On the Location of Ancient Greek Stress and its Relation to
Accent
1The problem of the existence, location and relation with the tonal accent of the natural rhythm of Ancient Greek2 has many decades and has seen multiple solutions arise.3 Not only the placement and structure of rhythm is a problem, but even its most basic nature, since it has been equally proposed that it is based in acoustic intensity (or “stress”4
) and that it is based in duration. The debate even seems to be otiose, given the substantial scarcity of evidence; however, I intend in this dissertation to compare three positions that have been developed on the subject in order to show that, besides the theoretical consequences that it has, there is an aspect of the question that is important for the reading of Greek texts in general.
The three selected positions are those that, in my opinion, given the amount of evidence they are founded on and the practicality of their application constitute today the best contenders in the debate. They are those of Allen (1973), Devine and Stephens (1994) and David (2006).5 The three approaches consist fundamentally in a set of more or less simple rules to analyze the rhythm of Greek. The most complex and complete one is, without a doubt, D.-S.’s. In practice, however, this complexity has its negative aspect: D.-S.’s theory is the most difficult one to use in the reading of texts. Nonetheless, once the rules they present for the mapping of Greek rhythm are isolated,
1 I have not been able to include, given the extension of this dissertation, a description of the problems that Ancient Greek tonal accent has to offer, on which, besides the main references in this work, see Probert (2006: 53-124). The conclusions of this dissertation are based in an extensive quantitative study of a sample of Greek poetry and prose; I have published all the data I have collected and used in greekmps.wordpress.com, and particularly the data used in this study in greekmps.wordpress.com/experimental-data/on-the-location-of-ancient-greek-stress-and-its-relation-to-accent.
2 From now onwards, “Greek”.
3 Throughout this whole dissertation I will refer with “rhythm” to the rules of alternation between prominent and non-prominent positions, whether in metre (“metrical rhythm” or “poetical rhythm”) or in the prosodical structure of words, not including tonal accent, although not necessarily independently of it (“linguistic rhythm”). On the problems of poetical rhythm, see Silva Barris (2011: 18-23). For obvious reasons, I cannot analyze the concept of metrical rhythm here, for the use of which the given definition should be sufficient. It must be noted, however, that, against the most extended tendency, I will use the old terms “thesis” and “arsis” to refer respectively to the prominent and non-prominent positions in metres. As to Silva Barris (2011: 10-11), the testimony of two Latin (!) grammarians of the IV and V centuries A. D. seems to me to be insufficient reason to contradict the long tradition of Greek grammarians in which the terminology is always consistent.
4
“Stress” as a term has many problems, on which see Probert (2006: 7-9). I will use it here to refer to acoustic intensity, as opposed to tonal variation and duration.
and ignoring the existence of different possible mappings for the same words,6 it is perfectly possible to demonstrate that, in general, these rules locate the prominent points of the words in practically the same places as the one proposed by A.7 D.-S. (123-9) propose a mapping system for the language of several rules, that can be schematized by pointing out that the proposed rhythm is essentially iambic, that a short syllable is basically non-prominent (what they call “arsis”) and that a long syllable is basically prominent (what they call “thesis”).8
A. (333), on the other hand, presents the following principles for the location of stress in Greek:9
1. A stress-matrix is constituted by (a) one heavy, or (b) two light syllables.
2. Words (or word-like sequences) longer than a matrix have internal contrasts of stress/non-stress.
3. If the final syllable is heavy, it is stressed.
4. If the final syllable is light, the next preceding matrix is stressed; except that in words of form the final disyllabic matrix may be stressed.
The radical difference between these approaches is not in the location of individual rhythmical prominences, but in the nature of the rhythm they propose. In A.’s theory, as can be inferred from the rules, a stress-matrix is the fundamental component of rhythm, while D.-S. present a feet-mapping system the internal differences of which are based on duration.
