1995.10.09, Devine/Stephens, Prosody of Greek Speech

Attempts to reconstruct the phonology of Ancient Greek, such as W. Sidney Allen’s classic Vox Graeca (Cambridge, 1968), have generally tried to pinpoint the exact phonetics of the Greek sound inventory (e.g. what kind of e an eta was). DS’s ¹ goal is different: their aim is to reconstruct the sound of Greek prosody. ² To this end, they employ a “rather different methodology”, which they describe as “a sort of archaeological laboratory phonology…. The premise of the method is that, although different languages have different prosodic systems, prosodic structure does not by and large vary crosslinguistically in a random, unlimited and unpredictable fashion” (vii). This method has in fact been employed and refined by DS in articles and books stretching back a good twenty years, ³ many of which adumbrate ideas that are here for the first time integrated into a comprehensive account of Greek prosody. The result is nothing short of a masterpiece. The authors proceed systematically from the smallest prosodic unit (the syllable) on up through the largest (the major phrase and utterance), devoting a chapter to each level of the prosodic hierarchy. The chapters typically adhere to the following format: First, DS provide background information for the prosodic unit in question, including extremely useful ⁴ and copious typological data from a wide range of languages, and results from experimental studies in acoustics and phonetics. This sets the cross-linguistic and theoretical context for an examination of the relevant philological data: “In the second stage, we proceeded to devise ways of testing the Greek texts, inscriptions and musical remains for specific prosodic properties in the general context of the background material we had assembled, taking care to ensure the statistical significance of the results obtained” (viii). The marriage of theory and philology that DS strive for, where both disciplines are given roughly equal weight, is a significant advance over previous studies, and allows them—so they claim—to give an account of Greek prosody which is “in some respects almost as complete as available accounts of the prosody of many languages that can be investigated in vivo in the laboratory” (ibid.).

Due to the scope and complexity of the book, the following summary of the contents is necessarily superficial; after that, however, I shall examine in more detail some of DS’s findings and methods. Chapter 1, “The Physiology of Prosody” (3-20), is an introduction to the anatomical basis of speech production, covering neurology, respiration, phonation, and the articulation of speech sounds. Chapter 2, “The Syllable” (21-84), concerns primarily the syllabification of consonant clusters, especially muta cum liquida (stop plus resonant, e.g., kr, gl, dm). The practice of comedy, where muta cum liquida is tautosyllabic (in contrast to tragedy and formal passages in comedy), reflects “the norm in colloquial Attic” (35). This is partially at variance with syllable divisions found in inscriptions, but these DS claim are due to the different syllabifications associated with the artificial slowing-down of speech that was used as a strategy “to align it with what is being written” (38). DS’s conclusions here are necessarily speculative, but linguists who have used the orthographic practices as evidence for particular phonological properties of Greek should take careful note and proceed more cautiously than has perhaps been done in the past. ⁵ In the remainder of Chapter 2, DS disprove the old claim that Greek poetry was sensitive to durational distinctions at the submoraic and segmental level, finding no statistical support for the metrical relevance of syllable onset length, intrinsic vowel and consonant duration (differences in the length of the same vowel/consonant in identical contexts), or contextual vowel duration (differences in the length of the same vowel/consonant in different contexts), and only limited evidence for a metrical distinction between ξ’ξ and C’ syllables and between heavy and superheavy syllables. If these findings are upheld, then they represent a very important contribution to the study of Greek metrics, upsetting as they do a number of very old and tenacious theories. ⁶

