Poetry, Prosodics and Masoretic Cantillation in Song of Songs

The Masoretic Hebrew text of the book שׁיר השׁיריםSong of Songs or Canticles consists of 117 פסוקים, that is a string of words marked at the end by the so-called סוף פסוק. This textual unity is called verse in our contemporary translations.

The word followed by the סוף פסוק is itself marked with the cantillation mark סילוק. This is a so-called disjunctive accent of the first level which I will mark as d1.Traditionally the Masoretic cantillation marks are divided up into disjunctive and conjunctive accents. The former group is itself split up into 4 levels, level 1 being the strongest level of disjunction. Of these 117 פסוקים in the Hebrew text of Song of Songs, there are 86 which are divided by the second of two cantillation marks of the level d1, the אתנח, creating two so-called half-verses within one פסוק.

In search of poetry in the book שׁיר השׁירים one of the most important impediments for the progress of the research into poetical units of this text is constituted by the lack of a clearly defined central unit of measurement that functions as a baseline. As a result of the absence of any demarcation criteria for poetical units, it is impossible to assert anything tenable with regard to the rhythmic of metrical nature of the text. In order to state anything about the type of metrical structures of a text, to assert whether a text is metrical or rhythmical in nature, or even to assert whether the genre of the text is in fact poetical, we need a baseline as a starting point that is used to count, for example syllables or words.

One of the elements of the Masoretic Hebrew text that is used to divide the text into units and sub-unites, is the so-called system of cantillation marks. Although this system is, of course, well known to every scholar of the Masoretic texts and treated more or less elaborately in any handbook of textual criticism, it remains a question whether the division created by these accents may serve as an analytical tool for the analysis of Hebrew poetry and the poetical structure of it.

Thus, the research question that I will address in this concise essay is: does the Masoretic system of cantillation marks indicate poetical units of some sort at some level? The null hypothesis is, therefore, formulated as follows: the Masoretic cantillation marks are indicators of poetical units. The statistical analyses that I will conduct, will verify or falsify this working hypothesis.

In the following paragraphs I will firstly try to gain insight into the division of the Masoretic texts by means of פסוקים form the perspective of textual criticism in order to linger on the origin of this textual division. Secondly I will elaborate on a contemporary view of the Masoretic cantillation marks that states that this system of marks is prosodic in nature. Thirdly I will discuss a recent proposal that views the inchoate levels of the Masoretic system of cantillation marks as poetic. After this preliminary research I will conduct the statistical analyses.

Function of the פסוקים

According to Tov (2004, 127) the division of the Hebrew Masoretic text into verses originates in the oral reading practice of the Torah.Tov, Emmanual. 2004. Scribal practices and approaches reflected in the texts found in the Judean desert. Leiden: Brill.

In particular, it remains difficult to know where and in which period the tradition of verse division developed. It is suggested here that the division into small sense units originated in conjunction with the public reading of Scripture (in the synagogue service).

Probably these public Torah readings have taken place from the second century BCE onward. Tov (2004, 127) constrats the origin of the division into פסוקים with the origin of the devision of the text into lager sections, the פרשׁיות, which stems from a writing tradition.

Thus, while the system of subdividing the text into open and closed sections reflects a writing tradition … the division into verses has its origin in the oral tradition of Scripture reading.

In the light of my research question, Tov (2004, 128) reaches an important conclusion with regard to the original function of the division of texts into פסוקים.

In that tradition, such a small unit was known as a פסוק, pasuq, that is, a unit after which one interrupts (פסק) the reading and leaves a pause, and which subsequently was indicated with a silluq accent. The indication of this accent, usually combined with a dicolon, at the end of a verse indicated the original oral division into verses and was therefore the end product of an exegetical procedure, rather than its beginning.

Tov’s textual critical point of view with regard to the division of texts into פסוקים means that this division is not of a poetic nature. Its origin nor its function bear a relation to the poetical aspects or properties of the texts. In stead it is of exegetical nature and as such not a part of the poetical starting point of a text but rather an end product of an exegetical tradition, which also accounts for the fact that many manuscripts have different divisions of the same texts.