While Da’s theory is close to A.’s in his proposal of stress as the basic rhythmical prominence of Greek, it is substantially different from all other approaches, for it links linguistic rhythm with tonal accent.10 Through an analysis of ancient testimony and of some previous formal approaches to Greek rhythm,11 Da. presents a system of rules that associate contextually an aspect of the tonal accent with a rhythmical prominence.12 As formulated by Da. (86) himself, the rules are the following:
6 It will be seen that this decision does not affect the conclusions of the present study.
7 They differ, actually, in their treatment of the succession of short syllables (A. 316-333 and D.-S. 129-135); however, this also does not affect the conclusions of this work.
8
It is symptomatic of the conjectural state of the theory that none of the six (seven, if the two possible ones for the analysis of the cretic ἄμβροτοι are counted separately) segmental analysis they present (pp. 122, 124, 130, 134, 138 and 140) has a trochaic ending.
9
I leave aside the rules for the location of secondary stress and for the location correction in pre-pausal syllables (A. 334).
10 The idea is not original in itself, but the directionality of the dependence certainly is. The authors who have proposed the existence of an accentual foot in Greek (Steriade 1988 and Sauzet 1989, among others; on the problem of accentual calculus from rhythm see D.S. 152-155) have usually tried to derive, with little success, the hypothetical tonal accent from a rhythmical mapping that would formally precede it. Da., on the other hand, presents a rhythm that is derived from the location of the tonal accent.
11 Specifically the ones of Sommerstein (1973) and Golston (1990). 12
1) Circumflex: stress strongly in relation to unmarked syllables in the word with a rise, a break, and a heavy fall in pitch (...).
2) Grave: leave unstressed, or lightly stressed in relation to unmarked syllables with a slight rise in pitch (...).
3) Acute: examine the following syllable; if it is
a) heavy, or prepausal, stress the following syllable heavily with falling pitch.
b) light, or non-existent, stress the acute itself sharply with rising pitch, or with a full contonation if the acute syllable is closed.
Now, to contrast these three positions requires, firstly, grouping them in such a way as to make the selection of a methodology possible. It seems clear that there are two different strategies in these three approaches: the linking of rhythm and tonal accent and the detachment of rhythm and tonal accent.13 A. and D.-S. propose what I shall call an independent rhythmical prominence (I.R.P.), while Da. proposes a dependent rhythmical prominence (D.R.P.).14 Two problems arise immediately when pretending to compare such approaches. The first one, the simplest one also, is that in practice Da.’s rules, combined with the limitation rules of Greek accent15, predict in almost all cases the same prominent places as the rules of I.R.P. authors. In words with a long final syllable in its dictionary form, there are no differences between the selected theories concerning the location of rhythmical prominence. There is one, though, in words ending in a succession of short syllables; however, given the lack of agreement between A. and D.-S. regarding such cases, these are not appropriate for study. Only words of trochaic ending allow opposing both approaches, since, while when they are properispomena, paroxytones or proparoxytones the three theories locate the rhythmical prominence in the same place, when they are oxytones the subordination of the rhythmical component in a D.R.P. approach makes it fall in the final syllable, notwithstanding the fact that it is short.16 The key to contrasting these theories, then, is an analysis of trochaic words in the Greek poetic corpus.
The second problem is that the simplest method for the comparison, an analysis of agreement between linguistic and poetic rhythm with each theory, is a priori useless as methodology. I.R.P. authors build their systems up from metrical evidence, that is, the foundation of their works has the direction metre language. Da., on the other
13 These two possibilities for Greek have been analyzed in detail by D.-S. (206-215). Their conclusion, however, depends on a good number of assumptions they never make explicit (mainly concerning the way we should interpret statistical tendencies).
14 From now onwards, when there is no need to mention them separately, I will refer to A. and D.-S. as “I.R.P. authors” or “I.R.P. approach authors”.