Chapter 3, “Rhythm” (85-156), presents and evaluates the data that can be, and have been, taken as evidence for forming conclusions about the rhythm of Greek speech at the level of the word. These sources of evidence fall into three types: phonological facts about the language, metrical texts, and musical settings. Among other things, DS posit an underlying rhythmic template onto which strings of syllables were mapped consisting of a long rhythmic unit followed by a short, and suggest various de- and re-footing rules that will be of interest to theoretical phonologists, particularly the claim that there was such an organizational principle independent of stress assignment and syllable structure. The musical testimony is treated in greater detail in Chapter 4, “Pitch” (157-194), which concerns itself with reconstructing the surface pitch patterns of Greek. In Hellenistic times, mismatches between accent and melody in songs are rare; unfortunately we have only conflicting evidence for the situation in the pre-Hellenistic age (166-7). The Hellenistic settings agree in showing a pre-acute rise and post-acute fall in musical pitch for medially-accented polysyllables; the interval of the fall is on average greater than the interval of the rise. This suggests a Mid-High-Low contour for such words (183f.). A particularly interesting finding is that the Delphic hymns show evidence for tonal effects of aspirated stops on following vowels. In a number of languages, the fundamental frequency of the beginning of a vowel following a voiceless aspirate is higher than after a voiceless unaspirated stop or a voiced stop, and it turns out indeed that in polysyllables of the type φιλόμαχον, whose first syllable is the preaccentual syllable and begins with a voiceless aspirate, the musical setting of the initial syllable is never more than a semitone lower than the pitch of the following stressed syllable (the peak) (179-80). It is nothing short of astonishing that such subtle information about ancient Greek phonetics is available to us; one merely needs to know where to look and how to devise the proper tests. ⁷

The results of Chapters 3 and 4 are integrated into a comprehensive theory of the prosody of the word in Chapter 5, “Word Prosody” (195-223). The task that awaits the researcher here is to determine “what were the phonetic properties of rhythmically prominent syllables, in particular whether they were stressed; whether the system of rhythm correlates or interacts with that of the word accent, and if not, how the two systems could coexist without tending to neutralize each other; and what were the consequences for the Greek word prosodic system of the emergence of a stress accent in later Greek” (195). Most of this chapter is devoted to examining possible typological parallels to their analysis of Greek: a language with a pitch-accent coexisting with a foot-based system of rhythm independent of the pitch-accent. Japanese, they claim, provides such a parallel: it is also a language with a pitch-accent and a quantitative rhythm independent of the pitch-accent (213).

The remainder of the book leaves behind the realm of individual words and considers the prosody of words and phrases in their syntactic contexts. Chapter 6, “Connected Speech” (224-284), concerns sandhi-effects at word-boundary such as resyllabification, crasis, hiatus, elision, and prodelision. DS contribute many important observations on the syntactic domains of these phenomena that considerably expand our knowledge of the interaction between phonology and syntax in the language; I shall give a sample. The domain of both coda-to-onset and onset-to-coda resyllabification ⁸ is less restricted in the freer styles of Euripides than in the stricter styles; this indicates that in colloquial speech, resyllabification was not limited to the same tight syntactically cohesive phrases as in more deliberate speech (240f.). The rate of R(-gemination after a short vowel (as inscriptionally ARGEMATA RRUMOIS IG I ².314.40) was found to be highest in more cohesive syntactic structures, especially those consisting of a prepositive nonlexical plus a lexical; and, as it is strongly avoided in tragic lyric, it would appear to have been a feature of colloquial Attic deemed more appropriate for comedy (249f.). The domain of elision is more difficult to establish since verse and inscriptions contradict each other: in prose inscriptions, for instance, elision of full lexical words is rare, whereas in verse there is no such constraint. For the situation in verse, DS note inter alia an attractive parallel from Kalmyk, where the domain of vowel harmony, normally the word, is extended artificially to the whole line in poetry. DS conclude (264) that elision was not confined to the minor phrase in fluent speech. Text editors should take note of the implications of these observations for evaluating emendations of various passages. ⁹