A final point to note is the fact that there have been, most likely, more systems of text division, parallel to that of the פסוקים, which originated in reading traditions as well, as Tov (2004, 131) notes.

Various early written traditions concerning the division of the text into small units (verses) have come down to us as described above. All these texts are based probably on an ancient reading tradition that initially was oral. Such an oral reading tradition was put into writing at a later stage, and integrated into the recording of the accents of MT. Within this tradition, each unit ending with a silluq is considered a verse. According to Revell, an additional, parallel, system for verse division was once operative, visible now only in the so-called pausal forms, occurring not only at the ends and in the middle of the Masoretic verses, but also in other positions.

The main conclusion to be drawn from Tovs (2004) analyses of the origin of textual divisions into פסוקים is the fact that it is the end product of a long exegetical tradition embedded in the reading practice that developed from the second century BCE onward. This, most likely, seems to contradict interpretations of the פסוקים as indicator of poetical units of the Hebrew text.And, as a consequence, any further subdivisions into smaller units by means of disjunction accents of the levels d2, d3, and d4 as well if interpreted as subdivision of a d1 level accent. Which leads me to conclude that it is, from the perspective of textual criticism, highly unlikely that the use of the פסוקים as demarcation of poetical units will lead to any tenable results.

Cantillation as prosodical system?

In recent decades the study of prosodic aspects of language saw an enormous increase in research output. This had its effect on the research into the system Masoretic cantillation marks as well. Dresher (1994) is, I think, a turning point in the research into the system of cantillation marks and their function. Dresher, Bezalel Elan. “The Prosodic Basis of the Tiberian Hebrew Systems of Accents.”. Language 70, no. 1 (1994): 1–52. Dresher (2013) presents the same research results as Dresher (1994) be it in more succinct style.Dresher, Bezalel Elan. “Biblical Accents: Prosody.” In Encyclopedia of Hebrew language and linguistics, edited by Geoffrey Khan, 288–96. Leiden, Boston: Brill, 2013.

In Dresher (1994, 6) the author points out that

… the accents serve as a guide to the proper phrasing of the text; that is, they comprise neither a semantic nor a syntactic representation, but a prosodic one.

When seen as linguistic object, the cantillation marks pose severe problems if interpreted as syntactical because the accents often do not correspond to syntactical nodes, e.g. clause or phrase edges, which is why Dresher assumes the cantillation marks to be prosodic in nature rather and logico-semantic.

In his research, Dresher’s starting point is the prosodic hierarchy as developed by Selkik, Nespor and Vogel, and by Hayes.I refer to the bibliography of Dresher (1994) for the 7 titels of the authors mentioned here. This model of prosodic hierarchy assumes the 4 top levels to be: utterance, intonational phrase, phonological phrase, phonological word. According to Dresher (1994, 8) these correspond to the following divisions in his research: verse, disjunctive phrase, conjunctive phrase, phonological word. Dresher’s divisions are, of course, based on the Masoretic cantillation marks. The utterance corresponds to the פסוק. The intonational phrase corresponds to a single word with a disjunctive accent d1, d2, d3, and d4. Although effectively of the two d1 accents only the אתנח occurs at a non-final position within the פסוק, the word with the d1 accent סילוק counts as a disjunctive phrase of its own right even though it is itself immediatly followed by the ‘colon’ of the סוף פסוק.


Figure 1 ⋙ The beginning of Song of Songs as written in the Aleppo Codex. The picture shows the last word from Song of Songs 1:1. The accent silluq is clearly visible underneath the penultimate letter as a perpendicular stroke.