15 See A. (236-9), Sommerstein (1973: 131-2), D.-S. (152-6) and Probert (2006: 60-69).
hand, starts with ancient testimony, and only in a later stage of his work, and even then not extensively, conducts metrical analysis using his theory.17 Nonetheless, the only direct source of evidence we have for the study of Greek rhythm is, precisely, Greek poetry. This implies that I.R.P. authors’ theories are, firstly, unverifiable, given the circularity of corroborating rhythmical rules extracted from metre through metrical analysis, and, secondly, a priori superior in an analysis of agreement to the proposals of a D.R.P. approach, since this perspective does not use the metre in the elaboration of its rules. In other words, the fact that using A.’s rules metrically prominent positions coincide with linguistically prominent position more often that when using Da.’s rules does not mean a thing, theoretically speaking.
It is therefore clear that it is not possible simply to oppose the levels of agreement from trochaic and trochaic ending words that each approach gives, because a priori it is evident that I.R.P. authors will get better results. On the other hand, in the particular case of trochaic words, there is the additional inconvenience of the nature of Greek metres, where the prominent part of feet is in almost all cases a long syllable, and thus a trochaic ending oxytone will never be able to have its linguistic rhythm in agreement with poetic rhythm.18 The mere introduction of these forms in poetry, then, seems problematic. A first test could be, therefore, to check if there is actually an avoidance of them in Greek metres. In order to do this, I have selected a sample of a little bit more than five thousand words from Herodotus’ Historiae,19 the Homeric Hymn to Hermes,20 a sample of almost two thousand trimeters from Sophocles,21 and a sample of around twelve hundred lines from the first twelve books of Iliad.22 In all cases, as it is obvious, I have only taken into account trochaic ending words. Naturally, the base for the comparison is the prose text, where there is no reason to avoid the studied forms. There is, of course, the problem that the foundation of the analysis is somewhat corrupt, since there are important differences in lexicon between Herodotus
17 Cuantitative approaches to Greek metre form Da.’s perspective have been realized, however. See Abritta (2010 and 2013).
18 Except at the ending of each line, given the brevis in longo principle, but only in cases in which the penultimate position can be occupied by a long syllable and the final position is prominent. I do not know of stichic metres that allow this, unless the iambic tetrameter catalectic (see West 1982: 92-3) or the anapestic tetrameter catalectic end in thesis, which does not seems likely.
19 Ed. Legrand (1930-1960).
20 Ed. Càssola (1997). From now onwards, HHHermes. 21
All the trimeters from Antigone, the first five hundred of Oedipus Rex and the first five hundred of Philoctetes; ed. Dain and Mazon (1955-1960). From now onwards, “Sophocles” must be understood as referring to this sample.
22
and the poets, not to say of age and dialect. However, lacking closer evidence, the only option is to relax expectations concerning the numbers we await.23
Table 1 shows the distribution by prosody of trochaic ending words in the analyzed corpora. The results are exactly the opposite of what was expected, and significant enough as to be worthy of trust.24 While in Herodotus there is almost one oxytone trochaic ending word per eight feminine words of the same kind, in the poets this ratio changes to an average of about one per four. There are 95% (Ω=1.9456) more chances of finding masculine trochaics words in the poets than in Herodotus.
Now, while at this point one might consider that the problem is solved, go back to the traditional idea that accent and rhythm do not mix,25 and declare that an I.R.P. approach must be more adequate than a D.R.P. approach, the regularity with which the poets differentiate from Herodotus is suspicious enough as to deserve further examination. In fact, the closest ratio between both types of trochaic ending words is in Sophocles, where iambic rhythm might be favoring masculine prosody. It is true that one would not expect that to happen in hexameter, but previous research26 suggests that in this metre the distribution of accents among line’s feet varies position to position, so maybe at some point in the verse poets may favor a group of words the prominent point of which actually fell in arsis. It is necessary, then, to perform a second test, in order to check if it is true that the accent has no bearing at all on metre. If that were the case, what we would expect (and this is hypothesis zero) is distribution of trochaic ending words to be approximately the same between feminine and masculine words in all positions of the metre. In other words, only shape to be significant in the distribution of words, and not accent.
