What are these symbols? Click to learn more...

Poetic Rhythm: Performance Patterns and Their Acoustic Correlates

Reuven Tsur
Tel Aviv University and Lancaster University
Voice and fax: 972-2-5611832
Email: tsurxx@post.tau.ac.il

Copyright © Reuven Tsur 1997
Received: 26 May 1997; Published: 17 June 1997

ABSTRACT: This paper assumes that when stress pattern and metre conflict in poetic rhythm, the reader may accommodate them in a rhythmical performance. Rather than in verse structure, the constraint for acceptability in versification is placed in the readerís "rhythmic competence", his ability or willingness to perform the verse line rhythmically, that is, in a way that both metric pattern and the linguistic stress pat-tern should be accessible to awareness at one and the same time. It suggests a cognitive mechanism that may render this feasible, and points out the principles of the vocal manipulations required. If the reader succeeds in such a performance, increased tension is perceived; if notóthe verse line disintegrates and tension ceases. Halle and Keyserís notion of stress maximum in a weak position is a powerful tool to describe a very high degree of deviance in versification; but it does not necessarily render a verse line unmetrical. The existence of a corpus of about 60 verse lines with a stress maximum in a weak position in major English poetry suggests that such verse lines may be acceptable on some grounds. The fact that about two thirds of the instances occur in precisely the seventh out of four positions available for "violation" suggests that these deviances are not random. The approach advocated here presumes to explain the cognitive rationale of this distribution. Two further points are demonstrated: that experienced readers tend to agree upon the kind of performance demanded by such verse lines; and that these performances are in harmony with the expectations of the theory propounded here. Moreover, when alternative mappings of stress pattern to metric pattern are possible in a verse line, some highly experienced readers may prefer a mapping that involves a stress maximum in the seventh position to some other, perfectly "metrical" mapping under the stress-maximum theory.

KEYWORDS: Versification, metricality, rhythmical performance, empirical study of metre, rhythmic competence, cognitive poetics, stress maximum in a weak position, iambic pentameter, perception-oriented theory of metre.

Poetic Rhythm: Performance Patterns and Their Acoustic Correlates

Theoretical Background

This paper assumes that the performance of poetry is not just a mediating agency between text and consumer; it is an essential dimension of any experiencing of poetic rhythm. It is conceived of here as a solution to a perceptual problem posed by the versification patterns of poems and the stress and intonation patterns of their language. Therefore, the rhythmic performance of poetry should be accorded much greater attention than it usually receives and, I should add, of a different kind. This paper draws upon a theoretical framework expounded earlier (Tsur, 1977; 1985a; 1985b; 1992: 155-179), and upon ongoing empirical research.1

Let us begin with two issues extracted from a recent "state-of-the-art" summary of the nature of performance, in the "Performance" entry of The New Princeton Encyclopedia of Poetry and Poetics (1993). The first issue concerns delivery style: "C. S. Lewis once identified two types of performers of metrical verse: 'Minstrels' (who recite in a wooden singsong voice, letting scansion override verse) and 'Actors' (who give a flamboyantly expressive recitation, ignoring meter altogether)" (ibid, 893). I shall argue that in between these two delivery styles there is a third one, which I call "rhythmical performance"; one, which is at the very core of poetic rhythm. The other issue concerns ambiguity. "Chatman isolates a central difference between the reading and scansion of poems on the one hand and their performance on the other: in the former two activities, ambiguities of interpretation can be preserved and do not have to be settled one way or the other ('disambiguated'). But in performance, all ambiguities have to be resolved before or during delivery. Since the nature of performance is linear and temporal, sentences can only be read aloud once and must be given a specific intonational pattern. Hence in performance, the performer is forced to choose between alternative intonational patterns and their associated meanings" (ibid. cf. e.g. Chatman : 1965, 1966). I shall argue that this is not so. I shall also argue that the two issues are intimately related.

Before proceeding, I wish to present the issue which, I claim, is at stake in the present discussion. "Iambic pentameter" means that there is a verse unit consisting of an unstressed and a stressed syllable, and that the verse line consists of five such units. In the first 165 verse lines of Paradise Lost, there are two such lines. Why should we speak at all, then, of "iambic pentameter"? Robert Bridges provided in his Milton's Prosody (1921) a list of "allowable deviations". These deviations were "allowable", mainly, on Milton's authority. Such a conception is outrageously unparsimonious, and counter intuitive in the sense that people don't read poetry with a list of allowable deviations in hand. Supposing that readers have internalised this list, it becomes psychologically unparsimonious. In the twenties and thirties of the present century the so-called "sound recorders" (whose work was summarised by Schramm, 1935), approached the issue empirically. This approach had the theoretical weakness that in many instances they mistook the structure of accidental performances for the structure of the poem; but much of their findings can be utilised for establishing the inventory of the reader's (or the vocal performer's) rhythmic competence. Another approach proceeded on the assumption that in the reading of poetry there are equal, or proportional time periods between stresses, or between "regions of strength"; one of the achievements of the "sound recorders" was that they failed to demonstrate the existence of such equal or proportional time periods. It was Morris Halle and Samuel Jay Keyser (1971) who proposed in their generative theory an exceptionally parsimonious criterion for distinguishing all "metrical" lines from all "unmetrical" lines, assuming that this criterion is internalised by the reader. The iambic pentameter line consists of an abstract pattern of regularly alternating weak and strong positions, upon which the sequences of linguistic stresses are "mapped". The "stress maximum" is a theoretical construct, defined as "a stressed syllable between two unstressed ones, within the same line and the same syntactic constituent". The phrase "a gárden" contains a stress maximum; "a bíg gárden" contains no stress maximum, because neither of the two stressed syllables occurs between two unstressed ones. All mappings are "allowable", except one: a stress maximum in a weak position, which renders the line unmetrical. What we are after here, then, is a systematic explanation of what is it that we perceive when we perceive a verse line as iambic pentameter, and, no less important, what are the constraints on this.2

Let us return now to C. S. Lewis' typology of performers, and Chatman's conception of intonation and ambiguity. In my approach, these two issues are intimately related in a complex manner: I have adopted from Wellek and Warren (1956, chapter 13) the assumption that poetic rhythm can be accounted for with reference to three dimensions: prose rhythm (linguistic stress pattern), metric pattern, and pattern of performance (the term "prose rhythm" is problematic; I am using it only because my sources use it; in my usage it will be short-hand for "the stress and intonation patterns of language"). In the above terms, the Minstrel subdues prose rhythm, and foregrounds the metric pattern; the Actor subdues the metric pattern in favour of the prose rhythm. For Chatman this may be a slight exaggeration, but in principle this is how things are and should be: when prose rhythm and metre conflict, "the performer is forced to choose between alternative intonational patterns". My position is that there is a third, "rhythmical performance", in which both metric pattern and linguistic stress pattern can be accommodated, such that both are established in the listener's perception. The same holds true of the conflicting intonation patterns articulating the linguistic unit (the phrase or sentence), and the metric unit (the line). This is precisely what the perceived rhythm of poetry is about, and by no means a side issue.

