Ancient Rhythmicians and Modern Prosodists: Searching for the Location of Meter

Steven J. Willett University of Shizuoka, Hamamatsu Campus
2-3-2 Nunohashi, Shizuoka Prefecture
Hamamatsu, Japan 432
Voice and autofax: (53) 457-4514
Email: Steven J. Willett

        Meter is measure, and what it has measured in the western poetic tradition for the last 2,800 years is not in serious doubt. We can trace with great linguistic precision the evolution of Indo-European syllabic meters into the fully-developed quantitative versification of Greek and Latin, follow the breakdown of quantitative versification in the third century CE when the phonemic distinction between long and short syllables is lost and then map the Great Resyllabization of south European verse as it mutated from the ruins of Classical metrics into the standard repertoire of Romance verse forms. The same analysis can be done for Germanic and Slavic versification. The initial results of this work in the relatively young science of comparative metrics are contained in Mikhail Leonovich Gasparov's Ocherk istorii evropeijskogo stixa, originally published in Russian in 1989 and now available as A History of European Versification in a brilliant translation by Gerry Smith and Marina Tarlinskaja. <1> It is the first attempt at a comprehensive synthesis of an enormous mass of research stretching from Sanskrit to verse libre, and covers all the principle Classical, Romance, Germanic and Slavic languages. If any book can be called "indispensable" to everyone in our field, Gasparov's gift to us—as I like to call it—qualifies. The findings of comparative metrics set out by Gasparov can be supplemented with over two millennia of diverse secondary evidence, including statements by the poets themselves concerning their rhythmic intentions, musical settings to poetry, observations by competent literary critics, writings of ancient rhythmicians and metricists, verse inscriptions and translations. Although difficult to interpret and often conflicting, much of it corroborates what the diacritical study of comparative metrics has told us. There two sources, comparative metrics and literary historiography, unite to give us a reasonably clear picture of the formal conventions that underlie versification at virtually every chronological stage of its development.

        Having said all of that, what have I said?

        On the empirical level, it is obvious that those who write on the theory of versification as opposed to narrow technical analysis should have a solid grasp of the complete western tradition, beginning with Greek and Latin. Critical writings on theoretical metrics are strewn with errors that stem from a narrow provincial knowledge of English and widespread ignorance of Classical, Germanic, Romance and Slavic languages. I cannot of course detail all these, but typical examples are the claims that (1) unrealized beats occur at the ends of the short lines in common measure, (2) metrical feet are not experiential, (3) the caesura is not a metrical phenomenon and (4) English is a stress-timed language. All are factually, and quite demonstrably, false. Let's briefly review the claims in order. (1) Whether unrealized beats exist or not, comparative metrics shows that common measure evolved from the medieval Latin goliardic 7d+6f line with a tendency to trochaic rhythm. In English the dactylic ending at the midline pause changed to masculine, while each hemistich acquired an unstressed syllable at the start to create a rough iambic meter. The second hemistich then became the shorter three-stress line when the meter evolved into oral folk poetry and, as Gasparov notes, lost some of its accentual-syllabic clarity, turning into a 4-3-4-3-stress dol'nik <2>. Since the even lines of common measure spring from a three-stress hemistich and always contained three stresses in origin, they cannot be considered truncated four-stress lines, thus they never ended with unrealized beats. (2) A large body of evidence from linguistics and cognitive science validates the concrete existence of prosodic feet as the source of metrical feet. On the grounds of historical linguistics, A. M. Devine and Laurence D. Stephens argue in their magisterial The Prosody of Greek Speech that "the rhythmic structures of Greek verse reflect, arise from and already exist in the rhythms of Greek speech and are not in principle the result of mapping the rhythms of Greek speech onto extraneous patterns, that is onto temporal patterns of nonlinguistic origin."<3> Prosodic feet are, they demonstrate, the basic constituent of the rhythm of the Greek language (as in many other languages) and provide the basis for metrical feet. On the grounds of cognitive science, Fred Cummins and Robert F. Fort at Indiana University have in their recent experimental work established strong evidence for the reality of the metrical foot as a well-defined unit in the production of speech.<4> (3) Whether the caesura is a metrical reality depends on definition and poetic tradition. In English poetry the caesura is widely considered nothing more than a pause created by the syntax, and thus a matter for oral declamation, but in Classical poetry it is a word boundary that intersects the metron and constitutes a crucial rhythmic fact to which the Greeks were acutely sensitive, as Devine and Stephens document in their earlier study Language and Metre: Resolution, Porson's Bridge, and their Prosodic Basis .<5> In Russian accentual-syllabic poetry, or syllabo-tonic poetry as they like to call it, the caesura is a word break that occurs at the same position in the line throughout the poem, often coinciding with a slight syntactical break.<6> It significantly affects the rhythm by forcing a stress on the following ictus to reestablish the rhythm after the break. Marina Tarlinskaja has even argued that the English caesura is not properly a pause but, like Russian, a word boundary that regularly occurs at the same position in the line, where it produces effects similar to those of Russian: the end of the first hemistich behaves metrically like line-end and the beginning of the second hemistich like line-beginning. And finally, (4) all experimental attempts to find stress-timing in unconstrained English have proven futile. Cummins and Fort in fact offer direct phonetic timing data in support of more varied, music-like rhythms in human speech than simple stress-timing. The elimination of stress-timing eliminates the explanatory rationale for many expressive effects that are supposed to result from collocations of stressed or unstressed syllables.

