Poetic Form

Geoffrey Russom, Brown University

In Russom, Evolution, I propose a general theory of poetic meter and explore its implications for metrical evolution. This article provides a brief summary of my findings as they relate to the study of literary universals. My initial hypothesis is that metrical constituents are abstracted from linguistic constituents: metrical positions from syllables, metrical feet from words, and metrical lines from syntactic constituents, preferably sentences. On this hypothesis it is obvious what constitutes the best match between abstract metrical patterns and the words chosen to realize them by the poet.

I think that my hypothesis qualifies as the null hypothesis but it is not uncontroversial. In Fabb and Halle, Meter in Poetry (11), the authors assume that “lines are sequences of syllables, rather than of words or phrases.” They reject “approaches which assume that literary language is a development of ordinary language, using the resources already available to it,” citing Hanson and Kiparsky, “Parametric theory,” as one of the language-based approaches they reject (pp. 46-8). According to an anonymous reviewer of this article, equating lines with sentences ignores the fact that syntax is asymmetric and hierarchical but meter is not. This generalization seems to be rather widely accepted but I regard it as an important error that has impeded metrical research. As will soon become apparent, I accept all regulated features of the poetic line as metrical. Regulated features of the line can include alliteration, rhyme, the word boundary, the major syntactic break (sometimes called the caesura), and syntactic parallelism. In the Celtic and Germanic alliterative meters I have studied, the constraints on syllabic patterns within the line are secondary consequences of constraints on morphological and syntactic structure.

In meters with long lines, there may be a metrical constituent between the foot and the line. In Germanic alliterative meters, the line consists of two verses and each verse contains two feet. The first verse of the line is called the a-verse; the second is called the b-verse. Important metrical constraints apply at verse level. I derive metrical constituents between foot level and line level from phrases between word level and sentence level. In meters that have a short line length relative to average word length, the line pattern might also be derived from a syntactic constituent below the level of the sentence.

All traditional meters employ the line. One early Irish meter specifies the number of words per line but imposes no constraints on syllables. This meter employs feet but no metrical positions. The Serbo-Croatian decasyllable counts syllables but imposes no constraints on word count, patterns of stress, or patterns of syllable length. This meter employs metrical positions but no feet. The meter of the Biblical psalms employs a line with neither feet nor metrical positions. The adjacent lines of a Biblical couplet match one another in syntax and propositional semantics, the characteristic features of the sentence as distinct from the word and the syllable. In this parallelistic meter, each line of the couplet takes the adjacent line as its semantic-syntactic prototype. Lines within a couplet are complex to the extent that they differ syntactically and semantically.

Certain features of the line can be fixed somewhat arbitrarily, for example the number of feet or the number of metrical positions. Many metrical rules are based on native-speaker intuitions, however, and operate below the level of consciousness, for the poet as well as the audience.

Metrical rules can be expressed as constraints on departure from a prototypical line pattern. The prototype for iambic pentameter can be realized by a line like

Refíned / gourméts // demánd / supérb / cuisíne.

As traditionally described, the iambic pentameter line has ten metrical positions organized into five iambic feet (separated by slashes). Each foot has two metrical positions, and the most prominent syllable within a foot is normally placed on its second position. In the prototypical line above, each metrical position is realized as a syllable with normal length, each iambic foot is realized as an iambic word, and the line is realized as a sentence with normal English word order, which is SVO (subject-verb-object). The major syntactic break within the prototype (notated by a double slash) falls after the fourth metrical position, between the subject (refined gourmets) and the predicate (demand superb cuisine). This is the normal position for the caesura in iambic pentameter. The caesura falls most often to the left of center because specifiers like subjects are shorter than predicates, which typically have a tensed verbal head and one or more complements (e.g. an object, a governed infinitive, and/or a prepositional phrase). Like sentence patterns, prototypical line patterns are hierarchically asymmetrical.

Linguistic rules applying at higher levels of structure can modify the output of lower-level rules. The most prominent stress assigned by a word-level rule can be subordinated within the phrase, for example by the higher-level rule for adjective-noun constructions. In a Modern English phrase like grèen grápes, the primary word stress of the adjective is subordinated to the primary word stress of the following noun. The most prominent stress in a small phrase can be subordinated by rules applying at the level of the sentence, for example by the nuclear stress rule, which subordinates every stressed syllable before the nuclear stress and provides the phonological basis for pronunciation. In a metrical line, similarly, line-level norms based on sentence structure exert more influence than foot-level norms based on word structure. Level-dependent effects were recognized within generative metrics as early as 1989, when Youmans observed that “higher-level boundaries are progressively more significant than lower-level ones” (“Milton’s meter” 376).