In Dresher’s article the phonological word is, de facto, a string of graphemes surrounded by spaces on both edges with מקף not counting as a space. This means that elements that are syntactically two seperate words, should, prosodically, be counted as one if clitisized or conntect by מקף. All types of enclitics, mainly particles, prepositions, and the article that are clitisized to their host element are not separate words but are, on a prosodical level, counted as one.This prosodic word is labeled as ω (Greek omega) in the literature. An example can be seen in the Figure 1. This fragment contains the last two syntactic words of Song of Songs 1:1, the preposition ל followed by the proper name שׁלמה. Both syntactic words form one prosodic word, which is reflected by the fact that is bears only one single accent.There are, actually, prosodic words that do have two cantillation marks. I will elaborate on this in another article.

The מקף is attested 111 times in Song of Songs. An example can be seen in Figure 2 where the מקף clearly joins two syntactic words, here the verb and its prepositional object, together.


Figure 2 ⋙ A single prosodic word consisting of two syntactic words, a verb and a prepositional object, from Song of Songs 1:6 as in the Aleppo codex.

The level of the conjunctive phrase is formed by one or more prosodic words with a conjunctive cantillation mark. The conjunctive phrase ends with the first prosodic word with a disjunctive accent following the last word with the conjunctive accent.

Dresher (1994, 10) states that three rules are operative the at prosodic level of the conjunctive phrase: spirantisation, gemination, and rhythmic stress shift. The first rule is clearly visible in Figure 2 where the first consonant of the prepositional phrase בי has a fricative (v) rather than plosive quality (b) because both words are closely connected through מקף.


Figure 3 ⋙ Two prosodic words from Song of Songs 1:4 (again taken from the Aleppo codex) of which the first has a disjunctive accent of the d3 level.

An example of the preposition ב that would have been fricative because it is preceded by a vowel, but is plosive nonetheless (the opposite of what is occuring in Figure 2) because of the פשׁטא, a disjunctive accent, on the preceding word ending with the vowel, can be seen in Song of Songs 1:4 in Figure 3. Also note the fact that this is a case where one would have expected an accent shift, the so-called נסיגה to be applied. The position of the accent though reveals us that, even though in Song of Songs נסיגה is utilized some 17 times to avoid an accent clash, in this particular case, it is not.

Dresher clearly shows that several linguistic processes are productive in Tiberian Hebrew on the prosodic levels of the phonological word, the conjunctive phrase, and the disjunctive phrase. Furthermore Desher substantiates the convergence of these prosodic levels with particular Masoretic cantillation marks of specific levels with ample empirical evidence.Dreshers hypotheses, as does any explanatory framework, poses some fundamental problems when confronted with Masoretic Hebrew texts but these must be addressed in a separate article.

Although Dresher and Tov approach the Tiberian Masoretic cantillation marks and the devision of the texts into פסוקים from different disciplines, a linguistic and a textual critical respectively, I think they both seem to verify the hypothesis that the Tiberian Masoretic cantillation marks are rhetorical and prosodic in nature, which is why this system of cantillation marks clearly reflects the fact that they originated in and continued to develop in the context of reading traditions in which the best delivery of the text was the primary goal.

Statistics

In the preceding paragraphs it was stated that the Masoretic cantillation marks are presumed to have originated from a reading tradition, and to have (in a later stage) developed to a fully fledged prosodic system in which not only the relation between words was laid down but the pronunciation of every single prosodic word as well. This had lead researchers like Tov en Dresher to their respective conclusions. These two interpretations, a textual critical and a linguistic respectively, go well together with the underlying variable of a reading practice, in which the best delivery in all circumstances was the goal of the system. Both conclusions are related to the prosodic aspects of language.

If the Tiberian system of cantillation marks is indeed of a prosodic nature, than, the question whether it indicates the poetic structure of texts as well should be raised. This assumption takes Dresher’s results a step further by assuming the cantillation marks to be a system that defines the poetical units and subunits of a text. Thus, this assumption obviously points to the same origin as the one proposed in Tov’s and Dresher’s research (the best delivery), but it nonetheless implies a quite different goal, namely the best rendering of the poetical structure of the text.

Thus to assume this we must use Dresher’s conclusion in a very specific manner, as it presumes Dresher’s conclusions to be statements about the prosodic nature of the poetic structure of the texts. The assumption that takes the cantillation marks to indicate poetic units though, reveals an underlying syllogism somewhat like the this: the texts are poetical with a prosodic structure, the cantillation marks reflect the prosodic structure of the text, thus, the cantillation marks are indicative of the poetic structure. But this might be a wrong inference as the prosodic structure of a text can have many different qualities, depending on their goal. After all, the prosodic structure as reflected in the cantillation marks may be of a completely different nature than the prosodic structure that would represent the poetic structure.

Shortly: delivery is not necessary scansion even though both imply explicit prosodical structuring of the text.

In order to reach any conclusion with regard to the prosodic nature of the Tiberian Masoretic cantillation marks – delivery or scansion – I think it would be necessary to start with a statistical analysis. Therefore, in this paragraph I will statistically analyse two Masoretic units, firstly the פסוק, and secondly the d1 level of the אתנח and the סילוק that both mark the end of the so-called half-verse.

Goal of the calculations

If the Masoretic cantillation marks are indicative of the prosodics of delivery but of scansion as well, than they obviously must resemble the poetic structure of the text. Now if we take the workingdefinition of poetry as some kind of rhythmic or metric structuring of language at some level of a text and we take the, according to the research literature two most likely, baseline units, namely the syllable and the prosodic word to be the basic building block of classical Hebrew poetry, than, as a consequence, some level of equal distribution of these basic building blocks within the units that are demarcated by a Masoretic cantillation mark should be quantifiable and hence statistically measurable.

This means that the initial research questions should be subdivided into 4 questions, because we have postulated 2 baseline units, the verse and the half-verse, and two levels of building blocks, namely the syllable and the prosodic word, which results in a matrix with four cells.

The hypothesis is, that if there is some form of metre, be it on the level of the syllable be it on the level of the prosodic word, than the data should reflect this by showing normally distributed data points with relatively small deviations of individual items relative to the mean and hence a small standard deviation.

The database

Before I can analyse any structural unit, we should bear in mind that there is an ongoing discourse with regard to a) the nature and quantity of the poetic units in classical Hebrew and b) with regard to the nature (metric or rythmic) of the criterion to be used to delineate these poetical units. It is clear that this a loop from which there a no escape: in order to know the baseline unit we need criteria to delineate the it, but in order to find this criterion (syllables, words, …) we firstly would need a baseline to which to compare the metric of rhythmic subunits.The database I used for the research is the Groves-Wheeler Westminster Hebrew Morphology (version 4.20) incorporated in the program Accordance as the HMT-4 module (version 1.8).

Before I could test the data, I had to create them. Thus I created a datasheet with the number of prosodic words and the number of syllables per verse and per half-verse. The criterion I used for the demarcation of the prosodic word is simple: every string of graphemes bound by a space on both the right and the left edge is taken to be a prosodic word, מקף not being a space. For the demarcation of syllables the criterion is equally simple: every consonant with a vowel, whether חטף or full or with שוא נע, is taken to be a syllable; the שוא נח and the furtive פתח are not regarded as a vowel.

Although one could define the prosodic word differently on the basis of an interpretational framework such as Harm van Grol’s or Dresher’s, I choose not to do so in order to be as empirical as possible. See, for example, Harm van Grol ####, and Dresher over Word in Tiberian Hebrew Equally one could delineate the syllable structure within the Tiberian Hebrew word differently than I do here.Hier nog Khan vermelden over syllable structure. For the purpose of a statistical analysis though, these differences in delineation of the word and of the syllable are insignificant because, we can assume that their distribution is normal across the corpus of classical Hebrew texts. As a consequence the absolute number of words or syllables will change, depending on the definition of these terms, but not the ratio of them within the verse or half-verse.