Metricality vs. Rhythmic Competence

It would appear that Wellek and Warren need the performance dimension in order to account for the fact that two unlike delivery instances may still be performances of the same metric structure, and to point out that some "sound-recorders" mistake in their analysis an accidental performance for the poem's metre. For my purpose, the notion of performance has more far-reaching consequences. In practice, all restrictions on metricalness have been violated by the greatest masters of musicality in poetry. Thus, for instance, Halle and Keyser and their critics in the 'sixties found 12 unmetrical lines under the stress-maximum theory. In an appendix to my 1977 book I have provided a list of over forty further instances in major English poems (and there appear to be many more). What is more, the distribution of the violating stress maxima seems far from being random. This, however, does not imply that "anything goes". It is, rather, that the constraints are placed not in the verse structure, but in the reader's resources of performing the line rhythmically. In cases of deviation, "prose rhythm" and metre may be conceived as analogous to the two incompatible terms of a metaphor. The reader registers their incompatibility and resolves them in a pattern of performance. The utmost limit of rhythmicality (as of the meaningfulness of a metaphor) is the reader's ability or willingness to cooperate, that is, to resolve the incompatibility of the two terms by a rhythmical performance (or, in the case of a metaphor, by a semantic interpretation). Thus, the present approach attempts to handle semantic and metric phenomena by a homogeneous set of principles. Consequently, the intonation contour or stress pattern of a performance does not reflect, as Chatman would have it, one or the other of the two conflicting patterns, but, rather, the solution of a perceptual problem posed by them. A rhythmical performance can be defined as the vocal conditions in which both prose rhythm and the metric pattern are established in the listener's perception.

According to Halle and Keyser, then, a stress maximum in a weak (odd-numbered) metrical position renders a verse line "unmetrical", as in the following line:

Quote No. 1


Since no criteria for metricality has yet been devised that were not violated by such masters of musicality as Milton and Shelley, I suggested (Tsur 1977) that the constraints should be located not in the verse structure, but in the reader's ability or willingness to perform it rhythmically. Nearly sixty instances of stress maxima in weak positions have been recorded in iambic pentameter lines, in major English poetry. Two thirds of them occurred in the seventh position (out of four positions available for violation). This suggests that the distribution is not random, and that poets like Milton, Shelley and Keats assumed that their reader's can perform such lines rhythmically. I predicted that experienced readers will tend to agree that a verse line with a stress maximum in the seventh position can be performed rhythmically. The solution to the perceptual problem will arise not with reference to the isolated stress maximum, but within a larger group. Such groupings are the performance patterns available to the reciter. If a reader is asked to read the line rhythmically, so as to preserve the stress pattern of the words and as much of the metre as possible, he is likely to group together, emphatically, the last four syllables and segregate the group from the preceding context. In other words, he will foreground in his performance a unified perceptual group called "stress valley". A stress valley is a cluster of four syllables; in terms of Gestalt theory it has a closed symmetrical (that is, "good") shape: two stressed syllables embrace two unstressed ones, as in "Píty the wórld", or "bóttomless pít". Such a perceptual organization may save mental processing space. A stress valley beginning in the seventh position ends in the tenth position, imposing upon the line an exceptionally strong closure. This paper will explore issues related to the stress valley, in an empirical study of recorded readings. In order to explain the rhythmic significance of the stress valley, I am going to present a short outline of the cognitive mechanism underlying poetic rhythm.

The present approach assumes that poetic rhythm is, essentially, an auditory phenomenon, though also affected by syntax and semantics. Its auditory qualities may be accounted for if we assume that it is processed in short-term memory, which functions in the acoustic mode, and is constrained by its limitations. The contents span of short-term memory is limited to seven monosyllabic words plus or minus two (Miller, 1970; that is why the longest verse line that can be perceived as a rhythmic unit without an obligatory break is ten-syllables-long). Its time span is roughly the period we can remember e.g. a telephone number without rehearsal. During this period short-term memory functions like an echo box. In order to render a verse line perceptible as a rhythmic whole, the reciter must manipulate his vocal resources in such a way that the verse line can be completed before its beginning fades out in short-term memory. When the immediately observable string of syllables deviates from metric regularity, the metric pattern may be perceived as reverberating in the background, provided that sufficient mental processing space is available. Training cannot expand these spans of short-term memory. The only thing one can do is to recode the verbal material in such a way that it occupies less mental processing space. Thus, for instance, "a man who sells goods" can be recoded as "merchant", and "a merchant who sells meat" can be recoded as "butcher". Such recoding is impossible in poetry, where the actual words may not be changed. Still, some mental processing space may be saved by two kinds of vocal manipulations: grouping and clear-cut articulation. Gestalt theory has laid down fairly rigorous rules of what facilitates perception: these include grouping and parsing (which is one kind of articulation). Accordingly, we may expect reciters to over-articulate, on the one hand, word and syllable boundaries (parsing) and, on the other hand, to group syllables and words in certain ways. Speech research of the past thirty years (cf. Lieberman, 1967) has established that in the flow of everyday speech we tend to rather careless articulation, and the listener has to do in the course of decoding a lot of subliminal guesswork. In conversational speech words are normally run one into the other in English (and more so in French), and it takes special decoding effort to determine the word endings. Taylor (1990: 212) suggests: "Say rapidly 'How to wreck a nice beach', and it will sound like 'How to recognize speech' [...]. The sentence illustrates the point that word boundaries are anything but fixed. Without boundaries words are hard to recognize". Thus, much decoding effort can be saved by clear articulation of word endings. Clear-cut articulation of phonemes and of syllable and word boundaries may save a lot of mental processing space. Intonation is a typical means of over-articulating syllable and word boundaries; but the over-articulation of the syllable (or word) final consonants too contributes to the over-articulation of boundaries.

I argued in my book that a distinction must be made between performance patterns and their acoustic and phonetic correlates. The relationship between, e.g., a stress valley and its acoustic correlates is similar to the relationship between a phoneme and its acoustic correlates. We are interested in the phoneme as an abstract category, and ignore the specific acoustic cues that are its exponents. Consequently, there is usually a trade-off between the possible acoustic correlates that may cue a certain phoneme. Thus, for instance, a voiced stop may be cued by the straightforward activation of the vocal folds, or by a lengthening of the preceding sonorant, or by reducing voice-onset-time, or by aspiration. Most language users would not distinguish between the various vocal devices; they merely perceive a unified abstract category, such as [b, d, g].

The same is true of the perceptual organisations required by the rhythmical performance of a deviant verse line. Consider the case of a stress valley, produced to accommodate a stress maximum in a weak position. Since a stress maximum occurs, by definition, in mid-phrase, or even in mid-word, the performer will face the following conflicting tasks: he must segregate the stress valley from the preceding context, but must preserve the continuity of the phrase, or even more so, of the word. The listener, or even the reciter himself, will be aware at best that the perceptual problem has been solved. On closer inspection, they might discern that a stress valley has been applied. On still closer inspection, they might even discern the opposite tendencies of continuity and discontinuity between the stress valley and the rest of the phrase or word. But it is impossible for them to discern by what phonetic means this has been accomplished. And, in fact, what matters for the solution is the abstract category "stress valley", and it is immaterial what trade-off between the various acoustic cues may take place.

The acoustic cue for continuity is usually quite straightforward: there is no measurable pause before the stress valley. Quite frequently also "an internally defined intonation pattern" (in Gerry Knowles's 1991 term) is assigned to the sequence of four syllables. It is more difficult to discover the cues for discontinuity. I contend that there is an open list of possible acoustic cues. Reciters display an astonishing degree of creativity; new performances provide acoustic cues some of which are quite expected, but some are entirely unforeseen. But as long as they generate a unified perceptual category and indicate the required segregation, listeners exposed to them for the first time immediately recognise them as appropriate (assuming that continuity is taken care of).

One more issue must be faced. Can we credit the non-phonetician listener with utilising such evasive phonetic cues as phoneme prolongation, glottal stop, or the absence of coarticulation in the processing of the rhythmical performance of poetry? There is compelling evidence that listeners are heavily relying on their linguistic knowledge in perceiving either linguistic stress, or continuities and discontinuities in the stream of speech ignoring, sometimes even overriding, phonetic cues. This may cast some doubt on the efficiency of "allophonic" manipulations of phonetic cues in the rhythmical performance of poetry. Using French and Hungarian judges, phoneticians and non-phoneticians, listening to a Hungarian prose passage, Kassai and Fagyal (1996) found evidence to this effect; but also to the effect that tampering with the stream of speech (by the omission of silent periods) may increase reliance on the more evasive phonetic cues, as pitch discontinuity or such segmental discontinuities as phoneme prolongation, glottal stop, or the absence of coarticulation. This may suggest that over-articulation and the reduction of speech rate in poetry recital serve, among other things, to force these phonetic cues to the listener's attention.

Test Case: Stress Maximum in a Weak Position

Consider the following verse line:

Quote No. 2

This line would be "unmetrical" under at least two different criteria, propounded by the various kinds of generative metrics. According to Halle and Keyser, a stress maximum in a weak position renders a verse line unmetrical, even if no polysyllabic is involved; according to Kiparsky, the stressed syllable of a polysyllabic in a weak position renders the line unmetrical, even if it is no stress maximum (cf. Barsch, 1987). Now cru- in the seventh position is both. Nonetheless, experienced performers seem to be in agreement as for the performance pattern that may render such a line rhythmical: a stress valley comprising the last four syllables. The two performers whose performances are to be discussed are PhD candidates at the department of Linguistics at Lancaster University.

Intuitively, in many performances it is an exceptionally high pitch that sets the syllable cru- apart from and, pushed forward, as it were, by the perceived leap to the high pitch. Indeed, in TB's reading the intonation contour on and rises from 120.492 to 121.823 Hz, then it resets to 156.383 Hz; then it falls to 117.287 Hz on -ci-.

In another delivery instance (by JH), exactly the same sort of stress valley is discerned ("crucifie mee"); again, it is segregated from the preceding conjunction by a conspicuous instance of pitch discontinuity (the jump to /u/, from which the intonatation pattern descends gradually). The contour of and falls from 87.849 to 79.603 Hz; on -u- pitch resets to 117.287 Hz; then, after some curling returns to the same peak (117.129 Hz), persisting until slightly after the middle of the vowel (this is what Knowles [1992] calls late peak, and according to Gestalt theory may be expected to exert a "forward-pushing perceptual force"). This seems to be just enough to suggest pitch discontinuity and some forward grouping of cru-.

Figure_1.GIF - 10.31 K

Figure 1 The wave plot and pitch contour of
"-corge and crucifie mee" in TB's reading3

Spectrogram No. 2

Figure 2 JHís first recording of "and crucifie mee"

JH recorded the verse lines twice: once in context, and once in isolation. In the latter recording, with reference to cru-, the peak is quite early in the vowel; but he creates a marked pitch discontinuity, by a big upward leap between the two words (the pitch of and swerves between 90.000-84.160 Hz, ending at 88.200 Hz, from which it leaps to 123.184 Hz, where a downward movement begins, falling to 92.259 Hz on ci).

In figures 1 and 3 one may notice another important point. Not only is and discontinuous with cru-, but it is also conspicuously continuous with scorge: the release of the affricate is run into the vowel of and; and the intonation of and is in the pitch region to which the long intonation contour of scorge falls (still, they are perceptually segregated by the long terminal contour articulating the boundary of scorge).