Realization of the foot as a word is clearly desirable in iambic pentameter, especially at the end of the line, where the principle of closure restricts metrical complexity. In his analysis of a large corpus of Milton’s unrhymed verse, Youmans observed that Milton often departs from ordinary word order to place an iambic word at line end, but rarely to remove an iambic word from that position.

Realizing the line as a clause or sentence is even more desirable than realizing the foot as a word, and occurs more frequently. In Shakespeare’s well-known sonnet 18, ten of the fourteen lines are realized as sentences or as clauses transparently related to sentences. In other sonnets, however, Shakespeare makes significant use of long, complex sentences that occupy two lines instead of one. The power of line-level constraints stands out with particular clarity when other constraints are loosened and compensatory measures are required to keep the metrical complexity of the line within tolerable limits. In the 404 iambic trimeters of Sir Gawain and the Green Knight, extrametrical words are added more freely than in iambic pentameter tradition (Russom, Evolution, chapter 10). To compensate, the frequency of trimeter lines realized as sentences or clauses is kept above 50%, a conservative estimate that can be validated by any reasonable counting procedure. Since these lines are significantly shorter than Shakespeare’s, we would expect a high frequency of sentences occupying two lines in Gawain, all other things being equal; but the Gawain poet keeps sentences short and simple to limit the total complexity of a line that is unusually complex in another way. Another compensatory device of the trimeter is a high relative frequency of closing feet filled by iambic words, as in Milton’s unrhymed verse.

The relation between sentence rhythm and the poetic line is highlighted by sound echoes like rhyme and alliteration, which are most effectively situated in prominent linguistic constituents. In SVO languages like Modern English, the last stressed word in a sentence normally has the most prominent phrasal stress, called the nuclear stress. The best site for the rhyming word in such languages is line-final position. Alliteration is universally associated with syllabic prominence (Kiparsky, “Role of linguistics” 231). In SOV languages like Proto-Germanic, the first stressed word of a phrase normally bears the most prominent stress; and the first word with metrically significant stress alliterates obligatorily in ancient Germanic meters, which include Old English, Old Norse, Old Saxon, and Old High German meters (Sievers, Metrik). The word at the end of a typical Proto-Germanic sentence has subordinate stress; and alliteration at the end of the line is systematically ruled out in the four alliterative meters. In a VSO language like Old Irish, the first stressed word of the sentence is typically a verb with subordinate stress. In the most common Irish alliterative meter, which resembles ancient Germanic meters in important respects, the first stressed word of the verse typically does not alliterate and the first alliteration falls on the second stressed word.

In meters with a single foot pattern and a single line pattern, metrical variety is achieved by carefully regulated departure from the linear prototype. In Old English meter, there is a foot pattern for every native word pattern. Most combinations of two feet qualify as verses and many combinations of two verses qualify as lines. In this meter, the many verse patterns and line patterns provide ample variety; and conformity to the prototypical two-word realization of each verse type is much more strictly regulated than conformity to the iambic pentameter prototype. Meters that switch unpredictably from one pattern to another, compensating with stricter conformity to patterns, provide especially valuable evidence for deep analysis of poetic form.

The most highly favored foot pattern in Old English corresponds to the pattern of trochaic words like hwīle ‘time,’ which has a stressed root syllable followed by an unstressed syllable. This is the prototypical Old English word pattern (Dresher and Lahiri, “Metrical coherence”). The most highly favored verse pattern has two trochaic feet. An example from Beowulf is lange / hwīle ‘a long time’ (the a-verse of line 16, notated as 16a). This verse prototype establishes a norm of two primary word stresses and four metrical positions for the verse. The prototype for the line corresponds to an SOV Germanic sentence, with a weakly stressed finite verb at the end preceded by more prominent words such as subject and object nouns. Lines ending in a finite verb are highly favored.

As with norms of iambic pentameter, norms of Old English meter exert more influence at higher levels. Consider for example two variants of verse type Da, one variant with four metrical positions and another with five. Both variants have a stressed word in the first foot and a compound word in the second foot. An example of the shorter variant is Beowulf 164b, fēond / mancynnes ‘enemy of mankind.’ An example of the longer variant is Beowulf 223a, sīde / sǣnæssas ‘large sea-nesses.’ The longer variant realizes the first foot with a trochaic word and is optimal in that respect; but it has five metrical positions, contravening the four-position norm for the verse. If the foot norm exerted more influence, the longer variant would have higher frequency; but in fact this variant has about half the frequency of the shorter variant and is vanishingly rare in the b-verse, where the principle of closure restricts complexity. The four-position verse norm exerts more influence than the foot norm.

In Beowulf, two-word verses like lange / hwīle, with each foot realized as a trochaic word, have highest frequency in the b-verse. When the first foot is realized as a group of constituents rather than a single word, the added metrical complexity inhibits placement in the b-verse, all other things being equal. Verses like 121a, fēond on / helle ‘a fiend in hell,’ occur 112 times in the a-verse but only 24 times in the b-verse. Distribution is very different, however, for verses like 45b, forð on-/ sendon ‘sent forth,’ which occur 11 times in the a-verse and 65 times in the b-verse. Such verses realize the first foot as a constituent group but end with a finite verb. The line-level preference for a finite verb at the end exerts more influence on verse placement than the preference for realization of the foot as a word.

For analysis of rule conflict in Germanic alliterative meters, it is helpful to think in terms of violable rules with varying influence. These rules are like the rules of Optimality Theory in important respects (Russom, Evolution 283). Unlike OT linguistic rules, however, metrical rules do not select one and only one linguistic pattern as the only acceptable output. Instead, optimal linguistic features are incorporated into prototypical line patterns. Departures from optimal form occur somewhat more freely in metrical systems than in linguistic systems, and under different conditions; but each departure from the metrical prototype increases metrical complexity, limiting the relative frequency of complex verse patterns and inhibiting their placement in the closing half of the line.

To be preserved for its cultural value over a long period of time, a meter must evolve to accommodate language change. English alliterative meter can be traced from its reconstructed origin in late Proto-Germanic (about 300 BCE) to its death in the sixteenth century CE. Evolution of this form involved abandonment of constraints that became unsustainable and introduction of new constraints based on traditional tendencies that were still perceptible. When the Old English inflectional system was vitiated by reduction of unstressed word-final syllables, unstressed function words were more urgently necessary, and the meter had to tolerate a larger number of unstressed syllables per line. In Middle English alliterative poetry, the metrical coherence of the line was preserved in part by stricter alignment of verses with clauses and by obligatory double alliteration in the a-verse. Effects of the principle of closure were reinterpreted as a rule requiring asymmetry between a-verses and b-verses. The more complex types were increasingly confined to the a-verse, and the b-verse was increasingly populated by types that ended with a trochaic word. A new rule was then introduced requiring a trochaic constituent at the end of the line. For thorough philological discussion of the new Middle English constraints and related bibliography, see Putter et al., Studies. By the late Middle English period, when SVO syntax was the norm, alliterative poets were using conspicuously archaic language to preserve a form almost two thousand years old that was designed for a different language type. After major language changes in early Modern English, poetic archaisms were no longer comprehensible and the meter died.

Borrowing from another language can complicate metrical history. During the early Middle English period, poetic culture was dominated by rhymed forms imported from France for the Norman French aristocracy. In the late fourteenth century, French was less widely spoken, but the prestige dialect of English had accumulated a vast store of words with the prototypical iambic pattern of French borrowings. Rhymed iambic meter must have seemed a natural choice for Chaucer, who stood close to the center of royal power. As the frequency of iambic words increased, the frequency of trochaic words declined. By Chaucer’s time, final -e in trochaic words was no longer consistently scanned as a metrical syllable. After massive loss of word-final unstressed syllables in early Modern English, the trochaic norm for native words had been replaced by a monosyllabic norm –– the stereotypical ‘four-letter word.’ The function of a lost inflectional syllable was usually taken over by a preceding monosyllabic function word such as a preposition, an article, or a demonstrative pronoun. The new stressed monosyllables had no inherent rhythm of their own and iambic meter was ideally suited to the proliferating iambic phrases that replaced trochaic words. Compare for example Beowulf 1507a, hringa / þengel, with its literal modern English translation, the lord / of rings. The Old English verse has two trochaic words; its translation has two iambic phrases.


  1. Metrical constituents are abstracted from natural linguistic constituents.
  2. Meters are not reducible to syllabic patterns, contra Fabb and Halle, ‘Meter in Poetry.’ Meters also employ morphological patterns, syntactic patterns, and perhaps other linguistic patterns as well, such as patterns of prosodic constituents between the syllable and the word.
  3. Meters exploit the native speaker’s ability to apprehend all aspects of linguistic form in real time with automatic facility.
  4. The only metrical constituent universally required is the line. The metrical position, the foot, and the verse (or half-line) may be employed but are not required.
  5. A meter can be represented as a prototype incorporating optimal features of the relevant language.
  6. Metrical constraints can be formulated as violable rules. A given rule can augment or diminish the effects of other rules.
  7. The influence of a violable rule depends on the metrical level at which it applies. Higher-level rules exert more influence than lower-level rules.
  8. A meter preserved through language changes may abandon unsustainable constraints and introduce new sustainable constraints based on statistical frequencies and stylistic preferences.

Topics for Future Research

Conclusions 1–8 above are based on meters I have been able to analyze. My frequency assessments are obtained from complete manual scansions of Beowulf and Gawain in sortable and filterable Microsoft Excel files. Researchers interested in obtaining copies of these files should send an e-mail to < geoffrey_russom@brown.edu >. In some cases, my conclusions can be sharpened in more than one way consistent with the data I have gathered. Language-based analyses of other meters by qualified researchers can impose useful constraints on progressively sharper formulations of universalist constraints. The following are just a few examples of relevant topics.

  1. Research on meters with short lines relative to average word length might reveal whether they are abstracted from phrases below the level of the sentence, as with Old English verses, or create allowable mismatches to lines abstracted from sentences.
  2. Short textual lines might be genuine metrical lines or artifacts of formatting conventions for manuscripts and printed documents. A Japanese specialist, for example, might consider whether the haiku is a genuine three-line unit or is best analyzed as a single line with three cola.
  3. I am willing to be persuaded that a meter can have an obligatory pattern of syllable stress or length that is in some sense ‘artificial’ and does not correspond to a natural linguistic pattern. Classical Greek and Latin quantitative meters may have these characteristics but I have not encountered a focused argument to that effect and would like to see one. I would analyze such an artificial pattern as one that has strong and weak metrical positions abstracted from strong and weak syllables but no ‘word feet’ of the kind discussed in this article. Artificial specification of stress or length patterns would be analogous to artificial specification of the number of feet or metrical positions, familiar subjects of conscious reflection by poets and their audiences.
  4. For language-based metrical theories, corpus-based studies can provide useful information about the relative frequencies of syllabic patterns, word patterns, and syntactic patterns. Studies of this kind may already be available for the researcher’s language of interest.
  5. Researchers who work on modernist experimental poetry might consider whether a given experimental form could be apprehended with automatic facility during oral performance; and if not on first exposure, whether subsequent exposure or conscious instruction could make the form intuitively accessible.
  6. The earliest meters provided important technologies of cultural preservation for customary law, religion, history, myth, and heroic legend. The cultural functions of poetic meter provide topics of great potential interest for psychologists and aestheticians.

Works Cited

(For additional bibliography see Russom, Evolution.)

Dresher, Elan, and Aditi Lahiri. “The Germanic foot: metrical coherence in Old English.” Linguistic Inquiry 22 (1991), 253–86

Fabb, Nigel, and Morris Halle. Meter in Poetry: A New Theory. Cambridge: Cambridge University Press, 2008

Hanson, Kristin, and Paul Kiparsky. “A parametric theory of poetic meter.” Language 72 (1996), 287–335

Kiparsky, Paul. “The role of linguistics in a theory of poetry.” Daedalus 102 (1973), 231–44

Kiparsky, Paul. “The Rhythmic Structure of English Verse.” Linguistic Inquiry 8 (1977), 189–247

Putter, Ad, Judith Jefferson, and Myra Stokes. Studies in the Metre of Alliterative Verse. Oxford: Society for the Study of Medieval Languages and Literature, 2007

Russom, Geoffrey. The Evolution of Verse Structure in Old and Middle English Poetry: From the  Earliest Alliterative Poems to Iambic Pentameter. Cambridge: Cambridge University Press, 2017.

Sievers, Eduard. Altgermanische Metrik. Halle: Niemeyer, 1893

Youmans, Gilbert. “Milton’s meter.” In Paul Kiparsky and Gilbert Youmans, eds., Phonetics and Phonology: Volume 1, Rhythm and Meter, 341–79. San Diego: Academic Press, 1989.