I created two datasets, the first with the פסוק as a unit of measurement the second with the d1 accents, סילוק and אתנח, i.e. the half-verse, as a measurement unit. The variables in both datasets are the same, except for the dichotomous variable accent, which is only used in the dataset for the half-verse and has the binary value 0 (Silluq) and 1 (Atnach).

Table 1: The variables in both datasets. Variable 1–4 are categorical, while variable 4 and 5 are continuous.

Measurement unit variable 1 variable 2 variable 3 variable 4 variable 5
Verse chapter verse   σ ω
Half-verse chapter verse accent σ ω

The analyses

Verse

In Table 2 the descriptive statistics of the number of syllables and words per verse are reproduced.

Table 2: Descriptive statistics of the number of syllables and words in the verse.

  N Range Min Max Mean St.Dev Skew Kurt
σ 117 38 9 47 26,91 9,560 0,268 -0,812
ω 117 12 4 16 9,70 3,257 0,259 -0,838

As to be expected, N=117 as there are 117 פסוקים in Song of Songs. Next the range, the minimum and the maximum are shown, followed by the mean (X̄) and the standard deviation (σ). The last two columns show the the levels of skewness and kurtosis.

Next I will elaborate on the two basic building blocks separately, starting with the syllable.

Syllables

Here I will firstly visualize the dispersion of the number of syllables around the mean.

As a first representation a histogram will clarify the distribution of the quantities of syllabes per vers across the 117 פסוקים in Figure 4.

Figure 4 ⋙ A histogram showing the distribution of the variable numbers of syllables (σ) per vers.

What is immediately clear, is the fact that the data are multi-modal and non-normally distributed on both sides of the mean.

Secondly, the relatively high standard deviation is striking. If there were some kind of regularity in the number of syllables per verse, than the standard deviation should have been far smaller.

The next chart (Figure 5) is a so called boxplot in which the black horizontal line in the green box designates the median (not the mean) which in this case is 26.00, one syllable apart from the mean number of syllables per verse.

Figure 5 ⋙ A boxplot showing the distribution of the numbers of the variable number of syllables (σ) per verse.

The upper and lower parts of the green box constitute the so-called inter quartile range (IQR), which in this case is 33,50 as the third quartile upper limit, and 18.00 as the lower limit of the second quartile. The first and the fourth quartile are represented by the so-called whiskers, the T-shaped lines above and below the green box.The software used to conduct these tests (SPSS) defines an outlier as a score greater than the upper quartile plus 1.5 times the IQR. An extreme score is defined as a score greater than the upper quartile plus 3 times the IQR. This also implies that the range between the upper limit of the third quartile (the top of the box) and the upper whisker is approximately 25% of the scores. The same applies to the bottom whishker. Even though there are neither outliers nor extremes the dispersion around the median is quite large, with a range of 15,50 within the IQR.

As a last chart, Figure 6 shows a so-called a P-P plot. In this plot the observed cumulative probability of a variable is plotted against the probability of a particular distribution, which in this case is, of course, the normal distribution. The better the dots are posited on the line, the more the data are normally distributed.

Figure 6 ⋙ A P-P plot showing the distribution of the variable number of syllables (σ) per verse.

As is easily observed, the number of syllables per verse are non-normally distributed around a mean. The line of dots is S-shaped which means there is a problem with skewness.

A last, and purely numerical representation of the (lack of) normality of data are the Kolmogorov-Smirnov test and the Shapiro-Wilk test.

Table 3: Numerical tests of normality for the number of syllables within the verse.

  K-S     S-W    
  statistic df Sig. Statistic df Sig.
number of syllables .095 117 .011 .956 117 .004

Both tests test the null hypothesis that states that the data are normally distributed, which entails that a significance of <.05 means that the alternative hypothesis is probably true and thus that the data are non-normally distributed. In this case on both tests the significance level is (far) below .05.

These two tests corroborate what the graphs already showed, the number of syllables per verse have a non-normal distribution and disperse greatly from the mean.

Words

I will now present the visualisation of the data with the variable word as a building block, while the unit within which the calculations are done is still the verse.