Results can be seen in tables 2, 3 and 4. Words are counted in the position of their final syllable, and I have only included positions where there is a significant amount of trochaic words. In the three cases it is almost impossible for the distribution to have arisen randomly. In Sophocles, where the chances are higher, it is of less than one in a thousand (p<0.001). In fact, a closer examination suggests there may be intentionality behind this distribution. Note, for example, that the fewer amount of
23 The same problem and a similar solution can be observed in D.-S. (1976: 160-1).
24 The chance of them being product of random variation is less than one in a hundred million. It should be noted that, given the fact that the sample’s (that is, the columns’) numbers are fixed for the purposes of the test, what we would have expected is an equal distribution of the total amounts of masculine and feminine words throughout the corpora. That is not what happens.
25
masculine words is in the sixth arsis, where such an accent is in strong disagreement with iambic rhythm.27 However, before entering such an analysis, it is important to confirm that these results are indeed the product of accentual variation. Note that, up until this point, I have spoken of words of trochaic ending, that is, of words of different shape. In the Iliad’s sample, for example, there are 459 trisyllabic words (331 of the
shape and 128 of the shape ) and 383 tetrasyllabic words (316 of the shape and 83 of the shape ), which, obviously, cannot occupy the same positions on the line. This suggests that the test’s numbers could be the result of something more than accent variation, provided that the alternation feminine / masculine were strongly associated with variation in word shape.28
Fortunately, the solution is very simple: it is enough to conduct the same test already done but now exclusively in disyllabic trochaic words. In this way, it is possible to eliminate the variable of shape, and therefore the impossibility of placing certain words in certain parts of the metre. The results can be seen in tables 5, 6 and 7. There is now a new problem, because the samples produce very different numbers. The possibility of a random distribution goes from 53% in the HHHermes to a less than 0.1% in Ilíad, with a 4.89% chance in Sophocles. Two out of three texts are significant enough as to seriously consider that accent is still an important factor in the distribution of words. What is more, the extreme cases are from the same metre, so metre is a variable that is also susceptible of being excluded as an explanation.
It should be noticed that the main reason why the HHHermes seems to show a random distribution of disyllabic trochaic is the reduction of feminine to masculine ratios, particularly in the third and sixth feet. The reduced range from the largest to the minor ratio in table 6 is what is causing the abrupt rising in the chance of random distribution.29 However, it was certainly predictable, considering carefully the change in values, that the ratios were going to diminish significantly. In the first place, as has already been observed (see n. 28), because the more syllables a trochaic ending word has the larger the possibility of it being feminine, and therefore, when eliminating all
27 Note also that of the nine cases four have grave marks, and four of the remaining five are followed by a clitic, which might suggests that there is a displacement of the accent to the last position.
28 As indeed it is. Pearson’s coefficient between the amount of syllables of a trochaic ending word and the ratio of feminine to masculine words is =0.91, so there is an almost perfect negative correlation between these variables. In other words, the more syllables a trochaic ending word has, the larger the possibility of it being feminine (actually, in the studied sample, when the word is tetrasyllabic, there is a chance of a 100%).
29
non-disyllabic words, the larger cut was going to be in the side of feminine words. Secondly, because by selecting only the disyllabic words the important type of post-acute emphasis (Da.’s rule 3.a.) is eliminated from the count, and it represents almost a 43% of all feminine words in the tables of hexameter and almost a 29% in trimeter. Again, as in the first case, the discarding of this type falls exclusively in the feminine line of the table, since there are no masculine post-acute emphasis in trochaic ending words. While it is true that by restricting the sample to disyllables a proportion of all accent types is eliminated, in this case all instances are removed. Therefore, the diminishing of the ratios was a priori inevitable.
It is necessary, because of this, to introduce some additional variable to disambiguate the results. Given the fact that shape has been eliminated, and not being possible to select types of accents, the only possibility left is syllabic structure.30 Given that a circumflex cannot fall in closed syllables with a short vowel, it can be deduced that the ratio between feminine and masculine words will be lower in trochaic disyllabic words with penultimate closed syllable than in those with a long vowel in penultimate position. The case is identical to the one of the elimination of post-acute emphasis, already explained. The results of this analysis in the HHHermes can be observed in tables 8 and 9. There is barely the possibility that the distribution is product of random variation in long-vowel penultimate words (p=0.03) and it has been greatly reduced in the other sample (p=0.24). The same happens in Iliad and in Sophocles31 with, respectively, a chance of 1.44% and of 0.93% of random variation in long-vowel penultimate words and a chance of 8.47% and of 10.92% in closed penultimate syllables. In the first case, the results suggests that accent is a determining factor in the placement of words; in the second one, while the results are much weaker, they are on the verge of statistical significance. This implies that the numbers of the HHHermes are only, for the purposes of this study, a marginal aberration.32
30 There are, actually, three more variable to take into account: ellipsis, clisis and word class. The elimination of the first two, however, does not convey a change in the results of Sophocles and Iliad, as can be checked in my blog (see n. 1), and there are only 11 trochaic accented clitics (see n. to table 1) in the HHHermes, which naturally are not enough as to modify the outcome of the test (it is interesting to note, however, that the elimination of those clitics reduces the chance of random variation to 36%). The third variable, word class, deserves more extended treatment, which I plan to give elsewhere.
31
I do not include the tables; they can be checked in my blog (see n. 1).
In the limits of this dissertation, I cannot dive into a more detailed analysis of the data. It seems clear that the tests performed are not fully conclusive, though they must be seriously taken into account, given the fact that almost all of them produced the same result. While at some points in this research the idea of abandoning the notion of a linguistic rhythm associated to tonal accent was appealing, the majority of outcomes suggest that it is probable that tonal accent is in someway linked to poetic rhythm. This implies, naturally, that it is also associated to linguistic rhythm, which is what Da. proposes, against the position of I.R.P. authors. It seems clear that the presented data can be better explained from a D.R.P. approach than from one that does not accept any link between accent and rhythm. Now, in this sense, and as was pointed out in the introduction, the goal of this work was never to provide a definitive answer on the problem of the natural rhythm of Greek, but to show through a comparison of theories that it is necessary to rethink the relation between metre and language. The evidence presented, that suggests that Greek tonal accent affects the placement of words in metre is, without a doubt, an important step in this direction.
Abritta, A. (2010) “Sobre la posibilidad de un análisis coral en Ilíada 53-305”,
Anales de Filología Clásica 23, 1-62.
Abritta, A. (2013) “Hacia una nueva musicalidad de la tragedia griega”, in Sapere et al. (eds.), Nuevas aproximaciones a la antigüedad grecolatina, vol. I, Buenos Aires: Rhesis, 9-22.
Allen, T. W. (1931) Homeri Ilias, Oxford: Clarendon Press.
Allen, W. S. (1973) Accent and Rhythm, Cambridge: University Press.
Càssola. F. (1997) Inni Omerici. Milano: Fondazione Lorenzo Valla. 1º ed. 1975.
A. Dain and P. Mazon (1955-1960) Sophocle, 3 vols., Paris: Les Belles Lettres [repr. 1967-8 (1st rev. edn.)]
David, A. P. (2006) The Dance of the Muses. Choral Theory and Ancient Greek Poetics. Oxford: University Press.
Devine, A.M. – Stephens, L. D. (1994) The Prosody of Greek Speech. New York and Oxford: University Press.
Devine, A.M. y Stephens, L. D. (1976) “The Homeric Hexameter and a Basic Principle of Metrical Theory”, Classical Philology 71, 141-163.
Golston, C. (1990) “Floating H (and L*) Tones in Ancient Greek”, Proceedings of the Arizona Phonology Conference 3, Tucson: University of Arizona Linguistics Department.