Spectrogram No. 3

Figure 3 JH's second recording of "and crucifie mee"

The reciter has another problem here: the last syllable of this line bears no content-word stress. The last position requires confirmation by stress; the stress valley demands that this stress be prominent indeed. Both TB and JH made a remarkable effort to face this challenge by a compromise between stressing and not stressing, but in different ways. In a reading in which mee is unstressed, one might expect the intonation curve fall on or immediately after cru- of crucifie, and keep the rest of the pitch sequence level at the base line. In a reading in which mee is stressed, one might expect the intonation curve rise on mee. In TB's reading neither of the two happens: the tone falls throughout crucifie, to 102.558 Hz; then, on mee it slightly moves up from 98.879 to 105.502 Hz, and then falls down to 69.778 Hz, and is perceived as an "allophonic" stress. In JH's first reading, pitch falls moderately on -fie mee and then, contrary to expectation, it rises again a few Hzs. Such a small rise usually goes unnoticed; here it is clearly perceptible, probably because a fall is expected. In his second reading there is a downward jump of intonation from cru- to -cifie. Normal expectation would be for a downward continuation on mee. Contrary to expectation, pitch falls on -fie from 89.271 to 79.603 Hz, and then rises again to 86.133 Hz; and then rises on mee from 81.667 to 89.271 Hz from where it has a long fall to 68.056 Hz.

In the first reading, this is reinforced by an uncommon manipulation of articulation (consistent with a similar manipulation in the next example). When isolated, the first vowel of the diphthong in crucifie is perceived as reduced; the vowel of mee, by contrast, is turned into a diphthong, with a released /j/ at the end. In the flow of reading this is not perceived. Together with the minute rise of pitch, it is merely sufficient to indicate that mee is more prominent than -fie, without being perceived as a stressed syllable. In the second reading, only the syllable of mee is turned into a diphthong.

Continuity and Discontinuity in Midword

Now consider the stress maximum in the seventh position of the second line of the following quote:

Quote No. 3

All the difficulties observed in the preceding example are present here, ruled as unmetrical both by Halle and Keyser, and Kiparsky: there is a polysyllabic here the most prominent syllable of which constitutes a stress maximum in a weak position. But these difficulties are heightened here to a considerable degree: the stress maximum occurs not in mid-phrase, but in mid-word. And if the preferred solution is to apply a stress valley that demands, among other things, segregation from the preceding portion of the line, the reciter is in real trouble: he must segregate the stress valley, but also see to it that the word is not disrupted.

Mere listening to JH's performance of this line reveals both opposite tendencies. It is clearly felt that the word is continuous, with no break after temt-; indeed, there is no measurable pause there. But there is also a very strong feeling that the last four syllables are segregated and performed as a cohesive unit. Intuitively, there is a sustained arrest on temt-, and then a sudden upward leap on -ation. First, a marginal observation: there is no acoustic trace of /p/, but this should be considered perfectly normal. The pitch contours in figure 4 may suggest some explanation for both the sustained arrest and the upward leap: the pitch contours of both his and tempt- are roughly in the same frequency region— hence the feeling of arrest; and pitch leaps from 77.098 Hz at the end of tempt- to 91.494 Hz at the onset of -a- which then falls and rises to 78.750 and 83.840Hz. Now a 14-Hz leap is not very big; but its significance also depends on the surrounding pitch ranges. In this hemistich, for instance, the longest contour is the last one, indicating the termination of the syntactic and the prosodic unit, and it falls from 89.634 to 70.447 Hz, that is, a span of 19 Hz all in all. The longest span in this line belongs to the contour of words, falling from 101.613 to 77.098 Hz (that is, 24 Hz); but here it has the function of compensating for the stressed these in the preceeding weak position. But the impression of "sustained arrest" on temt- has more to it. Intuitively, again, this syllable is unnaturally lengthened. Such lengthening is usually regarded as a powerful means of segmentation: it is usually felt that the following segment constitutes a "new start", even though there is no measurable break there. The trouble with such claims concerning lengthening is that it is almost impossible to substantiate them. They are only intuitive judgments. Although the length of the syllable can be measured, but how can we know whether it is longer or shorter than it ought to be? There is no standard to which it can be compared. In the present instance, however, an obvious measure for comparison is offered in the preceding verse line, the first syllable of Tempter. it is clear that the same sequence of sounds must be considerably longer in the stressed syllable than in the unstressed one. Here, however, the reverse is the case: temt- in temptation is marginally longer (0.5376 sec), whereas in tempter it is 0.5320 sec long. This is odd according to any standard. There is another oddity here. In both words, one would expect a short but unreduced vowel. In temptation, however, when tem- is isolated on the computer, its vowel is heard, clearly, as reduced. There is no linguistic justification for this, and is very odd indeed. In the present context it may be interpreted as part of an ingenious combination of vocal devices for generating discontinuity without a pause. Lengthening, as we have said, is an effective means to arouse a strong sense of discontinuity (required here by the need to segregate the stress valley).4 Duration, however, is also an effective acoustic cue for stress. By reducing the vowel, the reciter indicated that this cannot be a stressed syllable. And, indeed, it is perceived as unstressed. The reciter seems to have invented here a unique combination of phonetic devices, to which he may never have been exposed before; at least, no other use of it has ever been recorded. He had not the slightest idea to what vocal resources he had recourse; he only knew that he was trying to solve the perceptual problem posed by a stress maximum in a weak position, and, perhaps, that this demanded some conspicuous discontinuity after tem-, without disrupting the word. The reciter himself was most surprised when next day he was shown what he had done: he said he had thought he was doing something different, that he was trying to make tempt- comparable in stress to the preceding his.

Spectrogram No. 4

Figure 4 JH's recording in context of
"and with these words his temptation pursued"

Now this reciter recorded the verse lines with the stress maxima in weak position twice: once in context, and once in isolation. In isolation, of course, we could not compare the length of tempt- to its counterpart in tempter. The isolated syllable is shorter here then in the other performance. In other respects, however, this reading was consistent with his earlier reading: the vowel in tem- was reduced so much that it was not even certain that there was a vowel there at all. We did not expect a reciter to be as consistent as that. As for the pitch contour, two significant things should be pointed out in this reading. There is a relatively long terminal contour after the first syllable, falling from 106.010 to 79.891 Hz, and there is the reset of pitch from there to 108.088 Hz at the onset of -ta-, then falling again to 93.038 Hz. Thus, again, there is continuity and discontinuity at one and the same time in the middle of this word: there is no measurable pause between temp- and -ta-; but the terminal contour after the first syllable followed by a leap of pitch, and reinforced by segmental discontinuity, perceptually segregate the last four syllables from the rest of the line.

Spectrogram No. 5

Figure 5 JH's recording of the same verse line in isolation

For the same reason the verse line in quote 4 was chosen. The seventh position in it is occupied by a stress maximum in the middle of a polysyllabic.

