8

The Finnish national epic Kalevala was written in a certain poetic meter, which in some respects resembles the classical Greek and Roman meters. In addition to rules regarding syllable length and stress and the poetic rhythm, there is a strong tendency to place the longest words (by syllable count) in a verse at the end and shorter ones at the beginning. (This together with rhythmic restrictions heavily limits the long words one can use in this meter.) Ending a line in a one-syllable word is considered to be outright forbidden. This tendency is known as "viskurilaki" in Finnish, and it's not limited to poetry.

Does classical Latin poetry have a similar tendency? Does the length of a word correlate with its location in a verse in any way? Is this explicitly mentioned in poetry guides or has it been studied?

(I have limited this to classical Latin poetry. If the answer is different and interesting for Greek or later Latin, please leave a comment. Details should be taken to a separate question.)

  • Out of curiosity, what does viskurilaki mean? Laki is law, but I'm finding several possible translations for viskuri, none of which seem relevant. – TKR Apr 5 '17 at 3:59
  • @TKR Viskuri comes from the verb viskata, a derivative similar to the Latin -trum or -men(tum). So perhaps "throwing law", as the words are being thrown within a line. (I'm not sure if some other English verb would be best here: toss, hurl, cast...) – Joonas Ilmavirta Apr 5 '17 at 4:07
  • This basically screams for a corpus analysis. I'll see what I can do :) – blagae Apr 5 '17 at 7:13
  • @blagae A corpus analysis would be wonderful. If you can do or find one, please share. – Joonas Ilmavirta Apr 5 '17 at 14:51
  • 1
    @brianpck I mean syllable count. I'll edit that in. (In contexts outside the Kalevala meter viskurilaki deals with many kinds of length or weight, including syllable count, length (syllables, words, sentences, ...) and vowel quality.) – Joonas Ilmavirta Apr 5 '17 at 15:34
6

As promised before in a comment, I made a corpus analysis of whether word length, in terms of number of syllables, is correlated with position in the verse.

Spoiler alert: it isn't. brianpck has already done a basic corpus analysis with essentially the right conclusion, but I intended to do this as thoroughly as possible in order to hone my primitive skills as a statistician (meaning: please respond to any errors I may have made), and to make good on the aforementioned promise.

Input corpus

I used a hexameter corpus consisting of:

  • Vergilius, Bucolica
  • Vergilius, Aeneis
  • Ovidius, Metamorphoses

Caveats

Only verses that are scanned correctly by my Latin verse scanning tool are taken into account. This is 94.7% of the existing corpus (21525 out of 22723 verses) as of the time of this writing. There's a whole host of reasons why a verse wouldn't be scanned correctly, e.g.:

  • hiatus in the verse
  • hypermetricality (elision over verse boundaries)
  • unfinished verses
  • multiple possible scans
  • non-intuitive scansions of foreign words
  • other flukes

There are a small number of words that just cannot be determined correctly e.g. the word VOLVI can be either bisyllabic volvi (< volvere) or volui (< velle). There is no way to distinguish between these forms if the vol- part is in the second half of a foot: vo̱l|vi or vo̯lu̯|i, except by leveraging dictionary or even semantic knowledge. But the occurrence of this kind of problem should be rare enough to not influence the data significantly.

I didn't check for elided syllables, so e.g. multum ille et in Verg. Aen. I, 3 is counted as a sequence of words of 2, 2, and 1 syllables.

Data analysis

I compared the actual vs the expected word length of words in verses: this essentially normalized the data to always have an expected value of 1, in order to allow using the data set of the 4-word verses together with the 11-word ones.

For example, any word in a verse consisting of 5 words that is 15 syllables in total, has an expected syllable count of 3. A verse consisting of 10 words and 14 syllables in total will contain on average a word of 1.4 syllables.

By doing the calculation (actual * number of words / number of syllables), any word in any verse has an expected normalized weight of 1, no matter how many words and how many total syllables the verse contains.