The chart in Figure 7 shows the histogram.

Figure 7 ⋙ A histogram showing the distribution of the variable number of prosodic words (ω) per verse.

As with the syllable, the variable word deviates from a normal distribution, which results in a large standard deviation.

The boxplot in Figure 8 shows the median, which is 9 and the lower and upper limits of the second and the third quartile, 7 and 12 respectively. This means that the IQR has a range of 5 words. A number that is suspiciously high for a text with a metrical structure based on word-count.

Figure 8 ⋙ A boxplot showing the distribution of the variable number of prosodic words (ω) per verse.

In Figure 9 the P-P plot is shown.

Figure 9 ⋙ A P-P plot showing the distribution of the variable number of prosodic words (ω) per verse.

Again, the dots deviate from the normal distribution, the red line, and they have an S-shaped curve, meaning that this distribution is relatively skewed.

The Kolmogorov-Smirnov test and the Shapiro-Wilk test results are given in the table below.

Table 4: Numerical tests of normality for the number of prosodic words within the verse.

  K-S     S-W    
  statistic df Sig. Statistic df Sig.
number of words .113 117 .001 .960 117 .001

Compared to the tests of the syllable, this test shows an even lager significant deviation from normality: both the Kolmogorov-Smirnov and the Shapiro-Wilk test have a significance of .001.

Z-scores of skewness and kurtosis

As I showed in the paragraphs above, the S-shaped P-P plots of the distribution of the variable syllable and the variable word, implied problems with skewness. We can take the data of skewness and kurtosis a step further to analyze the relative deviation of normality in terms of these. For this we need the standard error of both skewness and kurtosis of our data. These are given in the table below.

Table 5: Skewness and Kurtosis and their respective standard errors of the number of syllables and words per verse.

  Skewness St. Error Kurtusis St. Error
σ .268 .224 -.812 .444
ω .259 .224 -.838 .444

To calculate whether the data exceed a certain limit, the skewness is divided by the standard error of the skewness. This basically gives us the z-score of skewness. The α-level I used for this research is the default .05, so the z-score that results from the calculation certainly should not exceed the value 1.96. Of course, the same can be calculated for the level of kurtosis.

Table 6: The z-scores of Skewness and Kurtosis of the number of syllables and words per verse.

  zskewness zkurtosis
σ 1,196 -.649
ω 1,156 1,887

Even though the z-scores do not exceed the α-level, they are relatively high for skewness and for the variable word on kurtosis, were it is even near the α-level.

Conclusions regarding the verse

I will conclude that the baseline unit verse shows no metrical structure. Neither the variable syllable nor the variable word have a normal distribution with a small relative standard deviation, and both variables have a range of the IQR that is far to large for a metrical text.

Half-verse

As for the verse, I will firstly show the descriptives in a table. In this case N=203, because there are 86 half verses with אתנח, so N=117+86=203.

Table 7: Descriptive statistics of the number of syllables and words in the half-verse.

  N Range Min Max Mean St.Dev Skew Kurt
σ 203 29 5 34 15,46 5,753 0,878 1,046
ω 203 11 2 13 5,59 1,971 0,916 1,243

The parameters shown in this table are the same as the ones in Table 2.

Syllables

Again I will start with a histogram to visualize the dispersion of the data around the mean.

Figure 10 ⋙ A histogram showing the distribution of the variable numbers of syllables (σ) per half-verse.

The dispersion around the mean is even larger than the dispersion we saw of the variable number of σ within the unit of the verse.

This is also clearly visible from the boxplot in Figure 11.

Figure 11 ⋙ A boxplot showing the distribution of the variable numbers of syllables (σ) per half-verse.

The median is 15.00 and the upper limit of the third quartile and lower limit of the second quartile is 12.00 and 18.00 respectively. What is even more remarkable is the number of outliers above the upper limit of the fourth quartile. Compared to the median, the range of the IQR, which is 6.0, is relatively large.

The last chart, the P-P plot is shown in Figure 12.