Quote No. 4

For some reason, reciters had a bigger problem with this line than with quote 3. They tended to stress rétributive, that is, to succumb to metre. So, some of the readings of this verse line are irrelevant to the present inquiry. But one of JH's readings stressed retríbutive. Intuitively, his solution here was very similar to that in (3). He appears to linger on ret-, more specifically on the /t/, as if to gather strength, and then by a sudden release, to leap to a relatively high -rib-. As for the sustained arrest on ret-, one can only rely on the intuitive judgment that it is lengthened beyond what would be natural. The /t/ is co-articulated with the subsequent /r/; but is also released. This is quite normal; still, it contributes both to continuity and segregation. The syllable re- is conspicuously continuous with the preceding the, and conspicuously discontinuous with -trib-. Its pitch rises from 90.000 to 98.438 Hz; then pitch resets to 122.500 Hz, generating discontinuity between the two syllables in the middle of the word. This discontinuity is reinforced by the artificial lengthening of /t/. Stops are abrupt, and obviously cannot be lengthened; but the release of the /t/ is preceded by an 81 msec pause, which is construed by the listener as an articulatory gesture, during which the tip of the tongue is pressed against the ridge, ending with the release. The release, in turn, is continuous with the /r/. Thus, the word retributive is uninterrupted but, at the same time, the segment -ributive hour is perceptually segregated from the rest of the line.

A closer scrutiny of the spectrogram reveals something that is quite unexpected: a double pulse, a "column" superimposed on the formants between the schwa and the /t/. This is clearly not an artefact of the machine, but produced by the speaker. It appears to be a glottal constriction, of the sort that recently has been observed in some London dialects to replace /t/. In the flow of speech it is not recognisable; it is merely perceived as a lingering on the /t/, without disrupting the continuity of the word. Such a lingering is usually perceived as a terminal feature, heralding a new beginning. Thus, again, the new beginning of a stress valley is foregrounded here, without disrupting the continuity of the word retributive. The lack of a pause reinforced by the co-articulation of /t/ and /r/ takes care of continuity; and the unexplained lingering on the /t/ and the pitch step-up takes care of the segregation of the beginning of the stress valley.

Performance Patterns and Phonetic Competence

Now consider this. In our visual imagination, a stress valley takes the following shape:Img00007.gif - 0.94 K . It consists of two stressed syllables enclosing two unstressed ones. The stress and intonation contours of speech never look like that. They consist of rising and falling curves, sometimes slopes. When subglottal air pressure builds up, the pitch sets up, or intonation begins on an initial high pitch. Subsequent pitch curves reflect a more or less gradual loss of air pressure. These are what Gerry Knowles calls "internally defined" intonation contours. In the present instance, our default expectations with reference to the phrase "retributive hour" would be for a gradual fall from -tri-, and then a rapid fall on hour. In such a reading, the reset from ret- and then the change of direction on -tri- presents this syllable as strongly stressed. The rapid fall on hour presents it as stressed. In JH's actual reading, by contrast, there is a rapid fall from -tri-;5 then a step up at the beginning of hour, and a gradual fall on it. It is this deviation from the default pattern that renders the first and last stressed syllables exceptionally prominent, causing the group to stand out as a perceptual unit, and serving as an effective closure to the stress valley and the verse line. Consider again the following verse line from Milton:

Spectrogram Nol 6

Figure 6 "Enduring thus, the retributive hour"
The markers indicate an 81 msec pause before the release of /t/

This too is unmetrical under both Halle and Keyser's and Kiparsky's generative theories; and, again, it is predicted that the rhythmical performance will consist in the foregrounding of the last four syllables "bottomless pit" as a stress valley, isolating it, in mid-phrase, from the preceding chunk of the line. Indeed, that is exactly what JH does in his performance. While there is no detectable pause between to the and bottomless, the isolation of the stress valley is indicated by a moderate pitch discontinuity, which is reinforced by a fairly late peak. The latter effects a moderate forward grouping of the syllable. The pitch on to the falls from 93.830 to 88.554 Hz, and then steps up to 101.613-105.502-101.147-102.558 Hz on bot. From here, an internally defined intonation contour descends, quite moderately, wavering around 96.288-99.324-90.741 Hz on tom, and around 94.231-89.271 on less. A terminal contour is assigned to pit, resetting to 107.561 and falling to 91.875. Thus, the stress valley even approaches to some degree the visual imageImg00007.gif - 0.94 K mentioned earlier.

Spectrogram No. 7

Figure 7 The waveform and pitch contour of "to the bottomless pit"
in JH's reading; the markers indicate the boundaries of /o/.
The peak occurs quite late.

Two important theoretical questions arise with reference to JH's rhythmic solutions. How does the listener know, what the "meaning" of a certain vocal manipulation is, and how does the reciter know what manipulations of his speech organs would achieve a specific effect. I propose to give here some tentative answers, derived from analogies with some better understood phonetic and phonological processes.

As for the first question, the distinction between the default and the actual intonation patterns assigned to the phrase "retributive hour" does have precedent in spoken English: Gerry Knowles suggests that the default pattern would be used by RP speakers, the other by Northern speakers (JH is clearly an RP speaker, but of Lancashire origin). So, the distinction is usually regional. However, Knowles played to his students, in a class-room situation, natural speech stretches, containing the afore-mentioned contrasting intonation contours. Native speakers of English identified the former intonation with a power position, the latter with a submissive position. So, how does the listener know that in the present instance the non-default intonation pattern indicates a stress valley rather than some regional distinction or some power relationship? Here an analogy with suprasegmental phonemes may prove illuminating. As pointed out by Lieberman (1967) and others, suprasegmental phonemes may be used to disambiguate utterances such as (They decorated the girl) (with the flowers) and (They decorated) (the girl with the flowers); or (light) (house keeper) and (light house) (keeper). Lieberman brings good experimental evidence to the effect that suprasegmental phonemes have no meanings of their own. At most, they may indicate which of two possible Phrase Markers is meant, if the listener is aware of two alternative Phrase Markers. In the present instance, no Phrase Markers are involved. When one is aware of regional dialects in the context, the two intonation patterns indicate the respective pronunciations; when power relations are involved, the intonation patterns may indicate a distinction between power attitudes; when the rhythmical performance of poetry is concerned, the respective intonation contours may indicate a distinction between phonemic and non-phonemic aspects of suprasegmental patterns.