My results are:

+---------------+-------+-------+-------+-------+-------+-------+-------+------+------+------+------+
| word position |   1   |   2   |   3   |   4   |   5   |   6   |   7   |  8   |  9   |  10  |  11  |
+---------------+-------+-------+-------+-------+-------+-------+-------+------+------+------+------+
| count         | 21525 | 21525 | 21525 | 21525 | 21456 | 18340 | 10133 | 3308 |  636 |   76 |    7 |
| average       |  0.95 |  0.92 |  0.97 |  0.98 |  1.03 |  1.07 |  1.12 | 1.17 | 1.20 | 1.17 | 1.43 |
| st dev        |  0.40 |  0.33 |  0.35 |  0.35 |  0.39 |  0.37 |  0.34 | 0.33 | 0.38 | 0.44 | 0.43 |
| variance      |  0.16 |  0.11 |  0.12 |  0.12 |  0.15 |  0.13 |  0.11 | 0.11 | 0.15 | 0.19 | 0.19 |
| # of st devs  |  0.12 |  0.25 |  0.07 |  0.07 |  0.08 |  0.19 |  0.35 | 0.52 | 0.53 | 0.38 | 0.98 |
+---------------+-------+-------+-------+-------+-------+-------+-------+------+------+------+------+


+---------------+-------+-------+-------+-------+-------+-------+-------+------+------+------+-------+
| word position |  -1   |  -2   |  -3   |  -4   |  -5   |  -6   |  -7   |  -8  |  -9  | -10  |  -11  |
+---------------+-------+-------+-------+-------+-------+-------+-------+------+------+------+-------+
| count         | 21525 | 21525 | 21525 | 21525 | 21456 | 18340 | 10133 | 3308 |  636 |   76 |     7 |
| average       |  1.01 |  1.24 |  0.94 |  0.94 |  0.96 |  0.96 |  0.91 | 0.87 | 0.82 | 0.83 |  0.66 |
| st dev        |  0.25 |  0.37 |  0.37 |  0.33 |  0.35 |  0.39 |  0.40 | 0.39 | 0.36 | 0.37 |  0.02 |
| variance      |  0.06 |  0.14 |  0.14 |  0.11 |  0.13 |  0.15 |  0.16 | 0.15 | 0.13 | 0.14 |  0.00 |
| # of st devs  |  0.06 |  0.64 |  0.17 |  0.18 |  0.10 |  0.09 |  0.21 | 0.34 | 0.49 | 0.45 | 16.77 |
+---------------+-------+-------+-------+-------+-------+-------+-------+------+------+------+-------+

word position -1 means the last word in a verse, -2 the second to last, etc

The last line of every table is the most important one, because it points to how significant the difference is between actual and expected data. In order to be statistically significant, we want the value in this row to be at least 2.

Conclusion

So the only place where the actual length of words deviates significantly from the expected value is the eleventh-last word in verses of 11 words long, so the first. All 11-word verses start with a monosyllable, but given that there are only 7 such instances in the entire corpus, this is negligible. If any of these 7 instances hadn't had a monosyllable as a first word, then the sigma value would've been 1.30 instead of 16.77.

All other values are not statistically significant, so no, there is no correlation between word position and word length expressed in syllables.

FYI I can provide the raw data on request

  • Great work, +1. Much more sophisticated than what I did. Perhaps for completeness you could check for any significant differences between Vergil and Ovid. The point, though, seems overwhelmingly clear. – brianpck Apr 7 '17 at 21:14
  • Thanks! This is very convincing evidence. The fact that the first word of a 11-word verse is always monosyllabic is not surprising even if it does not match the calculated expectation. But such edge cases are not really important for the overall conclusion. – Joonas Ilmavirta Apr 7 '17 at 21:14
8

After doing a rudimentary corpus analysis of the Vergil's Aeneid, my conclusion is that Latin verse does not show any meaningful relationship between syllable length and word position.

To test this, I wrote up a rudimentary Python program that went through the whole text of the Aeneid and found the syllable length of every word, grouping them by their position in the text. This is admittedly crude, but should give at least some indication of a tendency. Here are my results:

pos     words   min     max     avg
0       9896    1       7       2.41
1       9892    1       8       2.26
2       9873    1       7       2.35
3       9853    1       7       2.36
4       9813    1       7       2.52
5       8393    1       6       2.46
6       4429    1       6       2.36
7       1310    1       4       2.23
8       196     1       4       2.08
9       20      1       3       1.75
10      1       3       3       3.0

EDIT: I did a quick modification to measure just the last word, and the results are nearly identical:

pos     words   min     max     avg
-1      9896    1       7       2.43

Where:

  • pos = 0-indexed word position in a line (e.g. "arma" gets a 0, "virumque" gets a 1, etc.)
  • words = number of words in given position
  • min = minimum number of syllables of a word in given position
  • max = maximum number
  • avg = average syllable count in that position

As you can see, short 1-syllable words occur ubiquitously in any position. Longer words also seem to occur uniformly, though because very long 7-syllable words are so rare, they won't occur in smaller sample sizes because of standard deviation. The average too hovers at about the same value and only changes when the sample size gets too small.