        What do erroneous claims like these tell us? First, many attempts at a general theory of rhythm are torqued around modern English poetry, often quite contemporary, because the authors don't know any other languages—often not even Old and Middle English. That vitiates their claim to generality ab initio whatever other value the theories may have. Second, universal assertions, such as the nonexistence of the metrical foot, often founder on lack of familiarity with comparative metrics, historical linguistics and modern cognitive science. Theories built even partially on such assertions share the logical consequences. Third, any comprehensive theory must apply equally to the meters of all poetic traditions or abandon its claims to comprehensiveness.

        But to remain exclusively on the empirical level of rhythm is, in certain respects, rather like taking up residence in Edwin A. Abott's Flatland as a square. Flatland, you may remember, has only two dimensions, and social status is determined by shape. Squares appropriately are professors. If a sphere came down to visit us, and talked about the incredible romance of many dimensions, we should hardly credit it. Yet a slight shift of our gaze upward from the strictly measurable to the larger, overarching aesthetic domain reveals immediately some of the many problems with traditional metrics so forcefully detailed by Richard D. Cureton in his recent work.

        From the gravamen of Cureton's critique, two issues stand out in particularly sharp relief. Rhythm, he maintains, is persistently misidentified with, and therefore reduced to, a linear pattern of linguistic elements: syllables, stresses, tones, morae, phrases and the like. The formal patterns of conventional metrical analysis have not, however, succeeded except in local and sporadic ways in relating meter to specific semantic, iconic, rhetorical and esemplastic values. We lack, in short, a robust heuristics to show how meter means. Not having that, we also lack a key pedagogical rationale to justify repositioning versification at the heart of poetics. The heap of metrical commentary grows like a vast hill of shifting sand that supports nothing. The second issue follows from the first. Literary historiography tends to treat poetic texts, and assign them importance, according to external philosophical and cultural concerns rather than internal poetic language. Since we cannot relate the language of poetry transparently to these contextual concerns, our literary histories can include poetry only by reducing poetic achievement to artistically peripheral concerns. That deals a further devastating blow to pedagogy by encouraging teachers to present chronologically disparate poetry in an ahistorical vacuum or, more commonly, to teach only contemporary poetry.

        One response to this confusion has been the attempt to construct a temporal theory of rhythm employing a more-or-less close analogy with musical theory. The absence of semantic referentiality in music makes the project attractive because it permits the critic to free rhythm from its subservient role either as a mere intensifier of the narrative, dramatic and meditative paradigms that are thought to constitute a poem or as a signifier of aesthetic difference from prose. Both these roles reduce rhythm to an epiphenomenon of mimesis.