Figure 12 ⋙ A P-P plot showing the distribution of the variable numbers of syllables (σ) per half-verse.

Notice that the distribution is S-shaped again. This time even the middle part, the part where one would expect the dots to be mostly on the line, the dots are below the line of the normal distribution.

The Kolmogorov-Smirnov test and the Shapiro-Wilk test results are given in the table below.

Table 8: Numerical tests of normality for the number of syllables within the half-verse.

  K-S     S-W    
  statistic df Sig. Statistic df Sig.
number of words .142 203 .000 .944 117 .000

When compared with the test of the variable number of σ within the verse, the tests for this variable within the half-verse shows an even more significant deviation from normality, with both tests at a p<.001.

Words

The histogram of the number of words is shown in Figure 13.

Figure 13 ⋙ A histogram showing the distribution of the variable numbers of prosodic words (ω) per half-verse.

Again the data do not show a normally distributed and the histogram is positively skewed.

As with the syllables, also the variable words has a large number of outliers and an extreme score.

Figure 14 ⋙ A boxplot showing the distribution of the variable numbers of prosodic words (ω) per half-verse.

The median is 5, the lower limit of the second quartile being 4 and the upper limit of the third quartile is 6, leading to the IQR of 2. Here the upper whisker has the remarkable value of 9.0.

Figure 15 ⋙ A P-P plot showing the distribution of the variable numbers of prosodic words (ω) per half-verse.

The Kolmogorov-Smirnov test and the Shapiro-Wilk test results are given in the table below.

Table 9: Numerical tests of normality for the number of words within the half-verse.

  K-S     S-W    
  statistic df Sig. Statistic df Sig.
number of words .186 203 .000 .928 117 .000

When compared with the test of the variable words within the verse, the tests for this variable within the half-verse shows an even more significant deviation from normality, with both tests at a p<.001.

Z-scores of skewness and kurtosis

As for the unit verse, I will present the z-scores for the variables number of σ and number of ω.

Table 10: Skewness and Kurtosis and their respective standard errors of the variables number of syllables and number of words per half-verse.

  Skewness St. Error Kurtusis St. Error
σ .878 .171 1.046 .340
ω .916 .171 1.243 .340

To calculate whether the data exceed a certain limit, the skewness is divided by the standard error of the skewness. This basically gives us the z-score of skewness. The α-level I used for this research is the default .05, so the z-score that results from the calculation certainly should not exceed the value of 1.96.

Table 11: The z-scores of Skewness and Kurtosis of the number of syllables and words per half-verse.

  zskewness zkurtosis
σ 5.135 3.076
ω 5.357 3.656

Contrary to the levels measured at the unit of the verse, the z-scores of skewness and kurtosis of the variables number of σ and number of ω exceed the α-level to a considerable extent.

Conclusions regarding the half-verse

It is clear from the histograms, the boxplots, the tests of normality and finaly even the relative z-scores, that the unit of the half-verse shows an even greater deviation from normality than the verse.

One of the reasons for this, is the fact that the accent אתנח is not placed around the arithmetic midpoint (whether counted in syllables or in words) of the verse regularly. This will lead to very short and very long half-verses. This, in turn, may lead us to abandon the term half-verse as one that is misleading. Even though the accent אתנח is of the same level as the סילוק, namely d1, this, appearantly, does not mean that it splits up the פסוק in two more or less evenly long parts.

However this may be, I think it is safe to assume that the half-verse, like the verse, cannot be interpreted as a metrical unit, neither for the variable number of σ nor for the variable number of ω, as their quantitative distribution deviates by a far to great measure from the, to be expected, normal distribution with a small standard deviation.

Coefficient of variation

Another way of comparing distributions is by means of the coefficient of variation or CV, sometimes called relative standard deviation or RSD. The CV is calculated by the following formula: σ/X̄. We already calculated the standard deviation (σ) as well as the mean (X̄) for the variables number of σ and number of ω in both units, verse and half-verse.

