Ancient Scripts and Phonological Knowledge (Current Issues in Linguistic Theory)

ANCIENT SCRIPTS AND PHONOLOGICAL KNOWLEDGE

AMSTERDAM STUDIES IN THE THEORY AND mSTORY OF LINGUISTIC SCIENCE General Editor E. F. KONRAD KOERNER (University of Ottawa) Series IV - CURRENT ISSUES IN LINGUISTIC THEORY

Advisory Editorial Board

Henning Andersen (Los Angeles); Raimo Anttila (Los Angeles) Thomas V. Gamkrelidze (Tbilisi); John E. Joseph (Hong Kong) Hans-Heinrich Lieb (Berlin); Ernst Pulgram (Ann Arbor, Mich.) E. Wyn Roberts (Vancouver, B.C.); Danny Steinberg (Tokyo)

Volume 116

D. Gary Miller

Ancient Scripts and Phonological Knowledge

ANCIENT SCRIPTS AND PHONOLOGICAL KNOWLEDGE D. GARY MILLER University of Florida

JOHN BENJAMINS PUBLISHING COMPANY AMSTERDAMIPHILADELPHIA

The paper used in this publication meets the minimum requirements of Arnerican National Standard for Information Sciences - Permanence of Paper for Printed Library Materials, ANSI Z39.48-1984.

Library of Congress Cataloging-in-Publication Data Miller, D. Gary. Ancient scripts and phonological knowledge I D. Gary Millee. p. cm. -- (Amsterdam studies in the theory and history of linguistic science. Series IV, Current issues in Iinguistic theory, ISSN 0304-0763 ; v. 116) Includes bibliographical references and index. 1. Grammar, Comparative and general--Phonology. 2. Inscriptions, Linear B. 3. Cypriote syllabary. 4. Alphabet. 5. Language awareness. I. Title. 11. Series. P217.3.M55 1994 94-28635 414--dc20 ISBN 90 272 3619 4 (Eur.) 11-55619-570-2 (US) (alk. paper) CIP © Copyright 1994 - John Benjamins B.V. No part of this book may be reproduced in any form, by print, photoprint, microfilm, or any other means, without written permission from the publisher. John Benjamins Publishing Co.• P.D.Box 75577 • 1070 AN Amsterdam • The Netherlands John Benjamins North America • P.D.Box 27519 • Philadelphia, PA 19118 • USA

ACKNOWLEDGEMENTS Parts of this work were presented at conferences, and other parts read by friends and colleagues. Particular analyses and formulations have profi ted from comments and suggestions by Henning Andersen, Eimer Antonsen, Alice Faber, Susan Guion, Jay Jasanoff, Katherine Leffel, Marie Nelson, Steve Podlecki, Tom Sawallis, Rohert Scholes, Martin Schwartz, Nelleke Van Deusen-Scholl, and W. C. Watt. Thanks also to Susan Guion, JeanPierre Olivier, Rohert Scholes, and Roger Woodard for making available 10 me prepublished copies of their work. Special thanks go to Rex Wallace, John Stonham, and Elisa Maranzana for generous comments on the entire manuscript. Many useful discussions with Elisa helped c1arify numerous points. Eimer Antonsen and Jay Jasanoff provided valuable assistance on the chapter on the runic alphabet. Jean-Pierre Olivier also provided generous assistance by sharing his own Linear B fonts and advising me on several characters. My wife, Judith A. Miller, children 's book writer and illustrator, aided the final preparation by meticulously drawing the graphic symbols in Appendix A. Several generations of students who suffered through preliminary versions of this work also deserve a "thank you". Konrad Koerner deserves special mention for cheerfully fielding my Email pleas for help at all hours of the day or night, and for numerous valuable suggestions that enhanced the content of the manuscript as weIl as the style. Yola de Lusenet of Benjamins facilitated the technical issues of publication and advertising. D. Gary Miller Gainesville, Ra., March 1994

ABBREVIATIONSI

Archil. Alt. AUXlAux C CX>NJ qxl.

ablative accusative active agreement Ancient aorist Arabie An:adian arcbai.e Archilochus (mid Vll) Atlie auxiliary (verb) consonant conjugation (class) marker compound(ed)

Cret.

Cretan

ABL ACC ACT AGRlAgr Ane. AORlaor. Amb.

Are. arch.

Cypr. DAT dial(s)

Cyprian dative dialect(s) Empty Category Principle ECP FIFEM feminine future passive participle FPP FUf/fut future GEN genitive Germ. German Gk. Greek Gmc. Germanie Hdt Herodotus (ca. -484-424) Hesiod (ca. end Vlll?) Hes. Homer (ca. end Vill) Hom. Icel. lce1andie [idem] the same (meaning) id. Indo-European IE D. Iliad (of Homer) IMPV/impv imperative IND indicative ind.obj. indirect object INF infmilive Ion. Iooic/lonian Ital. Italian Knossos KN Lat Latin 1 Bibliographical

LOCIloc

MlMASC MEDPASS

MID Mod. Mye. Nln NOM NP o(t). O. OB/obj.

locative/locational masculine medi.opassive middle Modern Myc:enaean neuter

nominative nounphrase neuter

Old oiject

Odyssey (ofHomer) OldNorse optative plural perfect active participle participle passive personal correspondence PERFIPFCT perfect(ive) Particular Grammar ro Pindar (-518438) Pind Plato (ca. -427-347) R possessivelpossessed POSS pp perfect participle ppp perfectlpast passive participle prepositioo Prep PRFS present preterit(e) prel present participle Pd' Pylos PV q.v. [quod vide] whieh see s/sg. singular Sem. Semonides 01. mid Vll) Sim. Simonides (ca. -556-468) Thess. Thessalian Thuc. Thueydides (ca. -460-400) V vowel vl. variant (manuscript) reading VP verb phrase w.llt with llterature (i.e., references) Od. ON OPf/opt. pipI. PAP partie. PASS/pass p.e.

abbreviations are included in the References.

CONTENTS Aeknowledgements

v

Abbreviations

vi

O.

Prefaee 0.1 Objectives of the Study 0.2 Scripts as Representation 0.3 Plan of the Book 0.4 Dating Conventions

xi xii xv xvi

1.

Theoretfeal Prerequisites 1.0 Introduction 1 1.1 Scripts 1 1.2 The Syllable and the Sonority Hierarchy 2 1.3 Sonority Distance and English Words 4 5 1.4 Some Consequences and Predictions of the SH 7 1.5 Some SH Effects in Ancient Greek 1.6 The Sonority Hierarchy, PO Stipulations, and Syllable Parsing 8 9 1.7 Parameters of Syllable Head Projection 1.8 Directionality and Segments ofIdentical Sorority 10 1.9 Marked Syllable Structures and Change 11 1.10 Conclusion 12

2.

The Linear B Syllabary 2.0 Introduction 2.1 Linguistic Inventory of Linear B Signs 2.2 Mycenaean Syllable Structure and Spelling Conventions 2.3 Sonority Hierarchy and Mycenaean Syllabification 2.4 The Linear B Evidence 2.5 Glide-LiquidlLiquid-Glide Sequences and the Patterning of /w/ 2.6 Onset Clusters in Coda Position 2.7 Variant Spellings 2.8 Conclusion

13 13 15 16 18 22 24 25 26

The Cyrprian Syllabary 3.0 Introduction 3.1 Cyprian and Linear B: Continuity and Innovation 3.2 Script Innovation and Text-Type 3.3 The Cyprian Evidence 3.4 Counterconventions: The Writing of Stop-Continuant Strings 3.5 Morphological Input and Boundary Phenomena 3.6 Continuant-Stop Sequences 3.7 Exceptions and Other Conditioning Factors 3.8 Conclusion

27 27 29 29 32 33 34 35 36

3.

viii

4.

CONTENTS

The Greek Alphabet Introduction Transmission of the Northwest Semitic Script Nortbwest Semitic Scripts and the Adaptation to Greek The Case for the Early Borrowing of the Greek Alphabet Some Early Forms of Greek Letters Internal Greek Prehistoric Developments: The Vowels LaterlRegional Vowel Letters The Supplemental Consonants The Evolution of Qoppa Supplementals and the Sibilant Letters The Antiquity of Segmental Writing Adaptation and Development Phase of the Greek Alphabet Summary and Conclusion Interaction Between the Syllabary and Alphabet Traditions

4.0 4.1 4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 4.10 4.11 4.12 4.13

s.

6.

39 40 40 42 44

46 48 48 51 53 54 55 56 58

The Runle Alphabet 5.0 Introduction 5.1 The Older Runic Fupark 5.2 An Idealized Runic Abecedarium and its Divisions 5.3 Origin of the Runic Alphabet: Mediterranean Theory 5.4 Critique of the Latin Origin Theory 5.5 Runic as an Invented Script 5.6 Germanie Vowels and Runic Letters 5.7 The Thirteenth Rune 5.8 Phonological Basis of the Order of the Runic Letters 5.9 Class and Manner Projections of the Runic Matrix 5.10 Antiquity of the Phonological Matrix: Byblos 5.11 The Ras Shamra Matrix 5.12 Empty Cells and the Antiquity of the Runic Matrix 5.13 Conclusion APPENDIX: The Proto-Germanic Vowel System 5.14 Long Vowels 5.15 Core Short Vowel Reflexes 5.16 Short Vowels in Unstressed Syllables 5.17 New Long Vowels 5.18 The Status of ~ 5.19 Proto-Germanic Vowels and the Runic Alphabet

68 69 70 72 73 74 75 76

Literacy and Linguistlc Knowledge 6.0 Introduction 6.1 TheWord 6.2 Word Boundaries 6.3 Knowledge ofWord Constituency 6.4 Knowledge of Words in Antiquity 6.5 Acquisition of Morphology

85 85 frl 88 89 91

61 62 63 63 65 66

77 77 78 80 81 82

CONTENTS

6.6 6.7 6.8 6.9 6.10 6.11 6.12 6.13 6.14

7.

Word Salience: Conclusion The Syllable Segments and Phonemes The Role of the Phoneme in Language Change The Phonology/Orthography Interface Phonology-Orthography Mappings Spelling and Metaknowledge Implicit Segmental Awareness Conclusion

Implieations: An Ideal Seript? 7.0 Introduction 7.1 Realities 10 be Dealt With 7.2 Script Abstractness and Phonological Cues 7.3 The Alphabet Compromise VS. More Ideal Scripts 7.4 Devising an Eificient Syllabary 7.5 Reprise and Conclusion

ix 92 93 94

95 96 97 98 100

102

103 103 104 104 105

106

Appendiees Appendix A: From Proto-Sinaitic 10 Greek Appendix B: The Linear B Syllabary Appendix C: The Cyprian Syllabary

109

Referenees

117

General Index

137

113 115

o.

PREFACE

0.1 Objectives olthe Study The topic of this study is both more and less than the tide suggests. Since the target is the phonological knowledge that underlies segmental scripts, especially of the linear variety ('alphabets'), and, since the only verifiable instance of the (spontaneous) creation of an alphabet involves the Northwest Semitic script and its derivatives (Diringer 1968: 164,435; Sampson 1985: 77; R. Harris 1986:27; Sass 1988: 167; Cross 1989:77), our concentration will be on early Western seripts with segmental eoding. Beeause two syllabie scripts from Ancient Greece provide clues to the development of the Greek alphabet, they will also be examined. The objeetive in eaeh ease is to demonstrate the high degree of segmental awareness that was coded in the seripts and their orthographie eonventions. Some ancient Western seripts are thus examined for theoretical implications - assumptions about phonology that underlie them. It haB been the standard assumption of Western culture that the alphabet is the greatest invention sinee the wheel. 1 Some scholars, such as Roy Harris (1986:37), have ehallenged this "ethnocentrie bias of a European approach to non-European languages". Nevertheless, very few researehers have seriously disputed the prevailing notion that the alphabet is superior to other (e.g., syllabie) scripts; see the references in Coulmas (1989:44), who challenges this view as "simplistie". Yet that is not enough of a eritieism. Coulmas goes 1 The present work is exclusively linguistic. We will not get into the politics that bave attributed the technocratic supremacy of the West to the alpbabet (see Olson 1994). Already Plato (Phaedrus 274-275) bad the king respond to Thoth's legendary invention of writing tbat it would inhibit wisdom. For criticism of the still voguish view tbat the alphabet was literaUy responsible for the development of civilizatioo, see PoweU (1981), Fmnegan (1988), Bloch (1989), Coulmas (1989), Larsen (1989), Thomas (1992), and especially Maranzana (1993); cf. Barton (1994). As stated by Coulmas (p.l60), "Wbat is surprising about this approach is tbat it was ever taken seriously and discussed by serious scholars." For the development of writing, see, for the older literature, the indispensable studies by Gelb (1963), Diringer (1968), and Jensen (1969). More recent approaches are found in Naveh (1982), Sampson (1985), R. Harris (1986), Coulmas (1989), Healey (1990), and the articles in Senner (1989) and Hooker (1990). Technical studies include Dietrich & Loretz (1988), Sass (1988), Schmandt-Besserat (1992), Segert (1993).

xii

PREFACE

on to assert that the alphabet is "the most economical system" (cf. Diringer 1968: 13) but hastens to stress that it is better equipped to handle languages like Greek than tone languages. However, accentual distinctions in Greek, e.g., Glc; phas IrJ160sl "light" : 1Ile; phlJs IrJ106sl "man" are no better handled by any of the world's usual scripts. Sampson (1985:107) claims it does not matter because the FUNCDONAL YIELD of accent in Ancient Greek was low (which is not c1ear; see Bubenfk 1983: 134ft). That of course misses the point that in a language in which accent has a higher functional load , a strictly linear segmental script is woefully inadequate, as Sampson (pp. 37-38) notes in conjunction with English intonation. In fact, an alphabet is not the most economica1 system, nor is it necessarily the simplest to learn just because it contains the fewest symbols (Gelb 1963: 184ff; Diringer 1968:13; Sass 1988: 167-168, with reservations). To some extent, as several scholars have.tried to maintain, there is a trade-off in complexity with respect to linguistic (specifica1ly phonological/segmental) knowledge. That is, some have c1aimed that alphabets are harder to learn because segments are not salient. We will argue that (1) this is only partially tme, and (2) alphabets may be easier to learn but (for separate reasons) not simpler to read - despite the ostensible advantage adduced by Harris (1986: 119), that alphabets not only reduce the number of symbols but simultaneously "lose few or none of the facilities of 'word identification' which the previous writing system afforded." An important principle that will be employed throughout, though rejected by Gelb (1963: 140-143, 251), is the ACROPHONIC PRINCIPLE defined by Coulmas (1989:33) as the principle "whereby a word acquires the phonetic value of the beginning of the whole word for whose writing it was originally used," A weaker version is generally adopted, whereby "A as in apple" exemplifies a modified (non-iconic to the symbol) 'acrophonic principle, the source of the ancient letter-names (Diringer 1968: 168-169; Jensen 1969:53; Gessman 1975:14~ Sampson 1985: 78,101; Harris 1986:31), analogous to modem radio alphabets like able, bakeT, chaTlie, etc. (Gelb 1963:142; Faber 1992:126). Compare the Slavic glagolitic letter-names az "I", buky "letter", vati "knowledge", glagol' "speech", etc. (Gelb 1963: 141; Gessman 1975:75). 0.2 Scripts as RepTesentation What do scripts mirror? Coulmas (1989:47) challenges the view that scripts mirror speech, the underlying assumption of which is that "a good writing system is an isomorphie mapping of speech." That is, there is a oneto-one correspondence between sounds and signs (cf. Diringer 1968: 12-13, 163; Jensen 1969:583). Coulmas argues, following the tradition of Gelb

PREFACE

xiii

(1963: 15, 224ff), Jensen (1969:.583-586), and others, that this ignores three important points: (1) this ideal is probably not realizable for any script; (2) the script user is not interested in the same precision mapping as the linguist [and would actually be slowed down by it - D.G.M); and (3) orthography is normative. In actuality, there are no 'pure' scripts because of the conflict pinpointed by SPE (p.49): Orthography is a system designed for readers who know the language, who understand sentences and therefore know the surface structure of sentences. [...] It would be quite pointless for the orthography to indicate [... ] predictable variants. Except for unpredictable variants (e.g., man - men, buy - boughi), an optimal orthography would have one representation for each lexical entry. Up to ambigUity, then, such a system would maintain a elose correspondence between semantie units and orthographie representations. A system of this sort is of Iittle use for Olle who wishes to produce tolerable speech without knowing the Ianguage [...].