        I propose to examine some of the difficulties attendant on the attempt to locate poetic rhythm in a musico-temporal experience. I do so in a cooperative, not a destructive, spirit to advance our discipline by helping to clarify, point and tighten the most useful elements of temporal theories. A hard attack leads to a hard defense. Here are my objections in rank order of increasing seriousness.

Objection No. 1

        In a system of poetic rhythm based on music theory, the identification of meter with beating is an obvious gambit to circumvent the difficult fact that meter is an artistic convention for the intentional selection and patterning of linguistic phenomena. While it is quite correct to say that the psychological mechanism for perception of rhythm is not in the language, it is incorrect to say that meter is a beating independent of the language that elicits it, primarily physical or gestural in source, high subjective, rigidly binary—a kind of cycling wallpaper in effect—and unconstrained by lineation. There are two strong arguments against the identification of meter with musical beating.

        The first is the existence of separate metrical and beating systems in Greek poetry. Greek lyric monody and epic were sung to a lyre. Every Greek poet was also a composer. We have ample evidence to show that musical rhythms are faithfully reflected in the meters of the texts. Those meters were quantitative, that is, patterns of temporally long and short syllables that correspond to long and short notes. (See handout item #1.) Ancient Greek rhythmicians, who were theorists of rhythmic studies in contrast to metricists who were essentially just classifiers, noted that the princeps of lyric verse in the rhythm u u – u u – was shorter than the biceps by a quantity they could not measure, meaning that the ratio of u u to – was not an absolute 1:1l Various metrical features of the poetry suggest a ratio of long to short notes in sung lyric on the order of 5:3. This is probably close to the ratio in ordinary spoken Greek, though in lyrics written to accompany bodily movement like dancing or marching it was a stricter 2:1. The Greeks beat time to music. Each metron or measure was divided into 'up' and 'down' (anô, katô) sections, later called 'lift' and 'step' (arisis, basis) or 'lift' and 'placement' (arsis, thesis). These sections corresponded to raising or lowering of the foot, but were not themselves equal in length. Rhythmicians classified rhythms according to the ratio between the sections. I've included some common meters with their arsis and thesis marked in item #2 of the handout. It should be clear from the examples that the beat cycle comprises elements that could be of unequal duration. When unequal, the 'down' part of the metron was usually the longer. Here we have two interlocking but independent rhythmic systems that violate the identification of meter with beating.

        The second strong argument against the identification of musical beating with meter comes from language like Japanese and Chinese, which employ pure syllabic and syllabo-tonic versification respectively. Japanese possesses a complex intonation system that is partially phonemic but entirely stressless, and its poetry is strictly syllabic. The classical Japanese uta, also called waka or tanka, is a thirty-one syllable poem divided into five- and seven-syllable segments—which must not be confused with lines—and constitutes the foundation for virtually all poetry written in Japan for over a thousand years from 850 to 1900. The attempt to beat time with a stress-based language like English might produce some interesting results, since tactical beats must correspond to stresses or to silences construed as unrealized beats, but that is a linguistic impossibility in Japanese. And it becomes impossibility compounded if we try to impose the notion—which lies at the heart of Cureton's theory—that beats form a graded hierarchy with any three metrical levels making a metrical span or measure and a measure of tactical beats making a line. To begin with, Gerhard Widmer at the University of Vienna has shown that a musical measure may be subdivided into as many as 2^6 or 64 levels, not three. In practice, the hierarchy is quite arbitrarily defined as declining in prominence: the first beat is strongest, the third somewhat weaker and the second and fourth weakest. Let me read a famous poem by Sami Mansei from the eighth century as an example of how such an imposition fails. Please see handout item #3 with translation:

Yo no naka o/nani-ni tatoemu/asabiraki/koginishi fune no/ato naki ga goto.
(Being in the midst of life: what shall I compare it to—daybreak, emptiness
left behind a rowing boat.)