Chapter 7, “The Appositive Group” (285-374), continues the discussion of the interface of phonology and syntax by investigating the prosodic properties of the appositive group (their term for the prosodic domain between the word and the minor phrase, consisting of a pre- or postpositive and a full lexical item). Apparent violations of Porson’s, Knox’s, Hermann’s, and other bridges, as is well-known, typically involve a prepositive and a lexical word; as a prepositive was prosodically subordinate to the following host, the two words formed a single phonological word and are therefore not true exceptions to these bridges. ¹⁰ DS examine the occurrences of all types of mono- and disyllabic prepositives at bridges, and combining it with evidence from inscriptions argue for the existence in Greek of the appositive group as a prosodic domain. The fact that certain kinds of prepositives are admitted before metrical bridges more readily in comedy than in tragedy is taken as evidence that in deliberate speech these appositives did not form a prosodic unit with the following word, but did in more relaxed fluent speech; this has parallels in other languages such as Madurese (310). DS point out that the formation of appositive groups is not only conditioned by the non-lexical in question (for which they are able to set up a categorial hierarchy, p. 340), but is also sensitive to syntactic characteristics of the potential host material, such as branchingness. For instance, in the early plays of Euripides, a heavy monosyllabic preposition may stand at Porson’s bridge if it is in the discontinuous or head-interrupted configuration [N π ν] _PP (e.g. νερτέρων ἐκ δωμάτων Alc. 1073), but not if it flanks a branching NP, i.e. [P NP] _PP (e.g. ἐκ τῶν φίλων H.F. 1234). Thus “head interrupted structure provides an easier environment for subordination than the head flanking structure in branching prepositional phrases” (335); what DS posit here would accord well with many other languages, where clitics cannot flank a branching phrase. Among the myriad topics covered in the rest of the chapter are the prosodic status of proclitics: DS argue rather strongly, based especially on the musical evidence, that proclitics were not atonic, contra most traditional theories on the subject.

A similar set of metrical, musical, and inscriptional inquiries form the material of Chapter 8, “The Minor Phrase” (376-408), and Chapter 9, “The Major Phrase and Utterance” (409-455), which concern the final—and highest—levels of prosodic organization. DS show, continuing some of the investigations of Chapter 6, that onset-to-coda resyllabification of clusters consisting of s plus stop is constrained; it was licensed within the minor phrase (e.g. between verb and another part of the verb phrase) but usually not interphrasally (between words belonging to two different constituents); in those cases of interphrasal resyllabification, one of the two phrases was most commonly non-branching (383), a fact fully expected from linguistic universals. The examination of both metrical and inscriptional evidence shows that verbs were phrased with the subject only by default, that is, only when the verb was not modified by other material (392). Chapter 8 also considers what phonetic properties that minor phrase had; this is best done again by examining the musical settings, which show, not surprisingly, that the first peak in the phrase is pitched highest, followed by a gradual decline in pitch (downtrend or catathesis) with secondary rises for later peaks within the phrase (408). Chapter 9 further examines catathesis, this time over the domain of whole sentences, in its cross-linguistic context. Chapter 10, “Topic and Focus” (456-497), concerns the phonetic realization of these syntactic categories in Greek, and allied phenomena. The Delphic hymns show pitch rise correlating with emphatic focus (479). A particularly welcome section concerns discontinuous phrases in Greek and the interaction of discontinuity and resyllabification; DS’s results here should be extended by other researchers to Latin and the other ancient Indo-European languages (480f.). This final chapter is followed by a Bibliography (498-562) and Index (563-565). ¹¹

The careful, systematic, and penetrating evaluation of reams of philological data in the light of a highly sophisticated knowledge of phonetics and theoretical linguistics is the greatest contribution that this book has to offer. One is hard-pressed to think of questions that they did not ask. Where their treatment is shakier is in the handling of morphology and diachrony. As far as I can tell, in their stated attempt to reconstruct the prosody of Greek, DS never actually define what they mean by “Greek”—an omission that leads to some infelicities that I will talk about below. Most of the book, it turns out, is concerned with 5th-century Attic, but there is an uncomfortably large inclusion of data from other dialects and from other time periods that is all lumped together with the Attic data; the whole tends then to get analyzed, rather monolithically, as of a piece. DS are of course fully aware that “Greek” refers to any of several spatially and temporally heterogeneous entities, but when treating more than one variety of Greek, one must be careful to qualify the domain over which one’s conclusions are supposed to obtain. ¹²