The CVs are presented in the Table 12.

Table 12: The coefficient of variation for the number of syllables and words per half-verse.

  verse half-verse
σ .36 .37
ω .34 .36

It is clear that the CV’s are roughly the same for all the four elements.

Ovid Metamorpheses

In order to make all the data from the verse and the half-verse in the previous section more meaningful, I will compare them to some data from Ovid. Specifically the CV’s we be, as we shall see, usefull.

I have used the first 111 lines of Ovid’s Metamorphoseon. Liber Primus, a text of which we know with certainty that it has a metrical structure, the dactylic hexameter. I will visualise the variable number of σ only, because it is of now use te count prosodic words, as the syllable are counted in this poetry.

It is immediately clear that the data are dispersed around the mean more in accord with a normal distribution than any of the statistical measurements of Song of Songs.

Figure 16 ⋙ A histogram showing the distribution of the variable number of syllables (σ) per line in

Metamorpheses

.

This is reflected by the boxplot as well.

Figure 17 ⋙ A boxplot of the variable number of syllables (σ) per line in Metamorpheses.

It may look as if this distribution has more outliers than the boxplots of the number of σ and number of ω measured within the unit of the verse in Song of Songs, but this true only in an absolute numerical sense. The second quartile is absent, while the third quartile is only 1 syllable above the median, which is 15.

This is fact is also represented by the P-P plot.

Figure 18 ⋙ A P-P plot of the variable number of syllables (σ) per line in Metamorpheses.

The number of dots is only 6, because the total range is as less as 5, the lowest number of syllables being 13 and the maximum being 18.

The truly metrical structure of this text is reflected by the fact that the variable number of σ has a relatively small standard deviation of 1.035 around the mean of 15.40.

Coefficient of variation comparison

The CV’s from different samples are useful for comparison, which is why I will present the CV’s from the verse and the half-verse together with the calculated CV from the sample of Ovid.

Table 12: The coefficient of variation for the number of syllables per verse, per half-verse, and as measured in Ovid.

  verse h-verse ovid
σ .36 .37 .07

Again, only the variable number of σ is used for reasons that were formulated earlier in this paragraph.

What the comparison tells us, is that the deviation from normality relative to the mean is some 5 times greater in Songs of Songs than in the Metamorphoses. Given the fact that the Metamorpheses are written in the dactylic hexameter and thus do have a metrical structure, it is, to my opinion, implausible to view the verse or the half-verse as metrical based on the CV’s of syllable count.

Final Conclusions

Based on the distributions of the number of syllables and the number of words within the verse as well as the half-verse, I assume that it is reasonable to state that no metrical structure is present. Thus the null hypothesis, H0, which was our working hypothesis, should be rejected and the alternative hypothesis is likely to be true. The content of H1 could be anything but H0.

Neither the syllable nor the word are the building blocks of metrical structure of some sort within the units of the verse of the half-verse. Therefore, whether one assumes the Hebrew poetry of Song of Songs to be accentual of syllabic, neither the verse nor the half-verse can serve as units for poetical structure.

What the statistics do not exclude, and what might thus be taken to be a direction to the H1, is the possibility that the syllable or the word or building blocks of some metrical structure on unit other than the verse or half-verse.

Moreover, based on the statistics we cannot exclude the possibility that the syllable and the word are building blocks of a non-metrical type of poetry, that is rhythmic rather than metric.

Based on the statistic analyses, I suggest that exegetes in search of poetical structure of classical Hebrew texts should not feel constrained to the unit of the verse. As we have seen, the פסוק cannot be interpreted as a poetic unit within which metrical units are found. The same must be concluded with regard to the half-verse. Poetical units and metrical structures may converge with the פסוק or the d1 level of accents but merely as a coincidental outcome not as a result of the goal of the subdivision of the text by the Masoretic cantillation marks or the פסוקים.

Poetry, Prosodics and Masoretic Cantillation in Song of Songs - March 11, 2019 - Robert Voogdgeert