This brings us to our answer to the second question: how does the reciter know what manipulations of his speech organs would achieve a specific effect? That seems to be, precisely, what our phonetic competence is about. When people produce speech sounds, suprasegmentals and paralinguistic features, they are unaware of the combination of their muscle movements; in Polányi's (1967) term, they "attend away" from them, to their joint purpose, the phoneme /b, d, k/ or the suprasegmental phoneme (stress or intonation), or the emotive sound gestures. That is why we have only a "tacit knowledge" of our speech production. When confronted with the problem of a metrically deviant verse line, speakers mobilise the articulatory devices which they have acquired for phonological or paralinguistic purposes, and exploit them for the production of performance categories, such as stress valley.6 Fónagy (1971) gives some illuminating examples of the paralinguistic distortion of articulation for emotive purposes. Thus, for instance, a French mother may say "mai su" to her baby instead of "mais si", rounding her lips, indicating love. Such knowledge is unlearned and untaught (in Morris Halle's phrase), and human beings demonstrate considerable creativity in the invention of such vocal devices. I suggest, tentatively, that similar inventiveness may be active in the reciters' attempt to solve rhythmic problems.

Alternative Mappings and Stress Maxima in Weak Positions

One may summarise my foregoing argument as follows. The notion of stress maximum in a weak position is a powerful tool to describe a very high degree of deviance in versification; but it does not necessarily render a verse line unmetrical. Rather than in verse structure, the constraint for acceptability is placed in the reader's "rhythmic competence", his ability or willingness to perform the verse line rhythmically, that is, in a way that both metric pattern and the linguistic stress pattern should be accessible to awareness at one and the same time. I have suggested a cognitive mechanism that may render this feasible, and have pointed out the principles of the vocal manipulations required. If the reader succeeds in such a performance, increased tension is perceived; if not—the verse line disintegrates and tension ceases. The very existence of a corpus of about 60 verse lines with a stress maximum in a weak position in major English poetry suggests that such verse lines may be acceptable on some grounds. The fact that about two thirds of the instances occur in precisely the seventh out of four positions available for "violation" suggests that these deviances are not random. The approach advocated here claims to have explained the cognitive rationale of this distribution. I claim to have demonstrated two further points: that experienced readers tend to agree upon the kind of performance demanded by such verse lines; and that these performances are in harmony with the expectations of the theory propounded here. What is more, as we shall see, when alternative mappings of stress pattern to metric pattern are possible in a verse line, some highly experienced readers may prefer a mapping that involves a stress maximum in the seventh position to some other, perfectly "metrical" mapping under the Halle-Keyser theory. We shall now turn to such instances.

One important stage in metrical analysis is the establishing of the correspondence of syllables to metrical positions. This may grant the metrist or the reciter some flexibility in allowing or avoiding a stress maximum in a weak position in certain verse lines. Thus, for instance, there are verse lines in which there are more syllables than metrical positions available. In such verse lines two syllables may be assigned to one metrical position. Halle and Keyser (1966) have formulated the conditions under which this can be done, among them: when there are two consecutive vowels with no intervening consonant; or are separated by a sonorant or by a voiced fricative; sometimes two (unstressed) function words can be allocated to one position. I have elsewhere (Tsur, 1977; Tsur and Adam, unpublished) pointed out the performance rationale of the first three conditions: it is relatively easy to under-articulate in these conditions the boundary between the two syllables. The present theory predicts that the boundary between the two syllables occupying one position will be under-articulated; the boundary following the second syllable, coinciding with the position boundary, will be over-articulated. The over-articulation of the position boundary causes the reader or the listener to restructure the preceding two syllables, dimming in immediate memory the intervening boundary—provided that the appropriate phonetic and articulatory conditions are present. Shortening of the duration of such syllables may reinforce the impression of bisyllabic occupancy of one position.

When there are several instances of such conditions in a verse line, alternative mappings of syllables to metrical positions are possible. When one of the alternatives results in a stress maximum in a weak position, it might be quite reasonable to expect with Halle and Keyser that the other alternative(s) be preferred. But there are some illuminating precedents in which unquestionably competent readers prefer a mapping which results in a stress maximum in the seventh position, and apply to the line the performance strategies discussed above, with compelling results. Consider quote 5 from Keats's Elgin marble sonnet.

Quote No. 5

Under Halle and Keyser's conditions there are here three obvious candidates for bisyllabic occupancy of metrical position: "with a", "billo-", and "-lowy". If the syllables "with a" are assigned to one position, "bil-" becomes a stress maximum in the seventh (weak) position. The first syllable of "billo-" is stressed, and therefore under-articulation appears to be less plausible; so the most natural solution seems to be assigning the syllables "-lowy" to one position (as indicated in quote 5). In Douglas Hodge's performance of this sonnet, the unstressed syllables "-lowy" are considerably longer than the preceding stressed syllable "bil-" (145 and 164 msec respectively, as against 86 msec). When a word like "power" is squeezed down to one position, frequently there is little or no acoustic trace of the /w/. Here the /w/ is rather clearly articulated. On the other hand, there is no trace of over-articulation at the end of this word: the vowel is run into the following /m/, and the whole sequence "billowy main" forms what Knowles calls "an internally defined prosodic pattern". By contrast, the boundary after "with a" is conspicuously over-articulated by an intonation contour and by a 48 msec pause; whereas the /ð / appears to be more closely related to the ensuing "a" than to the preceding "wi-" (see the wave plot in figure 8), possibly indicating under-articulation between the boundaries of the two words. This analysis supports our intuition that in this performance, the syllables "with a" are more likely to be squeezed down in perception to one position than "-lowy", even though they are not of significantly shorter duration (133 and 150 msec). Such a parsing manipulates "bil-" into a weak position, rendering the verse line unmetrical, both under Halle and Keyser's and Kiparsky's theory of generative metrics (under the former, because it manipulates a stress maximum into a weak position; under the latter, because it manipulates the stressed syllable of a polysyllabic into a weak position). The present theory predicts that when a stress maximum occurs in the seventh position, it may be rendered, nevertheless, rhythmical, by certain vocal manipulations: if the last four syllables are emphatically grouped together, and are perceptually segregated (in mid-phrase or in mid-word) from the preceding unstressed syllable. One might also expect an over-stressing rather than playing down of the deviant stress, with a late peak on it's vowel or on the following sonorant, suggesting forward grouping. In the present instance this syllable is stressed, but not over-stressed; and is segregated from the preceding article by a 48 msec pause and by a conspicuous pitch discontinuity (resetting from 93.830 Hz at the end of a falling contour to 130.473 Hz at the onset of the next contour, peaking late, after the vowel, on the /l/; see figure 8). This pause is totally unwarranted from the linguistic point of view, and its only justification appears to be to satisfy the demands of two rhythmical solutions discussed above: the over-articulation of the position boundary where two syllables are squeezed into one position; and the segregation of the "stress valley" beginning with a stress maximum in a weak position.