An instance of this is their analysis of the syllabification of muta cum liquida in Chapter 2. I shall treat this in some detail, not to be picky, but because an account that is more sensitive to chronological layering has long been needed for this particular issue, and leads to some unexpected findings. When they arrive at their conclusion that tautosyllabic muta cum liquida was the norm for colloquial Attic, DS mention that variation from this norm “is due to the influence of non-Attic verse; in Homer, for instance, internal muta cum liquida is usually heterosyllabic, with exceptions involving liquids rather than nasals” (35). This is the only hint we get from DS that there was a chronological difference in the way muta cum liquida was syllabified; they conspicuously omit any mention of correptio Attica (the shift in the scansion of muta cum liquida from Epic to Classical Attic). This lack of attention to chronology, while not damaging to their central claims, leads to a rather haphazard account of certain sound laws and other material. On pp. 39-41, DS invoke a body of “linguistic evidence … almost all [of which] indicates unequivocally that the syllable divisions of normal speech are just those found in the metrical evidence” (39): the rhythm rule for formation of the comparative and superlative; three sound laws (Wheeler’s Law, the neuter nouns law, and Vendryes’s Law); and the reduplication of the perfect. As far as 5th-century Athens goes, Wheeler’s Law (whereby originally oxytone dactyl-shaped words or word-final sequences retract the accent to the penult, e.g. ποικίλος , cp. Sanskrit pe-alá- decorated’) and the neuter nouns law (whereby neuters in –ίον retracted the accent onto the antepenult if it was light, e.g. PUKTίον but θύριον) are irrelevant, since both were Common Greek developments.

The rhythm rule and the reduplication rule for perfects, it turns out, are rather more difficult evidence than at first appearance, and none of the difficulties has, to my knowledge, been sufficiently discussed in the literature. The rhythm rule states that the comparative and superlative endings are –ότερος after a heavy syllable, but –ώτερος after a light. DS note that for this rule, muta cum liquida is generally heterosyllabic, especially with core vocabulary (μακρότερος πυκνότερος but that “some marginal forms show variation across the corpus of Greek literary records”, for which they cite βαρυποτμώτατος Eur. Phoen. 1345 vs. βαρυποτμότατος Plu. TG 5, ἐρυθρώτερος Pl. Tim. 83b vs. ἐρυθρότερος Anaxandr. 22, and εὐτεκνώτατος Eur. Hec. 581 vs. εὐτεκνότατος Diod. Sic. 4.74. DS simply list these data without commentary, and it is not at all clear what conclusions are supposed to be drawn. First of all, when was the rule actually operative? Forms like μακρότερος πυκνότερος fly in the face of the view that muta cum liquida was tautosyllabic in 5th-century Attic, if the forms are at all relevant to 5th-century Attic; the fact that they are “core vocabulary” means that they have probably been lexicalized and preserve evidence for earlier phonological features (μακρότερος, for instance, is already in Homer and scans with heterosyllabic muta cum liquida as is normal in epic). The “marginal forms” are the best evidence for synchronic rules, as the comparative and superlatives of uncommon words are going to be created on the spot using productive phonology and morphology. As these “marginal forms” show variation, we have to ask what the variation means. One first notices a difference in chronology: βαρυποτμώτατος ἐρυθρώτερος εὐτεκνώτατος are between two and five centuries older than βαρυποτμότατος ἐρυθρότερος εὐτεκνότατος. ¹³ This makes it look like Greek started out with heterosyllabic muta cum liquida (μακρότερος Homer), then tautosyllabic after correptio Attica (βαρυποτμώτατος), and later heterosyllabic again (βαρυποτμότατος). ¹⁴ If so, then the 5th-century forms like βαρυποτμώτατος may be evidence for tautosyllabic muta cum liquida, assuming the rhythm rule was operative at the time. ¹⁵ However, these forms must be considered together with other unexpected comparatives and superlatives, and they are more frequent in the manuscripts than the handbooks tell us. Good manuscripts of Plato transmit ἐμμετρώτατος at Leg. 11.926a, and of other authors forms such as ἐλευθερότερος ἀνιῤώτατος φιλοκινδυνώτατος εὐδοξώτατος ὁμοιώτατος ἰσχπρώτατος; but all of these have been deemed to be faulty (see R. Kühner and F. Blass, Ausführliche Grammatik der griechischen Sprache ³ [Hannover, 1890] I ¹ 558-9). Have these unsavory variants been rejected on good philological grounds, or simply because they do not agree with a prescribed rule in the grammar-books? ¹⁶