Legrand, Ph.-E. (1930-1954) Hérodote. Histoires, 9 vols., Paris: Les Belles Lettres (repr. 1963-1970).
Probert, P. (2006) Ancient Greek Accentuation. Synchronic Patterns, Frequency Effects and Prehistory, Oxford: University Press.
Sauzet, Patrick (1989) “L’accent du grec ancien et les relations entre structure métrique et representation autosegmentale”, Langages 95, 81-113.
Silva Barris, J. (2011) Metre and Rhythm in Greek Verse, Wien: OAW.
Sommerstein, A. H. (1973) The Sound Pattern of Ancient Greek, Oxford: Basil Blackwell.
Steriade, Donca (1988) “Greek Accent: A Case for Preserving Structure”,
Linguistic Inquiry 19, 271-314.
Heródoto Ilíada HHHermes Sófocles Total
Femeninas 1039 1564 815 1514 4932
Masculinas 132 345 197 425 1099
Total 1171 1909 1012 1939 6031
Razón F/M 7,871 4,533 4,137 3,562 4,488
Table 1. Amount of words by author and by accent type.33
A2 A3 A4 A5 A6 Total
Masculinas 90 84 186 56 9 425
Femeninas 385 239 676 138 77 1515
Total 475 323 862 194 86 1940
Razón F/M 4,278 2,845 3,634 2,464 8,556 3,565
Table 2. Amount of trochaic ending words by accent type and by position in Sophocles.
A1a A2a A3a A5a A6 Total
Masculinas 40 13 45 53 46 197
Femeninas 87 26 308 163 231 815
Total 127 39 353 216 277 1012
Razón F/M 2,175 2,000 6,844 3,075 5,022 4,137
Table 3. Amount of trochaic ending words by accent type and by position in the HHHermes.
A1a A2a A3a A5a A6 Total
Masculinas 85 21 58 121 45 330
Femeninas 142 80 516 311 452 1501
Total 227 101 574 432 497 1831
Razón F/M 1,671 3,810 8,897 2,570 10,044 4,548
Table 4. Amount of trochaic ending words by accent type and by position in Iliad.
A2 A3 A4 A5 A6 Total
Masculinas 71 69 165 47 6 358
Femeninas 144 98 370 73 21 706
Total 215 167 535 120 27 1064
Razón F/M 2,028 1,420 2,242 1,553 3,5 1,972
Table 5. Amount of trochaic disyllabic words by accent type and by position in Sophocles.
33
A1a A2a A3a A5a A6 Total
Masculinas 40 12 40 51 42 185
Femeninas 87 18 75 79 59 318
Total 127 30 115 130 101 503
Razón F/M 2,175 1,500 1,875 1,549 1,405 1,719
Table 6. Amount of trochaic disyllabic words by accent type and by position in the
HHHermes.
A1a A2a A3a A5a A6 Total
Masculinas 92 21 53 123 37 326
Femeninas 142 45 153 154 72 566
Total 234 66 206 277 109 892
Razón F/M 1,543 2,143 2,887 1,252 1,946 1,736
Table 7. Amount of trochaic disyllabic words by accent type and by position in Iliad.
A1a A2a A3a A5a A6 Total
Masculinas 25 9 24 34 36 128
Femeninas 54 12 52 50 30 198
Total 79 21 76 84 66 326
Razón F/M 2,16 1,333 2,167 1,471 0,833 1,547
Table 8. Amount of trochaic disyllabic words with long vowel in penultimate syllable in the
HHHermes.
A1a A2a A3a A5a A6 Total
Masculinas 15 3 16 17 6 57
Femeninas 33 6 23 29 29 120
Total 48 9 39 46 35 177
Razón F/M 2,2 2,000 1,438 1,706 4,833 2,105
Table 9. Amount of trochaic disyllabic words with closed penultimate syllable in the