Spectrogram No. 8

Figure 8 Wave plot and pitch contour of "with a billowy main"

There is another curious instance in Hodge's reading of this sonnet and, again, precisely in the seventh position. In line 2 stress is displaced from the second to the first syllable of "unwilling". Semantically, this may be construed as an emphatic stress. Metrically, however, this means that the only regular part of the line is made irregular. A stress on "wil-" confirms metre in a strong position; a stress on "un-" infringes upon it in a weak position.

Figure_9.new.GIF - 6.90 K

Figure 9 Wave plot and pitch contour of "me like unwilling sleep"

What is more, it becomes an artificial "stress maximum in a weak position". A stress maximum in the seventh position constitutes an infringement upon metre, and there arises an urge to achieve again focal stability in the next stressed syllable in a strong position. This happens in the tenth position, achieving stability (and powerful closure) in precisely the last position of the line. The regular alternation of weak and strong positions is suspended during these four syllables; but by a series of vocal manipulations mental processing space can be saved such that the metrical set becomes available to consciousness. We have accumulated an ever-increasing number of instances in which this performance pattern is followed. In fact, this is precisely what Hodge too is doing here. In the present instance, "un-" is over-articulated by its excessive duration (relative to the other syllables), and by the elaborate intonation contour assigned to it, which includes the highest pitch peak in the phrase. Syntactically, the preposition "like" ought to be grouped with "unwilling sleep"; here, however, it is emphatically grouped with the preceding phrase: there is no measurable pause between them, and they have one continuous intonation contour. On the other hand, it is separated from its sequel, quite unusually, by a rather long, 94 msec pause; and the release of the /k/ preceded by a pause is construed as an over-articulation of the phoneme as well as of the word boundary, suggesting what Knowles has called "segmental discontinuity" before the stress valley. There is a jump from the falling intonation contour of "like" to the rising beginning of the high onset of the next intonation contour, indicating a new start, and forward grouping. Furthermore, this is one of the two places where we have encountered so far a "late peak" in Hodge's readings (that is, where the intonation peak occurs after the middle of the vowel), and a very late peak at that; this is usually perceived as a strong forward drive. It should be noted that in most instances of our corpus there is no measurable pause before a stress maximum in the seventh position, because it occurs by definition in the middle of a phrase or a word, and separation is indicated merely by pitch discontinuity, late peaking, and sometimes segmental discontinuity. In this instance, a syntactically unwarranted pause further enhances discontinuity. Hodge did not necessarily "know" what he was doing here; but, obviously, he seems to have had a very strong intuition as for how such a verse line can be rendered rhythmical.

Such instances as quote 5, where a stress maximum in the seventh (weak) position can be avoided or generated according to which one of two alternative pairs of syllables is assigned to one metrical position, are not without precedence in our corpus. Consider the first line of the following quote from Hamlet:

Quote No. 7

In the first line there are twelve syllables, in the second one eleven. The pairs of syllables marked by a horizontal bracket in quote 7 satisfy the phonetic conditions laid down by Halle and Keyser for bisyllabic occupancy, and the line may be ruled as "metrical". Nonetheless, in a paper on bisyllabic occupancy of metrical positions (Tsur and Adam, unpublished) we have found that at least one perfectly acceptable performance of the first line can be better accounted for if the first token of "devil" is treated as containing an extra-metric syllable at the caesura (alternatively, if the function words "and the" are assigned to one position), and the first syllable of the second token of "devil" is treated as a stress maximum in the seventh (weak) position. Here too, the reciter preferred to "violate" metre, and then to render the performance rhythmical by resorting to strategies predicted by the present study.


As I suggested (Tsur 1977: 3), the patterns of performance include some phonological categories, and are analogous to them in an important respect. In phonology, a great variety of phonetic features may be categorised as the same phoneme. Likewise, a great variety of phonetic and/or paralinguistic features (which constitute open sets) may be categorised as a limited set of patterns of performance. It is not yet clear whether they constitute a closed set, like grammatical categories. What is significant for metrical analysis is the category of performance rather than the features categorised. There is one important difference, however. In metrical analysis there may be good reasons for going outside the limited set of performance categories; the same noises (e.g. pitch-obtrusion) may be categorised as two distinct categories, stress and intonation, at one and the same time; this gives rise to important perceptual facts of metrical significance.7

According to the present conception there is no fixed limit of metricality: the utmost limit of acceptability is the readers ability or willingness to perform the verse line rhythmically. Jay Keyser suggested to me, back in the early seventies, that such a conception requires a systematic theory of the rhythmical performance of poetry. I attempted to propound, speculatively, such a theory (Tsur, 1977). The present paper has provided some empirical evidence for this conception.

The stress maximum in a weak position is an interesting test case. According to the Halle-Keyser theory, it renders a verse line unmetrical. As I have demonstrated above, various reciters tend to have recourse to the same performance patterns, when confronted with this kind of violations of metre. It is noteworthy that with the help of these performance patterns they feel they can perform rhythmically verse instances that are ruled unmetrical by various branches of generative metrics. Usually they do not know what they are doing and why, and are quite surprised to discover that, for instance, in stead of playing down a deviant stress, they over-emphasise it. These patterns have been predicted by the perception-oriented theory of metre, based on the hypothesis of limited channel capacity of human information processing, and on the hypotheses of gestalt theory. Such a conception grants the performer a great degree of creativity in his attempts to solve the perceptual problems arising from the conflicting patterns of stress and metre, while the dynamics of human perception provide the constraints that govern the solution. That is why experienced readers of poetry tend to agree on the performance patterns applicable to verse instances ruled unmetrical by some experts. And that is why listeners exposed to those solutions for the first time tend to recognise them as appropriate solutions to the perceptual problem concerned.