        The poem is a continuous stream of syllables without lineation. Even when printed with artificial lineation, as waka sometime are, it is sonically meaningless. Here the segments are marked by postpositives "o" and "no" in segments 1 and 4, by verb phrase in segment 2 and the noun "asabiraki" in absolute construction in segment 4. We find no phonological basis for beating, since the five- and seven-syllable segments have neither stresses nor repeating tonic sequences like Chinese, and cannot therefore elicit a four-tactus graded hierarchy. Might we nevertheless feel a musical beat, understood to be highly subjective and primarily physical/gestural in source, while reading the poem? If so, it would be a beating completely detached from both the linguistic prosody of Japanese and the traditional syllabic meter of the poem—nothing more than a distorting western habit imported into the poetry. A westerner might persuade himself that he feels a 1-2-3-4 binary beat, but no Japanese perceives or reads waka with one. The poem is too short for a cyclically invariant musical rhythm to establish itself, nor is there anything like an upbeat or downbeat of the type we have in Greek arsis and thesis to provide even the ghost of a subjective sense of beating.

        What is true for Japanese is equally true for many other languages with strict syllabic versification. If we turn to Chinese, the problems of trying to read the meter as musical beating are even more severe. I take my example from lü-shih or regulated verse, which produced most of the great poetry written by poets like Du Fu, Li Bo and Bo Juyi during the T'ang Dynasty from 618-907 CE. (See handout item #4.) Regulated verse was written in eight-line stanzas of five or seven syllables with identical rhymes on the even lines and an optional one on the first line. The four interior lines constituted two couplets, each of whose two lines were, word-for-word, syntactically parallel but semantically antithetical. More importantly, the four tones of the Chinese language were regulated by dividing them to two categories, level tones and deflected tones, and then specifying the deployment of the two categories to produce subtle patterns of repetition and contrast. Chinese has words of more than one syllable, but every syllable is an ideogram. So the number of characters equals the number of syllables, and that decides the rhythm. Against the linear, discrete, syllable-by-syllable movement of the line, we have the contrastive mirroring of tonal patterns within and between each couplet, a formal effect only perceived by vertical grouping. This means, among other things, that the syllable-delimited line is the integral unit which provides a structural foil for contrast: first lines mirror lines, then couplets mirror couplets. Like Japanese, one cannot beat time to lü-shih; there is no prosodic support for assigning a tactus, quite aside from the fact that Chinese music is based on completely different tonal and harmonic principles than western music.

        What these three examples show—and they could easily be multiplied—is that we have not solved the problem of creating a universally applicable explanation for the meaning of meter simply by redefining it as a linguistically independent musical beating. That approach illogically assumes the universal validity of western music for all poetries and compels us, if we follow it strictly, to beat time with poetic forms that entirely frustrate the imposition of a triply-graded stress hierarchy. There seems to me no way out of this fatal impasse: musical beating is not simpliciter meter. Its failure as a general explanatory hypothesis means that the motive engine to drive a system of poetic meter based on music fails, and with that failure all other high-level rhythmic components dependent on beating, which of courses measures tonal progression, fail equally.

Objection No. 2

        Lessing was the first critic to draw a precise contrast between the different aesthetic effects of a temporal medium like poetry and a plastic medium like painting or sculpture. In Laokoon, Über die Grenzen der Maleri und Poesie, his intent was to refute the widely-held belief that a poem is a speaking picture—ut pictura poesis in the common Latin formulation—and to establish the boundary between what was possible in painting and what in poetry. We need to do the same between reading poetry and listening to poetry. No one doubts that poetry is a kind of temporal medium, but what kind? If the musical analogy is anything more than a shallow, utilitarian comparison, then we must conceive a poem as progressing from start to finish in a continuous, rhythmically connected cascade of sound. But this is only true of listening to a poem, and even then most recitation lacks the inevitability of forward movement we experience in music simply because a speaking voice moves more irregularly than instrumental sound and is constantly making adjustments for personal emphasis, accent or nuance. Recitation is courser, more granular to use a phrase from artificial intelligence theory, than music. Reading, however, does not unfold in a continuous rhythmic stream from beginning to end. We read, stop, reread bits and pieces, look back over some earlier passage to clarify meaning, continue reading, glance ahead perhaps to confirm thematic direction, pause to reflect on moving aspects of the verse, reread memorable passages and constantly repeat all these disparate reading acts in every possible combination. Reading is, except for occasional moments, never temporally seamless. Listening to a poem and reading a poem are two very different cognitive activities. Reading a poem can almost be thought of as a sort of mental reappropriation. The text becomes continuously and fluidly interactive as we pull the words and phrases apart and rearrange them on our mental canvas in ways that the author could never have predicted—even when author and reader are the same. I shall return to this point in a moment, but it leads to my third, and in some respects most serious, objection.

Objection No. 3

        Despite the lingering belief that we think solely by the propositional syntax of lexical information, a belief the French deconstructionists did much to preserve with their language metaphysics, the recent work of Stephen Kosslyn and Roger Shephard among others has shown how crucial a role depictive imagery plays in our thinking, problem solving and communication. Using a battery of new techniques from positron emission tomography scanning to diagrammatic methods of exploring perceptual activities, they demonstrate that we think in both depictive and separate but conjunctive propositional representations that are syntactically separate from one another. Both are essential for cognitive processes, though substantial temporal restraints limit the processing of sentential information in ways that don't affect depictive representations. We can process the significance of different mental imagies almost simultaneously, but must process propositional syntax by much slower linear parsing. That limitation has a direct bearing on any theory of poetic meter that rests on a strong analogy with music. The continuous repetition in the flow of lexical meanings and the syntactical structures of language create patterns that are phenomenologically distinct from those in music, patterns that have a close dependence on cultural context and personal idiosyncrasies. Words possess semantic meaning, tones don't. Any chain of words, even when scrambled to eliminate coherent meaning in so-called language poetry, produces what I call semantic entanglement as the meanings and feelings evoked by the words interact in novel ways dependent on all the psychological variables of the individual. As a consequence, there are two necessary conditions if we are to compare the verbal chain of a poetic text with the tonal stream of music: (1) the poem must be listened to, not read, and (2) the poem must be in a language that the listener does not speak.

        By a slightly circuitous route, this brings us back to what happens when we read: the text, I said above, becomes highly interactive as we extract words, phrases and images for rearrangement on our mental canvas. Listening to a poem doesn't allow for the same sort of constant reorganization that occurs with reading. Poetry, with its metaphoric language, is image saturated. Perforce it conjures up images, which we in turn rearrange on that mental canvas in ways the author could not imagine because we "draw on" our personal experiences. Mental imagery and drawing are in many respects similar activities. In his book Image and Brain, Kosslyn notes that "one of the reasons that imagery is useful is that we can combine objects in novel ways. For example, one can imagine Charlie Chaplin riding a zebra, and 'see' whether he would have been able to peer over the top of the zebra's head. A theory of imagery must explain how familiar components can be arranged in novel ways in images."<13> He also goes on to say that we can visualize novel patterns that are not based on rearranging familiar components and mentally draw patterns that we have never seen. Which is precisely what we do as we read imaginatively: rearrange familiar components, visualize novel patterns of familiar objects and draw the never seen. Interactivity is one of the greatest, perhaps the greatest, joy in reading, but it is diametrically opposed to the way we experience music, where the unvarying forward momentum of tones militates against anything remotely like such dynamic reciprocity.

        A strong theory of poetic rhythm founded in musical rhythm must address the three problems I've raised: (1) poetic meter cannot simply be identified with a rigid musical beating that cycles like an endless wallpaper pattern, (2) reading is a radically different temporal experience than listening to music and (3) reader interactivity with the full semantic and pictorial abundance of the poetic text opposes the ceaseless forward movement of musical time.


1M. L. Gasparov, A History of English Versification, tr. G. S. Smith and Marina Tarlinskaja (Oxford: Clarendon P, 1996.

2A History of European Versification 179-80.

3 The Prosody of Greek Speech (New York: Oxford UP, 1994) 121.

4 Fred Cummins, "Synergetic Organization in Speech Rhythm" (1997). To appear in Proceedings of the Joint Conference on Complex Systems in Psychology.

5 Language and Metre: Resolution, Porson's Bridge, and their Prosodic Basis, American Classical Studies 12 (Chico: Scholar's P, 1984).

6 Barry P. Scherr, Russian Poetry: Meter, Rhythm, and Rhyme (Berkeley: U of California P, 1986) 15-17.

7 In "Formulas in Russian and English Verse" in Russian Verse Theory. Proceedings of the 1987 Conference at UCLA, ed. Barry P. Scherr and Dean S. Worth (Columbus: Slavica, 1989) 419-39, she shows convincingly that in the fully stressed form of English iambic pentameter the word boundary variation F-M-M-F-M tends to accompany the grammatical pattern adjective-noun-verb-adjective-noun. In Pope's verse, by her statistics, some 44% of the rhythmical line variant w S w/ S/ w S/ w S w/S is filled with the pattern Adj-N-V-Adj-N, and in Byron nearly half the lines with most of the rest closely resembling that pattern.
Such patterns have some resemblance to the effect of Classical caesura.

8 Fred Cummins and Robert F. Port, "Rhythmic constraints on stress timing in English" (1998). Journal of Phonetics. In submission.

9 M. L. West, Greek Metre (Oxford: Clarendon P, 1982) 20f., 36-39 and Ancient Greek Music (Oxford: Clarendon P, 1992) 127-35.

10 Steven D. Carter, Traditional Japanese Poetry (Stanford: Stanford UP, 1991) 1-15.

11 Gerhard Widmer, "The Synergy of Music Theory and AI: Learning Multi-level Expressive Interpretation," AAAI (94): 114-19.

12 James J. Y. Liu, The Art of Chinese Poetry (Chicago: U of Chicago P, 1962) 26-29, Burton Watson The Columbia Book of Chinese Poetry (New York: Columbia U P, 1984) 10f. and A. C. Graham, Poems of the Late T'ang (Harmondsworth: Penguin, 1965) 25-27.

13 Stephen M. Kosslyn, Image and Brain (Cambridge: Harvard UP, 1980) 286. See also the more recent Image and the Brain (Cambridge: MIT P, 1994).


1. In Greek scansion, u = a short syllable and – = a long syllable. The terms princeps (= –), biceps (= u u), anceps (= either long or short syllable, usually represented by an "x") and | (= word end) are used to describe a meter. While some meters like the dactylic hexameter can usefully be measured by feet, some of the most common meters like the iambic trimeter of tragic and comedic dialogue consist of dipodic metra in the form x – u – x | – u | – x – u – (that is, three metra each with two u –). Word-end inside the metron, or caesura, is most common after the sixth syllable and less common after the seventh syllable in this meter. The princeps of the first two metra can be resolved into biceps, something a fully quantitative meter can accommodate.

2. Arsis ("lift" = upbeat) and thesis or basis ("placement" or "step" = downbeat)

    Dactylic: u u – u u – u u – u u – u u – – x
    (Upbeat on u u, downbeat on –.)

    Iambic metron: x – u –
    (Downbeat on x –, upbeat on u –.)

    Trochaic metron: – u – x
    (Downbeat on –, and upbeat on – x.)

    Dochmiac: u – – u – (always used in urgent or emotional contexts)
    (Upbeat on u, downbeat on –, upbeat on – and downbeat on u –.)

    Ionic: u u – –
    (Upbeat on u u and downbeat on – –.)

3. Sami Masami waka:

    yo no naka o/nani-ni tatoemu/asabiraki/koginishi fune no/ato naki ga goto.
    In the midst of life: what shall I compare it to—daybreak, emptiness left in the wake of a rowing boat.)

4. Five-syllable regulated verse (one of two forms), + = deflected tone, – = level tone,
/ = pause and R = rhyme.

        – – / – + + (or – – / + + – R, if rhymed)
        + + / + – – R
        + + / – – +
        – – / + + – R
        – – / – + +
        + + / + – – R
        + + / – – +
        – – / + + – R.