allow lines to be fitted to the child's less developed working memory.