The conception of "allowable deviations" is clearly untenable. But I am also quite skeptical regarding the "rule-internalisation" conception of metricality. A weight-lifter need not internalise rules for "liftability"; if some weight exceeds his strength, he simply cannot lift it. By training he may move the utmost limit of "liftability" further away. The same seems to hold true of the utmost limit of "metricality". What readers may internalise are devices for the accommodation of the conflicting patterns of stress and metre (training). But even in this respect, I would grant the reader considerable creativity: for unforeseen instances he may invent new devices. The constraints on such inventiveness are the linguistic and metrical structure of the verse instance on the one hand, and such psychological constraints as the limitations of short-term memory and the Gestalt rules of perception on the other. The store of such internalised devices along with his ability to invent new devices constitute the reader's "rhythmic competence". The various devices can be arranged in scales of markedness, according to the relative difficulty of their application. The higher the device on the scale, the greater is the tension generated by the verse line that requires its application. On this view, the readers of Pope and Milton have not internalised different rules of metricality, but rather draw the utmost limit of metricality at different points on the same scales of markedness.


1.I am most deeply indebted to Gerry Knowles. My encounter with him ended twenty five years of agonising search for a method to obtain instrumental support for my perception-oriented theory of metre. This paper also contains some of his insightful comments on spectrograms and intonation contours of individual lines.

2. This paragraph does not presume to offer an exhaustive survey of twentieth century metrical theory; it is merely intended to introduce the problem to be discussed and a few attempts to handle it, in a roughly chronological order. Of the various types of generative metrics, for instance, only the Halle-Keyser theory is mentioned. Elsewhere (Tsur, 1977) I criticised at great length the Halle-Keyser theory; but most of this criticism is irrelevant here.

3. The lower window presents the wave plot display which shows a plot of the wave amplitude (in volts) as a function of time (in milliseconds); the upper window presents a fundamental frequency plot, which displays time on the horizontal axis and the estimated glottal frequency (f0) in Hz on the vertical axis.

4. Alternatively, one may regard this lengthened portion as creating a filled pause between the two syllables, generating continuity and discontinuity at one and the same time in mid-word.

5.The upper part of the pitch contour assigned to /ju/ represents the inherent frequency of the glide /j/, and does not belong to the perceived intonation contour. The peak of -trib- is at 126.446 Hz, from which the pitch falls to that of /u/ at 111.832 Hz.

6.Things are, in fact, less simple than that. These rhythmical solutions require quite experienced readers. Furthermore, they require two different kinds of experience. Professors of literature may have the relevant kind of experience with poetry and metric deviance, but lack the command of their vocal resources; whereas professional actors may have the command of vocal resources, but not a sufficient understanding of metric complexities. That is why Marlowe Society on the one hand, and J.H. on the other, are so rare informants for the purposes of the present research. They have both an intuition for metrical subtleties and a mastery of their vocal resources, to carry out their solutions of the perceptual problem posed by conflicting patterns of stress and metre.

7.This theoretical distinction between performance categories and acoustic cues is one central respect in which I beg to differ from some other empirical researchers.


Barsch, Achim (1987) "Trends in Rhythmic ó Language, Literature and Music". Poetics 16: 1-24).

Chatman, Seymour (1965) A Theory of Meter. The Hague: Mouton.

Chatman, Seymour (1966) "On the ëIntonational Fallacy'". QJS 52: 283-286.

Fónagy, Iván (1971) "The Functions of Vocal Style", in Seymour Chatman (ed.), Literary Style: A Symposium. (London: Oxford UP). 159-174.

Halle, Morris (1978) "Knowledge Unlearned and Untaught: What Speakers Know about the Sounds of Their Language",
in Morris Halle et al. (eds.), Linguistic Theory and Psychological Reality. Cambridge, Mass.: MIT.

Halle, Morris and Samuel Jay Keyser (1966) "Chaucer and the Study of Prosody", College English 28: 187-219.

Halle, Morris and Samuel Jay Keyser (1971) English Stress: Its Growth and Its Role in Verse. New York: Harper and Row.

Kassai, Ilona and Zsuzsanna Fagyal (1996) "Hogyan 'Eszlelik a Magyar Besz'ed Sz:uneteit Magyar 'es Francia
Anyanyelv"u hallgat'ok?" (How do listeners, native speakers of Hungarian and French, perceive pauses in Hungarian speech?).
Magyar Nyelv"or 120: 209-220.

Knowles, Gerry (1991) "Prosodic Labelling: The Problem of Tone Group Boundaries", in Johannson, Stig and Anna-Brita
Stenström (eds.), Enlgish Computer Corpora. Selected Papers and Research Guide. (Topics in English Linguistics 3.)
Berlin: Mouton de Gruyter. 149-163.

Knowles, Gerry (1992) "Pitch Contours and tones in the Lancaster/IBM Spoken English Corpus", in Gerhard Leitner (ed),
New Directions in English Language Corpora ó Methodology, Results, Software Developments.
Berlin: Mouton de Gruyter. 289-299.

Lieberman, Philip (1967) Intonation, Perception and Language. Cambridge, Mass.: MIT.

Miller, George A. (1970) The Psychology of Communication. Harmondsworth: Pelican.

Preminger, Alex & T. V. F. Brogan (1993) The New Princeton Encyclopedia of Poetry and Poetics. Princeton: Princeton UP

Schramm, Wilbur L. (1935) Approaches to the Science of English Verse. Iova City: Iova University.

Polányi, Michael (1967) The Tacit Dimension. Garden City, N.Y.: Anchor Books.

Taylor, I. (1990) PsycholinguisticsóLearning and Using Language. Englewood Cliffs, N.J.: Prentice Hall.

Tsur, Reuven (1977) A Perception-Oriented Theory of Metre. Tel Aviv: The Porter Israeli Institute for Poetics and Semiotics.

Tsur, Reuven (1985a) "Contrast, Ambiguity, Double-Edgedness". Poetics Today, 6: 417-445.

Tsur, Reuven (1985b) "Stress, Intonation, and Duration in Prose and VerseóSelected Issues in the Performance Dimension".
Eidos II, 1-2: 5-8.

Tsur, Reuven (1992) Toward a Theory of Cognitive Poetics. Amsterdam: Elsevier

Tsur, Reuven (forthcoming) "Douglas Hodge Reading Keatsís Elgin Marbles Sonnet" (in Style).

Tsur, Reuven and Galit Adam (unpublished) "Bisyllabic Occupancy of Metrical Position as a Perceptual Problem".

Wellek, René & Austin Warren (1956) Theory of Literature. New York: Harcourt, Brace & Co.

Recorded Reading

Hodge, Douglas (1995) John Keats. Hodder Headline AudioBooks HH 186.

This research was supported by the ISRAEL SCIENCE FOUNDATION administered by the

ISSN 1546-0401