Evidence for tautosyllabic muta cum liquida culled from the reduplication of the perfect—a subject which has received much attention in recent theoretical literature ¹⁷—should also not be invoked without some caveats. As is well known, the reduplicating syllable consists of the first consonant of the root plus -E- if the resultant reduplicating syllable would be light; otherwise the reduplicating syllable is simply ἐ (e.g. παίω πέ but πταίω ἔ). “Lighter”muta cum liquida clusters are tautosyllabic, to judge by κλέπτω κέ, τρέφω τέ, but “heavier”muta cum liquida clusters show both patterns (e.g. βλαστάνω βε and ἐBLA/STHKA). ¹⁸ Other consonant clusters also show both patterns: κτάομαι κέ but Ionic ἔ, πτήσσω πε Homer but Attic ἔ. DS conclude from this that “[s]uch forms are evidence that resyllabification failed across the reduplication seam, not that these clusters were regularly assigned to the onset word medially in prehistoric Greek.” If this be the explanation for κέ etc., it then becomes ad hoc to conclude that κέκλοφα etc. is due rather to tautosyllabicity of the cluster. If in the κέκτημαι cases, resyllabification of the cluster failed across the reduplication seam (certainly a possibility), then that is at variance with the scansion of these clusters in poetry (as heterosyllabic; see p. 32). It is also possible, though, that the κέκτημαι cases reflect a purely morphological, and not a morphophonological, process; some speakers appear to have overgeneralized Ce -reduplication (a morphological rule) at the expense of simple e- reduplication. This would have happened by their having failed to acquire the phonological rule that originally blocked application of Ce- reduplication before particular consonant clusters. Under this explanation of κέκτημαι etc., we do not run into conflict with the metrical evidence, as happened with the syllabification account—and the metrical evidence is after all supposed to reflect the actual Attic syllabification more closely than anything else, according to DS’s premises.

It appears that only Vendryes’s Law (whereby properispomena retract the accent to the antepenult if the antepenult is light, e.g. ἕτοιμος ) unequivocally supports DS’s position, as it is a specifically Attic rule, and apparently a fairly recent one (see p. 102). Forms like ἄγροικος ἄχρειος from ἀγροῖκος ἀχρεῖος do show that muta cum liquida, at the time Vendryes’s Law operated, was tautosyllabic in Attic. As I noted at the outset of this discussion, none of the problems with the other evidence vitiates DS’s claims; I merely wish to point out the delicacy with which such evidence, varying both chronologically and dialectally, needs to be handled.

Another unclarity that one encounters passim in PGS, and that I have already alluded to, is whether comedy is supposed to be the best witness to Attic pronunciation writ large, or just to colloquial, fluent, and rapid Attic pronunciation. Sometimes DS seem to be saying the former, as with the muta cum liquida cases, at other times the latter. If the former is their opinion, then by rights comedy should always reflect Attic phonetics better than tragedy; but a number of times it is rather tragedy that, by virtue of its stricter metrical rules, is claimed to evince particular prosodic phenomena most clearly. An example is Porson’s Law, which prohibits a break between the first arsis and the following longum if the arsis is long (so | – # – u x || is not allowed, where # is word-boundary). DS support the claim that these syllables must have been too long in speech to be able to be mapped onto an arsis, and they argue that this extra duration is due to prepausal lengthening of the same sort that accounts for brevis in longo, and that is independently evidenced by musical settings. So far so good; however, if the contextual lengthening of long syllables was a phonetic fact of colloquial Greek, then it would be expected that comedy should have the strictest constraints on the mapping of such overlong syllables, not tragedy and the iambographers; yet Porson’s Law is absent from comedy. To explain this distribution, DS claim that “it is probable that the constraint against full word boundary following a subordinated syllable is relaxed in the conversational speech styles on which the dialogue of comedy is based” (132, cp. also 280). ¹⁹ This sounds fine on its own, but has the unfortunate result of contradicting the earlier claim that tragedy was not an accurate reflection of 5th-century Attic muta-cum-liquida syllabification. The only way I can see to salvage the argument is to say that, as in some other languages, muta cum liquida in Greek was heterosyllabic in deliberate speech (cp. note 5 above) but tautosyllabic in fast speech; then tragedy, rather than reflecting archaic or dialectal pronunciation, would mirror the syllabification in careful, and not rapid, speech. Then one could say that comedy represents on all counts more rapid speech. But this approach has problems of its own; Euripidean superlatives like βαρυποτμώτατος and εὐτεκνώτατος could not then represent both the heterosyllabic muta cum liquida of careful speech and evidence of tautosyllabic muta cum liquida at the same time, unless one wanted to claim an artificiality of his morphology and phonology rivaled only by Homer’s.

My only other piece of criticism concerns the presentational style of the work. It is regrettable that a book of such great interdisciplinary interest and importance is so difficult to read. The prose is dense and often telegraphic; examples of the various phenomena are frequently given without commentary, leaving the reader to figure out how they exemplify what. DS assume a rather vast knowledge of both linguistics and Greek metrics on the part of the reader; I fear that its accessibility both to linguists who lack a strong Classical training, ²⁰ and to Classicists lacking linguistic training, will be impaired because of this. Instead of globally referring the reader at the outset to various standard reference works for definitions and explanations of the technical terms (x), it would have been far preferable to have provided a glossary, or to have at least systematically defined such terms as they appeared (which is only occasionally done). The long lists of cross-linguistic data, one of the book’s major strengths, are not always fully integrated into the surrounding discussion, and often not even relevant; at times DS seem to indulge in sharing such information just out of the sheer pleasure of sharing it. ²¹ I personally enjoy no-holds-barred narration of scientific and cross-linguistic trivia (in the word’s most non-deprecatory sense), for it is all fascinating to me; but it can be distracting and make the import of the broader context-setting hazy. The main conclusions of each section are not sufficiently flagged or highlighted, often buried in a wealth of surrounding detail; chapter summaries are sometimes given, sometimes not.

In spite of these few negative remarks and niggling over aspects of the methodology, I must emphasize that I enjoyed reading this book more than almost any other in Linguistics or the Classics in the past several years, and it is a spectacular piece of scholarship. PGS should be read by every Classicist even remotely concerned with Ancient Greek as the living entity it once was; an initial investment of stamina will yield great rewards. The linguistic and experimental background should be tremendously eye-opening and mind-broadening, and the amassing of all this material in one place is a remarkable feat, for which we should be very grateful. It is worthwhile for Classicists to know, for instance, that a phenomenon exactly comparable to verse-final brevis in longo is known from Hausa verse; both the Greek and the Hausa facts are but two examples of a very common cross-linguistic tendency for word- and phrase-final syllables to be generally weaker than word-internally (79f.). PGS should likewise be read by every historical linguist; it is a model for how to glean detailed prosodic and structural information from the records of a dead language. Although the details of surface phonetic implementations may appear at first glance to be inconsequential for linguistic theory, it is precisely this level which encodes an enormous wealth of information about the underlying machinery of a language, and it is this level which is transmitted from generation to generation. The efforts that have culminated in this volume provide a nearly bottomless source of real advances and ideas that should spawn years of fruitful research.