The conflict is thus whether a script is going to mirror the Jexical entry (semantic pole) or the output (phonetic pole) or some combination or a more abstract level of representation. Some scripts are elose to phonemic (e.g., Spanish, Latvian), some encode morphophonemic information (e.g., Dutch, German, Russian), and some contain a large amount of morphemic, lexical, and even heuristic information, e.g., English (cf. Sampson 1985:194-213; Coulmas 1989:175-176; §6.12 below). Spelling in English is often lexicalsemantic or logographic,2 viz. new "'- knew "'- gnu "'- pneu-, or main "'- malle "'mein",- Maine. Note also differences among identical spellings, e.g., -ambof bomb, tomb, comb, or the notorious -ough of through, though, thought, trough, tough, bough, hiccough. And so on. Since antiquity (e.g., Aristotle, de Interpretatione 1.4-6), it is customary to conceptualize writing as a representation of speech (see Harns 1986:83-86) but Harns (pp. 91-92) argues that writing cannot be simply a representation of speech because different writing systems are associated with different neuroJinguistic problems. For instance, Japanese dyslexics familiar with the syllabic kana script and the logographie (Chinese) kanji script do not have the same difficulties (but see Morais 1991:17-18; Mann 1991:57ff). More to the point, no system so far devised comes close 10 representing all of our phonological knowledge, presumably because most nonpictographic systems are linear attempts at representing something nonlinear/multiplanar. 2 On the loose use of the tenn logographic in connection with EngIish orthography, see, e.g., PoweU (1991: 75, 116); cf. Gelb (1963: 15) on 'visual morphemes'. For a useful overview of English orthographie conventions and their history • see Jespersen (1948: 146-149).

xiv

PREFACE

A body of evidence for the autonomy of writing systems is presented in Harris (1986:105-108). For instance, scripts may contain non-phonologically or non-morphologically realized (Le., unpronounced and/or unpronounceable) symbols, such as classificatory detenninative signs, extremely frequent in hieroglyphic and cuneiform systems, signs for people (male, female, queen, etc.), wooden objects; and so on. In our own system, botanists intersperse amid ordinary writing special symbols for male, female, etc.~ astronomers, chemists, mathematicians, and other professionals use special symbols (Gelb 1963: 15-20~ Gessman 1975:8). Differences between upper and lower case letters most frequentiy correspond to nothing in the spoken language at all. Occasionally, there is a semantic difference, as in Smith: smith, Democratic : democratic. AIDS: aids, etc. (cf. Gessman 1975: 15); a message in ALL CAPS may express the attitude or intonation of the speaker. And so on. Sequential restrictions, such as q only before u, are synchronically arbitrary and correspond to nothing Iinguistic (Harris, p. 115; more in §6.1l). Another factor, extensively discussed by Jensen (1969:587-592; cf. Gessman 1975:98102), involves stylistic developments, especially those motivated by considerations of speed, such as shorthands, abbreviations, cursive stenography, brachygraphy, tachygraphy. 'Fonts' constitute another stylistic difference (Gessman 1975: 15,87-95). All of this very strongly suggests that scripts have an autonomy of their own and are not merely representational sys.tems of spoken language. As emphasized by Harris (1986: 119), writing "as writing" has for millennia been "independent of the spoken word". While all of this is undeniable, one must not get 100 carried away wi th the independence of.scripts and language. The independence could allow us to forget that the entire point of a script is in fact to represent graphically some aspect or aspects of the linguistic knowledge of native speakers of a language (on which, see Chomsky 1986). The conflict will always be on what kind of knowledge will be mirrored, whether it will be exclusively phonological (and which aspects of that - syllables and/or segments), partly morphological, partly lexical or semantic. Thus viewed, writing systems are attempts at representing different, 'competing' aspects of language (more specifically, language knowledge), some phonetic (noncontrastive), some phonemic (contrast and opposition), some lexical/morphological (root or affix unity), some morphophonemic (in the broad sense). Such competing goals are apt to yield discrepancies and irregularities in graphie conventions. This study will investigate the properties of several ancient syllabic and linear segmental scripts to make explicit the aspects of linguistic knowledge which they are attempting to represent. Chapter 6 will present independent evidence for the types of knowledge identified in the previous chapters.

PREFACE

xv

0.3 Plan ofthe Book Recent work on scripts (discussed in chap.6) has tended to support the age-old prejudice that alphabets impart a knowledge of segments, and that people without alphabets have no knowledge of words, much less segments. Linguistic analysis of the Creek Linear B syllabary (chap.2) and Cyprian syllabary (chap.3) shows that nothing could be farther from the truth. The spelling conventions of these two ancient scripts are based on the Sonority Hierarchy (SH), and presuppose a sophisticated (at least implicit) knowledge of the arrangement of segments according to the SH. Specifically, the sophistication of developing and using a script based on the SH, consistently performing exhaustive SH analyses of each word and spelling individual segments according to their position in the SH, devising solutions to problems like SH onsets in coda position or codas in onset position, handling problems of syllable adjuncts and SH violations in the language, and occasionally trying to represent compositional information as weil, go lightyears beyond anything predicted by proponents of privileged alphabet knowledge. To assess the development of alphabets and their adaptation to particular phonological systems, the history of the Creek alphabet is explored in chapter 4. Its Phoenician source fits letter-by-Ietter into a phonetic matrix analogous to tbe Byblos and Ras Shamra matrices (chap.5), in which segments are arranged as follows: laryngeaIs > labials> alveolars> velars> dentals. Again, the question arises, how can there be a phonetic-order conception of segments without a concept of segments? Similarly, the Cermanie runic jupark (chap.5) fits a matrix arranged: lip-rounded> dental> (alveo)palatal > velar. The changes and adaptations in the creation and ordering of that script reveal a knowledge of segments and their phonetic/phonological properties. Chapter 6 discusses some recent experimental research which has denied the very linguistic knowledge demonstrated to underlie the ancient scripts. Independentcorroborative evidence for that knowledge is presented, forcing the conclusion that the experiments are faulty in their design and results. They do not begin to access the linguistic knowledge possessed by native speakers. Consequently, that type of research does a disservice to humanity and the scholarly community in claiming to validate age-old prejudices under a pseudoscientific guise. The concluding chapter discusses some implications for the design of scripts and for future experimental work.

xvi

PREFACE

0.4 Datlng Conventio", To avoid problems of time reckoning, I will follow Sampson (1985) in using a date like 1980 for the modem era (= AD.lC.E.) and -1980 for antiquity (= 1980 B.C.lB.C.E.). Also, for generic ancient dating, the convention of Roman numerals is used, e.g., Naxos, VI Naxos, sixth century B.C./B.C.E.,

=

or-6c.

1. THEORETICAL PREREQUISITES 1.0 Introduction After abrief discussion of the typology of scripts targets the hopeless ambiguity of the tenn syllabary, a theoretical perspective on the structure of the syllable is pursued. Segments are arranged in the syllable in the unmarked case according to the Sonority Hierarchy and independent feature geometry. The Sonority Hierarchy is shown to make substantial predictions about the sequencing of segments universally and in Ancient Greek, where it was a strong motivating factor in a number of changes. A theory of syllable parsing and syllabification is introduced which builds on principles and parameters of head placement. The syllable is claimed to be a Sonority 'Phrase' structurally. 1.1 Scripts A note on script types is in order. The term SYLLABARY is hopelessly vague, as emphasized by Faber (1992:122), who divides syllabically linear systems into those that are syllabically coded (Akkadian, Japanese [kanal 1) and segmentally coded (Hebrew, Aramaic, devanagari, etc.); segmentally linear systems she subdivides into complete (true alphabets: Greco-Latin and its derivatives) and dejective (Ugaritic, Phoenician, etc.). The Linear B script she considers to be "syllabically linear and incomplete" (p.I23). But it is also (partially) segmentally coded. Faber's classification has no category for 'exhaustively analyzed but inexhaustively coded' as a matter of choice, since she does not believe that people without an alphabet can exhaustively analyze words segmentally. As she explicitly states (1992: 112), given the evidence that is customarily considered,2 "segmentation ability, rather than being a 1 Poser (1992) argues tbat even tbis classification is too simplistic because the Japanese kana system, with symbols for vowellength, gemination, etc., is mora-based, not syllable-based (and Timothy Vance agrees, p.c.). I thank John Stonham for tbis reference. 2 This is a necessary qualification. In a letter (18 Jan. 1993), Faber writes, "My purpose my only purpose - in the Segmentation paper was to demonstrate that nothing in tbe interface between Canaanite and True-Greek necessitates tbe assumption that [tbe early users of True-Greek ortbography had an essentially modem segmental awareness]. The target audience was those psycholinguists who see in the stmcture of True-Greek orthography the

2

ANCIENT SCRIPfS AND PHONOLOGICAL KNOWLEDGE

necessary precursor to the innovation of alphabetic writing, was a consequence of that innovation." (This is a common misconception; cf. Bellamy 1989.) The Linear B script (chap.2) militates against that position. Complete onsets, nuclei, and codas must be exhaustively analyzed segmentally to detennine how a given string is written. This suggests at the very least an implicit knowledge of segments and their hierarchical organization. Alphabets, as noted in the Preface, are not only rare among script types (and all existing ones likely arose from West Semitic scripts), but constitute an attempt to represent phonological knowledge on a single linear plane, in apparent violation of our implicit knowledge of higher levels of organization. Alphabets are advantageous. Because of the limited number of symbols, they are easier to leam. However, they do not indicate any more awareness of segmentation-ability than a syllabary like the Linear B or Cyprian. An alphabet represents Iess knowledge than does the Cyprian syllabary because it codes only (an incomplete inventory of) segments and omits all reference to syllable structure and the SH. Faber, ironically like those she argues against, seems to give privileged significance to SEGMENfATION as defined by the separate linear representation of consonants and vowels, missing the point that (i) knowledge of the individual segments (consonants and vowels) is clearly necessary to code them properly according to the SHbased rules of the script, and (ii) there is no linguistic reason to represent segments in isolation; a syllabic representation contains more infonnation. One potential objection is that the Linear B and Cyprian syllabaries did not . factor out vowels as did the Ethiopic and Indic seripts (Sampson 1985:64ff; Faber 1992: 120). In fact, the Indie devaniigari(Diringer 1968:283ff), like the Greek syllabaries, represents word-initial vowels differently from those in CV syllabIes; it differs from the Greek syllabaries in factoring out vowels in CV combinations, but even there a is the default vowel and a special mark is needed to indicate the consonant alone (see Gelb 1963: 149ff). Moreover, the vowel indicators (before, after, or above the consonant) are conceptually syllabic (Diringer 1968:262-263; Faber 1992: 120, 1290.13). Thus, a range of phonological knowledge from segments to syllables is represented in scripts classified as primarily syllabic or primarily segmental (cf. Gelb 1963: 188). 1.2 The Syllable antI the SorwTity Hierarchy The system of syllabification adopted here is an elaboration of the nuc1ear structure theory of Leffel (1985) and Levin (1985), in whieh vowels only compelling evidence for the cognitive naturalness of sub-syllabic segmentation. and my intention was to demonstrate that this is at best a plausibility argument."

3

TIffiOREflCAL PREREQUISITES

(or some other segment, as a parameter) are projected as heads of syllabIes. In the conception here, the syllable is a Sonority 'Phrase', in which the most sonorous segment (defined by the Sonority Hierarchy) in a string projects as the head of the syllable (§ 1.6ff). In the standard version of the Sonority Hierarchy (SH) in (1),3 the onset is everything up to the nucleus (V), and the coda is the mirror-image counterpart after the nucleus. Tbe nucleus and coda together constitute an inner constituent (usually called the rime/rhyme from its metrical function), on which see Steriade (1982), Fudge (1987, 1989), Treiman & Zukowski (1991), Carlisle (1991), Kenstowicz (1994:252ff). (1) Sonority Hierarchy (SH)

SYLLABLE rime/rhyme

~coda

onset

nucleus

~

obst Nas stop cont m n (kpt) (s)

Liq Ir

G1 wy

V

~

GI

Liq

Nas

Obst

yw

rl

nm

contstop (s) (tpk)

Clements (1990:313) claims that the feature categories (obstruent, nasal, liquid, glide, vowel) are not to be further subdivided (viz. k > p > t; m > n; etc.), as theyare in (1), because pi ace is not part of sonority. It is true that the SH per se predicts tautosyllabic pw, bw, and tl, dl, which occur in many languages. Their absence from English (words Iike atlas syllabify hetere3 The Sonority Hierarchy is notbing new; it dates back at least to Thausing (1863). It was defined by Jespersen (1904:187-192) in terms of segments with increasing syllabic and tonehearing ('sonority') properties toward the center (nucleus) of the syllable and decreasing to the end (cf. Sievers 1893:182-190; Grammont 1933: 99, 110). Thanks to Laziczius (1966:171-226), Starnpe (1973), Bell & Hooper (1978:10-11), Kiparsky (1979, 1980), and others, the SB has received renewed interest in recent years. The most detailed evidence for the SH (for our purposes) is extensively documented in Steriade (1982), Booij (1984), Levin (1985), Harris (1985, ch.2), Hock (1985), Vennemann (1988), Murray (1988), BashfllU (1~), elements (1990), Rice (1992); cf. Kenstowicz (1994:254ff). My own view is that the SH ia of extralinguistic origin (most likely physiological) but interacts with phonological knowledge in crucial ways. It is universal but not so clearly an inviolable principle. It has exceptions. and it is not yet clear which are tolerated or what ia impossible. Repair strategies are natural (physiologically motivated) but apparently not obligatory.

4

ANCIENT SCRIPTS AND PHONOLOGICAL KNOWLEDGE

syllabically) must therefore be explained in some other manner. Rice (1992: 76) suggests a constraint against identica1 PLACE STRUCI1JRE. In that case, the fact that tr, dr, but not 11, dl, is tolerated implies that Irl and 111 do not share place structure. Rice suggests that English Irl lacks a place node. In Ancient Greek, by contrast, Irl evidently had a palatal place of articulation to judge, for instance, from phonological changes, such as Attic reversion conditioned by Ir/,/e/, and /i/ (see Meillet 1903:30; Sommerstein 1973:52ff; Miller 1976a). Significantly, 11 and dlwere possible (onset-type) clusters in Ancient Greek. In stop clusters generally, the coronal is second. According to Rice (1992:82-83), this follows from coronals having less (specified) place structure than the other places of articulation. In general, sonority and place sequences have different properties, and Rice (1992:87) concJudes that it is erroneous to build place into a universal SH. This is probablyon the right track, but since nothing crucial to our purposes depends on the theoretica1 rationale for the independent behavior of place features, (1) will continue to be assumed to be the basic SH.4 1.3 Sonority Distance antI English Wor.

To illustrate the SH, some basic English words are spelled out in (2). (2) English Monosyllabic Words and the SH OBST NAS UQ GL V GL UQ NAS OBST

cont. stop

stop cont. knight

ME: k brand trash

irons

x

n n

NE: b (5)

r r

a y re re a y

n

t t d

5

r

n

z

As a parameter, English also requires a MINIMAL SONORITY DISTANCE (Steriade 1982) of at least two 'slots' on the SH (e.g., stop + liquid but not 40ements (1990) and Rice (1992) also claim that continuancy is independent of the SH. It is true that continuancy constitutes a major theoretical problem in that continuants (especially [s)) crosslinguistically typically occupy positions before andlor after stops (see Fujimura & Lovins 1978; Kiparsky 1979:434-435). However, because of problems with reduplication of sC- strings in Sanskrit, Germanic, and other lE languages, violation of the Minimal Sonority Distance in sl-, etc., and for other reasons that will become clear in the course of this work, I provisionally accept the analysis of Levin (1985), who treats s in initial sC- and fmal -Cs strings as a syllable adjunct (cf. Kenstowicz 1994:258).

THEORETICAL PREREQUISITES

5

stop + nasal strings are tolerated in syllable onsets), which Clements (1990:317-318) declares the unmmked situation (cf. Rice 1992:67). The most frequent systematic exception to the SH is es], as in words like Eng. string (cf. p.4n.4), which may have to do with the unmarked status of coronals (discussion in Kenstowicz 1994: 285,516-521). The most frequent source of adjuncts is morphology (e.g., Eng. pI. -s, which probably remained because it is coronal). A word like six /slks/ is complex by (the standard version 00 the SH; with derivation [slks+p] (sixth) and inflection [slks+p+s] (sixths), the fonn is approaching upper limits on efficient pronounceability.

1.4 Some ConseqlU!nces and Predictions ofthe SH As shown in (2), core syllable and word structure is motivated by the SH. The order specified by the SH predominates for word-structure in all natural languages. 5 No language has just the marked orders (cf. Bell & Hooper 1978: 10-11; Clements 1990). Following are same typical consequences and predictions of the SH: 1) The SH provides a metric of relative syllable markedness (Bell & Hooper 1978). In English, [ayrnz] is a possible (monosyllabic) word, but there is no *[arynz], * [aynrz] , *[ayrzn], etc.; burn, harm are well-fonned, but *bunr, *hamr are not. 6 This does not mean there cannot be language particular exceptions (however motivated): Gothic akrs, Old Norse akr "field" (one syllable) are acceptable, but MARKED - in the technical sense: a child acquiring the language must specifically learn which marked orders are .tolerated by the language. 2) Vowels and glides interchange more easily than do vowels and segments of lower sonority, viz. [ey] > Ce], Ce] > [ey], as in the history of English, but a change of ren] to Ce] is not expected, unless [n] becomes a glide first or via nasalization. 5 John Stonbam (p.c.) fmds this an unfortunate formulation in tbat it "seems to weak:en the universal hierarchy to the status of a mere typologica1 tendency [... and] we lose much of the explanatory value. " On the Olle hand, it is naive to declare the SH a 'principle' because it has numerous exceptions ('adjunets'), however conceptualized, viz. licensed at some prosodie domain (see Kenstowicz 1994 §6.9). On the other, this does not entail that it is not universal or tbat it is a mere typologica1 tendency. As noted above, the motivation is physiologica1, providing a universal naturalness that determines core syllabification in all naturallanguages. 6 From the abstract sonority template (SR) can be derived language-specific constraints on possible morpheme/word, but one must be careful to sort out the irrelevant. For instance, it follows from the template tbat (monosyllabic) nbik is not a possible English word, but not tbat bnik is not; the latter is a parametrie detail due to the Sonority Distance requirement. For more diseussion of English syllables and what is or is not predicted by the SH, see Kenstowicz (1994:256-261); cf. §6.13.6 below.

6

ANCIENT SCRIPfS AND PHONOLOGICAL KNOWLEDGE

3) No language has diphthongs like ren] unless it also has [er] and [ey], assuming it has those sounds. As noted by Stonham (p.c.), a language might lack Irl but still have a diphthong like [en]~ cf. Stonham (1990) on metathesis in Straits Salish and variable-Iength vowels in Wakashan. 4) Segments of higher sonority become syllable nuclei (as a parameter) more readily than segments of lower sonority. Words like irons layrnzl with a heavy coda are subject to resyllabification. The SH predicts that of the liquid and nasal, both of which are in an environment to become 'syllabic' (syllable nucleus/head), the liquid, being more sonorous [closer to the 'center' of (1)], will syllabify first, yielding [ayrnz], that is, [aya-oz], as an alternate output. 7 5) The hierarehy of syllabification predicts greater instability of syllabic segments more distant from vowels on the SH: [$] will be more unstable than [Il} 1)], whieh will be more unstable than [r J). It has long been known (sinee Brugmann 1876, and Saussure 1878: 18) that the PIE syllabic nasals were not preserved in any IE language. 6) Nasals share glottal features (natural tendency to be voieed) with liquids (e.g., Crothers 1975: 159). 7) Nasals assimilate to the point of contact of an adjacent obstruent (ramp, rant, rank [rregk])~ liquids, on the other hand, seldom do (Ferguson 1975~ cf. Rice 1992:63-64). 8) Obstruents tend to assimilate to the voieing or nasality of nasals (Ane. Gk. TTEVTE [pente] --+ Mod. Gk. [pendel "five"), but only rarely assimilate to liquids and segments of more distant sonority (Ferguson 1975: 178, 182ft). 9) Marked sonority orders brought about by other ehanges are subject to assimilation, dissimilation, metathesis, epenthesis, resyllabification, or some other repair strategy to reinstate a more optimal order and/or minimize the violation of the sonority hierarchy. The proper generalization is that the more distant the exceptions are from the preferred (unmarked) syllable structure, the more unstable they are and susceptible to adjustment. Sinee SH-motivated assimilations and epentheses have been weil treated in the literature (from Hankamer & Aissen 1974 to Rice 1992:70-75~ see especially Steriade 1982~ Murray 1982), some other changes are sketched in (3)~ see Grammont (l923)~ Ultan (1971)~ Hock (1985); Stonham (1990). (3) SH-Motivated Metathesis and Resyllabification (a) OE setl --+ seid "throne", botl --+ bold "BUILDing", etc. (details in Brunnerl965,§183) 7 Some varieties of English retain the original dissyllabic [ayranz]; for the position of the vowel. compare the spelling irons and the German cognate Eisen.

TIIEOREflCAL PREREQUISITES

7

(b) Slavic migla "cloud" .... Czech mhla -Iunla - (dissyllabic) mlha;

Czech lZice - iice - 71ice "spoon", etc. (KnUnsky 1':176:67) (c) Old Iranian (e.g., Avestan) suxra- "red" ..... Mid. Iran. *suxr .... Iran. SUTX; caxr(a)- .... Iran. carx "wheel", ]:lfra- .... zarj"deep", asru- .... ars "tear", etc. (d) Zoque Iy+pata/ "his coat" .... pyata; Iy+nanahl "his mother" .... [nyanah] nanah (e) IE *agr-6- (Ok. aypoc; agr-6-s "field") .... Gmc. *00- .... OE re cer "field" (.... ACRE, and names: Whitt-aker [= Whit-field] "white field", etc.) (f) Modern Ellglish examples of resyllabification (prism, drizzle, etc.) are analyzed in Rice (1989) as [pn.zam], etc.

1.5 Some SH Effecu in Ancient GTeek In Greek there is abundant evidence for the operation of the SH (see esp. Steriade 1982). For the relation of Myc. ktoinii, a subdivision of the diirrws (Att.-Ion. Siilloc; ditnos) "deme", and KOlva kOinj "common", cf. xallat khama("on the ground" : X6wv khf!Ön "earth" (for discussion of the reconstructions, see Hajnal 1992). The root is *dh(e)ghom- (cf. Hitl tekan), which (regularly) became *f!1rftÖn in Greek, metathesized to 1rftf!tn ,just as /ti-t(e)köl (cf. aor. info TEK-E1V tek-eil) is metathesized to t(-kt-ö TlKT(j) "I beget". Here metathesis serves to keep the constituents intact by conforming to the ideal order for segments in the SH (or independent principles of feature geometry). If metathesis fails to apply, the violating stop is deleted, viz. *#'t1rftamaf -+ xajJ.a( khamat, *tkoini .... KOlva koini. A clearer case of the action of the SH is found in Osthoffs Law (OL) (Osthoff 1881: 1593): "every long vowel in the environment before aresonant (i = y, U = W, m, n, r, I) followed by another consonant within the same word is shortened in Proto-Greek." In other words, shortening takes place before clusters in the coda position of the SH but not before onset clusters. While many of Osthoffs examples no longer hold up (see the extensive discussion in Peters 1980:306-319), other examples have been added. lonic IlEaajJ.ßptlj mesambr(e (Archilochus 122.3 West +) "midday" (*meS-iim(b)r-iyii) requires shortening of /äl before its fronting to I~/~ cf. Rix (1':176 §64). For the underlying long vowel, cf. Doric, Aeolic, Cyprian lljJ.ap ämar "day". Note also alternations like 6aiJjJ.a f!aiima "marvel" (*thäw-ma) beside 6ljEOilal theeomai (Epic, Ion.) "I behold" (cf. Risch 1973:50). OL is unequivocally 8 Att. l1f:allI1ßp(cx mesembrtii supposedly has 11 erestored from 1ll1lpcx himerii"day", but nothing precludes insertion of Ibl in *(mes)ii.mri(y) ii after OL was lost in Attic.

8

ANCIENT SCRIPTS AND PHONOLOOlCAL KNOWLEOOE

required to derive forms like (-)Kpalpa -kraira "head" from an immediate *kr 8jra, in turn from *kr äh-ayra by contraction (Peters 1980:228-286). While this is not the place for a detailed technical treatment of OL, and I plan to discuss the problematic forms elsewhere, the essential point is that where a long vowel preceded an onset cluster of the SH, no shortening occurred. This is especially clear in KPllj.l VOI;' kremnos (11.+) "overhanging bank, cliff' (old derivative to KpEj.laj.lal kremamai "I hang up" (cf. Risch 1973:98): -mn- is a typical on set cluster in the SH (1).9 Compare l\iov Oion "egg" from an immediate *CXy)yon « *öwyon< *öw(i)yom; cf. Schindler 1969: 160-161, 165ff);10 -wy-, like -mn-, is an onset cluster according to the SH (or independent feature geometry). eaÜj.la thailna and j.lEOaj.lßplll mesambrfe point to OL at least as old as Ionian-Attic, before the fronting of läI to liJ I. I see no reason to dispute Osthoff s dating. The fact that OL may have been reapplied again within the history of particular dialects only shows that it was a natural process that could apply anytime the over-long conditions for it were met Its extinction in Attic followed the numerous contractions that created literally hundreds of long vowels before coda clusters, and shortly thereafter, the long diphthongs were either shortened or lost their glide component, following which the whole length contrast was given up (details in Threatte 1980; Allen 1988). The evidence for OL as a natural process is (i) its widespread appearance in natural languages, e.g., Germanic; partly Tocharian; Latin (see Allen 1973: 66-67), and (ii) its obvious origin in syllable patterns with the RITARDO of 'hypercharacterized' codas (cf. Allen 1973: 64-67, 177,222-223).

1.6 The Sonority Hiemrchy, PG StipulDlionl, and Sylloble Parling Most accounts of syllabification (since Kiparsky 1979, Steriade 1982) have somehow employed the SH in their algorithm. If the SH is universal, the child comes equipped with this knowledge to a particular language. All the child must leam are the marked PO (Particular Grammar) exceptions to it and the PO restrictions on the unmarked sequences. The syllable parser needs access to the same information plus a principle of maximizing onsets. That algebra is syllabified al.ge.bra in English follows from the fact that (i) 1ge is not a possible syllable by the SH and English contains no stipulation that 9 This phonological evidence is matched by word divisions in epigrapbic alphabetic texts and syllabary conventions (Hermann 1923; Steriade 1982; Morpurgo Davies 1987:100-101; Woodard 1993, 1994; and chaps. 2 and 3 below). 10 Wbile other scbolars differ on the Indo-European reconstruclion (e.g., Schrijver 1991:30, posits *hZ äliom). what concerns us bere is only the immediate proto-Greek ancestor.

THEOREI1CAL PREREQUISITES

9

permits it~ (2) bra is a possible syllable by the SH and English contains no stipulation disallowing it~ and (3), as John Stonham points out to me (p.c.), *alg.eb.ra would also obey the SH but must be excluded by a principle that maximizes onsets over codas. Arabic, on the other hand, which syllabifies al.geb.ra, must contain a stipulation that disallows complex onsets (Steriade 1982; Levin 1985; Abu-Mansour 1987). As a parameter, one onset consonant is permitted and required in Arabic. Only stipulations or parameter settings must be leamed by the child. Consequently, in the unmarked case (i.e., in the absence of a PO stipulation), the syllable parser will select the syllable structure that optimally conforms to the SH. This follows naturally from the conception of the syllable as Sonority 'Phrase' adopted here. Sensitivity to PO stipulations obviates the necessity of Steriade's proposal (1982:84-85) to apply the onset rule before the coda rule in English (to get al.ge.bra) but the coda rule before the onset rule in languages that syllabify al.geb.ra. This solution, however ingenious, misses the point that al.geb.ra is the marked case (in Oreek there was a shift from the marked to the unmarked syllabification; see Miller 1982, 1990), and makes an ordering problem out of what should be a question of PO permissible onset complexity. As noted independently by Leffel (1985) and Levin (1985), a version of X-bar theory predicts onsets over codas by the c-command relationship. Also, only the head is obligatory; some languages, Iike Arabic, may require an onset (as a parameter), but no language requires a coda: 1.7 ParameteTl 0/Syllllble Head Projection To illustrate the SH theory of the syllable, consider a word like simple. The parser sees /sympll (where y represents a lexically UNDERSPECIHED segment). The most sonorous segment (most vowel-like) is the y, which therefore projects as head, defining its vocalic/syllabic character. (In this system, as in that of Levin (1985), there is no need for a feature [±syllabic], since that follows trivially from its status as syllable head.) The s is picked up as onset. The remainder,/mpl/, violates the SH, and no PO stipulation permits it, so another head must be projected (or the form is discarded). Of that string, the most sonorous segment in a position to syllabify is the I, which therefore projects a head. Next, since mp is not a permissible onset by the SH and no PO stipulation permits it, only p can be selected as onset, leaving the residue, m, as coda of the previous syllabie. In French, where the m in simple signals a nasal vowel, word-edge -pi is permitted by stipulation (the /11 is devoiced as a repair), and the word is one syllable (the potential final [3] is irrelevant here). Consider an example from the Imdlawn Tashlhiyt dialect of Berber (Deli & Elmedlaoui 1985; cf. Levin 1985: 127ff; Kenstowicz 1994:278fl). Thba

10

ANCIENT SCRIPfS AND PHONOLOGICAL KNOWlEOOE

"cover" (3sgF) has two syllables [th.ba]. Again, one can assume that the first head projected is that of the most sonorous segment ta!. But then, thb is not a legitimate onset by the SH (stops precede continuants and glides), and, since Berber selects the parameter that disallows complex onsets, it is not permi tted by any sort of PO stipulation. Consequently, only Ibl can be taken as onset and another head must project. Since Berber fixes the parameter that allows non-resonants to be syllable heads, Ihl is the most sonorous segment of the remaining string, so it projects as head and Itl is picked up as onset. When a word contains two or more segments of equal sonority, it is reasonable to assume that they project as heads simultaneously, viz. algebra, or Berber Iyldy/, which yields ildi "pulI" (3sgM). Nevertheless, evidence that simultaneous projection is probably not a correct analysis will be presented in the next section. 1.8 Directionality and Segments o/ldentical Sonority

An interesting problem involves adjacent segments of identical sonority. It seems to be a frequently selected parameter that syllabification begins at the right edge and operates right to left (but left to right in Berber; cf. [{bayn]n] > [baynn], not *[baynn] , "appear" [3pIM]). Consider (with syllable heads in boldface) Eng. we Iwy/, ye Iyy/, you Iyw/, your Iywrl (glides are higher on the SH than liquids and therefore project first), kiwi Ikywy/, and Kikuyu Ikykwyw/. The last example is particularly interesting. The last Iwl is the first to project as head. Although the preceding Iwyl is a possible onset by the SH (or feature geometry), it is excluded, as are mn- and tl-, by the English Minimal Sonority Distance (§ 1.3), forcing tyl as the sole onset. The next segment that can project as head is the Iwl in the middle, which naturally picks up the preceding Iki as its onset, (k)yk being an illicit onset by the SH and not provided for by any PO stipulation. 11 The rightmost syllabification condition is frequent in other IE languages, as weil. While the particular details are complicated (different approaches in Miller 1974 and Schindler 1977), the general, or 'elsewhere', rule for IE syllabification is that the rightmost of two segments of identical sonority is the head, as shown in (4). 11 John Stonham reminds me of words like hymn, solemn, in which I should falsely predict that the final In! should syllabify. However, since -mn is not a possible coda (by the SH or feature geometry), it is subject to repair, in this case, deletion of In!. Tbe Minimal Sonority Distance (which he adduces) seems not to operate in codas (cf. -lI, -rl, etc.). When we compare words like cumin and denim which can reasonably be taken as underlying Ikywmnl, Idenm/, we see that the rightmost nasal correctly syllabifies. Tberefore, words like hymn constitute the 'particular' case as opposed to the 'elsewhere' case in Ikywmnl, etc.

TIIEOREI1CAL PREREQUISITES

11

(4) Syllabification of Rightmost Segment of Equal Sonority in JE (a) Latin 1) Iywng+öl- iungö [YUOgö] "I join" 2) Iwynk+öl- vincö "I conquer" (b) Sanskrit: Iyw-n-j+antil - yunjtinti "they join" (c) Gothic: Iyysl - jis "who" Since it is typica1 for segments to syllabify in descending order of their sonority (DelI & Elmedlaoui 1985; cf. Kenstowicz 1994:278ff), in (4a) and (4b), the glides must be given a chance to syllabify (project head) before the nasal since glides are higher on the SH than nasals. Then, of the glides, the rightmost projects as syllable head first, as most clearly in (4c), where the identity is absolute. Consider the development of Gk. ovolla 6noma"name" from *h3nh3mn (Beekes 1987:1-6; Schrijver 1991:24), if the reconstruction is correct. If the "Iaryngeal" *h3 was in fact a glide, they would both voca1ize first, then the final In!, yielding the correct output (for laryngeals as syllable heads, see § 1.7 above). 1.9. Mar1ced Syllable Structure, and Change Given that syllable heads can also be marked in the lexicon (cf. Levin 1985),12 a simple explanation can be offered for shifts Iike OE Tiwes-dreg, ME Tiw(e)s-dreg, to NE [tyii]S-day (Jespersen 1948:101ff). The string Ityws/, with the leftmost glide prespecified as head (and linked to two timing slots on the skeletal core), being marked (in the technical sense), is subject to loss of its lexical mark in language change. Loss of the lexical mark predicts that the default will take over and project the rightmost as head. 13 That is, Itywsl is replaced by Ityws/. Since the head is associated with two timing slots, the resulting change follows naturally.l4 One of the commonest forms for "she" in ME, namely s(c)ho, admits of the same explanation. From OE s«J, sv, one

12 John Stonham questions how this fonnulation can be any different from using the feature [syllabic]. In reality the problem is no different from that of a lexical mark for where tone or accent goes as opposed to a feature for tone or accent. A lexical diacritic is not a phonetic feature. For more discussion, see Levin (1985). 13 Rex Wallace (p.c.) objects that this is adescription rather than an explanation. Tbat is not true. Given that the motivation for the change was the marked character of the syllable structure. that predicts that the most natural change would be to lose the exception mark. The fonnulation above states this regularization formally. 14 For the stability of length, predicted by the nonlinear model, compare transposition word games in Bakwiri (a Bantu language), in which length remains in the same place, thongh the vowel is different, e.g., liiimgd > ngiiiUU "stomach" (Hombert 1986: 178. with data from other languages as weil).

12

ANCIENT SCRIPrS AND PHONOLOGICAL KNOWLEDGE

could expect (at least as one possible development) a form s 19 (where 9 represents an o-glide). Lexically, /sye/ requires the first glide to be prespecified as head. Loss of that lexical mark correctly predicts that the rightmost will syllabify, yielding [syö], whence [§ö] (length again being a property of head association). All of this was of course reinforced (or, according to BFitton 1991, triggered exclusively) by hro "she" > hyö> ~o rxYö] I [§ö], etc. (cf. Jespersen 1948:53). • The traditional problem of Sanskrit siV-ya-ti "sews" beside PPP sy ü-ta(Wackernagel 1896:91-92; Kurylowicz 1968 §278) can be explained the same way.1 5 In S iV-ya-ti /syw+ya+ti/, there is no problem since vy [wy] is a permissible onset by the SH, and therefore goes with the following syllable, leaving only the preceding y to project as head. In sy ü-ta- /syw+ta+/, on the other hand, of the string /sywta/, (sy)wt is not a permissible onset. Therefore, ta constitutes one syllable and /syw/ must project another head. By the headright parameter, /w/ becomes the head. As usual, the association with two timing slots remains a property of the head. 1.10 Concllllion Relating the syllable intrinsically to the SH goes a step beyond Leffel's (1985) and Levin's (1985) theory of the syllable as Nuclear Phrase and seems to simultaneously explain a number of problems involving syllabification. The SH interacts with other factors, such as feature geometry, which determines place features, among other things. There are also different parameter settings (requirement of an onset, disallowance of complex onsets, allowance of segments other than vowels to be syllable heads, minimal sonority distance of more or less than two, etc.). The SH is directly involved in the word structure of natural languages, and conditions numerous changes to rectify deviant sequences. To conclude this excursus, it seems clear that the SH makes substantial predictions about the sequencing of segments universally and in Ancient Greek. Consequently, it should not be surprising to find the SH underlying the syllabic spelling systems of the Linear B and Cyprian scripts. 15 Rex Wallace objects to this formulation. The problem traditionally involves the historical development of the laryngeals, pointing to an immediate *siHw-ye/o- vs. *syuH-tO-, with metathesis of the laryngeal in Olle of the forms. My assumption is simpler and involves no metathesis. HistoricaUy, *siHw-yelo- : *siHw-t6- should have given slVya-; *s lVta-. Altematively, assuming the regular development in sNya- (given that wy- was a possible onset, in the sequence *yH, only Iyl could have syllabified), *syHw-t6- might have given something like *siHu-tO-, whence *syuta-, or the like, in any event irregular. Even if there was a metathesis of the laryngeal (as a repair for the SH violation?), allowing a correct phonetic reflex, the problem remains a synchronie one of deriving syitd- from sN-.

2. THE LINEAR B SYLLABARY 2.0 Introduction The Minoans invented two syllabic writing systems, Cretan Hieroglyphic and Linear A, around the end of the -3rd millennium (cf. Heubeck 1979:23). The latest texts date to ca. -1450. All of the verifiably Greek texts are written in a syllabary known as Linear B, which was developed from Linear A around or after the -17th century and first attested ca. -1375 at Knossos (see Olivier 1986). It consists of fr7 signs. some of which are rare. In addition to clay tablets of the Mycenaean era (most of which date to ca. -12(0), there are inscribed sealings, labels. and ca 150 painted vessels. Most numerous are the texts from Knossos (3000-4000 complete tablets plus several thousand fragments) and Pylos (ca.I445). There are occasional documents from Mycenae, Thebes, Tiryns, Eleusis. and, since 1990. Khania (see. e.g., Olivier 1993). The Linear B tablets deal primarily with administrative matters involving inventories of personnel, supplies, equipment, parts of vehicles, etc., and are generally accompanied by a logogram and followed by a number. There are also tablets with tributes. ritual offerings, and land tenure, especially from Pylos. This chapter will demonstrate that the spelling conventions of the Linear B syllabary are based directly on the Sonority Hierarchy (SH) and presuppose a sophisticated (at least implicit) knowledge of the arrangement of segments according to the SH in order to spell words.

2.1 Linguistic lnventory 0/ Linear B Sigm The 'Mycenaean' Linear B script has syllabic signs only for syllable nuclei (vowels: a, e, i, 0, u) and nuclei plus onsets (CV) [Consonant-Vowel], including a few complex on sets (CCV: pte-, dwe-, etc.; cf. also qe [kwe]/[gWe]/[khwe]; ze [tse]/[d7e]). To facilitate referencing from other chapters, the inventory of signs in Linear B is reproduced in Appendix B. The linguistic layout of Stephens & Justeson (1978:277) in Figure 1 demonstrates the constitution of this script. As emphasized by Stephens & Justeson, it is evident that the Linear B syllabary does not reflect the Greek phonological constrasts among voiced, voiceless, and aspirated stops, nor the distinction between 111 and Ir/. The

14

ANCIENT SCRIPfS AND PHONOLOGICAL KNOWIEDGE

Indo-European labialized velars ('labiovelars') */kw gW gwhl were preserved in Mycenaean as lkw gW kwht (see Thumb-Scherer 1959:334-335; Lejeune 1972:43-53) and represented by the Jabialized series, perhaps fortuitously since Linear A also has signs for Iabialized and palatalized consonants. I plain C I paIatalized CY I i e a 0 u li e a 0 u STOPS I labial p/blph I pi pe pa po pu I - pte marked ephLI pU2 I - - dental t1th I ti te ta to tu I - - tja - marked (d) I di de da do du I velar k/g/kh_1 ki ke ka ko ku I - ze za zo zu NASALS I labial(m) lmimemamomu I - - - - dental (n) I ni ne na no nu I - - - - UOUIDS l/r I ri re ra ro ru I - - ra2 I'02 CONllNUANr sJ si se sa so su I - GLIDES I w lwiwewawo - 1- - - - Y I - je Ja jo ju I - - - - _---"h'--_~I_-_---=a~ - - I - -

labialized CW i e a 0 u

vowels

I I I I I

- - - - - twe - two - dwe - dwo qi qe qa qo -

diphthongs ai au

I I I I I

- - - - I I - - nwa - - I I - - - - - I ra3I swi - swa - - I I -

I I I -

- - I - - I -

- I

Fig.l: Inventory 0/ Mycenaean signs The precise nature of the z-series is disputed, but from the etymological point of view and from the evidence provided by their later reflexes, those signs are used for reflexes of Idl + Iy/, velars + Iy/, and initial */y/ (ThumbScherer 1959:336ff; Lejeune 1972:100-111). Their original value was likely palatalized, but the usual assumption is that in Mycenaean they represented affricates, perhaps Itß, dZ/ (cf. Morpurgo Davies 1988: 79-80, 105-106; Ruijgh 1985:105-126). Palmer (198O:31ff), while taking the position that the series continued to represent some sort of palatalized stops (cf. Crespo 1985), actually supports the affricate hypothesis. If indeed the ox-name a3zoro is Aiskhros "ugly", and assuming the equivalence of aketirija and azetirija for asketriai "(cloth) finishers", this would not be the only place in Greek where either a variable metathesis of sk to ks occurred (cf. altemations between o and tIJ, or tIJ and 0; see Lejeune 1972 §§61, 109), or, more likely, the same character can be used for affricates or clusters, including sibilant + stop or stop + sibilant; cf. the use of tIJ for both [ps] and [sp], paralleled by , for [zd] and [dz], and a character normally a variant of {, namely I, but used for

15

UNBAR B SYlLABARY

[zdldz] in Cretan (Bile 1988:78). Other examples in Nilsson (1918: 190ft). Potential parallels will be discussed in chapter 4. The important point for our purposes here is that there are at least two possibilities that would allow a symbol normally used for affricates to be used for a sequence that was similar perceptually. Stephens & Justeson try to ascertain the phonological properties of the Minoan language for which the Linear B script was devised. This issue will not concern us here. For additional discussion, see Hooker (1979), Heubeck (1982).

2.2 Mycenaean Sylkible Structure and Spelüng Conventions There are no syllabograms involving codas (-(C)VC, etc.), suggesting that the syllabary was devised for a language with open syllables only. With one exception (actually a compromise), no coda consonants are indicated at alt. The syllabary follows the universal algorithm for syllabification (§ 1.6ff) in selecting onsets before codas: a -VCV- string is therefore automatically syllabified -V.CV-, and that is paralleled by the syllabary conventions, e.g., podef "foot" (DATsg) is syllabified [po.dey] and written po-de (PY); /apu/ [apu] "from, off, away" is spelled a-pu (in many compounds) and /pantes/ [pan.tes] "all" (NOMplM) is writtenpa-te, as is "father" ([pa.~r]), given that all coda consonants, including those in word final position, are ignored. The syllable nuc1eus is a vowel. Onset and coda are defined by the SH (§1.2), reproduced in (1) for expositional clarity. (1)

Sonority Hierarchy (Schematized) SYlLABLE rime/rbyme

~coda

nuc1eus

onset

~

obst Nas stop cont m n (kpt) (s)

Liq 1r

GI wy

V

~

GI yw

Liq rl

Nas nm

Obst contstop (s) (tpk)

The onset is everything up to the nucleus (V), and the coda is the mirrorimage counterpart after the nuc1eus. Tbe nUcleus and coda together constitute

16

ANCIENT SCRlPfS AND PHONOLOGICAL KNOWIEDGE

an inner constituent (usually called the rime/rhyme from its metrical function) which determines syllable WEIGHf,1 In the ancient JE languages any coda cluster.long segment. or diphthong constitutes a heavy syllable. That is. any segment that can be associated with a timing slot on the skeletal core (or mora tier) is subject to lengthening by association with more than one timing siot. In the JE languages. phonologicallength (as opposed to phonetic Iength: the distinction was made by the Sanskrit grammarians; see Allen 1953:83-87) was binary: one timing slot versus more than one (any number). A syllable ending in a vowel linked to one timing slot was light. A syllable ending in a vowel plus any other coda segmentes) linked to one or more timing slots was heavy (i.e .• a long vowel was linked to two timing slots. as was a lang consonant; a diphthong could be light or heavy. i.e.• two segments linked to one or two timing siots; etc.). Onset clusters (defined by the SH) only made the preceding syllable heavy if the marked syllabification parameter (§ 1.6) was selected. as in Arabic. Vedic Sanskrit. and Homeric Greek (Miller 1990). in which case. of course. a consonant. not a (short) vowel. ends the syllable. 2 2.3 Sonority Hierarchy and MycelUUan SyllabiJication The manner in which complex onset consonant clusters are indicated is significant in that it demonstrates at least implicit knowledge of syllable structure and the SH. Consider the "tripod cauldron" TplnQC;" tr(pos (11. 22.164+). Since stops precede liquids on the SH. tr- is by definition a possible onset (from the SH point of view). The nucleus of the syllable is fit. Since a consonant cannot be written without a vowel in a syllabary. there is no way to write just the /tl. The /tl requires a 'dummy' vowel. and the one selected is a copy of the nucleus vowel: ti-ri-(po). Similarly. in BEollotc;; desmo B "with bands". continuants (s) precede nasals on the SH. making sma possible onset. Since the nucleus is 0 (oi contains a coda glide: [oy]). it is copied as the dummy vowel of the complex onset. viz. de-so-mo (to save 1 This is not a universal, as emphasized by John Stonham (p.c.). citing Zec (1988) who shows that in a number of cases weight is determined by the nature of the coda segments (e.g., only sonorants, never obstruents, in Kwak'wala). Similarly, Niang (1993) claims that Pulaar makes a four-way weight distinction in syllabIes. What all of this suggests is that the notion of weight is subject to parametri:lBtion (coda clusters only, certain coda clusters, etc.). See Stonham (1994) for a recent discussion of moraic phonology. 2 One other metrical possibility must be mentioned. A syllable that is heavy (by the deimition here) can be made light wben the metrical conventions permit the segment after the nucleus to be resyllabified as onset to a following V -initial word, as in Homer, where I10L lVVEtrE moi ennepe (Od. 1.1) "sing to me" scans, not as [moy.en.ne.pe] [-~-], but rather as if [mo.yen.ne.pe] [-~] (see Allen 1973: 142-143, 224-2~). For resyllabification across word boundaries in naturallanguages, see Kenstowicz (1994:280-285).

UNBAR B SYlLABARY

17

space.1 will write simply tiripo, desomo; dashes between syllable signs will be provided only when the division is potentially unclear or confusing). Conventions therefore factored out vowels even though the signs did not Consistent with the non-writing of coda consonants in the Linear B syllabary. geminates are not indicated either. It is generally agreed that the constituents of a geminate strand the syllable boundary, viz. t1T1TOt hfppoi [hip.poy] "horses". In Mycenaean, then. it is not surprising to find the ancestor of t1T1TOt written i-qo [(h)ik.kWoy] (KN. PY). Note also ze-u-ke-si (PY) [dZell.ges.si] "pair" (DATpl) to 'EiJyOC; 'lJ! iigos (n. 18.543+). E-ra-peme-na (KN) [eJ..Ja.pme.näl. 3 perfect passive participle of pa1TTlJl rhdptö "stitch". illustrates the problem of the conflict between the expected syllable division [eJJap.me.nä]. based on Homeric scansion in which the [p] of such a cluster is treated like a coda consonant for purposes of metrical weight. and the Linear B treatment of pm- as an onset. However. let us ignore metrical problems for now,4 and concentrate on the fact thatpm- is treated as an onset for the simple reason that, in terms of the SH, p, a stop, precedes m, a nasal, and therefore is universallya possible syllable onset. That does not mean that a language-particular stipulation cannot override it. but that is the marked case (in the technical sense): it has to be specially leamed by the languageacquirer who, in the default case, would treat any onset allowed by the SH as a legitimate onset.

3 There is little doubt about the meaning or formation of this word, but some details are unclear (e.g., Ruijgh 1985:124 and Duhoux 1988:45, eite as EPPaq,IiEVO: errhaphmenB); for other possibilities, see tbe discussioD in Garcia-Ram6n (1985:218-219). 4 I bave elsewhere (Miller 1982, 1990) supported the idea that epic meter is more conservative in terms of syllable division (e.g., Hom. pal.r( - vs. Attie pa.tri w "to father'') than Mycenaean. That Mycenaean bad already changed is Made probable by tbe same syllable division in Arcado-Cyprian and (tbe rest of) Ionic and Attic (Wathelet 1970). Woodard (1993, 1994) claims that Mycenaean and Cypriot SH spelling is independent of syllable structure. On the tbeory here, where syllable is a 'sonority phrase', that is impossible. Wbat is possible, however, is that the scribes aUowed their implicit knowledge of syllabification to override their actual syllabification, in which case the writing system is reaDy abstract, based exclusively on implicit knowledge (tbe claim to which Woodard is necessarily bound). Similarly Lejeune (1972:285) and Ruijgh (1985: 120ft) distinguish 'phonological/phonetic' from 'orthographic' syllabification. Steriade (1982) and Guion (1994), on the otber hand, project a syllable structure of maximized onsets (permitted by the SH) back to PIE, in which case Mycenaean spelling reflects actual syllabification, and the Vedic and Homeric scansions constitute a separate problem. The outcome of this divergence of opinion is irrelevant to om PfeSent discussion of wbat Mycenaean scribes at least implicitly knew about the SH, as evidenced by their spelling practices.

18

ANCIENT SCRIPrS AND PHONOLOGICAL KNOWlEDGE

2.4 The Linear B Evitknce Any sequence allowed by the SH (independently justified in chap.l) as a possible onset is spelled out in the Linear B syllabary, and any sequence specified by the SH as a possible coda is not written, according to the canons of Linear B spelling. (One interesting exception to this generalization and one compromise will be discussed below.) Consider the data. 5 ONSET

stop + stop [k >p >t] ko-to-(i)-na /ktoyni/ "plot of land" (KTolval klo hai Hesychius) tekotone Itektonesl TEnovEC; "carpenters" (11. 6.315+) ekoto (e'I::TUlP Hektör H. 1.242+) ponikipi Iphoynik P > t] stop + stop [t > P > k] potolise Ipt6lisl "city" [217.2+] [no examples?] totipeteraloipone Itö diphtheraloyphönl "the school-master" (GENsg)3 3 The alternate Cyprian GENsg -ötl (-ötl?) is of completely obscure origin and status. The -n form is most common at Edalion. and never affects the article to (fhumb-Scherer. p.165). Tbis example is from the epitaph of schoolmaster Onasagoras (lCS 143: Paphos). Though SLcI>&PaML~c;diphtheralophdr is ahapax. it is known from Hesychius. who glosses it YPltllllltT08L8cfaKltMC; grammatodiddskalos "schoolmaster" and explains it as a Cyprian word (cf. Masson. adICS 143).

30

ANCIENT SCRlPfS AND PHONOLOGICAL KNOWIEDGE

timowanakotose Itimowanaktosl "of Timowanax" [150,405.1, etc.]

stop + continuant

[see below]

stop + nasal punutakoro- "Pnutagoros" [403.1]

continuant + stop nasal+stop [not written; cf. §3.1 (A2)]

kasikeneta lkasign~täI "si ster" in a digraphic inscription (ICS 164) [KaatyvtlTTl kasign«~ H. 4.441+] terekiniya lterkhni(y)a/ "plants" [217] [exc.: ikimamenose likmämenosl "wounded" (AccplM) [217.3-4] (probably perfect: ThumbScherer, 169; Ringe 1984: 129)

continuant + nasal inalalisimena linalilismenänl "engraved" (ACCSgF) [see Masson, ad ICS 217.26] (the form shouId be -li-se-Tne-, but see §3.6)

nasal + continuant weretase probably Iwretasl and not Iwritans! (ACCplF)4 "pacts" [217. 28,29] ; cf. AU. PllTpa rh«"ii, Ion. PllTPTl rhi!r~(Od. 14.393+)

nasal + nasal (m > n)

nasal + nasal (n > m)

memanamenoi Imemnämenoyl "remembering" (pr. medpass. part, 1.1. Eil VIlIlEVOt) [261]

NOMplM =

nasal + liquid

liquid + nasal kumerenai Ikumemahil "guide" (3pl pres.) [264.4: Neumann 1974] (formerly read kurner ifilai)

4 Arcado-Cyprian apparently did not have the Second Compensatory Lengthening (-Vns- -+ -V :8-); cf. Are. 1I'IlJlO'CXl; pdnsas [DGE 665 A12] "all", o4li~V01. ophellonsi {DGE 665 A23] "(they) owe" (see Wyatt 1973:41); -ns- remains except syllable8: Are. TOC; tOs .« *tons: Att.-Ion. TOU,. toUs) "the" (ACCplM); 'll'Civauc; pans "air' (ACCpIF *pan(t)sans: Att.-Ion. 1I'clauc; p'&äf); see Dubois (1986 §38ft). Cyprian is ambiguous, but, based on the r than ltonsl, ItösJ, or any Arcadian evidence, it is generally agreed that to-se is Itösl other of the theoretical possibilities (cf. Bechtel GD 1.416; Thumb-Seherer. p.161; Scbmitt 1977:94). Viredaz (1983: 186,206) signals Paphian spellings like -aus (*-ans) as evidence of nasalization or nasal es without eIosure. Whatever the details, retained - Vns is most unIikely.

inEnaI

CYPRIAN SYLLABARY

31

paramenone "Parmenön" [154.2] aramaneuse /armanewsl "(son) of Armanes" (see ad ICS 217.21)

autarami lautar mi! "but me" [235] stop + liquid pilokuporone Iphilokuprönl "of Philokypros" [217.1]

tamatiri /dämätri/ "to Demeter" [182] turumione /drumionl [217.19] Apul1wc: Dnlmios (ACC) "Copse Stream"

etewatoro /etewandrö/ "of Etewandros" [176]

nikokelewese /nikoklewes/ "Nikokl(ew)es" [6.1.7.2+] [exc.: ekerato lekhrätol "used" (?) [306.2] (other readings possible)]

liquid + stop talatone ldaltonl "(writing-)tablet" [217.26] (8EATOC: diltos Batr. 3+) arakuro largUröl "of silver" [217.6+] kateworokone IkatcSworgonl "besieged" (3pl . aor. ) [217.1] (see Masson 1983:265-266) sunorokoise /sun (h)6rkoys/ "with oaths" [217.28] (OplCoc; Jujrkos Hom.+)

continuant + liquid esolo leslo-I "Eslo-" [ISa, 327+ variants in Viredaz 1983: 193]

pereseutai lpersewtäyl "Perseutas"

nasal + gUde

glide + nasal

[no examples?]

gilde + liquid [like Linear B] weretase Iwretasl "agreements" (Accpl) [217.28,29]

e(u)weretasatu le+wretisatol "agreed, contracted" [217.4, 14] (with variable resyllabification [ew.re.tä.sa.tu]6) zowara[- Id~wni[lios]! Name (ZwF paAtOC: Zöwrdlios) [327.4 (Bulwer Tablet)]

liquid + contlnuant (DAT) [181.3] (hero: 2 alph. texts)

liquid + gilde [cf. Linear B] alawo-/alwo-I "vineyard" (ACc?)5 arawasatu larwisatol "prayed" [343a 4] (apaOl1al ar ~mai Horn/poet) [ambiguous as to onset or -xxla treatment, but -rw- codas are to be expectedl puruwoso Ipurwo-sol "Pyrwos" [198 (2x)] (but probably Eteocypriot)

5 The problems with this word (3x in ldal. Bronze: Iines 9, 18,21) involve its Case (GEN or ACC) and its derivationaI morphology (see Masson, ad loe.); the fIrst three segments of the word. which are aII that concem us here, are not in dispute (pace Beekes 1971 :350-352). 6 Compare Buck 1955 §§ 55, 70.3, and Cypr. a-ro-u-ra (e.g., ldal. Bronze 20) "land"; note the consistent spelling from Mycenaean (§2.4, last entry) to the C1assicaI period (~upa:).

32

ANCIENT SCRIPfS AND PHONOLOGICAL KNOWLEOOE

stop + gilde

[no examples?]

gilde + stop

gilde + gilde

[no examples?]

gilde + gUde

continuant + gilde

[no examples?]

gilde + continuant

So far, precisely the same strings are treated as onset or coda as in the Linear B syllabary. Especially noteworthy is the consistent exception to the SH involving glide and liquid sequences, confinning that something general (non-idiosyncratic) is at issue here. 3.4 Counterconvention,: The Writing ofStop.continutJnt String' As noted in §3.1 (Ca), there are two ways of writing continuant-stop clusters. The same is true of certain strings involving stop plus continuant. Consider first the status of [ks].

(3) [ks] as Complex Sign (a) e-xe leks/ "from" (E:~( h:) [217.6.7.24 (2x).etc.] (b) e-xe lo-ru-xe leks=oniksel "banishes" (aor. subj. 3 sg. of E:~­ Opt'W ex-or(z ö); cf. Att. eeopt0'O exorfsei (Thumb-Scherer, 135, 157,160,170; Cowgilll964:358; Masson, ad !dal. Bronze 12) (c) wa-nD-xe Iwanaksl "lord" (avat dnax) [211.1.220.2.264.1+] (d) e-we-re-XQ lewerksa/ "I did" (aor. act. 1 sg. of F€ Py- = erd ö Hom.+ Ion., poet) [261] (=Pfohll966, #11: Golgoi, -V) (e) ka-ru-xe-e-mi = Kap~ lud kärux emf "I am Karyx" (digraphic stele [260]: Golgoi, mid -VI) = Att.-Ion. "i'ip~ kerux "herald"7 (4) Normal Onset Treatment of [ks] (a) e-u-ka-sa-me-no-se lew.ksa.me.nosl €uedjJ.€Voc:; "praying" (aor. mid. part. NOMSgM of €UX0jJ.al elikhomai) [181.2] (b) to-ka-sa-to-ro Idoksandröl "of Doxandros" [16&] (c) ta-pi-te-ki-si-o-i It'amphideksföyl "Amphidexios~ (DATsg) [335.2] Since complex cluster signs in Mycenaean (§2.1) are onl~ for onsets and alternate with regular conventions for spelling onsets, assuming contin!lit)!:... (3) and (4) are both ORset treatments. It is then clear from these core examples that [ks] is treated as an onset, as the SH predicts. Why then are two different 7 The authority on the accentuation is Herodian (Gr. 1.44, etc.). On clitics with such words, see Smunerstein (1973:176fl); Scbrijver (1991: 95. 112.128,219) ciles the form as thirllt.

CYPRIAN SYlLABARY

33

graphic representations used? xV and kVsV should be equivalent alternative spellings, but the xV signs are normally reserv€;d for word-final position. Why should that be the case? I submit that this was originally another ingenious way of resolving the dilemma of what 10 do with theoretically possible onsets that happen to be in coda position (cf. §2.6): those that are in coda position were written with the xV series, which simultaneously insists on their onset value; those in onset positions are written by the usual convention for onsets. Another stop + sibilant string involves [ps] in o-pi-si-si-ke 16psis kel "whoever" (= An. öonc: av Mstis an) in the Idalium Bronze [217.29].8 By the SH, [ps] is a possible onset, and the [sk] in the same construct is a coda. Both are correctly written.

3.5 Morphologicallnpwand Boundary Phenomena The one (verifiable) exception is e-ke-so-si /(h)eksonsil "will have" (3pl) in the ldalium Bronze [217.31], where a coda spelling is found. Since the root is (h)ek h- "have", morphological spelling (cf. Guion 1994) is more likely than a difference in syllabification proposed by Viredaz (1983: 188), unless such a resyllabification could itself be motivated by the presence of the morpheme boundary. A similar case of compositional spelling is directly verifiable in (3b), where a word divider separates the particle and the verb. Given the normal Greek treatment of preverbs. one might expect (*}e-ko-so-ru-xe 8 For the variety of interpretations of this construct. see Masson (ad ICS 217.29). The reading adopted here is that of Thumb-Scherer (1959: 100. 160. 168. 174). The meaning is clear. Like all curses. it must begin "whenever someone or whoever breaks this law. on him [...]". The phrase therefore contains Cypr. sis (= Att-Ion. ne;" tis) "someone" and modal Ire. But what is oP(i)? Even if it were unequivocally established that Myc. opi can mean "when" (cf. Thess. O1tE( KE 0 I [K ]al.pOC;; KC:TEV~K€L opeC /ce ho I [kJairos kateni/cei [Buck 33.26-27: Larissa. II] "when the time arrives"). the opi- adopted by Wathelet (1970:84 w. lit.) and Viredaz (1983:191) is not without problems. since the ldalium Bronze uses (h)6re (lhE) for "when". as does Mycenaean (and most of the rest of Greek). If the curse were to begin "whenever someone [...J", we might then expect *(h)Ote-sis-/ce (or the like). One more-orless expects the curse to begin "whoever [.. .]" (AttlSaTLC;; hOstis). Various dialects bave a -ttform (e.g., Lesbian lhnc;; 6Uis) generalized from the neuter ~Tn 6Ui (Sappho, etc.), Hom. ihn MUi< *yod-kWid (Wackernagell885:89ff; Jacobsohn 1910:114-124; cf. Bechtel1.78); cf. Myc. jo-qi (PY 318) ly6kWkwil "whieh". This is the formation posited by Thumb-Scherer (p. 160) to underlie Cypr. 16p-sis-keJ « *h6d kWis ke), comparable to Att.lSanc; lfv hOs-tis ein (cf. Beekes 1971:341). Masson (ICS, ad loc.) disputes this on the grounds that it is "phonetil~, OTPOcj>lYYOC; strophigx, GEN strophiggos (Euripid.+), unless to be read as storphi(Ei (see Viredaz 1983:191); in either event, the stremains treated as an onsel.

=

~~as~ (a) a-ri-si-to-se /aristosl "Aristos" [102]

.

)

(b) a-ri-si-to-/co-ne laristÖkhönl "Aristokhos" (GENsg) [181.1] (c) a-ri-si-to-ke-le-we-i /aristoklewey/ "Aristoklewes" (DAT) [352.4] (d) mi-si-to-ne /misthön/ "payment" (GEN) [3x in Idal. Bronze: 217.4, 5,15] (= ~to66c; misthos "recompense" 11. 10.304+) (e) ko-ra-sa-to-se /korastös/ "richly" [264.2] (hapax: Neumann 1974) (f) e-pi-si-ta-i-se lepfstahisf "care, attention" (hT(OTaotC; ep(stasis) [264.3] (Neumann 1974) (g) e-se-ta-se festäse/ "stood; ereeted" [118,163.2] (= Att.-Ion. E'0TllOE estese) and related forms:

CYPRlAN SYlLABARY

35

1) e-se-ta-sa-ne lestäsanl "set up; erected" (3pl) [261] (= Att.Ion. E'oT1'\Oavestesan) 2) e-pe-se-ta-se lepestäsel "erected" [103+] (freq.) 3) ka-te-(e-)se-ta-se Ikatestäsel "set up; erected" (to Ka6(OTlll.ll kath(steni) [6.2+] (freq.) (h) a-ku-we-u-su-ti-ri-yo "Akuweustrl(y)ö"? [327.11 (Bulwer Tablet); see Viredaz 1983: 188, 194 w. lit] It should be clear from (5) and (6) that the treatment of [sC] as onset or coda is in no way haphazard. Word-internally, the coda treatment is regular, as predicted by the SH.9 The onset treatment is regular only in word-initial position. Mirror-image 10 the problem of theoretical onsets in coda position (above), these cases involve SH codas in onset position (cf. Guion 1994, with 'extrasyllabic' s). What does one do with a coda cluster in onset position? It could be treated as a coda and ignored, as in Pylian Linear B tu-ru-pte-ri-ja Istrupterl(y}äs/ "of alum" (alphabetic oT(phJ1JTllPW st(r)upurfä), but Cyprian writes codas. That poses a conflict since coda segments copy the (preceding) nucleus vowel, but word-initially there is no preceding vowel. The conflict was resolved in the only feasible manner. The dummy vowel (barring the somewhat bizarre alternative of generalizing the sign with diunmy e from final position) can only come from the syllable nucleus, which aligns such examples with onsets. This does n'ot mean that they were regarded as onsets. Syllable adjuncts (Levin 1985) pose a serious graphic problem for strictly SH-based syllabic representations. That the Cyprian writers were (at least implicitly) aware of the problem is indicated by the fact that in non-initial position they maintained the SH-sanctioned treatment. That they were able to go beyond the confines of a SH-based syllabic representation and write the adjunct Isl at all illustrates a clear conception of the segments involved and their relation to the SH.

3.7 Exceptions antI OtheT Conditioning Factors Rare exceptions occur in both directions. A rare (regional?) variant of the (e-p)e-se-ta-se I(ep)estäsel class (6g) is e-sa-ta-se [92.2] and e-pe-sa-ta-se [93.1] (Salamiou). Since the root is st~ "stand", the spelling is most likely motivated morphologically (cf. Beekes 1971:341; Viredaz 1983: 191). 9 In this context, observe the spelling of inalalisimena linalälismenänJ "engraved" in §3.3 (conlinuant + nasal). This is consistent with the adjunct representation of s in sC clusters, except that Isl should be lower in sonority than ImJ, wherefore Guion (1994) posits morphological spelling. viz. [[-lis]menän].

36

ANCIENT SCRIPI'S AND PHONOLOGICAL KNOWIEDGE

The major (verifiable) exception in the other direction involves a wordinternal onset treatment in ka-ra-si-ti Ignisthil "eat" (I es 264.1 Neumann 1974). The form is imperative to (the rare verb) ypaw grdö "gnaw, eat", (probably colloquial; cf. ypaoTl4;' grdstis "fodder"), parallel to po-ti lpöthil "drink" in the same line (cf. Thumb-Scherer 1959: 168).10 As emphasjzed by Beekes (1971:341), the syllable boundary cannot be [grasthi] because it is unlikely that a dactylic hexameter would begin with a short syllable (but see Miller 1982:53f1). The spelling is exceptional even for that text; cf. ko-ra-sato-se Ikorastösl (6e), e-pi-si-ta-i-se lepfstahisl (61). I suggest that this is another attempt at compositional spelling. The root is synchronically gra- and the imperative is -thi, but what is -s-? Historically, it was part of the root, but synchronically a number of -s- formations in which [s] became [h] and then disappeared were reanalyzed as vowel sterns (see Miller 1982, chap.5, § 106, etc.). Synchrot}ically, the problem is similar to Herodian's dilemma regarding the quasi-compounding -s- in words like [e e ~], [05 Q] [a ä]. Additionally, it had long been realized that the signs representing Iyl and Iwl could be used for lil, lul (cf. already the Ugaritic script of -1400). Greek no longer had Iy/; which had been lost by this time; in Mycenaean already, Iyl variably altemated with Ibl or 0 (see Wyatt 1968; Miller 1982: 144ft). Therefore First, I [y] served for li/, liI. Second, Y (originally [w]) presumably, like Latin V (Nilsson 1918: 186ff), also served for luf, lül (cf. (?) rare Cretan spellings like aFTov awt6n "him(self)", etc.). The form was stylized to F; it defies common sense to imagine that the resemblance to Linear B (+ luf is accidental (cf. Cypr. 'Y' lu/). Because of the stylization of waw to F, the contemporary Phoenician form Y could be reborrowed for lu!, lül (cf. Diringer 1968:361; Heubeck 1979:89-90). As a late borrowing, it was placed at the end of the alphabet -

TIIE GREEK ALPHABET

47

the first of the supplementals. 6 Incidentally, older and younger forms of the same letter in an alphabet presuppose prehistoric evolution - a problem for Powell 's hypothesis. Reborrowing Phoenician y provided for a unique letter for each of the five cardinal points of vowel articulation (quantity and tenseness (referred to by some theoreticians as advanced tongue root) were not distinguished): ii

ee

uü 05Q

~

aä

It is not likely to be coincidental that the five vowel points that are distinguished in the Greek alphabet are the very same ones that were already distinguished in all of the older (Near Eastern and Greek) syllabaries (cf. Gelb 1963:120-176; Jensen 1969:456-457). This militates against both the Accidental Development hypothesis (e.g., Praetorius 1908:284; Daniels 1992; Faber 1992) and the Unique Adaptation hypothesis that the alphabet was invented first and exclusively in Euboea to record Homer. Gelb (1963:181182) and others (cf. Heubeck 1979:86) have claimed that, given the long tradition of representing those same five vowels, combined with the availability of the old laryngeal and pharyngeal letters, the development of the vowel signs in the Greek alphabet was no accident. Also against the accidental invention hypothesis is the fact that vowel letters were independently invented by the Aramaeans in the -11th century, using w for 6 For paralIeIs to the placement of new letters at the end, see Diringer (1968:151), Segert (1993:86). Why wasn't waw displaced to the end? Quite simply, the letter was reborrowed, not as a consonant, whose place was defined by the model abecedaria, but as a vowel, which had no place in the Phoenician prototypes. Another possibility is that F developed from a cursive Phoenician form (EG 1.77; Wachter 1989:37ff). This is generally regarded as not likely (LSAG 1990:427). In either event, F reßects the name and place of the Phoenician consonant while the clearly related vocalic alternant was placed at the end of the alphabet. This does not (pace Wachter 1989:72ff) entail that the Greek alphabet was ever vowelless. To the contrary, it seems likely that the rationale for the alphabet was to represent vowels in a simpler manner than in the syllabaries (§4.13), but nothing precludes an early use of waw for both Iwl and Iu!. One other possibility must be addressed, that waw originally represented both Iwl and Iu! and then, in the position after E, evolved to F. This, however, will not account for the positional differentiation. If there were just Olle form Y for Iwl and Iu!, there would be no need for two positions. Therefore it can only have been after Y was stylized to F in its basic position that the förm Y could have been reborrowed in a new position (first of the supplementals), creating a formal split between F Iwl and Y lu/. Powell (1991: 31,43) claims that the adapter created a split between F Iwl and Y Iu!, but that in no way motivates the particular forms. His "simple variations" does not explain why it is just the way the Evolution Theory predicts and not the other way around, or some other possibility.

,r'" I

I I

48

ANCIENT SCRIPfS AND PHONOLOGICAL KNOWLEDGE

lül, Y for IV, and h for other final vowels (Cross 1989:86; Faber 1992: 130); cf.

Middle-Late Kingdom Egyptian (Gelb 1963: 168ff; Davies 1987:37). Finally, it hardly seems accidental that the only sign reborrowed as a vowel (distinct from the corresponding consonant) has a strikingly identical form in the Cyprian syllabary, where otherwise the vowelletters are quite different from those of the alphabet (see Appendix C).

4.6 LoteTIRegional Vowel Lette" In Fast Ionic, which lost initiallhI, hila evolved into eta, which provided a letter for leI (hy the acrophonic principle). That left an asymmetry - a gap in the corresponding 15/ slot (cf. Nilsson 1918:189), for which a letter was provided by opening 0 (omicron) at the side or base, thereby creating n (omega), the first example of which is disputed, depending on the relative dating of the aOAH1NoI7 Doll:;nos inscription from Old Smyma (LSAG 341,345(69), pI. 66; Heubeck 1979: 93, 126: end of the -7th cent.?) and the graffito abecedarium from Samos (EG 1, fig. 42; LSAG 420, 471, pI. 79; Powell 1991: 157: ca. -660-650). According to LSAG (1990:428), the abecedarium contains the oldest omega, but Powell (1991: 61n.178, 157n.87) follows Guarducci (00 1.159-160) in assigning to a Parian sherd with omega the date ca. -700, making that currently the earHest example of omega (if the date is correct). The new letter n was added at the very end, as the most recent supplemental. Eta continued to occupy the position of its source heta. The Ionic alphabet (featuring eta and omega) was in -403/2 (archonship of Eucleides) adopted at Athens whence it became, during the -5th and-4th centuries. the official alphabet in many other areas of Greece as weil (cf. Threatte 1980:26-27). In Athens. the Ionic alphabet began to be used ca. -450 in State Decrees and was completely adopted by -403/2: "No public documents can be found in the Old Attic alphabet after this date. and scarcely any trace of the older script can be found in any text" (Threatte, p.27). 4.7 The SupplementoJ. Consonants The supplemental consonant signs constitute by far the most complicated development. Part of the problem is that letters for the aspirated series were created, parallel to the paired T Itl, ® Ithl - itself not without problems. Heubeck (1979:89) and Powell (1991:39-40 w. lit.) reiterate Praetorius' (1908:285f1) putative contradiction in borrowing T as Itl rather than Ithl (since it was aspirated in Semitic), and pharyngealized (not velarized, as Powell calls it) ® as Ithl rather than Itl. when the pharyngealized stops were not 7 The last

0

and the -s are secondarily inserted (see Wachter 1991:63).

1HE GREEK ALPHABEf

49

aspirated. However, Heubeck and Powell miss the point that stylistic and distinctive features are not the same. The plain stops in Semitic were aspirated as a non-distinctive feature, just as they were in Ancient Greek and (to some extent) all other languages (cf. below). One must be careful to distinguish the non-distinctive aspiration of the unmarked phoneme ItJ from the systematically marginal character of the marked phonemes IV and Ithl (marked in the sense of characterized for additional information and more difficult to acquire). Tbe letter T associated with the unmarked Itl in Semitic retained the unmarked association in Greek, while the letter ®, associated with a marked value in Semitic, was applied to the corresponding marked segment in Greek (cf. Nilsson 1918:184) and in the Indic Brähmi script (Nilsson, p.I94) .. The systematic association of letters with unmarked and marked values across languages presupposes, with Naveh and others, a bilingual environment, as in Ancient Cyprus or Crete, and precisely not the sort of superficial value associations (predicted by Powell) that are characteristic of casual contact situations (e.g., with an informant), as shown by Kiparsky (1973). This provided a sign for Ith/, but there was none for Iphl or Ikh/. These were the next signs filled in: cf> , X .8 Very early inscriptions from Crete, Thera, Melos, Sikinos, and Anaphe do not use these supplementals (cf. Powell 1991:49). The question is, why not? There are only two logical possibilities. Either (1) the original Greek alphabet had the supplementals and some areas (Crete etc.) lost them, or (2) the original Greek alphabet had only one supplemental, 1'; the 'primitive' scripts of Crete etc. would preserve that situation. These polar extremes are recently represented by Wachter (position 2) and Powell (position 1), each with an impressive barrage of support from scholars of the past. Needless to say, it matters less what anyone 'thinks' than 8 It must be explained why, in all Greek alphabets, VII upsilon is the first of the supplementals (i.e., #23), and why cl» invariably represents Iphl and is normally the 24th letter, displaced to 25th only in the 'central' cuhure area (Euboea, Boeotia, Athens), where a form of X (= [ks] in Euboea, Boeotia, [kh] in Athens) occupies 24th position (see Wachter 1989: 31). Most scholars (even,limitedly, Powell) assume different layers in the development of the Greek alphabet. Wachter (1989:40-48) assumes first I, then cl»X - except that he takes the Athens type XcI»as primary. Just why X has displaced cl» in that region is impossible to determine for certain. Wbat is certain is that (pace Wachter) the displacement cou1d not have taken place the opposite way in the other areas because (1) the value of cl» is coosistent while the value of X is not; (2) X is not always the 25th letter when cl» is 24th. As noted by Praetorius (1902), the indication is that cl» is older than the other supplementals (except for 1'). evidenced by the inconsistency in their values and positioning. Wachter cannot explain the invariant association of cl» with Iph/. To explain that requires that cl» be more established than the rest. On the 'primitive' alphabets see the main text below.

r I

50

ANCIENT SCRlPfS AND PHONOLOGICAL KNOWLEDGE

whether or not there is any empirical evidence for either of these positions. The most empirical evidence adduced so far has been by Powell (1991:57), who shows that it is likely that the earliest Cretan abecedarium had at least because it occurs in Eteocretan inscriptions from the -6th cent., e.g., in the name of the town - within but across alphabetic traditions 13) about the form - X~' \V. A - and value ([ps], [kh]) with regard to kh(e)i - W, \V, - might be explained by the assumption oftwo very similar letters, initially distinct, one an affricate, the other a velar displaced by the reborrowing of kappa. Suppose the former was an older borrowing but an unneeded letter. Then, as the quasi-affricates (clusters of stop + s) received letters,14 it was used for that purpose. What about the order? Why did it not just go in the pI ace of san? There was general agreement among the Greeks that san had the shape M (whatever its source - a recent reborrowing causing displacement of older 'V?). Whatever the details, it could not occupy the position of the current conception of san, and fit formally and functionally with the other supplementals at the end of the alphabet. 15 For a radically different proposal, see Wachter (1989:49-61).

4.10 The Antiquity ofSegmental Writing The judicious survey of the history of literacy in India by Patel (1993) concludes that writing may have existed in India since the Indus civilization and that it was practiced in the Vedic period and during the time of Pä!)ini. (See §l.l, fordevanagarias a segmentally coded script.) The absence of extant writing from before the -5th century is explained by the fact that "early Brahmanicalliterature was written on frail and perishable leaves, birchbark. 13 I owe this formulation to Rex Wallace. 14 Why a language should have cluster-symbols

for [les] and [ps] has occasioned considerable speculation. For instance, Sampson (1985: 103) attributes their existence to the fact that those are the ouly clusters that could occur at syllable and word end. More to the point, they are the only onset clusters that could occur in coda position. The unitary letter reflects awareness of the old convention of providing complex signs only for onset clusters. 15 One other possibility (from Praetorius 1902:677ff, 1908:287-288) is mentioned by Jensen (1969:463 w. lil) and Faber (1992: 125, 131n.27), who discuss early Sonth Arabian letters that closely resemble phi and psi (variant form and function of khi). Faber speculates that the old Phoenician alphabet from which both descended (Sass 1988: 166-167, dates the South Arabian planned creation of the alphabet to the -1l/IOth century) must have bad those letters (cf. Praetorius 1908:288). The letter that ends up as psi, as noted above, does have a correlate that could easily be confused with and/or split off from the old form of IkJ, which agrees with the S. Arab. value as a pharyngeaI fricative. The variant of phi with the v8lue Iwl or lvi is again not too difficult to motivate if also derived from qoppa, which had always been associated with lip-rounding. Already Nilsson (1918:183) emphasized that Praetorius' hypothesis of Greek letter borrowing from S. Arab. was "sicher unrichtig", emphasizing internal paralleIs.

1HE GREEK ALPHABET

55

and later on hand-made paper" (Patei 1993:202). Similarly, Old Canaanite written on papyrus did not survive (Segert 1993:87). Mycenaean documents were probably transferred to perishable materials (Olivier 1986). Continuity of the Cypro-Minoan script in the Cyprian syllabary (§3.0-3.1) presupposes interim documents on perishable materials (cf. Heubeck 1979:73; Olivier 1986). As noted above and inchap.5, similar arguments have been made for Greek and Germanie. Nevertheless, it must be emphasized that this has no hearing on the literacy of Homer (against Bellamy 1989, see Miller 1990) or of the public in general. William Harns (1989, e.g., 101ft) finds that literacy in Ancient Greece was restricted largely to a privileged minority and coexisted with an oral culture (cf. Andersen 1989; Thomas 1989, 1992). The absence of documents, then, is not a major obstacle to the general idea that alphabetic writing in Greece can antedate the earliest epigraphic monuments. Another parallel with India can be adduced. Just as the oral mode of transmitting Vedic literature prompted development of grammatical analysis (Scharfe 19n, Patel 1993), so the orality of tbe very popular Homeric texts may have underlain the shift from a 'folk' grammatical tradition (Morpurgo Davies 1987) to a (more) professional guild of grammarians, whose main function seems to have been textual exegesis, as stated fairly explicitly in the introduction to Dionysius Thrax's TEXVll ypaliliaTtKll "Art of Grammar" (see Kemp 1987: 172-173), and directly manifested in the copious Homeric scholia (e.g., Erbse 1969-) and other ancient commentators. 4.11 Adoptation anti Developmeni'Phase ofthe Greek Alphabet Another argument for greater antiquity is that the Brähmi script of the -3rd cent. Ashokan inscriptions represents the Sanskrit phonological system so weIl that it must have had a long history of development (Basharn 1967:394; Patel 1993:203). The Greek alphabet, on the other hand, appears to be less weIl adapted to the phonological system of Greek, but the opposite has also been argued; cf. Coulmas (1989: 162): The Semitic alphabet applied to a non-Semitic language could not be used to represent the sounds of that language without significant adaptations. The lack of signs for vowels was crucial here since, in contrast to the Semitic languages, vowels in Greek occupy a position on a par with consonants. By finding a solution for the problem of vowel indication the Greeks overcame this obstacle, thus making the alphabet more suitable for both their language and other non-Semitic languages.

It has been argued (e.g., Harris 1986:120) that the North Semitic alphabet may have ignored vocalic differences in reducing the earlier

.56

ANCIENT SCRIPfS AND PHONOLOGICALKNOWlEDGE

syllabary, thereby adapting better to the word structure of the Semitic languages. So, by (re)developing symbols for the vowels, the Greek alphabet was undergoing adaptation to the Greek morpho-phonological system. Sampson (1985:101) emphasizes that Greek had many lexical contrasts with vowels, wbich were "important for communication" (cf. Diringer 1968:263, 435). Moreover, words frequently begin with vowels in Greek but not in Semitic, and sequences of vowels are virtually unknown in Semitic but frequent in Greek (e.g., Atala "Aeaea", Circe's island). Daniels (1992:97) sees consonant clustering in Greek as a factor. Other changes (development of separate aspirate symbols, addition of other vowel letters, etc.) were all attempts to better adapt the alphabet to the Greek phonological system. Although the adaptation was never complete(d), the process of change and adaptation at the dawn of documentation seems to be continuing a prebistoric process, from which it is reasonable to conclude that, like the Indic scripts, the Greek had a long bistory of development Coulmas (1989: 164) affirms: In principle the Greek alphabet was suitable for representing all the phonemes of the Greek language [...]. Systematic vowel indication is attested in the earliest Greek documents; no developmental state with defective vowel writing is known.

This suggests aperiod of evolution and development of orthographie norms and conventions (cf. Gelb 1963: l8Off). There were also choices made. In the Indic linear scripts, syllable structure was partially coded by different representations of onset vs. coda resonants (Mahulkar 1981:49); cf. the Greek syllabic scripts (chaps. 2,3 above). But the Greek linear script was more strictly segmental in that little reference to bigher levels of organization was made. While the Greek tradition was weil aware of organizational units beyond the segment, the decision was made to represent little more than linear segmental units. Therefore, it is pointless to argue that the Greek alphabet was poorJy adapted to the phonological system. All of this presupposes at least implicit knowledge of segments and their organization into bigher linguistic units. That is, of course, another area of major controversy, which will be treated in Chapter 6. 4.12 Suminary and Conclusion

Powell (1991) unfortunately ignores completely the evidence for an early borrowing of the Greek alphabet: (i) the letter-forms and their ProtoCanaanite prototypes; (ii) internal evolutionary evidence, such as Y to F; (iii) evidence for reborrowings by comparison of 'duplicate' letters with their earlier forms and contemporary Phoenician counterparts. Moreover, (iv), he

TIIE GRFEK ALPHABET

57

fails to motivate the order of the supplementals and the constant association of cl> with Iph/, and (v) his theory of an adapter (as opposed to a bilingual, Hterate environment) fails to motivate the matching of unmarked and marked values across the languages. The stages in the development of the Greek alphabet reconstructed here are the following: (1) assignment of old 'pharyngeal' letters to vowels; (2) evolution of Y to F; (3) reborrowing of Yat the end of the alphabet (first of the supplementals) as a vowel sign sanctioned by the corresponding syllabary traditions; (4) 1> evolved to IrJIl via the normal phonological development of lkw(h)o/a] to [p(h)o/a]; (5) for the non-aspirate Ip/, 1T was naturally used, leaving the letter cl> exclusively for the labial aspirate Iph/, all of which alone explains the consistent association of cl> with labials; (6) the reborrowing of contemporary (qoppa) from Phoenician entailed a displacement of the sign phi with the marked value to the end of the alphabet; (7) evolution of W to X and 'V provided a possible separate sign for the marked value Ikh/, which was exploited by the reborrowing of 'modem' kappa for the unmarked value IkI, as provided for in the model abecedaria; (8) regional differentiation of signs for affricates and affricate-like clusters; (9) rejection by conservative 'southem' areas of supplementals not sanctioned by the syllabary tradition (thus l' was allowed to remain, but not the supplemental consonant letters); (10) the East Ionic evolution of eta and (within the historical period) creation . ofomega. Powell may be right that the spread of the alphabet in Euboea had to do with recording Epic, but Thomas (1992:56-65) rightly assails this view as "romantic", emphasizing that most of the earHest Greek alph~betic writings are not poetic at alt. More likely, it was the high level of cultural activity at that place and time that prompted a renaissance of interest in the Greek alphabet, as a consequence of which (at least short) examples of epic verse were recorded, along with a wide array of other things. In any event, Powell is surely wrong that the Greek alphabet is the product of a single adapter at that time and place, though a single adapter may have ultimately been responsible for the initial creation of the alphabet. The two views, of course, make entirely different predictions. Powell predicts that no very early inscriptions, say, using 1> for Ikw(h)1 should ever turn up and that, if earlier inscriptions and abecedaria do turn up (and the date of Homer would - circularly - have to be moved back), regardless of where they are found, they should have the full set of supplementals he reconstructs. My prediction is the opposite, of course. To the extent that it is plausible for inscriptions a century or two earlier than the current corpus to

«

r

58

ANCIENT SCRIPfS AND PHONOLOGICAL KNOWLEDGE

I

ever turn up, neither of these hypotheses is properly testable (empirically verifiable or falsifiable). Hopefully time will teil. 4.13 Interaction Between the Syllabary anti Alphabet Tradition. Another issue involves the assumption of some continuity between the syllabary and alphabet traditions. Such continuity is typically denied (e.g., Stroud 1989: 110). Faber writes (letter of 18 Jan. 1993): I'll grant you your demonstration that Mycenaean spelling conventions probably reflect sub-syllabie awareness. if you will grant me that there could not have been complete eultural continuity between users of Linear B and of early True-Greek orthography. If there had been such continuity. one would expect to see some reflexes of it in early True-Greek orthographie convention [...].

Of course there was no complete continuity, or there wouid have been no change. Yet there was continuity, and it has been documented by Hermann (1923), Morpurgo Davies (1987), and others. Additionally, everyone grants that there had to be continuity of the syllabary tradition from Cypro-Minoan to the Cyprian syllabary of the historical era, despite the absence of Greek documents in any related script during the 'Dark Age' (cf. Heubeck 1979: 65ff, 85ff; see §3.0-3.1 above). There was continuity of the conventions of syllable division based on the Sonority Hierarchy (and feature geometry) from Mycenaean on in both the syllabary and a1phabetic traditions. This was carried down to the detail that s plus stop received special treatment. The alphabetic tradition continued the recognition of the special problem of onset clusters in coda position. Meister (1894: 185) noted the agreement of the syllabary and the alphabet in having a unitary sign for [ks], and Nilsson (1918: 184) further noticed that the form of X = [ks] is paralleled by the form of the Cyprian sign for [ksa]. This was generally taken to imply influence of the alphabet on the syllabary (cf. Masson, ad ICS §28.7a), but it is actually the other way around, since the only clusters that traditionally (Mycenaean+) received special letters were onsets, and that is inherently a syllabic notion. Another syllabic coding in the Greek alphabet was the nonlinear (and nonsegmental) representation of aspiration (see Steriade 1982). Daniels (1992: 197) makes the astute point that one of the reasons (if not the main one) for the shift to the alphabet had to do with the extremely awkward representation of Greek consonant clusters. For example, str6phigx "pivot" in some tradition might have been (*)so-to-ro-pi-ni-xe (cf. §§3.4, 3.6). To that one can add the problem of identifying where the vowels were.

1HE GREEK ALPHABEf

59

For instance, se-pe-re, as in se-pe-re-ma Isperma! "seed", could also be read [sepre], [spre], [sep-e], [spre], etc. These motivations also presuppose interaction between the syllabary and alphabetic traditions. This will be elaborated in Chapter7. The representation of the same five syllabary vowels was continued. This is important in response to Gelb's point (1963: 182) that it is improbable that one person developed the exceptionless use of vowelletters using as a model the linear Semitic scripts with their highly irregular vowel notation. Since syllabaries invariably represent vowels (by definition!), which is natural, that being head of the syllable (chap.l), this feature of the alphabet was another point of contact between the two traditions. Moreover, at least one of the vowel symbols exhibits identity across the systems - conspicuously the only one of the supplementals to be permitted in the 'south' where the syllabaty tradition of no separate signs for the aspirates prevailed (except for e which was sanctioned by the Phoenician script). Evidence has been presented that at least one symbol (with amazingly close forms of similar function in Linear B) evolved over time by means of changes in the phonological system of Greek. In the earliest inscriptions, the same assumptions about words. clitics, and word-divisions were maintained in the alphabetic tradition that prevailed in the syllabary tradition (cf. §6.4). Finally, Heubeck (1979: 67-68,86) makes the interesting point that a very archaic Cyprian inscription (ICS 174: Paphos, second half of the -8th cent.) is atypically retrograde (to-ro-to-so-si[- i.e., [Lü]si-stortö "of Lusi-stortos" [Neumann's restoration, accepted by Viredaz 1983: 191]), and claims that this can most easily be explained by interaction with the Semitic script. Moreover, the Cyprian syllabary and the alphabet both begin to thrive around the -8th century, suggesting a renewal of interest in both forms of writing. Any of these points in isolation is subject to challenge, but the composite picture that emerges is one of considerably more interaction between the syllabary and alphabet traditions than is typically granted (except by Nilsson 1918). The composite evolutionary theory points to a compromise between the vowelless Semitic script and the superfluous vowels of the syllabaries to create a script with the advantage of indicating where and what the vowels were (chap.7). Influence from the syllabary tradition came both in the form of particular symbols and orthographic conventions. Given the continuity from the Linear B to the Cyprian syllabary, there is nothing inherently bizarre about the idea that the alphabetic tradition developed concurrently and that scribes familiar with both scripts transferred syllable-based conventions to the alphabet, or, in the case of consonant clusters, deliberately distanced themselves from the awkward vowel repetition. Sometimes, as noted, the influence

r

60

ANCIENT SCRIPfS AND PHONOLOGICAL KNOWLEDGE

went the other way. This Greek-internal evidence accords better with the findings of Semitic epigraphers who recently opt for a -10th (or -9th) century date for the origin of the Greek alphabet

5. THE RUNIC ALPHABET 5.0 Introduction The Germanie peoples used a 'runie' alphabet from around the first century 10 the Middle Ages. The word rune has occasioned much speculation. Gothic riina can tmnslate Lat. mysterium "mystery; secret". Old English run means "mystery; counsel; discussion; word" (Fell 1991). There is one mention of runic letters (run-stafas Accpl) in Beowulj 1695, in a description of inscribed golden hiltplates on the captured sword that Beowulf gives 10 Hrothgar (cf. Elliott 1989: 17). Runic letters (runstajum DATpl) are themselves equated with magie (dr ycrmft) in JElfric, Homilies 2.358 (cf. Elliott 1989:81). The 'secret' was important enough 10 be borrowed by the Celts, e.g., O.lr. run "seeret" (cf. Elliott 1989: 1ft). Just why runes were surrounded by so mueh mystery and seerecy is itself a mystery. An10nsen (198Oa; 1988; 1989: 140ft) takes a very strong position against magical theories of the runes. It is true that the development of the runes need not be further obfuseated by the fantastic, but nothing preeludes associations with ritual and magie (Düwel 1983: 111ff w. lit.). as in Ancient Greece (Thomas 1992:78-88). In any event. as Antonsen insists, that is just one USE of the script that has no bearing on its creation. original funetion(s). or the original meaning of the word. which may have had to do rather with scratching (Morris 1985). but see Fell (1991); early runic rfln was "message; text" (Antonsen 1990:314). Since runes were the stock-in-trade of the writers in runes,l the very knowledge of the letters in a largely illiterate society could have prompted the interpretation as "mystery; seeret". more-or-Iess as in ancient Babylonia the 'supreme seeret'. the key 10 the uni verse. that the god Ea taught his son was the concept of the number 1 The word erilaz is sometimes translated "nme-master", which EImer Antonsen (lener of 24 Feb. 1993) desaibes as "a stab in the dark. All we know about it is that it is used in parallel with terms like gudija 'priest' and pewaz 'servant' ." Antonsen (e.g., 1981:56-57) translates erilo:t. simply "eril" and suggests to me the phrase writers in runes for the present context. It has also been suggested that the term originated as a tribal name, Heruli, of people skilIed in nmecraft (cf. Elliott 1989: 11-12 w. lit.), but there is no evidence for that (Antonsen 1990: 314). 1 wisb to take this opportunity to thank Professor Antonsen for several sets of extensive comments on this chapter. Marie Nelson and W. C. Watt also read an earlier version, and Jay Jasanoff sent me detailed comments on the Gennanic Vowel System (Appendix).

62

ANCIENT SCRIPfS AND PHONOLOGICAL KNOWLEDGE

(Hopper 1969: 12). More directly relevant is the point made by Watt (1989: 92n.3) that from at least the' -6th century 10 the Middle Ages mystical power was associated with the correct recitation of the abecedarium. There are in the neighborhood of 5000 runie inseriptions, some 3000 in Sweden alone, ca. 1100 in Norway, some 700 in Denmark, around 60 in England, and so on. In general, the farther south one goes, the more rare they become (Düwel 1983:3). This creates a problem for Italic theories of the origin of the runic alphabet. This chapter presents additional evidence for the archaic Mediterranean theory of the origin of the runic alphabet, and shows that it was created along phonetic parameters analogous 10 those underlying the ancient scripts of Byblos, Ugarit, and the Phoenician script.

5.1 The Older Runic Fu park There are extant some 250 early Germanic inscriptions in the older runic alphabet (Düwel 1983:123), though only a little over 50 have more than two identifiable words (Antonsen 1980b: 1). An idealized version of the 'older jupark' (named from the first six letters), which had 241etters, is presented in (1), following Antonsen (1989: 142). (1) Older Runic Alphabet (Idealized)

f,,1>

~ ~


w

rz ss [>3 ZlS s16 X/h9 syllabic: U2 Cl4 in i:J13 sonorant: rs Ws nlO Y12 stop: vcl.: ~ P14 vcd.: g7

d24

It cannot be accidental that, on a sequential projection from 11 to du, the letters all fall in place as natural classes with respect to their major category features. The odd letter ;8, conspicuously hearing the number , fonns a kind of dividing line. The continuants to the left of 'center' are arranged front to back; the most strident (sibilants) stand alone on the right. The vowels to the left of center occur in two clusters, back lu/lai before front lil Ii:J/, which in turn subdivide into high before low. The mid vowels leI 101 occur (mirrorimage) front before back to the right of center. lO The resonants are more 10 Altematively, one might assume, following the discussion of Watt (1989:81-82) of midpoint divisions in Canaanite abecedaria, that was the dividing line, in which case the mysterious vowel should be another mid vowel. As noted above, that is a possibility.

THE RUNIC ALPHABEf

73

difficult and suggest a projection that is not yet clear. As they stand, there is a curious i nterspersing. Possibly Irl is 'placeless' (§ 1.2) and stands alone, followed by front to back Iwllnl/yl (glide-nasal-glide) on the left and Iml 111 101 (nasal-lateral-nasal) on the right (on the patterning of nasals and laterals, see Rice 1992:62).1l Perhaps most interesting is that the arrangement follows a principle identified by Watt (1989:71) for the Ras Shamra Matrix (§5.11 below), altemating similar and dissimilar classes. The stops subdivide very naturally into voiced and voiceless, and in both cases, the order is velar > labial> dental, conforming to the SH and/or independent feature geometry (§ 1.2). Moreover, dentals also come last in the Canaanite matrices (below).

5.10 Antiquity o/the Phonological Matrix: Byblo. The phonological knowledge underlying the runic fupark is of greater antiquity than the runic script, as shown by Watt (1987, 1989) in his analysis of the Phoenician Byblos Matrix (ca. -1000), and even older cuneiform Ugaritic script of the Ras Shamra Matrix (ca. -1400), on which see Dietrich & Loretz (1988), Segert (1993). Both are descendants of an earlier ProtoCanaanite script whose organization can be assumed to have had a similar phonological basis (see Segert 1993:87fO. The Byblos Matrix (Watt 1987:2) is reproduced in (7), with phonological interpretation in (8).12 (7) The Byblos Matrix

I ~

~

11

1

Y ~''l

0

III

J1

'I-

IN

~

r

4..J

IV

V

7

L\

~ \j/

cp

$

~

"f ? -I-

11 Tom Sawallis suggests to me (p.c.) that tbe sonorant series might be arranged acoording to descending tongue height, hut simultaneously cautions regarding tbe subde nature of such an observation, especially in the absence of detailed phonetic information. On tbe otber band, if the observation is correct, that would provide detailed phonetic information. Needless to say, nothing can be based on such circular reasoning. 12 The interpretation in (8) is m08tly from Watt (tbc; matrix and all of its categories), but partly from Faber (1981, 1990, 1993), especially for the interpretation of the sibilants.

74

ANCIENT SCRIPfS AND PHONOLOGICAL KNOWlEOOE

Like the Runic Matrix (5), the Byblos Matrix has 'gaps' due to the lack of all the phonemes necessary to complete a feature matrix. Empty cells are necessarily characteristic of any phonetic-feature-based script (cf. Allen 1953 on the Indic prätisäkhya tradition). Watt (1987: lfO states: (1) the number of empty cells is minimal; (2) the chance of just any random order of letters fitting such a grid is infinitesimally smalI; (3) reversing any two letters would complicate the matrix; and (4) statistically, it would take about a trillion random arrangements of tbe letters before accidentally bitting on tbe one that fit tbe Byblos Matrix. (8) Phonological Interpretation of Byblos Matrix III I TI LARYNGEALS 13

1) 2)

Bll...ABIALS

1 h

b w

~

m p

3) 4) 5) 6)

IV

V

ALVEOLARS VELARS DENTALS

g

d

dZ Y n

1;1

~

k

1

~

q

ts

r t

5

5.11 The Ras ShamraMalrix A Ugaritic abecedarium (9), from ca. -1400, exhibits the same phonologica1 motivation for the ordering of letters, as clarified by Watt (1989:62). (9) Phonologica1 Interpretation of Ras Shamra Matrix (Watt) EXTREME BACK EXTREME FRONT

n

I

MIDDLE

III

Laryngea1sl Labials & Interdentals Alveolars Pbaryngea1s & Palatals non-frlc. frlcalive

1)

1a

2) 3) 4) 5) 6)

h

~

b w m p

BACK

DentoAlveolars stop cont frlc. non-fric.

g

p

V

Velars

z

a

FRONT

IV

II

d ~

1;1

Y n

k

~

q

5

Z

1 tS

r y

13 'Laryngea1', as Watt (1989) explains in considerable detail, is merely a traditional coverterm for glottals and pharyngeals, i.e., 'extreme back' sounds (pharyngea1 cavity, in current feature geometry). Watt also provides extensive discussion of the other categories.

1HE RUNIC ALPHABET

75

Wbile tbe Byblos Matrix has 22 letters with 8 empty cells, tbe Ras Sbamra Matrix bas 27 letters witb 21 empty cells. Watt's divisions could potentially be simplified to reduce tbe number of empty cells (more easily in the 1987 version), but Watt (1989:83-84) defends tbe empty ceHs and their distribution, based in part on the rhythm of pronouncing the letter names. In fact, Watt claims, tbe arrangement was set up to be recited vertically as weil as horizontally. Also, as noted by Watt, tbe system is pattemed in terms of contrasts - (extreme) back/front> middle > back/front (1989:71).14

5.12 Empty CeUs anti the Antiquity ofthe Runic Matrix Recall tbat tbe Runic Matrix (5) bas 24 letters with 11 empty cells, five double occupancies, and one slot witb tbree phonemes. This can be reduced by recognizing, witbin a given place feature, independent projections of consonants and vowels, a non-problem for tbe Proto-Canaanite scripts. It can hardly be accidentaI tbat, of tbe 6 multiple occupancies, four are paired C and V sets, viz. jlu, alr, nli, and y/ &8. Tbe suggestion is that tbe inventor(s) of the runic alphabet viewed c1ass and pi ace features almost as independent planes. One can forever debate tbe precise categories and projections of these matrices, but (i) there is always going to be a trade-off between empty cells and tbe number and type of categories recognized, and (ii) the leading idea must not get obfuscated among mounds of semi-irrelevant details. Wbatever class, place, and manner categories are ultimately adopted, the essential point remains tbat a number of scripts since tbe middle of tbe -second millennium bave employed essentially tbe same pbonological knowledge in their construction. Tbat knowledge includes words, syllabies, segments, and the organization of segments, not only according to the Sonority Hierarcby, but also by place features and manner of articulation. Tbe runic script itself bears evidence of antiquity, botb in tbe epigraphic factors in §5.3, and in the letter orders. At first glance, the Canaanite matrices appear to be quite different from tbe runic in (5). However, recall that the projection in (6) exhibits the stop order witb dentals at tbe end, as in (8) and (9). Moreover, tbe laryngealletters in tbe first slot of the Canaanite matrices had long since been reanalyzed as vowels (see cbap.4). When the laryngeal category is disposed of, next in order are tbe labials, witb wbich the Runic Matrix begins. Then, P follows tbe original interdental position in (9). A 14 Given the grammatical treatises of Sumerian formatives in Old Babylonian from ca. -1700 (Jacobsen 1974), it is ciear that there was an even oider tradition of intricate grammatical analysis. which. although morphological. indicates that the tools oflinguistic theorizing were availabie.

76

ANCIENT SCRIPfS AND PHONOLOGICAL KNOWLEOOE

number of letters also occupy the sarne relative position as in the ProtoCanaanite script (Appendix A). Most conspicuous are the 'middle' four (p z s t): Ipl before /21 is a perfect match with Proto-Canaanite and , as also Isl before Itl matches Proto-Canaanite and . The major divergence from the Canaanite scripts, then, involves the realization that after labials, the rest of the dentals can be grouped with the interdentals, in which case the minimally-differentiated front-back system of labials> dentals> velars follows naturally. In several places, Watt (1989) seems to larnent the demise of the matrix. But perhaps the matrix (in modified form) did survive, in the Runic Matrix. 5.13 Concbuion

The anangement of the Runic Matrix may not appear to be economical, but there is a good deal that is not known regarding the assumptions about phonological theory made by the inventor(s) of the runic alphabet. If there was a 3-dimensional conception that would enable vowels to be projected on a separate plane from consonants of the same place features, some of the abenances (such as double-filled cells) disappear. The sarne might hold for the curious interspersing of nasals and non-nasals (§5.9). If somehow nasals were conceptualized to project onto an independent plane, that seemingly uneconomical anangement (by some current criteria) would also vanish. It is statistically significant that the letters of the jupark, in order, fit a phonetic matrix precisely, allowing for 'gaps' in the system. The empty cells are due to the phonological system of Proto-Germanic rather than to deficient knowledge. One need only consider how many empty slots a phonemic script for English would have, if based on class, place. and manner features. The 'secret' of the runes may have been the metaknowledge underlying the arrangement of the letters, the matrices and projections in (5) and (6), and other interconnections and generalizations that remain to be established. In a cultural tradition of interest in numerology, astrology, and the calendar (see Hopper 1969), is it coincidental that Iy/. as in year, is the twelfth letter, and Id/. as in day, the twenty-fourth? (On the Germanic names of the runes, see Düwel 1983:106-110 w. lit.; Elliott 1989, chap.5; Polome 1991; cf. Page 1987: 14ft). As to historical continuity, it is of interest that the first letter is If/, as infee, Gmc. *fehu (Goth. faihu) "cattle; goodS", corresponding culturally to Proto-Canaanite 'blp- "ox-head". Such interconnections may turn out to motivate the derailments of economy, as interesting as the phonological grid underlying the script. All of this constituted the privileged information of the rune-carving guild, the knowledge that made the runes a ·secret'.

77

1HE RUNIC ALPHABEr

APPENDIX: The Proto-Germanie Vowel System

5.14Long VoweLr

Germanic inherited the basic vowel system (10) from Pre-Germanic (Le., following the merger of *laI with *Ial , vowel lengthening in laryngeal environments, and ignoring special conditioning factors, such as accent, etc.). (10) Pre-Germanic Vowels

u o

e

ü

ö

a

ä

Germanic then shifted *Iöl to löl and *101 to Ia/: *oktow> Goth. ahtau "eight"; *bhr4er> Goth. bropar (0 =löl) "brother". Presurnably this 'shift' was phonetically motivated by higher and lower allophones. The so-called *eJ (PIE *leI) had lower reflexes (löl, leI, läJl) while *e2 (see below) had higher reflexes (lei, lie/). This suggests that in Germanic, as in Greek, the IE vowels */e/, */öl were (phonetically) lower, as shown in (11). (11) Proto-Germanie Lang Vowels (Stage I) Ü

e

Ü

ö-.

e

(ä)

ö

That this was the ease is clear from loanword evidence: Germanie I öl was borrowed into Latin as löl (*bök-, e.g., OHG Buohhunna) > Latin (si/va) Bäcenis, wbile Lat.löl and löl were borrowed as Gme./ü/, löl respectively; cf. Lat. Röma "Rome" > OS, OHG Ruma; Lat. Remitli "Romans" > Goth. Rumoneis (see Streitberg 1896, §59; Antonsen 1972: 134 w. lit.). In some natural way, this aceounts for the reflexes of PIE *leI as weil as for the rnerger of *Iöl with löl (a phonetie trend continued in English where OE stdn > stone, Iuim > horne , etc.), since löl was fairly low, viz. [öl. 5.15 Core Short Vowel Reflexes IE */el tended to have bigher reflexes and merge with */i/, while */01

went in the opposite direction (see Antonsen 1972:132-134; 1982:10-12): (12) Proto-Germanic Short Vowels (Stage I) u

u

e

o a

e
OF mede, OHG mete, OE me(o)du "mead" vs. *widhu- (Gall. Vidu-, O.lr. foi "tree") > OS widu, OHG witu, OE (widu »/wudu "tree, spear, etc.", Eng. wood. Occasionally there are doublets, like OHG skiflslreJ (OE scip > ship), OS, OF, OE wuifbeside OHF woif "woIr', whieh, according to Hock (1973), are due to the coa.Iescence of, e.g., NOM *wuifaz > *woifaz (> wolf) and VOC *wulfe (> wulj). This leaves the Gennanie languages with doublets to (re)distribute. Signifieantly, in neuters like OHG Jel "skin", berg "mountain", horn "horn",joh "yoke", etc., where there was never a distinction between nominative and vocative, no doublets exist (*fil, *birg, *hurn, *juh), despite other places where an eli- or olu-alternation occurs in these words, as infillen "to skin", gibirgi "mountain range" (Gebirge), etc. 15 5.16 Short Vowels in Un,tressed Syllßbles The fate of leI and lil in unstressed syllables is less certain. OHG 2pl. pres. beret« *1Jher+e+te) "you (p.) bear, carry" may be erucial evidence, if it is not leveled. Antonsen (1972: 123, 138-139) takes it for original and contrasts it with 3sg. birit, from *1Jher+e+ti, Gme. *~ritXi).16 Antonsen claims that unstressed */el did not affect the vocalism of root syllabies. The idea that leI became [i] in unstressed positions has been signalled as the explanation of alternations like iklek "I" (runie ek/ik, OS ec/ik, ON ek, OE je, etc.) and putative derivations like *seghes (Sb. sdhas "power") > *se yiz > *si}iz > Goth. sigis, ON sigr, OS, OHG sigi, Germ. Sieg "vietory". Meid (1967 §111) derives the final -i- from the *-is- of Skt. arc-($- "ray, beam", ete., and Antonsen (1972:139) diseusses e-raising before */z/. Hollifield (1980:34) reformulates as e-raising to lil in unaccented syllabi es exeept before Ir/. In the case of OHG (etc.) sigi, there is another possibility - a Caland compound 1S lbis is dialect-intemal. Cross-dialectally, there Me exceptions, pointed out to me by Antonsen (FAX of 30 Nov. 1993): Germ. Gold, Eng. gold vs. Dan. guld; Eng. ward « IwurdJ) beside Germ. Wort, and even OHG skijlskef, Dan. skib, Swed. skepp "ship", etc. 16 Jay Jasanoff (p.c.) thinks that 2pl. beret is most likely leveled from the a1ready leveled form berat. Hollifield (1980:34-35) claims that unstressed Ie/ (except before Ir/) became lil in Germanic, wherefore he believes that the Monsee-Vienna fragments preserve the Gmc. 2 pI. in quidit "speak, say", etc., forcing the conclusion that forms like beret Me leveled from leveled berat. Antonsen (FAX of 30 Nov. 1993) finds it unlikely that berat would have been leveled to beret in light of 1 pI. beram es and 3 pI. berant, and reiterates the problem of explaining the absence of umlaut in the 2 pI. pres. ind. versus its presence in 2 sg. biris, 3 sg. birit. He challenges the lengths to which scholars go to maintain a traditionaI rule ("all unstressed PIE leis become lil in Gmc.") Suffice it to say, there is little agreement on whether or not all cases of unstressed lei became lit. Nothing here depends crucially on that detail.

TIffi RUNIC ALPHABEf

79

form *segh-i- (extensive discussion of such formations in Bader 1962, chap.l; Nussbaum 1976; see also Szemerenyi 1990:204-205 w. lit.), beside the neut. -s-stem *segh-es-, in names like Sige-ricus (cf. Goth. reiki, OHG rlrhi, Germ. Reich "kingdom"), Germ. Siegreich; Germanic-Latin Segi-m erus,etc., and these names show contamination with sigis; cf. Se/igis-mundus (6th cent.), Sigi(s)-bertus (6th cent.), Sigis-meres (5th cent.), etc. Since there is independent evidence for a later preservation of final lil than lei, forms like *se y-i- and *peretJi suggest a possible solution. Suppose lei became [i] before lil of the following syllable (cf. Streitberg 1896 §63). Such a rule has a considerable amount of support from early attestations. To begin with, the early evidence for ist (ca. 350) from *esti "is",l7 and Sigi- in runic Ssigaduz « *Sigi-hajJUZ) [K 47: Svarteborg, Sw., ca. 450], agrees with the onomastic evidence, pointing to an e-raising rule as early as the 1st century; cf. (ca. 100) Segi-merus (Tacitus, Annals 1.71 [2x]), Segi-mundus (Tacitus,Annals 1.57), but Sigi-m~us (Velleius Paterculus 2.118.2 [fl. ca. 30]). Velleius Paterculus' Sigi-merus is particularly interesting because it appears beside Segestes (2.118.4); cf. (-1st cent.)Segestes in Strabo (7.1.4) and in Tacitus (Annals 1.55 [3x], 57 [4x], 58, 59 [3x], 60, 71). Given older Seges- and Segi-, there is no reason for later Seges- beside Sigi- unless Sigireflects a change in Germanie about that time. 18 By everything known about Germanie compounding, there is no way the radical *e of Segi- could not have been stressed, viz. Segi-merus (cf. Streitberg 1896: 53, 55, 121, 142; Bennett 1972: 104). The change of Segi- 10 Sigi- around tbe 1st century agrees with that in the name of the Finns: Latin (ca. 100) Fenni (Tacitus, Germans 46x) but Finni, 4>(VVOL (2nd cent.: Ptolemaeus, Geographia 2.11.16, 3.5.8); cf. ON Finn(a)r, OE, OS, OHG Finn, and runic Fin(n)ö "Finnish woman" (K 86; ORI 74, Berga stone, Södermanland, Sweden, ca. 500). Another reasonably early example of radical e-raising is found in tbe word for "friend": runic uiniz (K 135: S0nder Rind, Denmark, 500; cf. earlier ekwinai "1 for a friend": 17 A number of factors presumably played a role hefe, e.g., the normal cUtic status, plus the fact that fif was probably lost (after dentals and therefore?) after ftl on verbs, as in Latin; cf. "'mari> Lat. mare "sea", Gmc. "'mari (Goth. mari-saiws "sea", OS, OHG meri, OE meri/mere) "lake; sea", but *eti > Lat. et "and", Goth. i p "yet, but", ambiguous because Gothic lost finalfif in absolute fmal position (cf. Streitberg 1896:54f1). Hollifield (1980: 175) also accepts the Idea that *-i, "though generally retained in Proto-Germanic, was lost in the personal endings of the verb at least as early as CODlDlon Germanic." 18 Not everyone agrees on the validity of the loanword evidence, as Antonsen (FAX of 30 Nov. 1993) points out to me, citing Marchand (1959). My point is, simply, that where there is DO conflict between the internal and extemal evidence, there is no reason to doubt their mutual corroboration.

80

ANCIENT SCRlPfS AND PHONOLOGICAL KNOWLEOOE

ORI 12: Rogaland, Norway, 300), which yields ON vinr, OS, OHG wini, OE wine, Eng. (Good)-win. Since the Indo-European root is *wen- (Lat venus "love", Skt. vtinas "lust", van(- "desire''), the Proto-Oermanic fomi would have been something like *wen+i+s, and the change to Iwin-i-zI obviously occurred weH before 300, since there is no trace of the inherited radical *e anywhere in Oermanic.

5.17 New Long Vowell Following yet another e-raising rule, this one in nasal clusters (cf. *wendaz [Lat ventus "wind"] > Ooth. winds, OE, Eng. wind), a sequence of the type *-Vox-Ioses the nasal with compensatory lengthening. The historical sequence was something like (13) for Ooth. peihan "to thrive, prosper" (cf. O.lr. con-tecim "coagulate") and Ooth., OS, OHO /iluin "to catch" (cf. Lat. pang-ö "fasten, fix, settle"). For discussion, see Vennemann (1971:102ff); Hollifield (1980:32); Voyles (1992:60-6i). (13) Same Early Germanic Phonological Changes 1) Pre-Oermanic *tegk+on+om 2) Orimm 's Law *"eox+onom 3) Omc. vowels (etc.) *.,eux+an(an) 4) e-Raising before Nas. *.,iUX+an S) Nas. Deletion *llix+an 6) Other tnh+an

*paok+on+om *faux+onom *faux+an(an) *fäx+an fäh+an

Nasal Deletion (13-5), despite much support in the literature (some references above), ultimately brought new läl into the system, but the precise dates of that change are unclear. As Jay Jasanoff insists (p.c.), the reflexes of Nasal Deletion remained distinct form the lil reflex of *eJ (OE säton "we sowed") into Old English, where it fell together with the reflex of *-ans- (etc.); cf. *gans (OHO gans etc.) > OE gos "goose", like ptJhte «*"aux-tön [Hollifield 1980: lSOff, 160ft]; cf. OHO dit:hta) "I thought" (cf. Streitberg 1896:76; Antonsen 1972:127). This suggests that Nasal Deletion was in fact rather a nasalization process and that its output was a nasal(ized) vowel. Moreover, this new vowel did not fall together with new läI in Ioanwords in some dialects; cf. OHO suochtlri (= Ooth. sokareis) "seeker" vs. OE (Beowulj 253) (leas)-sceaweras "(deceitful) observers; spys" (NOMpl) with shortening of - ä(from Latin -tlrius). Adducing shortening in the extreme northwest corner of Germanic (Lowe 1972:214) does not explain why that never affected the reflex from the nasal, again pointing to a distinctive contrast in NW Gennanic between long läI and long nasalized 1,1. What lends this hypothesis same

THE RUNIC ALPHABEf

81

credence is that the 3 sg. pres. of ON fd "to get, grasp" (= Goth. filtan, OE fon "to seize") is given by the First Grammatica1 Treatise (§5.5 above) as f~r, i.e., If~/, showing that in the 13th century, the vowel of fK.h)- was still nasalized. Therefore, Nasal Deletion should be refonnulated as a Nasalization process (cf. Streitberg 1896 §93).

5.18 The Stattu 0/ i! Another potential problem for the segment-letter match in §5.6 involves the status of the peculiarly Gennanic *iZ. There is no problem with *eI wbich bears the reflexes of JE *leI (*se-ti-s [cf. Lat s e-vi"I sowed", semen "seed"] > Goth. se JlY, OE s iiJd > seed). Throughout Northwest Gennanic the reflexes of this vowel were kept distinct from a new bigher vowel, traditional *iZ. It is fair to say that there is extremely little agreement among scholars on the origin and development of this vowel in Gennanic, and this is not the place for a lengthy digression on this topic, so I will concentrate on areas of general agreement. It has long been rea1ized that this high, tense, elose */el was a Germanie innovation (cf. Streitberg 1896 §79). As noted by Streitberg, the new vowel was categorially limited. It occurs in only two words with any claim to antiquity within Germanie (Streitberg 1896:65): Goth. fera, OHG fera, feara, fiara "side", of unknown origin, and *hir "here" > Goth., ON, OE, OS her, OHG hiarlhear. Kurylowicz (1952) has explained the latter essentially as a new lengthened grade based on alternations like OE se/se "tbis; he", he/he "he", etc., whence the vowel of her was held in place by the alternation he : M(r).l9 The second category is the NW Germanie preterit of the 7th class, e.g., OE llitan : lit "let", replacing an older reduplication pattern, in Goth. letan : lailot. One possibility here (so already Streitberg 1896 §79(3); cf. Kurylowicz 1952) involves contraction and/or compensatory lengthening, viz. *he-hait (Goth. hai-hait) "named" > OE heht/Mt, etc. The details of this formation remain murky despite much recent attention (e.g., Fulk 1987; Kortlandt 1991). Nevertheless, one of the sources of the new vowel is unequivocally compensatory lengthening:*mizd ö (Goth. mizdo) "reward" > 19The traditional account deriving *Mr from *ktiir (cf. Streitberg 1896 §79.1) is reiterated in Voyles (1992:72-74. and §2.33). Jay Jasanoff (E-mail message of 22 Oct 1993) supports a variant of the morphologica1 solution, comparing OHG th .., OE par "there" vs. (with short . vowel) Goth. (?), ON par "there",OHG thara "thither". "The long li, then. must have originally been an expressive variant [...]." OHG hera "hither" (> Germ. her) has a short Je! vs. the *e of her/hiar/hier "here". "The conclusion naturally suggests itself that [ ...] Gm bilabials > alveolars> velars > dentals, and has 22 segments and 8 open slots (gaps in tbe pbonological system). The Ras Sbamra Matrix has 27 letters with 21 empty cells (cf. the Indic and Korean Han'gül scripts, for arrangement according to pbonetic features). The Runic Matrix is arranged labial> dental> alveopalatal > velar, and has 24 letters with 11 empty cells, five double occupancies, and one slot witb three phonemes. This can be reduced by recognizing, within a given pi ace feature, independent projections of consonants and vowels, a nonproblem for the Proto-Canaanite scripts. It can hardly be accidental tbat, of tbe 6 multiple occupancies, four are paired C and V sets, viz. flu, alr, nli, yloe. The triply-filled slot bas tbe two strident sibilants Is zI along with the coronal stop Itl. Tbe suggestion is that tbe inventor(s) of the runic alphabet viewed class and place features as independent planes. 6) Different scripts bave different advantages and disadvantages. The best evidence suggests tbat a script must contain lexical information, but how that is to be accomplisbed is not clear. If we opt for a script that also codes some phonological knowledge, the ideal appears to be one based on the

r 108

ANCIENT SCRIPfS AND PHONOLOGICAL KNOWLEDGE

syllabic principle (arrangement of segments according to onset and coda position), but without the customary disadvantages of syllabaries. A truly complete script, representing all of our phonological knowledge, even if theoretically possible, would be too cumbersomely inefficient and confusing to read. Since reading is primarily done with long-term memory, in which a word (or some other unit) is treated as a visual gestalt (Henderson 1992; Taft 1992; Maranzana 1993; articles in Willows, Kruk & Corcos 1993a), all that is needed is a quick phonological c(l)ue to the recognition of the gestalt. 7) The alphabet was a compromise between the vowelless West Semitic (Phoenician) script and the very cumbersome syllabaries that duplicated the nucIeus for each cIustering consonant and failed to factor out identical segments. But it was not a very good compromise. One problem with the alphabet is that there is more phonological information than is needed and it appears to inhibit the speed of gestalt recognition. (For the largely irrelevant details it contains, note the frequent reaction on writing a word and deciding that it does not 'look right'.) Another problem with the alphabet, signalied throughout, is that it attempts to represent segments on a strictly linear plane in violation of our (to some extent explicit) knowledge of their organization into syllabi es.

,.

I:;

----

------------... ---

y

AM: "-_

APPENDIXA From Proto-Sinaitic to Greek earlv name

Proto-Sinaitic NW Semitic/Canaanite

1>

'hlp- "ox-head"

(y ~

2>

bet- "bouse"

0

3>

gaml-"throwstick"

~

4>

digg- "fisb" ?

C>"Q

4}

5>-

110- "man callimz" ?

1:

)

fs>

wo (waw)

7>

ze(n-) ??

8>

.{J€(t-) "fence" ?

9>

{i(t-) "spindle" ?

"mace"

LJ

Z q

ld

\ >

--0

tJ IJ '1

T ~.

?

1

t:::>

a

da7

1 B

B

ßilTa

r

ya 11 l1a

t::..

BEATa

E

E

E

TJ7QP'[

F

B(yal1l1a

I

Z

'ijTa

H

(h)ijTa

e

aiiTa

11\»«(

'\

~

E y

Y

IIr

1\

::f

?

4;

; qu(p-) ?

CJ::',)

8

L../

~\/

~~ ~

1

15> (FfIk- 1)

1& -

'*

;;'1.

S,

1=

(-10th c.)

Y

)~ )

L---.../

t

-+

\

~ (Biblos LfJ) 'f....

1? 5\ p p p ?( 2{ 55

p

pw

I

OlW a

;-

T

T

Tau

Y 'IV

r

y

v (4JlMv)



cj>(Elt

X

X(Elt

'"

4J(elt

n

W (~Eya)

t

cp

(b) [= ]

(rB)

(c) [= ] (cf. 11)

'-11' "V J. X (KB)

(d) [=] (cf. 18)

(cp i ) '-11' "V ~

(e) [= ]

1

7

()

Q

NOTE: The letter forms in tbis Appendix are essentially schematic and idealized (cf. Naveh 1982: 25, 180). For details and actual forms by place and date, see Sass (1988: 183-184), Morris (1988), LSAG (1990).

--

r

APPENDIXB The Linear B Syllabary

IJI a

A e

, i

11

f'

,-

~

p

lt2

ai

au

0

u

m y t1

=t=

D ~

~

pa

pe

pi

po

C

*

Ati T