Ancient Greek Pitch Accent, Not Stress *

Ancient Greek accent involves both stress and tone (Allen 1973, Steriade 1988). We present here a new piece of evidence for the details of the tonal part, based on a lowering process whereby a H tone on the final TBU of a word lowers if the word is followed by another tonic word: H...H → L...H. This is essentially what is reported for Rimi, a Tanzanian Bantu language, where HH loses the first of its H tones (Olson 1964, ‘Anti-Meeussen’s Rule’). If correct, this motivates three tonal classes in Ancient Greek: HL* in words like basiléˈàà ‘king.acc’ (Sauzet 1989, Golston 1990) and two additional classes, H*L in basiˈléù ‘king.voc’ and H in basiˈleús ‘king.nom’. 


Introduction
A long line of traditional and theoretical work has made it increasingly clear that Ancient Greek had a mixed accentual system: the location of the accented syllable is determined by a metrical procedure, which counts syllables, is sensitive to syllable weight rather than number of moras, and builds metrical constituents.Later rules interpret metrical prominence tonally.(Steriade 1988: 271) We present here additional evidence for pitch accent in the language.Traditionally, Ancient Greek accent is categorized five ways, depending on the orthographic accentuation, as shown in below. (1) Accentuation and nomenclature acute on the ultima πολιτικός po.lii.ti.kós oxytone acute on the penult πολίτης po.lií.tɛɛsparoxytone acute on the antepenult ἄνθρωπος án.t h rɔɔ.pos proparoxytone circumflex on the ultima πολιτικοῦ po.lii.ti.kóò perispomenon circumflex on the penult πολῖται po.líì.taiproperispomenon A natural interpretation is that the accent regularly marks the stressed syllable, but this is not the case.Allen shows clearly that the pitch peak does not always align with the stress peak: 'the statistical trend is for the acute [H tone] to occur in weak position, whereas the strong position tends to coincide with the circumflex or the syllable following an acute... with a falling melodic pattern in either case ' (1973: 262).Words in Ancient Greek had either 'recessive' accent, where the accent falls as early in the word as the above possibilities allow, or non-recessive accent, where the accent consistently falls on the ultima or penult.The frequent dissociation of high pitch and stress in Greek leads Sauzet 1989 to treat the recessive accent as HL*, where the L tone attaches to main stress (*) and the H tone attaches to the tone-bearing unit preceding it, as shown below for ἄξιος 'worthy'.
( Golston & Paulsen discussion.) We assume here that there are actually three tonal classes in Ancient Greek.In addition to HL* (2), we find surface classes of H and HL*, the latter being especially common on words that have lost a stem-vowel (traditionally 'contract-vowels').The lost vowel is not synchronically recoverable, so the surface H*L accent must be part of the grammar.
( We pursue the details of these accentual classes in other work in preparation.For the present, we focus on words with accent on the final vocalic mora (traditional oxytones) and what transpires when such words Golston & Paulsen are followed by other accented words in the postlexical phonology.

Word-final change of acute to grave
Ancient Greek grammarians note that oxytones change from orthographic acute (´) to grave (ˋ) when immediately followed by another accented word in the phrase; perispomenon, paroxytone, and the rest have no such change (see Probert 2003).Our evidence for a tonal interpretation of Greek accent involves this change of word-final acute to grave.Word-initial acute (6a) and word-medial acute (6b) appear unchanged regardless of where the word occurs in its phrase.Word-final acutes, on the other hand, only surface as such phrase-finally (6c) or preceding an enclitic (6d); otherwise, they appear as graves (`), no matter how many syllables intervene between the two accents (6e-j): (