N.B. I have never formally studied statistics, so I am happy to improve the metrics if someone can provide a better heuristic. I can also come up with some graphs if that would be helpful.

Also, as a caveat, my syllable counting program is the slip-shod work of two minutes: it works correctly for the first 10 verses of the Aeneid, but obviously a better implementation could catch more edge cases.

Here is my program for reference:

import re
from collections import defaultdict

text = """full text with line breaks here. I took text from https://www.gutenberg.org/files/227/227-h/227-h.htm and removed chapter headings"""

text = ''.join(c for c in text if c.isalpha() or c.isspace())
def syllables(word):
    return len(re.findall(r'^i[aeiou]|a[eiu]|o[ei]|[aeiouy]', word.lower())) - word.count('qu')

verse_lengths = defaultdict(list) 
for verse in text.split('\n'):
    for i, word in enumerate(verse.split()):
        verse_lengths[i].append(syllables(word))

print('{}\t{}\t{}\t{}\t{}'.format('pos', 'words', 'min', 'max', 'avg'))   
for pos, word_lengths in verse_lengths.items():
    avg = round(sum(word_lengths) / len(word_lengths), 2)
    print('{}\t{}\t{}\t{}\t{}'.format(pos, len(word_lengths), min(word_lengths), max(word_lengths), avg))
  • +1 Thanks! This is very interesting. Verses have varying amounts of words, so the way you quantify position does not allow to see if words at the end of a verse tend to be short or long. Would it be possible to add another table with position measured from the end of the verse? (Something that might be interesting but I don't want to push you into: Perhaps one could scale the position so that 0 is always at the beginning and 1 always at the end by dividing by word count. Then one could see how many words of different lengths appear in a given bin of width 0.1 or so.) – Joonas Ilmavirta Apr 5 '17 at 15:44
  • Added with just the last word, which confirms the general hypothesis. – brianpck Apr 5 '17 at 15:47
  • Great! This is very convincing. Something is slightly off, but it's not important: There are 9896-9892=4 verses with only one word, but they can only have up to 7 syllables each. These are either Virgil's incomplete verses or some noise like headings. (There are only 9896-9813=83 verses with four words or less, so it doesn't really affect the averages.) – Joonas Ilmavirta Apr 5 '17 at 16:00
  • 1
    Those are incomplete verses: "Dardanidae" (1.560), "audierit" (2.346), "abluero" (2.720), and "rumpite" (3.640) – brianpck Apr 5 '17 at 16:08
4

The Wikipedia page on "Dactylic hexameter" (¯˘͟˘ | ¯˘͟˘ | ¯˘͟˘ | ¯˘͟˘ | ¯ ˘˘ | ¯˘̱) says

Like their Greek predecessors, classical Latin poets avoided a large number of word breaks at the ends of foot divisions except between the fourth and fifth, where it was encouraged. In order to preserve the rhythmic close, Latin poets avoided the placement of a single syllable or four-syllable word at the end of a line.

More relevant to the Finnish pattern you mention, although less relevant to Classical Latin, is the following quotation:

In the late empire, writers experimented again by adding unusual restrictions to the standard hexameter. The rhopalic verse of Ausonius is a good example; besides following the standard hexameter pattern, each word in the line is one syllable longer than the previous, e.g.:

Spes, deus, aeternae stationis conciliator,
si castis precibus veniales invigilamus,
his, pater, oratis placabilis adstipulare.

Unfortunately, the relevant sections of the Wikipedia article are unsourced so it's hard for me to say how accurate this is.

The phenomenon of caesura also seems like it could be relevant; apparently a caesura is always located at a word break, and occurs within a foot. The "principal caesura" of a line usually comes in the third or fourth foot according to Joel Derfner's answer to the linked question.

(Please downvote this answer if you have reason to believe the statements from the Wikipedia article are incorrect or if you are unsure about them. I don't want to propagate misinformation; I'm just posting this to get the ball rolling.)

  • Thanks! This is useful, but I think I will wait for something better sourced before accepting. The Kalevala meter also has a caesura, and it limits long words in the middle of a line. – Joonas Ilmavirta Apr 5 '17 at 14:49

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.