Acute → grave as destressing
If the change of word-final acute to grave involves destressing, it is highly unusual in the languages of the world, as it targets stress when followed by another stress at an any distance (6e-j).Destressing is usually due to the avoidance of stress clash (Selkirk 1984, Nespor & Vogel 1989), which targets only adjacent syllables, such that we might expect (6e), clash avoidance between stresses on adjacent syllables across words, but not (6f-j), which are increasingly unlikely candidates for stress clash as the number of syllables intervening between the two accented syllables increases.
Another possibility is that some kind of Rhythm Rule (Liberman & Prince 1977) is responsible for the change in accentuation.In English, we see àbstráct árt realized as ábstràct árt, and this can happen even to stresses that are not adjacent, as we see when Sànta Mónica shifts it primary stress leftwards in Sánta Mònica Fréeway.But this should shift stress, not delete it, so that we would find politikós pánt h ɛɛr realized as, e.g., polítikos pánt h ɛɛr, with accent shifted rather than lost.
Gussenhoven (1991) argues that the Rhythm Rule is actually accent deletion in English rather than stress retraction, that an accent deletes before a following accent: This is a more promising approach to the Greek facts, since the grave may well represent no tone at all instead of a lowered tone.Still, it lacks some detail when applied to Greek, where only word-final H tones delete/lower.Why should this be the case?For that we need a more fully articulated tonal analysis.

Acute → grave as lowering
As is well known, the Obligatory Contour Principle (OCP, Leben 1973) prohibits tier-adjacent identical tones-and these need not be adjacent in terms of syllables (see Myers 1997) When tonic words come together, then, there are nine tonal possibilities (9-11).Word-order is extremely free in Ancient Greek (Devine & Stephens 2000) due to phonological movement (Agbayani & Golston 2010), so the following sentences are all grammatical despite the sometimes odd word order.We begin with cases of H + any accent, which always results in lowering of the first H tone to L. That is, word-final H is always followed by another H tone if it is followed by any tonic word, explaining why word-final H lowers except when phrase-final or preceding an enclitic.Non-final H is always followed by a L tone in Greek since only H*L and HL* place a H tone non-finally; since non-final H is always protected from a following H tone by an intervening L, the OCP never applies and all non-final Hs surface as is.

A parallel in Bantu
A number of Bantu languages have processes that eliminate the first or second H in a sequence of Hs.Best known are cases where HH → HL, which Goldsmith (1984ab) calls Meeussen's Rule in honor of A. E. Meeussen's work on Bantu.The opposite occurs as well, where HH → LH, in languages like Rimi, spoken in Tanzania (Olson 1964, Myers 1997).This is Anti-Meeussen's Rule.
Our proposal is that Ancient Greek has a form of Anti-Meeussen's Rule that lowers a H before a H across words to respect the OCP.The details of application fall out immediately given the independently needed accentual classes proposed in 2.
We'd like to thank Lee Bickmore, Martin Krämer, Zach Metzler, and audiences at UCLA and Tromsø University for helpful discussion.Mistakes and stupids are of course our own. * Our third tonal class occurs mostly as exceptions to the falling accents.Words with a simple final H accent throughout are limited to the so-called Attic declension and their accentuation is not firmly established; i.e., there may be no nominal or verbal paradigms in the language with final H everywhere. Final H accent is, however, firmly established as a sub-pattern that interrupts the falling H*L and HL* patterns.The pattern for ἀγαθός 'good', for instance, has H in the nominative/accusative and H*L in the genitive/dative: ACC.PL khruuˈsóòn khruuˈsóòn khruuˈsóòn GEN.PL khruuˈsóìs khruuˈsáìs khruuˈsoìs DAT.PL Note that both HL* and H*L are falling accents, echoing Allen's 'falling melodic pattern'.PL Fully mixed paradigms include words like βασιλεύς 'king', which has H in NOM.SG, HL* in ACC and GEN, and H*L in DAT and NOM.PL: . If the tonal analysis we sketched in 2 is correct, Greek words end either in H or in one of two falling tones, HL* and H*L: The reason should be clear: the first H tone is in each case followed by another H tone, which triggers Golston & Paulsen lowering to satisfy the OCP.If the first word is H*L, as we see in the three cases below, its final L tone prevents lowering due to OCP no matter what the following accent is: